<%BANNER%>

An Assessment of Deterrent and Labeling Effects for Violent Offending Subpopulations

Permanent Link: http://ufdc.ufl.edu/UFE0042876/00001

Material Information

Title: An Assessment of Deterrent and Labeling Effects for Violent Offending Subpopulations A Contemporary Methodological Approach
Physical Description: 1 online resource (303 p.)
Language: english
Creator: WARD,JEFFREY T
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2011

Subjects

Subjects / Keywords: COURSE -- CRIMINOLOGY -- DETERRENCE -- DEVELOPMENTAL -- LABELING -- LIFE -- METHODOLOGY -- QUANTITATIVE -- SANCTIONS -- VIOLENCE
Sociology and Criminology & Law -- Dissertations, Academic -- UF
Genre: Criminology, Law, and Society thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Traditional criminal justice responses to violations of the law remain paramount in efforts to control violence. Two competing theoretical propositions predict opposite effects of official intervention on crime. Specific deterrence theory suggests that official intervention results in the reduction of criminal behavior, whereas labeling theory suggests that it ultimately results in the amplification of criminal behavior. Using this theoretical backdrop, scholars have attempted to discern whether deterrence or labeling theory is correct which has resulted in mixed and inconclusive empirical findings. More recently, scholars have acknowledged that both labeling and deterrence theories may be relevant explanations of behavior since the effects of sanctions may be contingent on offender characteristics and may also differ in the short-term and long-term. However, key methodological problems and exacting data requirements have resulted in few empirical studies that have been able to generate much needed practical knowledge about how best to control violence with traditional criminal justice measures. The current study places deterrence and labeling theories in a life-course and developmental context to study how official intervention may serve as a potential turning point in a criminal trajectory. Informed by theoretical advances in life-course criminology, the study empirically determines for which violent offending subpopulation(s) official intervention is more likely to result in an increase in violent behavior and for which subpopulation(s) it is more likely to result in a decrease in violent behavior. Moreover, the current study assesses differences in sanction effects in the short-run and long-run across different offending subpopulations. Using data from the Rochester Youth Development Study, the current study employs both propensity score matching to account for possible selection artifacts and latent class growth analysis to bypass problems of subjective classification. Findings indicate that three trajectory groups emerge from the data including high offenders, non-offenders, and low offenders. Propensity score matching was highly successful in removing bias in covariates for the total sample as well as for the non-offending and low offending trajectory groups. It was not successful at removing pre-matching covariate imbalances, however, for the high offending trajectory group. The treatment effect estimates provide partial support for short-run labeling effects for the sample as a whole and for the low offending trajectory group. Limitations of the study and directions for future research are discussed.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by JEFFREY T WARD.
Thesis: Thesis (Ph.D.)--University of Florida, 2011.
Local: Adviser: Krohn, Marvin D.
Local: Co-adviser: Gibson, Chris.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2013-04-30

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2011
System ID: UFE0042876:00001

Permanent Link: http://ufdc.ufl.edu/UFE0042876/00001

Material Information

Title: An Assessment of Deterrent and Labeling Effects for Violent Offending Subpopulations A Contemporary Methodological Approach
Physical Description: 1 online resource (303 p.)
Language: english
Creator: WARD,JEFFREY T
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2011

Subjects

Subjects / Keywords: COURSE -- CRIMINOLOGY -- DETERRENCE -- DEVELOPMENTAL -- LABELING -- LIFE -- METHODOLOGY -- QUANTITATIVE -- SANCTIONS -- VIOLENCE
Sociology and Criminology & Law -- Dissertations, Academic -- UF
Genre: Criminology, Law, and Society thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Traditional criminal justice responses to violations of the law remain paramount in efforts to control violence. Two competing theoretical propositions predict opposite effects of official intervention on crime. Specific deterrence theory suggests that official intervention results in the reduction of criminal behavior, whereas labeling theory suggests that it ultimately results in the amplification of criminal behavior. Using this theoretical backdrop, scholars have attempted to discern whether deterrence or labeling theory is correct which has resulted in mixed and inconclusive empirical findings. More recently, scholars have acknowledged that both labeling and deterrence theories may be relevant explanations of behavior since the effects of sanctions may be contingent on offender characteristics and may also differ in the short-term and long-term. However, key methodological problems and exacting data requirements have resulted in few empirical studies that have been able to generate much needed practical knowledge about how best to control violence with traditional criminal justice measures. The current study places deterrence and labeling theories in a life-course and developmental context to study how official intervention may serve as a potential turning point in a criminal trajectory. Informed by theoretical advances in life-course criminology, the study empirically determines for which violent offending subpopulation(s) official intervention is more likely to result in an increase in violent behavior and for which subpopulation(s) it is more likely to result in a decrease in violent behavior. Moreover, the current study assesses differences in sanction effects in the short-run and long-run across different offending subpopulations. Using data from the Rochester Youth Development Study, the current study employs both propensity score matching to account for possible selection artifacts and latent class growth analysis to bypass problems of subjective classification. Findings indicate that three trajectory groups emerge from the data including high offenders, non-offenders, and low offenders. Propensity score matching was highly successful in removing bias in covariates for the total sample as well as for the non-offending and low offending trajectory groups. It was not successful at removing pre-matching covariate imbalances, however, for the high offending trajectory group. The treatment effect estimates provide partial support for short-run labeling effects for the sample as a whole and for the low offending trajectory group. Limitations of the study and directions for future research are discussed.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by JEFFREY T WARD.
Thesis: Thesis (Ph.D.)--University of Florida, 2011.
Local: Adviser: Krohn, Marvin D.
Local: Co-adviser: Gibson, Chris.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2013-04-30

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2011
System ID: UFE0042876:00001


This item has the following downloads:


Full Text

PAGE 1

AN ASSESSMENT OF DETERRENT AND LABELING EFFECTS FOR VIOLENT OFFENDING SUBPOPULATIONS: A CONTEMPORARY METHODOLOGICAL APPROACH By JEFFREY T. WARD A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2011 1

PAGE 2

2011 Jeffrey T. Ward 2

PAGE 3

To my father, the late Edward David Ward, Sr. 3

PAGE 4

ACKNOWLEDGMENTS I thank the Harry Frank Guggenheim Foundati on for providing funds to support my dissertation research and also for instilling confidence in me as a young scholar by seeing the merits and practicality of my research. I owe great thanks to Kerri Ward, my beloved wife, for her unshakeable patience and support throughout the dissertation proce ss specifically and my graduate career more generally. I could not imagine any one being quite as caring about both my general well-being and life goals. I also gr eatly appreciate my family, friends, and colleagues for their support. Perhaps most importantly, I thank the me mbers of my dissertation committeeDr. Marvin D. Krohn, Dr. Chris L. Gibson, Dr. Lonn Lanza-Kaduce, Dr. Ronald L. Akers and Dr. Walter L. Leitefor thei r guidance throughout the research and writing processes. While each committee member made contributions across all of the components of my dissertation, each individual member provid ed certain guidance on the dissertation and taught me specific lessons t hat will have lasting influences on my career; these warrant specific mention. Dr. Leite provided invaluable met hodological guidance to ensure a highly complicated statistical analysis was competently executed. I owe much of my advanced methodological skills to the teachings of Dr. Leite. Dr. Akers ensured my empirical research was grounded in the appropriate theoretical perspectives. It has been a great honor to discuss deep theoretical issues wit h Dr. Akers and his open door policy has left an indelible mark on my approac h to teaching and mentoring. As has been typical throughout my early career, Dr. Lanza-Kaduce challenged me to think clearly about the rationale for my research as well as think thoroughly th rough its implications. 4

PAGE 5

Dr. Gibson helped me to deeply consider ho w my substantive research on violent crime can best be answered methodologica lly. In addition, he has contributed tremendously to my growth as a scholar by treating me as a colleague and ensuring that I always understand his thought processe s and rationales for his decisions. Dr. Krohn has taught me to always step back and look at the bigger pictur e of my research and to focus on areas where I can make impor tant substantive contributions to the discipline, which are lessons learned that wil l have lasting positive effects on my career trajectory. His ability to give honest and detailed feedba ck overnight is uncanny. I thank Dr. Krohn and Dr. Gibson for embracing my tendencies to strive for perfection but for also teaching me there is a time when one can safely let go. Finally, I am indebted to Dr. Krohn and his colleagues (Dr. Terence P. Thornberry, Dr. Alan J. Lizotte, and the rest of the re search team) for allo wing me to use the Rochester Youth Development St udy for my research. 5

PAGE 6

TABLE OF CONTENTS page ACKNOWLEDG MENTS..................................................................................................4 LIST OF TABLES............................................................................................................9 LIST OF FI GURES ........................................................................................................11 ABSTRACT ...................................................................................................................12 CHAPTER 1 INTRODUC TION....................................................................................................14 Youth Violence and Crime Cont rol..........................................................................14 Criminal Traj ectori es...............................................................................................16 Official Intervention as a Turning Point ...................................................................18 Bettering or Worsening Violen t Crime Trajec tories?.........................................20 Bettering and Worsening Violent Crime Trajec tories?......................................21 Present Study .........................................................................................................23 2 THEORIES ON SAN CTION EFFE CTS..................................................................31 Deterrence Theory..................................................................................................31 The Origins of Deterrence T heory ....................................................................32 Free Will, Pleasur e and Pa in............................................................................35 Certainty, Celerity and Se verity of P unishm ent................................................37 Specific and Gener al Deterr ence.....................................................................40 Absolute and Restrictive Dete rrence................................................................42 Tipping Po ints...................................................................................................43 Modern Deterrenc e Theor y..............................................................................44 Perceptual De terrence.....................................................................................46 Labeling T heory ......................................................................................................49 The Roots of Labe ling Theor y..........................................................................50 Emergence of the Labelin g Perspect ive...........................................................52 Nine Tenants of Labeling Th eory.....................................................................54 Primary and Secondar y Devianc e....................................................................57 The Causal Pathways to Secondary Deviance.................................................63 3 THE DEVELOPMENTAL AND LI FE-COURSE PERS PECTIVE.............................68 The Age-Crim e Curve.............................................................................................68 The Birth of Developmental and Life-course Cr iminology.......................................70 Dual Taxonomy Theory....................................................................................73 Age-Graded Social Bonding Th eory.................................................................76 Trajectories, Transitions and Turning Points..........................................................78 6

PAGE 7

Distinct Criminal Tr ajectory Groups? ......................................................................82 4 SANCTION EFFECTS IN A DEVELOPMENTAL AND LIFE-COURSE PERSPECTIV E.......................................................................................................86 Official Intervention: Life Tr ansition or Turn ing Poin t?............................................86 Turning Points within Trajectory Groups.................................................................89 Turning Points in the Shortand L ong-runs ............................................................95 Types of Official Interventions.................................................................................97 5 REVIEW OF SANCTION E FFECTS LITERA TURE..............................................100 Sanction Effects in the Aggregat e.........................................................................100 The Contingencies of Sanction Effects.................................................................106 Key Limitations of Prior Res earch.........................................................................110 Data................................................................................................................110 Offender Categoriz ation.................................................................................112 Selection E ffects.............................................................................................113 6 DATA ....................................................................................................................114 Overview of Data..................................................................................................114 Analysis Sa mple...................................................................................................117 Measures ..............................................................................................................120 Behavioral Trajectory and Outc ome: Violent Crime........................................121 Treatment Conditions: Offi cial Interv ention.....................................................122 Covariat es......................................................................................................123 Demographi cs..........................................................................................123 Neighborhood charac teristic s...................................................................125 Family dyna mics......................................................................................127 School factors..........................................................................................130 Peer associ ations .....................................................................................131 Values and mental states.........................................................................133 Prior delinquent behavior .........................................................................134 Prior official in terventio n...........................................................................136 Descriptive St atistics.............................................................................................137 7 METHODOL OGY.................................................................................................147 Method ..................................................................................................................147 Latent Class Growth Analysi s.........................................................................148 Conceptual ov erview ................................................................................148 Technical de tails...................................................................................... 151 Model estima tion......................................................................................155 Propensity Score Matc hing Analys is..............................................................157 Conceptual ov erview ................................................................................157 Technical de tails...................................................................................... 161 Model estima tion......................................................................................166 7

PAGE 8

An LCGA-PSM Int egrated Met hod.................................................................171 Analytic St rategy...................................................................................................172 8 RESULTS: VIOLENT CRIME TRAJECTORIES OF OFFENDING .......................176 Model Select ion....................................................................................................176 Assessment of Model Adequacy...........................................................................179 Trajectory Group Char acteristic s..........................................................................181 Trajectory Group-specific De scriptive Stat istics ....................................................184 Trajectory Group-specific Interven tion Experiences and Outcom es.....................187 9 RESULTS: SHORT-TERM AND LO NG-TERM SANCTIONS EFFECTS.............200 Bivariate Relationships between Offici al Intervention an d Violen ce......................200 Bivariate Relationships in the Shor t-run.........................................................201 Bivariate Relationships in the Long -run..........................................................204 Covariate Balance Assessments of Treatment Models .........................................205 Treatment Model for Arre st.............................................................................207 Total sample............................................................................................207 Trajectory groups..................................................................................... 211 Treatment Model for Police Cont act...............................................................217 Total sample............................................................................................217 Trajectory groups..................................................................................... 219 Treatment Effect Estimates of Official Intervention on Subsequent Violence.......223 Treatment Effects in the Short-run.................................................................224 Treatment Effects in the Long-ru n..................................................................227 10 CONCLUSION S...................................................................................................252 Research Su mmary..............................................................................................252 Limitati ons.............................................................................................................257 Discussion and Imp lications ..................................................................................263 Directions for Futu re Resear ch.............................................................................272 LIST OF REFE RENCES.............................................................................................279 BIOGRAPHICAL SKETCH ..........................................................................................303 8

PAGE 9

LIST OF TABLES Table page 6-1 Comparison between analysis samp le and individuals excluded from the analysis sample across valid cases fo r key demographic factors and violent behavior at wa ve one. ......................................................................................143 6-2 Descriptive statistics for valid case s in the analysis samp le (N=595)...............144 6-3 Number of males committing zero thr ough six distinct acts of violent crime before and after the tr eatment per iod...............................................................145 6-4 Descriptive statistics for vari ables with imputed dat a (N=595)..........................146 8-1 Model selection criteria for five different Poi sson LCGA models estimating violent crime traj ectories ...................................................................................191 8-2 Model selection criteria for two alternative three class Poisson LCGA models with different orders of polynomials across the trajecto ry groups.....................191 8-3 Mean posterior probabilities for trajec tory group assignment for the selected three-group Poisson LCGA model ....................................................................191 8-4 Additional diagnostics for the selected three-group Poisson LCGA model.......191 8-5 Total number and proportion of indi viduals within each violent crime trajectory group (N=595). ..................................................................................191 8-6 Parameter estimates for the selected three class Poisson LCGA model..........192 8-7 Trajectory group-specific descripti ve statistics and AN OVA result s..................193 8-8 Trajectory group-specif ic mean outcomes in the s hort-run and long-run..........195 9-1 Bivariate associations between arrest and violent crime in the short-run for the total sample and by trajectory group........................................................... 229 9-2 Bivariate associations between police contact and violent crime in the shortrun for the total sample and by trajectory group...............................................229 9-3 Bivariate associations between arrest and violent crime in the long-run for the total sample and by tr ajectory group................................................................. 230 9-4 Bivariate associations between polic e contact and violent crime in the longrun for the total sample and by trajectory group...............................................230 9-5 Treatment and control group covariate means and bias statistics before and after PSM for the arrest treat ment (total sample)..............................................231 9

PAGE 10

9-6 Treatment and control group covariate means and bias statistics before and after PSM for the arrest treatment (High o ffenders)..........................................233 9-8 Treatment and control group covariate means and bias statistics before and after PSM for the arrest treatment (Low offenders)..........................................237 9-9 Treatment and control group covariate means and bias statistics before and after PSM for the police contact treatment (total sample).................................239 9-10 Treatment and control group covariate means and bias statistics before and after PSM for the police contact treatment (High offenders).............................241 9-11 Treatment and control group covariate means and bias statistics before and after PSM for the police contact treatment (Non-o ffenders).............................243 9-12 Treatment and control group covariate means and bias statistics before and after PSM for the police contact treatment (Low offenders)..............................245 9-13 Post-matching short-run treatment effect s of arrest for the total sample and by trajectory group............................................................................................247 9-14 Post-matching short-run treatment e ffects of police contact for the total sample and by traj ectory gr oup........................................................................ 247 9-15 Post-matching long-run treatment effects of arrest for the total sample and by trajectory group.................................................................................................247 9-16 Post-matching long-run treatment effe cts of police contact for the total sample and by traj ectory gr oup........................................................................ 247 10

PAGE 11

LIST OF FIGURES Figure page 1-1 Illustrative model of official inte rvention as a potential turning point in a criminal traj ectory...............................................................................................28 1-2 Illustrative model of official intervent ion as a potentially different turning point for divergent criminal trajecto ries........................................................................29 1-3 Illustrative model of official intervent ion as a potentially different turning point for divergent criminal trajectories wit h short-term and long-term differences......30 7-1 Illustration of three alter native matching st rategies ...........................................175 8-1 Estimated trajectory of violent offending during adolescence for the total sample (c =1).....................................................................................................196 8-2 Estimated trajectories of violent offending during adolescence for the three latent classe s (c=3)...........................................................................................197 8-3 Number of individuals experiencin g arrest in waves seven and/or eight across the three violent crim e trajectory groups...............................................198 8-4 Number of individuals experiencin g police contact in waves seven and/or eight across the three violent crime trajectory groups......................................199 9-1 Short-run violent crime mean comparis ons of those arrested and not arrested for the total sample and by violent crime trajectory group before matching......248 9-2 Short-run violent crime mean compar isons of those with and without a police contact for the total sample and by violent crime traj ectory group before matching. ..........................................................................................................249 9-3 Long-run violent crime mean comparisons of those arrested and not arrested for the total sample and by violent crime trajectory group before matching......250 9-4 Long-run violent crime mean comparisons of those with and without a police contact for the total sample and by vi olent crime trajectory group before matching. ..........................................................................................................251 11

PAGE 12

Abstract of Dissertation Pr esented to the Graduate School of the University of Florida in Partial Fulf illment of the Requirements for t he Degree of Doctor of Philosophy AN ASSESSMENT OF DETERRENT AND LABELING EFFECTS FOR VIOLENT OFFENDING SUBPOPULATIONS: A CONTEMPORARY METHODOLOGICAL APPROACH By Jeffrey T. Ward May 2011 Chair: Marvin D. Krohn Cochair: Chris L. Gibson Major: Criminology, Law and Society Traditional criminal justice responses to vi olations of the law remain paramount in efforts to control violence. Two competi ng theoretical propositi ons predict opposite effects of official intervention on crime. S pecific deterrence theory suggests that official intervention results in the reduction of cr iminal behavior, whereas labeling theory suggests that it ultimately results in the amp lification of criminal behavior. Using this theoretical backdrop, scholars have attempted to discern whether deterrence or labeling theory is correct which has resulted in mixed and inconclusive empirical findings. More recently, scholars have acknowledged that both labeling and deterrence theories may be relevant explanations of behavior since t he effects of sanctions may be contingent on offender characteristics and may also di ffer in the short-te rm and long-term. However, key methodological problems and exacting data requirements have resulted in few empirical studies that have been able to generate much needed practical knowledge about how best to control violence with traditional criminal justice measures. The current study places deterrence and labeling theories in a life-course and developmental context to study how official intervention may serve as a potential turning 12

PAGE 13

13 point in a criminal trajectory. Inform ed by theoretical advances in life-course criminology, the study empirically det ermines for which violent offending subpopulation(s) official intervention is more likely to result in an increase in violent behavior and for which subpopulation(s) it is more likely to result in a decrease in violent behavior. Moreover, the current study assesses differences in sanction effects in the short-run and long-run across different offend ing subpopulations. Using data from the Rochester Youth Development Study, the cu rrent study employs both propensity score matching to account for possible selection ar tifacts and latent class growth analysis to bypass problems of subjective classification. Findings indicate that thr ee trajectory groups emerge fr om the data including high offenders, non-offenders, and lo w offenders. Propensity score matching was highly successful in removing bias in covariates for the total sample as well as for the nonoffending and low offending trajectory groups. It was not successful at removing prematching covariate imbalances, however, for t he high offending trajectory group. The treatment effect estimates provide partial support for short-run l abeling effects for the sample as a whole and for the low offending trajectory group. Limitations of the study and directions for future re search are discussed.

PAGE 14

CHAPTER 1 INTRODUCTION Youth Violence and Crime Control Violent crime rates have begun to stabilize in recent years, following roughly a ten year period of steady declines (Rand, 2008). While national trends over the past two decades would not appear to be especially alarming, conc erns over violent crime remain consistently high over time (Zimring, 1998) and recent years have been characterized by a relentless worry over the coming generation of super-predators (McCollum, 1996, p. 3, emphasis in original ). Interlaced within this pervasive disquietude over violent crime are specific point s of emphasis that typically wax and wane over time. In the mid1970s, youth violence took cent er stage during a brief threeyear period (Zimring, 1998). A second coming of anxiety over the violent crime of adolescents has reemerged during the 1990s (Z imring, 1998) but, unlike its antecedent version, it has persisted as a topic of seri ous social concern today (e.g., see MacDonald et al., 2009, p. 1). The problem of youth violence presents unique challenges for public policies, which have to balance concerns over y outh development and crime control. Adolescence is viewed as a time period of extensive personal growth and development. It is during this time frame that individuals acquire the bulk of the social and human capital necessary to successfully navigate the adult social wo rld. Recognizing adolescence as a critical period in the devel opment of an individual, the legal system has considered youth to be less culpable for their actions and the juvenile justice system was formed to correct antisocial beh avior and foster prosocial development (Kupchik, 2006). Some have maintained that this intermediary form of justice is 14

PAGE 15

problematic, however, and that society should leave the kids alone whenever possible (Schur, 1973, p. 155). Drawing on the labeling perspective (Lemert, 1951), radical nonintervention suggests it is best to do nothing at all as in tervening may not only fail to alleviate, but actually worsen, the crime pr oblem by disrupting the acquisition of human and social capital and embedding youth in devi ant social networks (Schur, 1973). The rehabilitative ideals of juvenile justice, and most certainly the concept of radical nonintervention, have been largely set as ide in favor of a legal system that gets tough on juvenile crime. T hese changes are, in part, due to a perceived inevitability of a major violent crime wave spearheaded by a new breed of violent youth (Zimring, 1998). Unlike conceptions of developing yo uth as normal or indi viduals in need of support and guidance, notions of youth as super-predators suggest adolescents will almost assuredly be antisocial as adults. The implications for formal crime control endeavors are resoundingly clear. To the extent that juveni le violence is indeed the product of individuals who are largely unchangeable, policies focusing on reducing violent victimizations through get tough strategies may be appropriate. However, to the extent that engaging in juvenile violence is a conscious decision made by an individual, the goal of official intervent ion is not the control of behavior through incapacitation but rather t he control of subsequent behavi or through the calculated utilization of punishment experiences. Strate gies that maximize deterrent and minimize labeling effects would appear to be the most appropriate formal intervention approach to controlling future acts of violent crime. The discourse on youth violence and crime c ontrol brings two key issues into the limelightstability and change in behavior. These ideas naturally implicate the 15

PAGE 16

developmental and life-course perspecti ve, which focuses on understanding the unfolding of behavior over time and seeks to di scern the effects of life events such as criminal justice system contact on behaviora l trajectories (Farrington, 2006). Despite the fact that there is much to be gained theoretically and practically by placing sanction effects in a developmental context (e.g., see Moffitt, 1993; Sampson & Laub, 1993; 2003), surprisingly little research has addressed this critical area of inquiry (Huizinga & Henry, 2008). To evince the importance of a developmentally informed investigation of sanctions effects for the study of violent crime, offending trajec tories and the role criminal sanctions may play in alteri ng them are now disc ussed. Criminal Trajectories The presence of stability in criminal behavior is undeniable. Adult antisocial behavior virtually requires childhood antisoc ial behavior (Robins, 1978, p. 611). While there has been a plethora of theories t hat have advanced our ability to explain individual-level criminal behavior (e.g., see Agnew, 1992; Bur gess & Akers, 1966; Hirschi, 1969; Gottfredson & Hirschi, 1990), the best predictor of future behavior remains past behavior (White, Mo ffitt, Earls, Robins, & Silv a, 1990). The notion of stability in offending behavior does not imply, however, that individuals do not change their level of involvement in crime altogether; rather, stability in offending refers to the fact that there is consider able between-individual continui ty in behavior over time (Farrington, 1989; 1992; Tracy & Kempf-Leonard, 1996). Criminal trajectories chart the course of offending over the life-course. A nalogous to specifying the future location and speed of a projectile by knowing its origin and path, an individuals prospective behavior can be projected, at least in theory, from their retrospecti ve criminal trajectory. 16

PAGE 17

The importance of criminal trajectories to the modern era of criminology can be traced back to Hirschi and Gottfredsons (1983) discussions of the most well-known offending trajectory in crim inologythe aggregate age-crime curve. They claimed that the age distribution of crime could not be accounted for by existing criminological theories, which helped to spark a number of explicit attempts to devel op explanations of offending over the life-course (see Lahey & Waldman, 2005; Moffitt, 1993; Catalano & Hawkins, 2005; Thornberry & Krohn, 2005; Farrington, 2005; Sampson & Laub, 1993; 2003; LeBlanc, 2005; Wikstrom, 2005; for a revi ew, see Farrington, 2006). Several of these theories essentially claim there are of fenders who cluster into trajectory groups but whose behavior, when viewed in conjunc tion, creates the illusion of a single agecrime curve (Moffitt, 1993; Thornberry & Krohn, 2005; LeBlanc, 2005; Lahey & Waldman; 2005). The existence of distinct gr oups of offenders is one of t he most contentious issues in developmental and life-c ourse criminology (Farrington, 2006). The bulk of the controversy centers on the interpretation of whet her criminal trajectories are deflectable. Developmental criminology, in practice if not in theory, tends to emphasize the notion that people get locked into certain traj ectories (Sampson & Laub, 2005a, p. 14). Some view this as a critical mistake which characterizes humans as robotic and inappropriately deemphasizes the importance of change in offending behavior over time (Sampson & Laub, 1993; 2003; 2005a; Laub & Sampson, 2003; Laub, Sampson, & Sweeten, 2006). Sampson and Laubs criticis ms are well received; scientists should indeed be hesitant to ex plicitly or implicitly deny the relevance of human agency in offending a priori. However, characterizing criminal trajectories does not automatically 17

PAGE 18

preclude one from assessing the relevance of important life events such as justice system contact. In fact, as will become clear there appear to be important benefits to assessing intra-individual change in offendi ng behavior using the gr oup-based trajectory framework (Nagin & Tremblay, 2005; Haviland & Nagin, 2005; Haviland, Nagin, & Rosenbaum, 2007: Hav iland, Nagin, Rosen baum, & Tremblay, 2008).1 Official Intervention as a Turning Point While most antisocial adults were antisocial as children, it is also the case that many antisocial children do not become ant isocial as adults (Robins, 1978). Put differently, evidence suggests there is subs tantial between-individual stability in behavior over time (Farrington, 1989; 1992; Tracy & Kempf-Leonard, 1996) but withinindividual change in criminal behavior can coexist with this between-individual continuity (Farrington, 1990; Sampson & Laub, 1992; Ver hulst et al., 1990). Embedded within long-term behavioral trajectories are life ev ents (e.g., marriage, employment, justice system contact) that unfold over relative ly shorter periods of time (Elder, 1985). Collectively known as transitions, thes e key life events that are nestled within trajectories of behavior may gene rate a turning point in one s path (Elder, 1985). Like a sudden force that changes the speed or dire ction of a project ile, key life events experienced by individuals can serve to deflect ones criminal trajectory from its likely course. A number of social factors have been shown to be relevant to altering criminal trajectories including marriage (Laub, N agin, & Sampson, 1998; Sampson & Laub, 1993; Sampson, Laub, & Weimer, 2006), employment/work (Uggen, 2000; Uggen & 1 Behavioral trajectories have become increasingly popular in recent years and they have provided for new and exciting research questions across a wide variety of disciplines (for a review, see Piquero, 2008). 18

PAGE 19

Staff, 2001; Sampson & Laub, 1993; 2003; cf. Wright & Cullen, 2004) and military involvement (Elder, 1986; Sampson & Laub, 1993; 1996; 2003). Being subject to formal sanctions is also thought to be a critical life event that can serve as a turning point in an offending trajectory (Sampson & Laub, 1997); one which warrants furt her attention. The importance of studying how official intervention may shape an offending trajectory is amplified when c onsidering the stark r eality that reactive criminal justice responses such as arrest and incarcerati on remain politically popular and are very difficult to supplant. While there is widespr ead recognition that the causes of violent crime are highly complex (see Farrington, 1998; Reiss & Roth, 1993) resulting in the need for comprehensive strategies to contro l violent crime, there is a pervasive disconnect between what research shows an d what policy makers do (Austin, 2003; Tonry, 2006). In recent decades, there has been unremitting growth in expenditures across all three components of the crimi nal justice system. Police expenditures continue to lead the way topping 100 billion dol lars, constituting a 420% increase over roughly a twenty-five year period (Perry, 2008). While judicial and correctional expenditures are each approxim ately half that of police expenditures, growth in spending on courts and corrections represent even larger proportional increases (Perry, 2008). Given that funding for justice system functions is at an all time high and traditional criminal justice responses continue to play a salient role in effo rts to control violent crime, it becomes increasingly important to more fully understand the role official intervention plays as a turning point in violen t crime trajectories. Indeed, the extent to which criminal justice sanctionsfoster reci divism or help lead to the termination of 19

PAGE 20

criminal activity is a central one [questi on] in criminology, and takes on even more importance given the recent incarceration increases in the United States (Bhati & Piquero, 2008, p. 218). Bettering or Worsening Viol ent Crime Trajectories? While official intervention may play no appr eciable role in influencing subsequent offending behavior one way or the other, two competing theor etical perspectives predict opposite effects of official intervention on subsequent offending behavior. Forming the backbone of criminal justice system operations, specific det errence theory suggests that formal sanctioning results in the reduction of cr ime. Individuals who experience certain, swift, and adequately severe sanctions are thought to align their future behavior with societal expectations due to the threat of future punishments (Bec caria, 1963; Gibbs, 1975; Zimring & Hawkins, 1973). Converse ly, labeling theory suggests a dismal outcome; official intervention is hypothesized to set in motion a chain of events that actually leads to the worsening of criminal behavior (Becker, 1963; Lemert, 1951; 1967). Specifically, those who are labeled by forma l sanctions are likely to be blocked from conventional opportunities, form negative self -identities, and acquire supportive deviant others which embeds the individual into a car eer of crime (Paternoster & Iovanni, 1989). Then, controlling crime may be best accomp lished by either applying formal sanctions to engage criminal deterrence or, perhaps somew hat ironically, by withholding official intervention to prevent triggering the labeling process. On their face, deterrence and labeling theori es would appear to be irreconcilable. The central question underlying sanction effects research in this case is a relatively simple one: does official intervention result in either a deterrent or labeling effect (or neither)? Figure 1-1 provides a concrete example of official intervention as a potential 20

PAGE 21

turning point in a violent crime trajectory. Illustrated here is an individual who is tracking along a specific criminal trajectory and, at some point in ti me, experiences an official intervention. When formal sanctions are in itiated, one of three possibilities emerges: the individual may stay the course (null effect), the i ndividual may worsen his or her behavior (labeling effect), or t he individual may decrease his or her offending (deterrent effect). From a deterrenc e and public policy standpoint, nu ll and labeling effects are both undesirable to a lesser or greater degree, res pectively. Official intervention is only useful as a violent crime control strategy when experience with justice system contact alters ones offending trajectory to ward conforming behavior. Bettering and Worsening Viol ent Crime Trajectories? The seemingly incongruent propositions of labeling and deterrence theories can compel individuals to make an artificial choice between the two theories (Wellford & Triplett, 1993). Most studies assessing the e ffectiveness of controlling crime through official intervention have taken for gr anted that there is a single homogeneous population that is equally sensitive to sancti on effects over time. The assumption of uniform susceptibility to sanctions appears unt enable. We should not expect labeling [nor deterrent] effects to be invariant across subgroups (Paternoster & Iovanni, 1989, p. 381). In other words, bot h labeling and deterrence theor ies may simultaneously be relevant explanations of offending behavior since the effects of official intervention may be contingent on the characte ristics of offenders (Huizi nga & Henry, 2008; Thorsell & Klemke, 1972). Despite the acknowledgemen t by academics of the need to explore contingent labeling and deterr ent effects since the 1970s (e .g., see Thorsell & Klemke, 1972), there have been only a handful of empiri cal studies that have heeded this advice (Huizinga & Henry, 2008, p. 249). This is an unfortunate empirical omission as 21

PAGE 22

developmental and life-course theories ant icipate that the size and direction of treatment effects are cont ingent upon ones behavioral traj ectory (e.g., see Moffitt, 1993). Figure 1-2 is a schematic demonstrating diffe rential sanction effects for individuals following two hypothetical offending trajectories While the outcome possibilities for the individuals following along each trajectory rema in the same (i.e., null, deterrent or labeling effects), there may be fundamental differences in the actual outcomes for different groups of offenders. For example, Moffitt (1993) puts fo rth a dual taxonomy theory of criminal behavior where she suggests that there are two fundamentally different types of offenders. Adolescencelimited offenders engage in acts of crime and delinquency for a brief period of time as an expression of independ ence brought on by a maturity gap between their biological and so cial ages. Life-course-persistent offenders commit acts of crime and delinq uency relatively consistently over time as a result of neuropsychological deficits and exposure to criminogenic environment s. Moffitt (1993) predicts that adolescence-limit ed offenders would be subject to labeling effects since official intervention would interfere with a normal desistance process. However, lifecourse persistent offenders, w hose etiology of offending is rooted early in life, would be unlikely to experience offi cial intervention effects altogether (Moffitt, 1993). In addition to different types of offender s having differential susceptibility to sanction effects, Sampson and Laub (1997) ar gue that it is impor tant to not only consider the immediate crime enhancing or reducing effects of official intervention but also examine longer term outcome s. While there may be imme diate deterrent effects of official intervention, being subject to forma l sanctions may nevertheless set in motion 22

PAGE 23

the labeling process ultimately resulting in deviance amplification (Sampson & Laub, 1997). Figure 1-3 adds this possibility to the diagram for the two hypothetical individuals following along distinct offending traj ectories (cf. Figure 1-2). To complicate matters, then, both deterrent and labeling effects may be possible for the same individual. Findings supporting initial deterrent effects of arrest followed by subsequent labeling effects have been observed (e.g., see S herman et al., 1991). In sum, labeling and deterrent effects may differ for different types of offenders and /or both theoretical processes may be relevant for the same type of offender leading to initial deterrent effects and subsequent labeling effects. These possibilities emerge when official intervention effects are considered within a developmental and life-course framework. Present Study Though there is much to be gained from lif e-course informed investigations of deterrent and labeling effect s, rarely is there an exam ination in a developmental perspective that includes early delinquency/ crime, arrest, justice system sanctioning, subsequent delinquency/crime and longer-term outcomes over differ ent phases of the life course (Huizinga & Henry, 2008, p. 223). Taking a practical approach to the control of violent crime, the present study draws on theories of deterrence, labeling, and lifecourse criminology to examine how, if at all, official intervention may be most effectively employed to reduce subsequent offending behavior. More specifically, this research seeks to determine for which violent offending subpopulation(s) official intervention is more likely to result in an immediate increase in violent behavior and for which subpopulation(s) it is more likely to result in an immediate decrease in violent behavior. As a related point of investi gation, the proposed research ai ms to uncover the duration of any initial labeling and det errent effects and determine if the length of the effects 23

PAGE 24

differ by violent offending subpopulations The present study tackles two key problemsclassification of offenders and sele ction artifactsthat st and in the way of addressing these critical empirical questions. To assess whether official intervention serves as a turning point in the criminal trajectories of offenders, a classification procedure for diffe rent types of offenders is necessary. While individuals prior record s have been used to classify persons into offending subpopulations in previous research there is a critical problem with this strategy. For example, two indivi duals who have been arrested once would be classified similarly despite the fact that one may be a chronic offender and the other an exceptionally infrequent offender Then, classifying individuals based on their prior records as opposed to their offending behavior is somewhat imprecise if the goal is to assess contingent effects based on behavior. As an alternative to classification based on prior record, Huizinga and colleagues (1986) classified individuals according to a delinquency typology (e.g., non-delinquent, ex ploratory delinquent, non-serious delinquent, serious delinquent). However, there are critic al problems with subjective classification procedures as well. First, a subpopulation can be subjectively classified as distinct but only represent random vari ation. Second, unusual subpopulations can fail to be classified at all. Third, statistical tests examining differences across subpopulations may be undermined by uncertainties about the reliability of subpopulation assignment (see Nagin, 2005). Fina lly, subjective classification schemes may fail to consider trajectories of off ending behavior over time, which can limit their ability to assess whether official intervention serves as a turning point in certain criminal trajectories. 24

PAGE 25

In addition to the problems associated with subjective classification, methodological limitations of previous studi es have resulted in the inability to discern between labeling and deterrent effects for offending subpopulations with any definitiveness. One key problem with prior empirical work is the use of imprecise methodological designs that result in an inabil ity to disentangle selec tion artifacts from real official intervention effects (Smith & Pa ternoster, 1990). Wi th some exceptions (e.g., Chiricos, Barrick, Bales, & Bont ranger, 2007; McAra & Mc Vie, 2007; Smith & Paternoster, 1990), research designs have failed to employ strong methodology that addresses the problems of making causal inferences with non-experimental data. However, the classification and selection ar tifact problems can be resolved given the availability of high quality long itudinal data for official intervention effects research (Huizinga & Henry, 2008, p. 248) and count erfactual methodological techniques designed to approximate exper imental designs with observ ation data (Haviland et al., 2007; 2008). In sum, the cu rrent work advances research on the utility of controlling violent offenders through tradition al criminal justice measures by clarifying the theory underlying differential sanction effects using a developmental and life-course framework and by conducting an advanced and rigorous empirical analysis using data from a major longitudinal panel study. In doing so, the study provides answers to two urgent academic inquiries (i.e., what the effects of o fficial interventions are for different violent offending subpopulations and how long these effe cts last) that have sweeping practical implications for the control of violence. The current research begins by reviewi ng the relevant history, development and present state of thinking r egarding the key theories of sa nctions effects (Chapter 2). 25

PAGE 26

Next, the widespread influence of the development al and life-course paradigm is noted along with a detailed explanatio n of what is meant by key life-course ideas such as criminal trajectories, turning points and st ability and change in offending (Chapter 3). Sanction effects are then placed in the devel opmental and life-course perspective to illuminate differential sanction effects hypot heses between offending trajectories and also, within offending trajectories, across the shortand long-runs (Chapter 4). Subsequently, literature on the effects of official intervention is reviewed with emphasis placed on what is known from studies ex amining contingent deterrent and labeling effects as well as what remains unknow n or unclear due to data or methodological limitations (Chapter 5). Next, the data section deta ils the Rochester Youth Development Study (RYDS), details the analys is sample and the measures used in the study, and provides descriptive statistics (Chapter 6). This is followed by the methodology section which includes an ex planation of the conceptual underpinnings and methodological details of propensity score matching, latent class growth analysis and their methodological integration (Chapter 7). The next chapter focuses on the LCGA model selection and provides a detailed description of the number, shapes and proportion of individuals belonging to trajec tory groups. In addition, similarities and differences between the violent offending subpopulations with respect to official intervention experiences, covariates, and viol ent crime outcomes ar e reported (Chapter 8). Constituting the main analyses, result s of the integrated LC GA-PSM methodological procedures are detailed in t he next chapter. First, biva riate relationships between official intervention and subsequent viol ent offending behavior are reported for both types of official intervention in both the shortand long-runs. Next, the successfulness 26

PAGE 27

of the propensity score matching procedure is documented for the tr eatment of arrest and police contact. Following t he matching procedure, treatment effect estimates of the official interventions are examined in bot h the short-term and l ong-term (Chapter 9). Finally, a summary of the principal empi rical findings is offered with a detailed discussion of the implications of the current research for the effectiveness of violent crime control with traditional cr iminal justice approaches. Bo th the limitations of the current study and directions for future research ar e also considered (Chapter 10). 27

PAGE 28

Criminal Trajectory Official Intervention Labeling Effect Deterrent Effect Null Effect Time Criminal Behavior Figure 1-1. Illustrative model of official intervention as a potential turning point in a criminal trajectory. 28

PAGE 29

29 Time Criminal Behavior Criminal Trajectory Official Intervention Labeling Effect Deterrent Effect Null Effect Figure 1-2. Illustrative model of official inte rvention as a potentially different turning point for divergent criminal trajectories.

PAGE 30

30 Time Criminal Behavior Criminal Trajectory Official Intervention Labeling Effect Deterrent Effect Null Effect Figure 1-3. Illustrative model of official inte rvention as a potentially different turning point for divergent criminal trajectories with short-term and long-term differences.

PAGE 31

CHAPTER 2 THEORIES ON SANCTION EFFECTS Chapter 2 provides a comprehensive re view of the two major theoretical frameworks from which the orig inal competing sanction effect s hypotheses are derived. First, the classical school of criminology is traced back to 18th century writings to articulate the foundation of det errence theory, on which the modern day criminal justice system still rests. Focusing primarily on specific deterrence, key advances in the theoretical thinking about deterrence are explored which have served to clarify the mechanisms through which official intervention is hypothesized to reduce subsequent offending. Second, the emergenc e of the labeling perspective from its roots in symbolic interactionism is reviewed. Expanding on label ing theorys general notion that sanctions may ultimately increase subsequent offending, theoretical work that has explicated the causal pathways of the deviance amplificat ion hypothesis is thoroughly explored. Deterrence Theory This section details the very long and decidedly influential history of deterrence theory in criminology and crim inal justice. First, the or igins of deterrence theory are elucidated by tracing historical beliefs about crime causation and the corresponding brutal nature of criminal punishments. Key changes in thinking about human behavior and the important legal and penal reform effo rts that unintentionally led to the formulation of deterrence theory are discussed. Second, the theoretical assumptions regarding free will and hum an behavior reveal that choice governed by the principle of utility, is the chief cause of crime. Thir d, the crux of deterrence theory is outlined, highlighting how certainty, severity and ce lerity of punishment are hypothesized to control ones choice to engage in crime and viol ence. Fourth, attention is paid to the 31

PAGE 32

important distinctions dr awn between specific and general deterrence and, fifth, consideration is given to the differences between absolute and restrictive deterrence. These subsections underscore how deterrenc e may operate at two interrelated levels and address the issue of what exactly constitu tes evidence for a deterrent effect of criminal sanctions. Sixth, the notion of ti pping points is explained and their applicability to specific deterrence is noted. Seventh, recent advancements in the theorizing about deterrence are described and the necessary restriction of deterrence theory to the study of formal punishments is not ed. Finally, the significanc e of perceptions to the deterrence process is emphasized, which raises some new theoretical dialogue regarding possible outcomes of cr iminal sanctions. The Origins of Deterrence Theory Throughout history leading up to the 17th century, religious views maintained a stranglehold on the meaning of both the physical and social world. For more primitive societies, spiritualistic beliefs afforded humans a simple cause-and-effect explanation of events that they otherwise did not have t he means or capacity to understand. The phenomenon of crime was no differ ent and violations of the law were conceived of as being the product of demonic possession. Spiritualism was characterized by an ongoing battle between two forcesgood an d evil (Tannenbaum, 1938). Individuals were largely denied human agency, being r educed to pawns in a larger-than-life strategic game between opposing spiritual forces Ironically, the flesh of human bodies was held accountable for the actions of int angible, possessive demons. By todays standards, the methods employed to both determine guilt and punish criminal offenders were heinous and nonsensical. Considered within historic al context, however, they 32

PAGE 33

were not especially surprising given prevailing thoughts on the transcendental nature of criminal conduct. Criminal trials were known as ordeal s which involved placing offenders in situations where there was the expectation of harm to the accuse d under the belief that god would intervene to protect those who we re innocent. While there was regional variation in the particulars of ordeals, a common version of a trial by fire involved an offender walking a number of paces grasping a red-hot iron bar weighing one or three pounds, depending on the severity of the offense (Perkins 1926). Those who emerged unscathed from an ordeal were found innocen t. Unless an offender was sufficiently wealthy and able to successfully bribe the o fficiator to stage the outcome of the ordeal, there was always evidence of guilt. Whats more, for many offenders the determination of guilt and the punishment of death were i nextricably linked. For example, some offenders were tied up and tossed into rivers to be deemed proven guilty if they actually drowned (Newman, 1978). In short, a true tria l by ordeal essentially guaranteed that an offender was going to be found guilty and so metimes meant that one would pay immediately with t heir life. Beginning in the mid 1600s, a marked sh ift in thinking about human behavior occurred and crime began to be viewed in a se cular context. As opposed to being the work of spirits, criminal behavior was believe d to be the result of ordinary and natural human agency. This revolutionary view point saw men and women as persons who pursued their own interests without concerning t hemselves with the ramifications of their actions for others (Hobbes, 1967[1651]; Locke 1988 [1689]; Rousseau, 1968 [1762]). Given peoples preoccupation with self-inter ests, Hobbes (1967) argued that a social 33

PAGE 34

contract was essential to living in an envir onment that was free from anarchy (see also Locke, 1988; Rousseau, 1968). A social contra ct involves the relinquishment of some personal freedoms in exchange for societ ys protection from the harms and wrongdoings of other me mbers of society. Importantly, individuals should willingly give up only the smallest portion of liberty absol utely necessary for adequate protection from other citizens (Beccaria, 1963 [1764]). The success of a soci al contract rests upon the development of a state capable of upholding its responsibi lities to citizens without infringing upon their rights.2 Thus, the biased, capricious and arbitrary natures of the legal and penal systems, which characterized trials and punishments during the reign of spiritualism and bled over into the 17th and 18th centuries, became a focal point of concern. Primarily interested in reforming Eu ropean penal and legal systems, two 18th century utilitarian social philosophers set out to provide a philosophical rationale for 2 The relevance of the civilizing process (Elia s, 1939a [1978]; 1939b [1982]) for the theoretical applicability of deterrence to violent crime should not be overlooked. Since the safeguarding of individual rights and liberties requires an entity capable of upholding its end of the social contract, it is not surprising that the process of state formation was set in motion during this time period. Detailing the state formation process in Europe, Elias (1939a; 1939b) offered one of the most important social-psychological explanations of violence to date. Brought on by in ter-group tensions and rivalries, the process of civilization emerged in order to diffe rentiate the behavior of courtly soci al groups from ot her segments of society. With increasing human interdependencies t hat resulted from a modernizing world, manners and civilized behavior spread from the courtly social grou ps to other members of society. More important here, however, are the ramifications the civilizing pr ocess had for the character and quality of violence. The civilizing process results in a ri se in repugnance threshol ds for violence, increas es in drive-controls and guilt, and a removal of violence from everyday life. The process also entails the formation of a state monopoly on violence with respect to both the means and use of physical force (Elias, 1939a; 1939b). The monopoly of violence reserves violence for specific individuals such as the police and/or for specific time periods as in wartime (Elias, 1939a; 1939b). Resulting from these social changes in violence is the taming of expressive forms of violence. However, while use of violence for emotional satisfaction is lessened, the use of violence as a means to attain a specific goal is augmented. The deterrence doctrine maintains that the threat of legal sanctions is most effective in curbing instrumental crime (see Chambliss, 1967; Parker & Smith, 1979; Thomas & Williams, 1977). While it would be a gross misstatement to say that all violence in America is instrumental, the civ ilization process ensures that deterrence based criminal justice system responses have, at least in theory, relevance not only for crimes against property and society but also for violent crimes against persons. Violations of the law and violence can both be understood in instrumental terms and concepts of ra tionality are critical to understanding these behaviors (see Felson, 2009). 34

PAGE 35

punishment and eliminate the arbitrari ness and irrationality that plagued the administration of justice. Presented in t he form of a collection of concise essays, On Crimes and Punishment remains one of the most influentia l scholarly works in the field of criminal justice (see Beccaria, 1963). In his book, Beccaria (1963) called for a number of reforms including easily accessi ble and interpretable codes of conduct, abolishment of the death penalty, impartial application of the law, and sufficient resources to mount a defense against a criminal accusati on, to name a few. The rationales provided for legal and penal system modifications were well grounded; for example, Beccaria (1963, p. 32) argued that torturous methods of guilt determination should be abolished because its application ensures that t he very means employed to distinguish the innocent from the guilty will most effect ually destroy all difference between them. In fa ct, the reform efforts of Cesare Beccaria and Jeremy Bentham were so well formulated and commonsensical t hat the two utilitarian social philosophers articulated a theory of crim inal deterrence which remains the driving force behind the practice of modern day criminal justice in America as well as many nations across the globe. Indeed, unlike many t heories of criminal behavior, deterrence theory provides both a straightforward explan ation of crime and an unequivoca l solution to the problem (Pratt, Cullen, Blevins, Daigle, & M adensen, 2006). To understand the elegance and explanatory appeal of deterrence theory, it is necessary to explicate key assumptions about human choice and its prescript as well as the phenomenon of crime. Free Will, Pleasure and Pain Most theories of crime are born out of the positivism trad ition of science and attribute crime to causes that are internal and/or external to the individual but, importantly, largely out his or her direct control (e.g., se e Agnew, 1992; Shaw & McKay, 35

PAGE 36

1942). Deterrence theory takes a different a pproach and rests on the assumption that individuals have the ability to exercise fr ee will and directly control their behavior (Beccaria, 1963). While there may be certain lim its to the ability of humans to reason as well as individual differences in this c apacity (Cook, 1980), deterrence theory assumes that individuals are rational actors and freely have the ability to act on their own accord (Beccaria, 1963; Bentham, (1948) [1789]; Gi bbs, 1975). Thus, the cause of violent crime is as simple as a mere choice by the offender to engage in the antisocial behavior. Individuals do not aimlessly make choi ces; rather, the theory of deterrence suggests that an individuals course of action can always be known. Nature has placed mankind under the governance of two sovereign masters, pain and pleasure It is for them alone to point out what we ought to do, as well as to determine what we shall do (Bentham, 1948, p. 14, emphasis in original ). As the concepts readily infer, pleasure refers to the experience of happiness, benefit, or advantage whereas pain refers to the experience of mischief or unhappiness (Bentham, 1948). All individuals inherently seek to experience pleasure and to avoid pain; this is known as the principle of utility and is assumed to be relevant at all times and dictate every decision making process (Bentham, 1948). Systems which attempt to ques tion it [the principle of utility], deal in sound instead of sense, in caprice instead of reason, in darkness instead of light (Bentham, 1948). While Bentham makes a d auntless claim regarding the relevance of the principle of utility for human behavior, it is undeniably logical and enticingly parsimonious. 36

PAGE 37

In addition to human free will and the prin ciple of utility, deterrence theory assumes that crimes have certain pain and pl easure values which are inherent in the acts. Specifically, Bentham (1948) argues th at the pains and pleasures of single acts vary according to their intensit y, duration, certaint y, and propinquity. Intensity refers to the strength of the pains and pleasures that result from engaging in a criminal act and duration concerns the length of time in which these two forces are experienced. Certainty has to do with how likely pains and pleasures are to occur if one decides to commit a crime. Propinquity pertains to the time that elapses before pain and pleasure is experienced once the criminal act is completed. In addi tion to these four elements that characterize the quality of pains and pleasures, two other factors help guide the maintenance of behavior overtime including fe cundity and purity. The former re fers to the chance that an act will be followed by sens ations of the same kind, whereas the latter refers to the chance that an act will not be followed by sensations of the opposite kind (Bentham, 1948). Individuals who commit crimes receive pleasure from doing so and the pleasure associated with a given crime is thought to be similar for all individuals (Beccaria, 1963). Consequently, a scale of cr imes can be formed t hat orders criminal acts based upon their gravity or, in other words, their pleasure invoking quality (Beccaria, 1963). Given the interlo cking assumptions of free will and human governance by the principle of ut ility, the tactics necessary to formally control crime and violence emerge as relatively straightforward. Certainty, Celerity and Severity of Punishment Stemming from conceptions of men and women as rational and self-interested actors, the crux of deterrenc e theory rests on the propositi on that punishment that is certain, swift and adequately severe will decreas e the occurrence of crime (Beccaria, 37

PAGE 38

1963; Gibbs, 1975). The certainty of punishm ent refers to the likelihood of one getting caught and receiving punishment for committi ng a particular offense (Gibbs, 1975). The theory maintains that if the ce rtainty of punishment is great er, individuals should be less likely to commit crime. An especially high d egree of certainty of punishment is thought to create a feeling of omnipr esence such that the offender believes he or she could be under surveillance at any given time and, thus caught for his or her offense. In the theory of deterrence, certainty takes a state of primacy over celerity and severity of punishment and a large body of empirical resear ch supports this theoretical assertion (Nagin, 1998). Celerity, or swiftness, refers to the dur ation it takes to experience the punishment following the commission of a crime (Gibbs, 1975) Punishments that are applied swiftly are thought to be more effective in reduci ng crime because they yield a clearer and stronger association between t he crime and punishment (Becca ria, 1963; Gibbs, 1975). Despite this logical assertion, celerity has not been found to be especially relevant to deterring crime and violence (Clark; 1988; Nagin & Pogarsky, 2001). The severity of punishment refers to the amount and type of punishment experienced by the offender for committing a crime (Gibbs, 1975). Recall, deterrence theory is a byproduct of the di rect response by reformers to the biased and brutal nature of the administration of just ice. While deterrence is best achieved by maximizing certainty and celerity of punishment, severi ty of punishment should not be maximized but be commensurate with the crime (Beccaria, 1963; Gibbs, 1975). There is also an interesting interaction between the certai nty and severity of punishment. More specifically, punishments that are more seve re tend to be applied with less certainty, 38

PAGE 39

whereas punishments that ar e applied with a low degree of certainty must be more severe to invoke deterrence (see Beccaria, 1963). While the principles of dete rrence are largely responsibl e for the administration of justice today, the theorized relative importance of certainty, severity and celerity to the deterrence process did not take hold immedi ately. For instance, the conceptualization of the human as self-serving coupled with the emergence of new forms of wealth resulted in a dramatic surge in the number and types of offenses that were punishable by death in England (McLynn, 1989). During the reign of the Bloody Code in England (1688-1815), the same cr ime could be prosecuted under to tally different statutes and penalties [and] often a multip licity of laws bore on the sa me crime. The theory and practice of criminal law were light -years apart (McLynn, 1989, p. xiii). To state that the gap betw een theory and policy has been closed entirely or even that policies are no longer enact ed solely to influence the severity of criminal sanctions would be incorrect. Many statutes currently on the books in America lead to punishments that do not appear to be commens urate with the criminal act. For instance, Zimring, Hawkins and Kamin (2001) detail Californias three strikes law in which the focus on controlling habitual off enders leads to instances of individuals receiving twenty-five years to life for larceny. Not surprisingly, three strikes legislation has resulted in a swelling of the prison popul ation (Zimring et al., 2001) and some court districts such as those in the San Francisco region are reluctant to apply the penalty, resulting in a reduction in certainty of punishment (Males & Macallair, 1999). These examples underscore the point that while deterrence theory suggests a greater emphasis on certainty of punishment, often in creases in the severity of sanctions are 39

PAGE 40

initiated to engage criminal deterrence. It must be understood, however, that this approach may have negative feedbacks on the cert ainty of punishment (e.g., selective enforcement and application of t he law) as well as the celerity of punishment (e.g., lengthy appeals processes for death sentences).3 Despite potential problems with the practical application of the pr inciples of deterrence, the t heory logically suggests that manipulating certainty, severity and celerity of punishment will reduce criminal behavior and it is consistent with prev ailing notions of individual culpability. The proposed solution to the crime problem that is derived from deterrence theory is thought to both prevent criminal behavior from occurring in the first place and correct it if it has already transpired. Specific and General Deterrence The sanctioning of offenders engages a deterrence process at two levels simultaneously (Zimring & Hawkins, 1973). When an offender is caught and punished in a swift fashion and with adequate severity, he or she is thought to reduce their future offending behavior due to the fear of future sanctions. This process is known as specific deterrence and it is so named because it s referring to the effects of punishment for the specific offender who committed the cr ime. Relatedly, attempts to deter the future behavior of a specific individual th rough formal intervention also serve as examples to the general public of the ra mifications of committing a crime. The apprehension and formal sanctioning of crimi nals are widely covered by news and 3 Most research on deterrence theory focuses on the addi tive effects of certainty, severity and celerity of punishment. While there is a general rank-ordering of the importance of the various factors of punishment as well as some important interrelationships among them such as those discussed above, one should not lose sight of the fact that deterrence theory does suggest that all three elements must be present to some extent for deterrence to be successful in reducing criminal offending (Stafford et al., 1986). Hence, if there was a high d egree of certainty and celerity, for ex ample, but essentially no severity of punishment, the utility of deterrenc e to control crime and violence would be miniscule at best. 40

PAGE 41

media organizations (Jewkes, 2004). This process of reducing crime is known as general deterrence because the appl ication of formal sanctions to individuals can also influence perceptions of certainty, celerity and severity of the general public as well. Differences between general and specific deterrence are often noted in research but subsequently disregarded (Zimring & Hawk ins, 1973). Stafford and Warr (1993) note the distinctions draw n between general and specific deterrence have muddied the waters and maintain that individuals experiences with punishment and punishment avoidance are multifaceted. To Stafford and Warr (1993) specific deterrence is conceptualized to not only consist of direct experiences with punishme nt but also direct experiences of punishment avoidance. Recognizing the potent ial importance of knowledge of the punishment of others, they redefine general deterrence to be indirect experiences with punishment and punishment av oidance. The point to be emphasized is that in most populationswhether memb ers of the general publ ic or punished offenderspeople are likely to have a mixture of indirect and direct experience with punishment and punishment avoidance (Stafford & Warr, 1993, p. 126, emphasis in original). This reconceptualization of deterrenc e theory is helpful here because it guides the expected effects of direct experiences of punishment across different types of offending populations. Specifically, St afford and Warr (1993) argue that direct experiences of punishment and punishment av oidance should be most relevant to the behavior of habitual offenders and less relevant to novice offenders with little or no criminal justice contact (i.e., the latter gr oup is driven more by their notion of general deterrence). While it is k nown that deterrence theory pred icts an inverse relationship 41

PAGE 42

with direct and indirect punishment experiences the question as to what exactly would constitute evidence in support of a deterrent effect has not ye t been broached. Absolute and Restrictive Deterrence A more unconventional distinction, but one of great importance to assessing the efficacy of deterrence theory, has been drawn between absolut e and restrictive deterrence (Gibbs, 1986).4 Absolute deterrence refers to whether the fear of punishment leads individuals to refrain fr om committing offenses altogether (Gibbs, 1986). In other words, an offender is deterred from crime only if they enter into a state of complete offending inactivi ty. As a more temperate in terpretation of the success of official intervention, restrictive deterrence re fers to the extent to which fear of legal sanctions reduces the overall number of crimes (Gibbs, 1986). Individuals need not cease their criminal offending but it should decrease following punishment experiences. While the main goal of formally sancti oning an offender is undoubtedly to achieve absolute deterrence, it may be impractical to attain this objective. Indeed, it would be unsagacious to view a substantial reducti on in violent crime among chronic offenders attributable to official intervention as a total failure. Moreover, the idea of restrictive deterrence is more in line with views of criminal desistance as a dynamic or developmental process that unfolds over time (see Bushway, Thornberry, & Krohn, 2003). Restrictive deterrence acknowledge s that punishment experiences may have 4 An important distinction has also been drawn betwe en absolute and marginal deterrence (Gibbs, 1986). In this case, absolute deterrence refers to t he criminal deterrence that occurs by having a system of legal punishments in place as compared to a system in which no legal punishments are in place. Most concur that the existence of sanctions in general plays an instrumental role in reducing crime but many scholars contend this is a relatively uninteresting phenom enon. Marginal deterrence refers to the extent to which changes in legal punishments increase or de crease participation in crime. Scholars maintain that understanding marginal deterrent effects, not absolute deterrent effects, should be the principal aim of deterrence research. 42

PAGE 43

greatly increased the awareness of the pains of crime making it considerably less likely in many situations, yet not sufficiently enough such that the pains outweigh the pleasures in all situations. Tipping Points Every change in the certainty, severity or celerity of punishment may not be accompanied by a change in offending behavior. It is conceivable that there is an optimal range in which marginal deterrence (i.e., changes in behavior associated with changes in legal punishments) occurs, but outs ide of which there is no relationship between punishment quality and criminal behavior. In deed, penalties that are inappreciable to offenders may fail to have any deterrent value (Tittle & Rowe, 1974; Brown, 1978; Yu & Liska, 1993) and punishm ents that approach certain levels may have diminishing returns (Yu & Liska, 1993). Deterrence theory explicitly predicts some of these floor and ceiling effects. For exam ple, deterrence theory argues that it is necessary that punishment reaches a cert ain degree of severity to engage criminal deterrence but need not be harrowing to do so (Beccaria, 1963). In theory, making punishments more severe than necessary fa ils to have any added deterrent benefit and could actually reduce the certainty of punishment (see Beccaria, 1963). Moreover, punishments such as the death penalty may trigger a brutalization effect, where there is an increase in offending among t he general public following the execution of an offender (Bowers & Pierce, 1980). Then, there may be a lower threshold of punishment severity that must be reached in order fo r deterrent effects to occur at all. Similarly, there may be an upper threshold of punishment that, if exceeded, does not result in any additional gains in criminal deterrence and po ssibly could make matters worse. 43

PAGE 44

Much of the empirical work on tipping poin ts has focused on certainty of arrest within the context of general dete rrence. This research has revealed that the effects of tipping points may differ across divergent demographic groups. For instance, Yu and Liska (1993) note that Blacks ar e subject to higher arrest cert ainty; they find that tipping points for certainty of arrest, especially ceili ng effects, are especially pronounced for this demographic group. This idea can easily be applied within the context of specific deterrence. Because of differences in moti vation, perceived rewards or the social costs of committing a crime, individuals may requi re different punishment experiences to reach their tipping point. Iss ues of equity in the administration of justice aside, uncovering information regarding the penalties necessary to trigger specific deterrence for certain types of offenders could be highly useful information for reducing crime and violence. Modern Deterrence Theory While the essential idea of deterrence theor y remains that increases in certainty, severity and celerity of punishment lead to decreases in criminal behav ior, the theory is considerably more nuanced and detailed than the original statement s of the theory lead one to believe (Gibbs, 1986). Gibbs (1986) maintains that deterrence theory can be broken down into three premises and two coro llaries. The premises state there is a direct relationship between objective properti es of punishment and deterrence, there is a direct relationship between perceptual pro perties of punishment and deterrence, and there is an indirect relationship between deterrence and crime (Gibbs, 1986). The two corollaries state there is an inverse relati onship between the perceptual properties of punishment and crime and, secondly, an inverse relationship between the objective properties of punishment and crime (Gibbs, 1986). While this conceptualization of 44

PAGE 45

deterrence incorporates both perceptions and objective ex periences, it also expands deterrence beyond certainty, celerity and severi ty of punishment into the realms of nonlegal sanctions There has been a push to extend deterr ence theory beyond legal sanctions. On the heels of strong academic interest in soci etal reaction theories of crime, a renewed enthusiasm over the deterrence perspec tive occurred in the 1970s and 1980s when economists began applying rational choice pers pectives to explain criminal behavior (e.g., see Becker, 1968; Heineke, 1978). So me contend that the reach of deterrence theory extends beyond formal sanctions (Zimri ng & Hawkins, 1973). In one view, formal sanctions are most likely to have deterrent effects through informal social sanctions (Andenaes, 1974; Gibbs, 1975; Blumstein & Nagin, 1976; Williams & Hawkins, 1989; Zimring & Hawkins, 1973). In other words, formal sanctions trigger informal, negative reactions from parents and community member s which are ultimately responsible for any observed reductions in offending. Alter natively, rational choice theory subsumes formal sanctions, adding them to a conglomerat ion of costs and rewards of crime that are included in offender decision making (Cornish & Clarke, 1986; Paternoster, 1989a; 1989b). While the inclusion of considerations of perceptions in deterrence theory is most appropriate, the inclusions of non-legal s anctions and other elements of punishment have been argued to be logically inconsistent with the theory of deterrence as its explicit focus is on the costs of formal sanctions (A kers, 1990). Akers (1990) also notes that extending the theory beyond formal sanction s encroaches on other theories such as social learning theory (i.e., differential reinforcement) (see also Akers & Sellers, 2009). 45

PAGE 46

Thus, one view holds that to t he extent that informal sancti ons are incorporated as part of the deterrence process, the theory of deterrence upon which the criminal justice system is based is, strictly speaking, no longer being evaluated (Akers, 1990). However, some formidable scholars have praised efforts at connecting formal and informal sanctions. For instance, Nagin (1998) states In my judgment the most important contribution of the perceptual dete rrence literatureis t he attention it has focused on the linkage between formal and informa l sources of social control (p. 19). The basic idea is that the deterrence process is invoke d when an individual refrains from committing a crime because they fear arrest may bring about adverse social consequences (Williams & Hawkins, 1986). These ties between formal and informal sanctions may be important for policy considerations. Regardless of whether one agrees that info rmal sanctions belong in the realm of deterrence theory, the consider ation of them provides a mechanism which links justice system contact to decreased future crime. Significantly, this is a unique mechanism providing a different rationale for why arrests should be associated with decreased crime and violence. The other major mechanism, which is without controversy, is that of perceptions and the importance of them to modern deterrence theory is now discussed. Perceptual Deterrence While objective levels of certainty, severity and celerity of punishment are important to deterrence, per ceptual deterrence theory holds that individuals make decisions based upon their knowl edge or perceptions of these objective realities (Gibbs, 1986; Nagin, 1998). In other wo rds, objective properties of punishment should influence sanction perceptions which should, in turn influence decisions to engage in criminal behavior. This point would be moot if individuals were perfect perceivers and human 46

PAGE 47

calculators of objective information. This, however, appears not to be the case; individuals do not accurately perceive thei r objective risk of punishment (Cook, 1980; Kleck, Sever, Li, & Gertz, 2005) Moreover, calculations of objective risk of punishment and the acceptability of perceived risks are not systematically the same across persons (Clarke & Cornish, 1985; Cornish & Clark e, 1986; Cook, 1980; Gibbs, 1986; Nagin & Paternoster, 1993; Nagin & Pogarsky, 2001; 2003). There are also individual differences in exposure to information about risks of punishment from a variety of sources including media, visibility of law enforcement, direct and indirect experiences with punishment, and daily conversations (Cook, 1980). The validity of rational decision-making theories does not requi re a perfect correspondence between contingencies and perceptions of them, but does require some correspondence if the theories are to have any explanatory or predictive power (Kleck et al., 2005, p. 627). The realization that perceptions hold the key to whether objective properties of punishment influence criminal behavior shifted the focus of deterrence research from macro-level research and interrupted time-ser ies impact studies to micro-level studies focusing on the effects of individual perceptions on crime (Nagi n, 1998). In general, this line of research has revealed that certaint y perceptions do influence criminal behavior as predicted by deterrence theor y, but the effects appear to be rather modest (Pratt et al., 2006). More specifically, the relev ance of perceptions for deterrence has been studied with cross-sectional, scenario-bas ed, and panel survey approaches (Nagin, 1998). Both cross-sectional and scenario-b ased studies lend regular support for the relationship between risk perceptions and self -reported criminal behavior or criminal intentions (Nagin, 1998); however, panel studies and those with stronger 47

PAGE 48

methodological designs typically find much weaker support for deterrence theory (Nagin, 1998; Paternoster, 1987), thereby highli ghting the importanc e of using strong methodology. Panel studies attempt to address the important tempor al ordering question of whether risk perceptions cause crime or whether crime causes risk perceptions (Greenberg, 1981). Non-offender s have been thought to overestimate their sanction risks, which are lowered to more reasonable levels once the individual begins engaging in crime without being subject to punishment (Paternoster, Saltzman, Chiricos, & Waldo 1982; Paternoster, Saltzman, Waldo, & Chiricos, 1983a; 1983b; Saltzman, Paternoster, Waldo, & Chiricos, 1982). Some scholar s maintain that the observed negative relationship between risk perceptions and sanctions is actually evidence for an experiential effect, not a deterrent one (Paternoster, 1982; 1983a; 1983b; Saltzman et al., 1982). However, evidence for an experientia l effect does not nece ssarily contradict deterrence theory. Recall, Stafford and Wa rr (1993) redefine specific deterrence as direct experiences with both punishment and punishment avoidance. Those individuals who elude detection would be more likely to continue to offend, or have lower risk perceptions, precisely because they have dodged authorities in the past (Stafford & Warr, 1993). In testing the effects of direct and indirect experi ences with punishment, Piquero and Pogarsky (2002) unexpectedly found an em boldening effect, in which the behavior of those with punishment exper iences appeared to actually worsen following official intervention. Offering alter native explanations for this finding, Pogarsky and Piquero (2003) propose that a positive relationship between punishment experiences and risk 48

PAGE 49

perceptions could simply mean that punish ment experiences identify those most committed to offending or that humans are ill ogical and subject to the gamblers fallacy (see also Piquero & Pogarsky, 2002). Termed the resetting effect, they suggest that offenders who have a recent experience with puni shment may think they are unlikely to get caught again in the near future (i.e., t heir risk perceptions actually decrease). Interestingly, the hypothesis of the augmentation of deviant behavior following a criminal sanction ordinarily falls under the purview of labeling theory. Labeling Theory This section provides an in-depth treat ment of the one-hundred year history and development of labeling theory, which is an al ternative theory of criminal sanctions. In doing so, it touches on the cyclical nature of labeling theory in academia which has involved states of development, prominence, irrelevance and revival. First, labeling theorys roots in the symbolic interactionism perspective are discussed, which provides the foundational ideas which were necessary for the emergence of a theory of criminal sanctions that anticipates problematic behavioral outcomes. Second, the basic idea of deviance amplification is traced back to the init ial formulations that specify the effects of tagging one as a deviant. Third, the depth and breadth of labeling theory is considered by formalizing and discussing ea ch of the theorys nine tenants. This discussion makes clear that labeling is a theory of criminal behavior which is independent from labeling as ei ther a theory of law making or law enforcing. Honing in on the deviance amplification hypothesis (i.e., labeling as a theory of criminal behavior), fourth, important clarificati ons about the character and qual ity of deviance both before and after a labeling event are reviewed; the dev elopmental nature and the critical role of formal labels in the transition from primary to secondary deviance are noted. Finally, 49

PAGE 50

given the emphasis here on the criminal s anction as an independent variable, important developments roughly twenty years ago that have specified the causal pathways to secondary deviance and have sparked renewed interest in labeling theory are thoroughly examined. The Roots of Labeling Theory The roots of labeling theory are deeply entrenched in the sociological perspective of symbolic interactionism, which holds that social interaction and the communication of ideas, through language and symbols, is the lifeblood of human m eaning (see Blumer, 1969). An individuals values, attitudes and self-identity all can be understood only in the context of social interactions with other members of so ciety. Symbolic interactionism hinges on three premises: h uman beings act toward things on the basis of the meanings that the things have for themthe meaning of such things is derived from, or arises out of, the social interaction that one has with ones fellows [and] these meanings are handled in, and modified through, an interpretative process used by the person in dealing with the things he encounters (Blumer, 1969). While the first principal idea of symbolic interactionism describing human action is quite commonsensical, it is often neglected in modern sociological theories of human behavior (Blumer, 1969). Symbolic interacti on theorists place considerable emphasis on understanding meaning as a key determinant of human actions and consider it misguided to focus research efforts on unc overing specific factors that influence behavior without considering meaning. According to the perspective, regarding meaning as a trivial intervening process or having causative factors envelop meaning unduly demotes its importance to behavior (Blumer, 1969). 50

PAGE 51

In discussing the second key idea of symbolic interactionism, Blumer (1969) notes that meaning is neither physically inherent nor psychologically determined through human perceptions; rather, t he meaning of activities, situations and persons is constructed through a process of social intera ction. In addition to the importance of social interaction in creating meaning, the th ird chief notion of the perspective holds that a process of interpretation is critical. Occurring in two stages, an individual first communicates with his or herself and considers the things that have meaning and then handles meaning through interpretation. T he actors selects, checks, suspends, regroups, and transforms the meanings in the light of the situat ion in which he is placed and the direction of his action (Blumer, 1969) In short, meanings have their influence on behavior through processes of social as well as self-interactions. The process of self-reflecting over social interactions can be clearly seen in Charles Cooleys (1902) looking-glass self, which is a prominent concept within the symbolic interaction tradition that is highl y relevant to the labeling perspective. According to Cooley (1902), our idea of our selves unfolds in a three-part process involving the imagination of our appearance to the other person; the imagination of his judgment of that appearance, and some sort of self-feeling, such as pride or mortification (p. 152). Importantly, T he comparison with a looking-glass hardly suggests the second element, the imagined judgment, which is quite essential (Cooley, 1902, p. 152). Hence, the consideration of others judgments of our appearance, not simply our appearance to others, is ultimately what drives self-feelings. The looking glass self draws attention to the import ance of an individuals interpretations and expectations of social interaction as the ke y factor in eliciting a self-feeling that is 51

PAGE 52

ultimately tied to ones behavioral response. In sum, the theoretic al ideas developed within the symbolic interaction tradition of sociological thought set the stage for the emergence of a theory of labeli ng, which links criminal behavior with societal reactions or, more precisely, with perceptions and interpreta tions of social intera ctions. Emergence of the Labeling Perspective Drawing on the idea that self-conceptions stem from individual interpretations of ones beliefs regarding the way other individuals view him or her (Cooley, 1902), Frank Tannenbaum (1922; 1938) articula ted the earliest versions of the labeling perspective and advanced the idea that labels play a key role in the development of anti-social behavior. Interestingly, his theoretical insights that have been firmly imprinted on the field of criminology grew, in part, out of his own intimate experiences on the wrong side of the law. Less than a decade after serv ing a year-long imprisonment sentence, Tannenbaum (1922) writes of the ramifications of incarcer ation maintaining that it is when an individual is tagged as a criminal and imprisoned that ones selfand socialworlds crumble beneath him or her. More specifically, individuals lose their own ambitions and interests and have conventional opportunities such as work slip away, impelling a change from a conformist to a deviant identity (Tannenbaum, 1922).5 These initial formulations of the labeli ng perspective drawn out of Tannenbaums early writings on the American prison system (Tannenbaum, 1922) were more fully advanced later and applied to understand the in fluence of justice system contact on adolescents (Tannenbaum, 1938). In the midst of emerging perspecti ves of criminal 5 Other scholars also noted the potential corrupting role of corrections endeavors in America (e.g., see Bonger, 1916; Shaw, 1930; Lombroso-Fe rrero, 1911) but it was Tannenbaum s (1922) work that placed strong emphasis on these forces helping to birth the labeling perspective. 52

PAGE 53

behavior that challenged the reign of biologi cal positivism in the early to mid-1900s (Sellin, 1937; Sutherland, 1937; Shaw & McKay, 1942), Tannenbaum s (1938) labeling perspective judged criminal behavi or not to be chiefly born at birth; rather, criminals were thought to be largely born out of a proce ss of societal reaction to their behavior. The first dramatization of t he evil which separates the child out of his group for specialized treatment pl ays a greater role in making the criminal than perhaps any other experienceThe process of making the crimi nal, therefore, is a process of tagging, defining, identifying, segregat ing, describing, emphasizing, making conscious and selfconscious; it becomes a way of stimulating, suggesting, emphasizing, and evoking the very traits that are compla ined of (Tannenbaum, 1938). Put pl ainly, an adolescent who has been tagged as a deviant assumes the menacing role (Tannenbaum, 1938). Labels possess their power because they are st igmatizing; deviant labels remain highly influential and consistently denouncing because negative images of criminals encompass society through various sources of media on a daily basis (Becker, 1963; Goffman, 1963). Tannenbaums (1922; 1938) work is especially important because it laid the foundation for a theory of human behaviorimplicating formal intervention efforts in the production of anti-social conductand high lighted the importanc e of the labeling process for the lives of adolescents. The labeling perspective as a whole, however, extends well beyond a theory of criminal behavior. Before honing in on the major works that have expanded on the basic idea of devianc e amplification, which is most relevant to the current study, each of the nine core ideas of l abeling theory is outlined. Understanding these tenants helps to demonstrate that labeling as a theory of criminal 53

PAGE 54

behavior (i.e., deviance amplification) is entir ely separable from labeling as a theory of law making and law enforcing. In part, it is this very distin ction that is responsible for the acceptance of deviance amplific ation as a mainstream criminologi cal theory today. Nine Tenants of Labeling Theory Formulating nine key statements that summa rize the key ideas of labeling theory and capture the breadth and depth of the perspective, Clarence Schrag (1971) writes: (1) no act is intrinsically criminal, (2) criminal definitions are enforced in the interest of the powerful, (3) a per son does not become a criminal by violation of the law but only by the desig nation of criminality by authorities, (4) due to the fact that everyone conf orms and deviates, people should not become dichotomized into criminal and non-criminal categories, (5) the act of getting caught begins the labeling process, (6) getting caught and the decision-making in the criminal justice system are a function of offender as opposed to offense characteristics, (7 ) age, socioeconomic class, and race are the major offender characteristics t hat establish patterns of differential criminal justice decision-making, (8) the criminal justice system is established on a freewill perspective that allows for the condemnation and rejection of the identified offender, (9 ) labeling is a process that produces, eventually, identification with a deviant image and subculture, and a resulting rejection of the rejector s (Schrag 1971, pp. 89-91; see also Wellford, 1975, p. 333). Wellford (1975) rightly notes t he heart of labeling theory lies in the first, sixth, and ninth assertions; taken together, these assertions formulate a theory of law making, a theory of law enforcing, and a theory of criminal behavior, the la tter of which is especially relevant to the current inve stigation which seeks to examine the effects of criminal sanctions on violent offending behavior. T he nine propositions can be usefully divided into those most relevant to law making and enforcing and those more material to explaining criminal and dev iant behavior. Several theoretical statements directly pertain to labeling as a theory of law making and law enforcing. The first and second statements of labeling theory remind us that crime is a socially defined phenomenon and that the enactment and enforcement of 54

PAGE 55

laws is guided by power relations. The sixth claim of labeling theory forms the basis of the status characteristics hypothesis, which st ates that certain types of individuals will be more likely to experience deviant labels r egardless of their off ending behavior (see Paternoster & Iovanni, 1989). Tied to conf lict and critical criminological thought (Bonger, 1916 [1905]; Marx, (1964) [1844] ; Quinney, 1980), this proposition holds that who gets arrested, prosecuted and punished is driven largely by irrelevant, nonbehavioral characteristics. An extension of the sixth assertion, the seventh tenant specifies which characteristics enter into the decision-making processes of criminal justice professionals which include an i ndividuals gender, race, and socioeconomic status. Any in-depth discussion of labeling theor y necessitates consideration of the aforementioned propositions which are directly pertinent to labeling as a theory of law making and law enforcing; it is these more critical assertions that are partially responsible for the academic dismissal of labeling theory in general during the 1970s and 1980s (Wellford & Triplett, 19 93). Critical scholars c harged labeling theorists with focusing on non-powerful, easy targets such as nuts, sluts, and preverts (Liazos, 1972). According to critical scholars, the wo rk of labeling theorists invoked issues of power and conflict incompletely (Best, 2004). Despite only borrowing key ideas from critical perspectives and refraining from becoming engulfed in ideology, labeling theory as a whole nevertheless becam e seen as wedded to conflict and critical thought greatly contributing to its rejecti on among mainstream scholars during the 1970s and 1980s (Wellford & Triplett, 1993). 55

PAGE 56

Several tenants point to the importance of official labeling for characterizing one as a criminal and highlight the ramifications of the labeling process. Collectively, the third and fourth tenants suggest that si nce crime and deviance is fair ly common in society, it is the act of officially stamping one as a deviant which is relevant to offender categorization. That is, classifications based on rule-violating behavior are argued to lack utility. It is seen as somewhat problem atic that a person who has violated the law in which no social reaction occurs is not c onceived to be a deviant, but a person that is falsely accused of violating the law in which a social reaction occurs is thought to be a deviant (Gibbs, 1966). This criticism is ce rtainly warranted and it makes little sense to contend that studying rule-br eaking behavior is unimportant.6 At the same time, these two propositions do make clear the emphasis that labeling theory places on tagging and the importance of this event for des ignating one as an outsider. Gibbs (1966) claims that labeling theorists have never specified exactly what kind of reaction identifies deviant acts (p. 13). While ther e is some degree of imprecision, the fifth proposition does state t hat getting caught is the key ev ent that sets in motion the process of labeling. Given the emphasis that is placed on official acts and labeling by individuals in positions of authority (Be cker, 1963; Garfinkel, 1956), getting caught by the police serves as a very logical point of origination of t he labeling process. Once an individual has been caught (i.e., an official label has been applied), the ninth proposition maintains that the end resu lt is the worsening of delinquency. This statement is most important and forms the basis of what is known as the deviance 6 Indeed, the very idea of deviance am plification requires that rule-break ing behavior is an important area of study in criminology. Once an individual is labeled, it is their rule-breaking behavior that is ultimately thought to increase (Paternoster & Iovanni, 1989). In other words, it seems decidedly less fruitful to study the effects of labeling on labeling (as opposed to the effect of labeling on rule-breaking behavior). 56

PAGE 57

amplification hypothesis, wh ich specifies that the application of a deviant label ultimately leads to secondary deviance (Sch rag, 1971; Paternoster & Iovanni, 1989). The adverse ramifications that official l abels may have for offending behavior are made possible by the eighth tenant. This propos ition is simply an acknowledgment that operations of the criminal justice syst em are predicated on human free will and, consequently, rejection of the individual does not conflict with prev ailing thoughts on law enforcement. Recall, the i deas of free will and rationality are the same notions upon which deterrence theory is based (Bentham, 1948). While criminal justice system operations may be based on theories of criminal deterrence, deviance amplification remains a possible outcome of official intervention because criminal justice system professionals do not contend that the experience of getting caught (i.e. being labeled a criminal) is a problem in and of itselfit only becomes problematic if it leads to adverse outcomes consistent with the deviance amplification hypothesis. Making explicit the ideas implicit in the ninth tenant of labeling theory, the theoretical literatu re that is responsible for the development and eventual formulation of testable hypotheses of devi ance amplification is now reviewed. Primary and Secondary Deviance The labeling perspective became considerab ly clearer with the advent of Lemerts (1951) distinction betw een primary and secondary dev iance coupled with his explanation of the tr ansition between these two types of deviance (i.e., a process of individual action and societal reaction unfol ding overtime). Primary deviance is characterized by anti-social behavior brought on by a wide variety of circumstances including those that may be sociological, cultur al, psychological, biological, or situational in nature. While the initial sources of anti-social behavior are many and varied, primary 57

PAGE 58

deviance is rationalized or otherwise dealt wit h as functions of a socially acceptable role (Lemert, 1951, p. 75). Put differently, an individual perceives his or her deviant behavior to be justifiable or not necessarily inconsistent with societys behavioral expectations. Thus, participation in pr imary deviance does not have a sizeable influence on an individuals sense of self or his or her perceived role in conventional society (Lemert, 1967). While changes in attitudes, behaviors and roles are not ev ident with primary deviance, these features form the hallmark of secondary deviance. According to Lemert (1951), participation in secondary deviance takes place when a person begins to employ his deviant behavior or a role based upon it as a means of defense, attack, or adjustment to the overt and covert probl ems created by the consequent societal reaction to him (p. 76). At this stage, the multitude of possible causes that could have driven patchy acts of primary deviance bec ome largely inconsequential and societal reaction replaces these factors as the preemi nent cause of more in tense and structured nonconforming behavior (Lemert, 1951). Sharing the common cause of formal societal reaction, it is secondary devi ance that proves to be most problematic; thus, the very behavior society was attempting to regulate c onspicuously takes a turn for the worse. Understanding the character of primary and secondary deviance is important but it is of equal significance to understand how indivi dual actions and societal reactions bring about this marked shift in the quality and charac ter of deviant behavior. Outlining the transition from primary to secondary deviance, Lemert (1951) sketches an interaction sequence that progresses in roughly eight steps: (1) primary deviation, (2) social penalti es, (3) further primary deviation, (4) stronger penalties and rejection, (5) further deviation, perhaps with 58

PAGE 59

hostilities and resentment beginning to focus upon those doing the penalizing, (6) crisis reached in tole rance quotient, expressed in formal action by the community stigmatizing the deviant, (7) strengthening of the deviant conduct as a reaction to the stigmatization and penalties, (8) ultimate acceptance of deviant and soci al status and efforts at adjustment on the basis of the associated role (p. 76). This description of the typical labeling proc ess brings two ideas to light, which are slightly at odds with one another. First, the passage from primary to secondary deviance usually does not take place ov ernight. There is an ongoing interaction between the individual and society that occurs over a period of time, implying that labeling is a developmental pr ocess. However, following cycles of acts of primary deviance and societal reactions to them is the momentous event of being officially labeled a deviant. The problematic behavior reaches a point where it is no longer tolerated and a formal intervention occurs th at officially stamps an individual as a deviant. While the transition to secondary deviance tends to unfold over time (Lemert, 1951; 1967), transformations in identity are most likel y during status degradation ceremonies especially those that involve an official act (Garfi nkel, 1956). Second, then, key stigmatizing events by formal institutions play a decis ive role in an individuals behavioral conversion (Becker, 1963).7 Importantly, it is forma l interventions such as arrests or convictions that serve as official labeling events; indeed, the criminal justice system maintains exclusive rights over whom is to be considered criminal (Garfinkel, 1956). These pivotal formal labeling events are of primary interest to labeling scholars studying how sanctions may influence the transition from primary to secondary deviance. 7 It is important to note that formal stigmatization is most likely to have a problematic effect on behavior when there is no attempt made to reintegrate the nonc onforming individual back into conventional society, which is typical of formal intervention effort s in the United States (Braithwaite, 1989). 59

PAGE 60

One of the chief criticisms levied against labeling theory is that it provides no explanation for criminal or deviant behavio r in the first place (Akers, 1968) and labels may be neither necessary nor sufficient conditi ons for the maintenance or amplification of criminal behavior (Mankoff, 1971). Davis (1961, p. 461, emphasis in original ) notes that labeling theorists who claim that reactive processes of society provide the causal factor in deviance are thereby providing only a partial view of a highly complex problem. At the same time some maintain that there has been a plethora of theories that have been advanced to explain specific aspects of anti-social behavior, with few attempts to explain criminal behavior in any general sense (Lemert, 1951). This has been occasioned in no small way by the preo ccupation with the orig ins of pathological behavior and by the fallacy of confusing original causes with effective causes. All such theories have elements of tr uth, and the divergent viewpoints they contain can be reconciled with the general theor y here if it is granted t hat original causes or antecedents of deviant behaviors are many and diversified (Lemert, 1951, p. 75 emphasis in original ). In other words, while ther e are numerous divergent primary sources of violent and anti-social behavior, t he labeling process is a general explanation of behavior as it results in the intensificati on and/or long-term stabilit y of the problematic behavior irrespective of the initial origins. As will become clear, this is a critical theoretical notion which direct ly implicates formal intervention as a potential turning point, albeit in a socially undesirable direction, in a viol ent behavior trajectory. The involvement of formal institutions in the labeling process is more nuanced than simply stamping rule-breakin g criminals as deviants; formal institutions first define what is right and wrong and then utilize their aut hority to select what codes to enforce 60

PAGE 61

and which individuals to label as deviant (Becker, 1963). In one of the most recognizable assertions of t he labeling perspective, Becker (1963, p. 9) states that deviance is not a quality of the act the per son commits, but rather a consequence of the application by others of rules and sanction s to an offender. The deviant is one to whom that label has successfully been appli ed; deviant behavior is behavior that people so label. In other words, specific behaviors and certain individuals are not necessarily inherently deviant. Instead, acts and actors become deviant when society defines them as such. Rule-breaking behavior involves violating a code of conduct but being a deviant requires the acquisition of a label. Beckers (1963) delineation between rule-b reaking behavior and the affixation of deviant labels creates an important two-by-two labeling typology He refers to individuals who do not engage in rule-breaki ng behavior and do not have deviant labels as conformists. This represents the majori ty of people in society. Individuals who become involved in rule-breaking behavior and are appropriately labeled for doing so are referred to as pure deviants. For so me individuals, however their rule-breaking behavior and their assigned label or lack thereof does not neatly align. Specifically, those individuals who violate codes of conduc t but are not labeled deviant are known as secret deviants and those who do not viol ate rules but are never theless labeled are known as falsely accused (Becker, 1963) Labeling theory sugges ts a worsening of behavior is likely for those labeled who were originally involved in primary deviance (pure deviants) and the onset of real problematic behavior for those labeled who were initially engaged in convent ional activities (falsely accused) Absent the attachment of a formal label, the theory predicts deviance am plification processes will not likely be 61

PAGE 62

triggered; and, this is thought to hold irrespec tive of whether the individual actually engages in rule-breaking behavior (i.e., whether an individual is a secret deviant or conformist). While the experience of being formally labe led is paramount to the transition to secondary deviance, some individuals may accept and others may reject their newly assigned statuses (Becker, 1963; Davis, 1961; Turner, 1972; see also Paternoster & Iovanni, 1989). Becker (1963) refers to indi viduals who are demarcated as deviant and view themselves as different from convent ional society as out siders. Importantly, outsiders are individuals whose deviant ident ity becomes their master status meaning that its effect is all-enc ompassing and overbearing (Becke r, 1963). Outsiders reshape themselves in light of their new trenchant identity and simultaneously have their social experiences and opportunities remolded ar ound them. More specifically, the acceptance of a deviant label causes a break down in the social resources necessary to successfully navigate the conventional social world and pushes the individual toward deviant groups that further so lidifies the transition to a devia nt career (Becker, 1963). It is the acquisition of deviant peer groups, following both the experience and acceptance of a deviant label that mar ks the final step in the develop ment of a career criminal (Becker, 1963; see also Paternoster & Io vanni, 1989). While the works of Becker (1963) and Lemert (1951; 1967) are among the most important contributions to the labeling perspective, their efforts fall short of fully explicating a testable theory of labeling. The causal pathways to secondary deviance became formal hypotheses some twenty years after these import ant contributions were first introduced to the field. 62

PAGE 63

The Causal Pathways to Secondary Deviance Following a period of intense skepticism about labeling theory t hat centered on its definitional weaknesses (Gibbs 1966; 1972), its perceived strong ties to conflict theory (Wellford & Tripplett, 1993), and its key theoret ical problems (Akers, 1968), new life was given to the theory with the publication of tw o seminal articles in 1989 (see Paternoster & Iovanni, 1989; Link, Cullen, Struening, Shrout, & Dohrenwend, 1989). Together, these two studies helped to clarify the specif ic causal pathways that lead to secondary deviance and to discharge those criticisms le vied against labeling theory which were unsubstantiated. Paternoster and Iovanni (1989) maintained that the prototypical causal pathway to secondary deviance begin s with a public labeling event which leads to, in order, the exclusion from conventional activities, the al teration of self-identity, the acquisition of supportive deviant others and, fi nally, secondary deviance. If a label is hidden from public view, however, they claim that it is unlikely to set in motion the sequence of events that lead to secondary devianc e. Just as if an individual was not labeled at all, a private label will typica lly have no influence on the individual and the result of the labeling event will be uninterrupted access to conventional activities as well as maintenance of ones identity and supporti ve conventional others. In general, labeling theory suggests that much depends upon whether a label is made public or kept private. The roots of labeling theory in symbolic in teractionism remind us that there may be more to the labeling proce ss than the objective labeling event. One thing that is resoundingly clear from prior empirical work on labeling theory is that the specific causal pathways thought to occur when labels are either public or private do not always occur (Paternoster & Iovanni, 1989). In other words, an individual can disregard or, ultimately, 63

PAGE 64

accept a deviant label in cases that each of these main pathways does not anticipate. In instances where a label is made public, tw o possibilities arise in which an individual may not become involved in secondary deviance. First, if societal reactions to a public labeling event are such that they are ultimately inclusive, this can result in the reintegration of the individual back into society (Braithwaite, 1989). Reintegrative responses seek to prevent the labeled i ndividual from being excluded from normal routines and conventional activities, which can thwart secondary deviance (Paternoster & Iovanni, 1989). Unfortunatel y, the United States genera lly does not engage in the reintegrative shaming of violent and other criminal offenders (Braithwaite, 1989). Second, if an individual is excluded from nor mal routines and conventional activities, the individual can potentially reject their ascribed status in a process known as deviance disavowal (Davis, 1961). Despite being excl uded from society, the individual maintains their conventional self-identity and, therefore, does not progress to secondary deviance. Just as a public label may not always l ead to the problem of secondary deviance, a private label may not always prevent it. If a label is kept private, an individual may nevertheless reorient his or her identity in accordance with it in a process known as deviance avowal (Turner, 1972). Paternoster and Iovanni (1989) argue that even if a private label helps an individual escape social exclusion from conventional life, the individual alters his or her identity anyway th rough a process akin to self-labeling (see Thoits, 1985). The private labeling event influences the persons identity directly, guiding the individual toward deviant peer s and secondary deviance. Thus, the deviance amplification hypothesis is a probabili stic statement of causation which holds that a labeling occurrence may block one from conventional activities and normal 64

PAGE 65

routines, may alter ones of self-identity, and may lead to the acquisi tion of supportive deviant others, which could ultimately resu lt in secondary deviance (Paternoster & Iovanni, 1989). While additional empirical work on the labeling process is needed, some evidence suggests that labels in fluence subsequent behavior and that deviant peers and blocked access to conventional opportunities are especially important intervening mechanisms (Ber nburg, 2002; Bernburg & Krohn, 2003; Bernburg, Krohn, & Rivera, 2006). During the peak of labeling theory in the 1960s, the scholarly id ea that mental patient labels were directly responsible for underlying chronic mental illnesses was advanced (Scheff, 1966). This bold assertion on the etiology of mental illness was argued to be inaccurate since many indivi duals suffered severe mental illnesses irrespective of any secondary deviance e ffects (Gove, 1980). Individuals undergo systematic behavioral and/or mental health testing prior to being labeled suggesting that serious mental illness is a preexisting condition (Gove, 1980). Link and colleagues (1989; see also Link, 1982; 1987; Link et al., 1987) saw merit in both claims about the contributions of labels to behavioral problems and critiques regarding the direct effects of labels. Noting that most prior research focused on the direct effects of labels or on the indirect effects through self-identity only, Link (1982) first drew attention to other potential intervening mechanisms such as educ ation and employment. To the extent that labels influence the ability to secu re employment, earn sufficient income, and achieve educational goals, additional stre ssors may be placed on the individual resulting in increased mental health problems (Link, 1982). Link (1982) found that being labeled a mental patient had adverse affect s on work status and income, controlling for 65

PAGE 66

psychiatric problems among other variables. Five years later, Link (1987) discovered that, in addition to problematic income and work outcomes, labels led to selfdevaluation. Drawing on his prior research on mental patients, Link along with his colleagues (1989) developed a modified label ing theory of mental pati ents, which has important relevance for investigations into the infl uence of sanctions on subsequent criminal behavior (Krohn, Lopes, & Ward, in press). Their theory sugges ts that individuals first consider what it means to be a mental patient and think through the anticipated discrimination and devaluation t hat accompanies such a label. If an individual is labeled, then these societal conceptions become directly relevant to the mental patient but if the individual is not labeled then these impressions are immaterial. Their model suggests that individuals who ex perience an official label will respond to it by keeping it a secret, withdrawing from social life, or educ ating members of societ y. These tactics, especially withdrawing fr om social interactions, can lead to adverse consequences for ones social network ties, earning power, and self-esteem. Regardless of ones coping response to being labeled, this experience can also directly result in problematic self-esteem, occupation and social network outcomes. That is, an individuals perc eption of the discrimination and devaluation experienced by mental patients leads to an expectation of reje ction which is responsible for the social and occupational difficulties (Link, 1987; Li nk et al., 1989). Importantly, selfdevaluation, loss of earning power and problem s with prosocial network ties are thought to ultimately exacerbate the underlying m ental illness (Link et al., 1989). Thus, secondary deviance arises indirectly and self-esteem, earning power, and social 66

PAGE 67

network ties are the important mediating variables which tr ansmit the effects of deviant labels, irrespective of whether coping strat egies such as withdrawal are invoked in the causal chain (Link et al., 1989; see also Paternoster & Iovanni, 1989). In sum, Paternoster and Iovannis (1989) t heoretical clarifications of the labeling process for criminals and Link and colleague s (1989) discussions of the effects of mental patient labels suggest that official labels ultimately lead to secondary deviance and point to similar intervening variables as the ones that are integral to the labeling process. However, there is one major point of contention; Link and colleagues (1989) do not specify a specific causal chain for the mediating variables and, instead, suggest each one of the intervening variables may be rele vant in its own right. While empirical research is needed to test these slightly diffe rent assertions, the deviance amplification hypothesis is clear in its imp lication of access to conventional activities, sense of selfimage, and social network ties as the causal factors that lead to secondary deviance. Nearly one-hundred years of refinement has resulted in testable hypotheses and a theory of criminal behavior that, not only stands apart from labeling as a theory of law making and law enforcing, but is gaining traction among mainstream scholars as a serious explanation for criminal behavior. This is due in large part to its ability to neatly fold into existing theories of crime and dev ianceespecially thos e theories aimed at explaining crime over the lifecourse (Sampson & Laub, 1997). 67

PAGE 68

CHAPTER 3 THE DEVELOPMENTAL AND LI FE-COURSE PERSPECTIVE Chapter 3 provides an overview of t he origins and major principles of the developmental and life-course perspective. First, the emergence of the developmental and life-course criminological perspective is traced back to the intriguing relationship between age and crime. Next, t he formal birth of life-course criminology in the early 1990s is explored focusing on the timely arrival of both state dependence and dualtaxonomy approaches to explaining the meaning of the aggregate age-crime curve. While both theoretical explanations of offending focus on stability and change, they do so in decidedly different ways which has import ant implications for the current research. The major building blocks of long-views of crimetrajectories, transitions and turning pointsare then presented. Finally, the important and highly controversial topic of whether distinct criminal trajectory groups exist is discussed. In short, Chapter 3 provides the necessary background of the major ideas and theories of the developmental and life-course perspective fo r the consideration of the effects of sanctions over time across different offending subpopulation sthe focus of Chapter 4. The Age-Crime Curve One of the closest things in the field of criminology to an empirical fact is the existence of the aggregate agecrime curve, which appears as a bell-shaped distribution that is positively skewed (Hirschi & Gottfredson, 1983). Crime incr eases swiftly during the early teen years, reaches a peak dur ing late adolescence, undergoes a sharp decline in early adulthood and then continue s to gradually fall off through adulthood (Hirschi & Gottfredson, 1983). Gi ven these characteristics, it comes as no surprise that most theories of crime, and t he empirical efforts to test them, have focused almost 68

PAGE 69

exclusively on the offending behavior of adolescents and young adults (Sampson & Laub, 1993). It is these individuals who commit the bulk of property and violent offenses, therefore providing the largest c oncerns for criminologists and policy makers alike. Nevertheless, focusing on this gr oup neglects the significance of childhood and adulthood as periods relevant to explaining criminal be havior (Sampson & Laub, 1993). The relationship between age and crime has been argued to transcend crime types, cultures, places and times (Hirsch i & Gottfredson, 1983). In a noteworthy assessment of long-term historical trends in vi olent crime, Eisner (2003) finds there is strong similarity in the age patterns of viol ent crime across six different European areas spanning four centuries. Still, not all empirical evidence points to the position that the age-crime curve is invariable. For instanc e, Monnkonen (2001) ex amined violent crime patterns over two centuries and concluded that the especially large number of offenders in late adolescence/early adulthood appear s to be a relatively new phenomenon. Moreover, other research has found that the age-crime curve appears to differ for divergent crime types. For example, viol ent criminal offending peaks later than nonviolent crime and decays less rapidly (Blumste in & Cohen, 1987; Blumstein, Cohen, & Farrington, 1988; Farrington, 1986). This research has cast some doubt on the uniformity of the age-crime curve throughout history and the extent to which age is similarly related to participation in various fo rms of crime. Nevertheless, the relationship between age and crime in the a ggregate remains one of the most consistent empirical findings in criminology. Most early attention given to the robus t relationship between age and crime turned out to be more practical and descriptive in nature than theoretical (see, however, 69

PAGE 70

Greenberg, 1977; Farrington, 1986). For inst ance, two National Academy of Science panels in the late 1970s and 1980s focused on the practical issues of deterrence and incapacitation (Blumstein, Cohen, & Nagin, 1978) and on describing and defining the importance of criminal car eers (Blumstein, Cohen, Roth, & Visher, 1986). In speaking on the use of the career concept in crim inology, Gottfredson and Hirschi (1988, p. 38) state that Blumstein et al. do not bring to the career concept meanings derived from a conception of crime or a theor y of criminality. Rather, they simply announce a decision to apply it to crime. Ten years follo wing Hirschi and Gottfredsons (1983) seminal article, the formal birth of lif e-course criminology occurred as the most notab le attempts to make theoretical sense of the aggr egate age-crime curve and explain criminal behavior at various stages of human devel opment were advanced (see Moffitt, 1993; Sampson & Laub, 1993).8 It is these initial efforts, which remain leading explanations nearly twenty years later, which paved the way for a number of additional developmentally sensitive explanations of criminal behavior (see Lahey & Waldman, 2005; Catalano & Hawkins, 2005; Thor nberry & Krohn, 2005; Farrington, 2005; LeBlanc, 2005; Wikstrom, 2005). The Birth of Developmental a nd Life-course Criminology It has been found that adult antisocial behavior virtually requires childhood antisocial behavior (Robins, 1978, p. 611) but it has also been found that most antisocial children do not becom e antisocial as adults (Gov e, 1985, p. 123). These 8 Moffitts (1993) and Sampson and Laubs (1993) theories were not the first to highlight the developmental nature of criminal offending. For exam ple, Thornberry (1987) proposed an interactional theory of crime and delinquency, which suggested that learning and control factors affect crime and delinquency all in an interacting fashion over the lifecourse. While this theory is important and has been applied to explain stability and change in offending over time (see Thornberry & Krohn, 2005), the formal arrival of the life-course criminology paradigm is probably best traced back to 1993. 70

PAGE 71

classical statements highli ght the two fundamental iss ues the developmental and lifecourse perspective attempts to addre ssstability and change in behavior. There is remarkable stability in offending over time; i ndeed, the best predictor of future behavior remains past behavior (White et al., 1990). Ho wever, stability in crime and violence is not perfect and is more nuanced than at first glance. More specifically, stability refers to between-individual continuity in behavior as opposed to absolute stability in behavior (Farrington, 1989; 1992; Tracy & Kempf-Leonard, 1996). In other words, there is considerable consistency in the rank ordering of individuals level of criminal involvement but behavioral changes within-individuals are real and likely possibilities (Sampson & Laub, 1992). Taking decidedly different approaches in t heir attempts to explain the age-crime curve and to address issues of stability and change in criminal behavior over time, the publication of two original works (Moffi tt, 1993; Sampson & Laub, 1993; see also Sampson & Laub, 1992) thrust the life-course perspective into the criminological spotlight in the early 1990s.9 On the one hand, Moffitt (1993) maintained that two 9 There are three major types of explanation for the aggregate age-crime curve, which include population heterogeneity, state dependence and taxonomy approaches (Ezell & Cohen, 2005). While there are several specific theories that fall within these thr ee categories, Gottfredson and Hirschis (1990) selfcontrol theory, Sampson and Laubs (1993; 2003) age-graded social bonding theory, and Moffitts (1993) dual-taxonomy theory are the most notable and probably also the most representative of the types of accounts. The latter two theories represent life-co urse approaches to the study of crime and are discussed in this chapter at-length. While not a lif e-course theory, Gottfredson and Hirschis (1990) selfcontrol theory warrants mention as a competing expl anation, or more precisely, an alternative idea. Gottfredson and Hirschi (1990) draw an important distinction between criminality and crime and propose that criminality (i.e., ones propensity to engage in cr ime) remains relatively stable throughout the lifecourse whereas participation in crime changes over time in accordance with the aggregate age-crime curve. Gottfredson and Hirs chi (1990) claim that the relationship between age and crime cannot be explained by current criminological theories and, inst ead, suggest that age has a direct effect on crime (see also Hirschi & Gottfredson, 1983). However, explanations cent ered on maturation are not really explanations at all as they simply claim that things just unfold naturally over time (Dannefer, 1984). Thus, the population heterogeneity account highlights potenti ally important and persistent differences between individuals who may help to explain stability in crime but fails to explicitly address the meaning of the aggregate age-crime curve and change in offending behavior. 71

PAGE 72

fundamentally different types of individ ualslife-course-persistent and adolescencelimited offenderswere characterized by stabili ty and change in offending, respectively. On the other hand, Sampson and Laub (1993) argued that all individuals experience both stability and change in behavior over time. Opposing the idea of different types of offenders, they have maintained over the years that the field of lif e-course criminology would benefit from a more pr ocess-oriented, generalized a ccount of within-individual behavioral stability and change over time (Laub, Sampson, & Sweeten, 2006, p. 329). The concurrent arrival of these developmental theories with different emphases helped to create a synergistic effect that propelled life-course criminology to the forefront of criminological thought; despite some noteworthy theoretical advances in the field since then (e.g., see Akers, 1998), the perspective has remained the prevailing organizing principle of modern criminology. It is worth noting that some claim the attention given to developmental and life-course explanations is largely unfounded. For instance, it has been argued that developmental and life-course theories merely borrow concepts from other criminologi cal theories failing to offer any truly unique explanations of criminal offending (Akers & Sellers 2009, pp. 319-320). While life-course criminological theories do use some of the same variables as existing theories of crime, they apply them in different ways resulting in novel explanations of offending over the life-course. For instance, some theories suggest that the causes of crime and delinquency are fundamentally different for cert ain types of individuals. Additionally, some theories anticipate that the causes of crime for the same individual may vary over time. The idea of studying crime and viol ence over different developmental stages across the life-course has firmly taken hold, serving to both invigorate numerous 72

PAGE 73

substantive areas of criminol ogical research as well as spark heated theoretical debates on the meaning of criminal involv ement over time. Dual Taxonomy Theory Foreshadowing taxonomic theoretical effo rts, Hirschi and Gottfredson (1983) write: Taken at face value, the career criminal notion suggests division of the criminal population into discrete categorie s: those who offend occasiona lly or sporadically for a typically brief period of time and those who offend regularly over an extended period of time. If the division proves valid, two gener al age-crime distributions could be extracted from the data (p. 574). Moffitt (1993) was the first to run wit h this basic idea, claiming that there are two f undamentally distinct types of cr iminalslife-course-persistent and adolescence-limited offenders. Life-course-persistent offenders constitute only approximately ten percent of all offenders but exhibit the most problematic behav ior. These individuals are troubled with neuropsychological deficits stemming from a host of sources that include, but are certainly not limited to, maternal drug abuse, child abuse/neglect and exposure to toxic agents in the womb. Moffitt (1993) maintains there is an important overlap between individuals who have neuropsych ological deficits and those who have high exposure to criminogenic environments. It follows that children who are most at risk for neuropsychological deficits are often also si tuated within environment s that make them least likely to overcome these problems. T hey tend to have parents that lack resources and cognitive abilities to cope with their diffi culties. Moreover, these individuals may reside in dysfunctional homes that also lay in neighborhoods with in sufficient social and community resources. Given the experience of both biological and sociological setbacks, evocative interactions are seen as an important contribut ing factor in the 73

PAGE 74

development of life-course-persi stent offending. An evocativ e interaction occurs when a childs initial problematic behaviors, which may stem from neuropsychological deficits, elicit destructive responses from parents or other individuals who serve to further draw out the childs problematic behavior. In short, possessing neuropsychological deficits and being exposed to problematic environment s often co-occur; the compounding effect of these two factors during the formative years provides the recipe for the emergence of the life-course-persistent offender. Once the path of a life-cour se-persistent criminal trajectory has been charted, these offenders maintain their status thr ough both reactive and proactive interactions (Moffitt, 1993). The former refers to differences in the way these individuals experience and interpret interactions with others whereas the latter refers to self-selection into environments that support ones criminal lifestyle Their responses to social situations combined with their choices for social interaction keep these individuals locked into their chronic offending trajectory. Evocative, reactive and proactive interactions all lead to adverse ramifications for the individual across the life-course. Firs t, life-course-persistent offenders experience contemporaneous consequences by carrying with them into specific situations the very traits that have caused trouble in the past. Second, these individuals experience cumulative consequences by having prior diff erences and problems snowball over time. Importantly, both contempor aneous and cumulative consequences are thought to narrow options for change. Specifically, Mo ffitt (1993) suggests that opportunities for change will often be actively transformed by lif e-course-persistents into opportunities for continuity. Hence, life-course-persi stent offending is an ongoing developmental 74

PAGE 75

process in which offenders actively participate, even if unwittingly, in the maintenance of their criminal identity. A second group of offenders, who constitute the overwhelming majority of all individuals, are known as adolescence-limit ed offenders (Moffitt, 1993). The crime committed by this group has a distinct etiol ogy and their offending is characterized as being a normal part of human devel opment. Indeed, a small, thir d group of individuals, who abstain from crime altogether, are the men and women who are thought to be abnormal (Moffitt, 1993). The adolescence-limited group neither engages in crime during childhood nor offends during adulthood. By definition, this group fills out the bell shape characteristic of the age-crime curve (t heir life-course-persist ent counterparts are responsible for the long tail that gradually decreases thr ough late adulthood). During the teenage years, adolescence-limited offenders mimic the behaviors of life-coursepersistent offenders because they experience a maturity gap between their biological and social ages. Youths decisions to commit minor crimes and status offenses as well as use drugs are reinforced by the fact t hat these behaviors prov ide adolescents with a way to assert their independence and demonstrat e their readiness for adult activities. Unlike life-course-persistent offenders, adolescence-limited o ffenders are able to desist from crime because they are responsive to changing reinforcement contingencies. While offending during the late teenage years serves a function during this time period (i.e., the assertion of i ndependence), they abandon it fo r conventional pursuits once they reach adult status when the reward s and benefits associ ated with fulfilling conventional adult roles become apparent. 75

PAGE 76

In sum, Moffitts (1993) theory as originally formulated predicts two qualitatively distinct types of offenders whose distribut ions, when superimposed form the aggregate age-crime curve. Her theory explains both t he large number of offenses committed by adolescents and the sudden drop and apparent gr adual decline in offending throughout adulthood, which are characte ristic of the relationship between age and crime in the aggregate. The dual taxonomy suggests that adolescence-limited i ndividuals undergo change in criminal behavior over time w hereas life-course-persistent individuals experience stability in offending across the lifecourse. As will be shown, categorization of offenders yields curious hypotheses about the influence of sanctions for different offending subpopulations. Age-Graded Social Bonding Theory Noting that prior research has suggest ed the existence of both continuity and discontinuity in offending, Sampson and Lau b (1993) develop a theory that seeks to explain both of these empiri cal facts. Unlike Moffitt s (1993) theory which is characterized by stability in one gr oup and change in another, Sampson and Laubs (1993; 2003; see also Laub & Sampson, 2003) age-graded social bonding theory stresses stability and change within every indi vidual. Borrowing ideas from Hirschis (1969) social-control theory and Shaw and McKays (1942) social disorganization theory, they maintain that structural contex t (e.g., low family SES, residential mobility, mothers employment) influences crime and delinquency through its effects on informal, family and school social controls. Given t here is considerable stability in structural context over time (Shaw & McKay, 1942) and t hus constancy in informal social controls (see Bursik, 1988; Sampson & Groves, 1989), t hey argue that there is continuity in behavior throughout the life-course (Sampson & Laub, 1993; 2003). While anti-social 76

PAGE 77

behavior in adolescence tends to carry over into adulthood, acquiring prosocial social bonds in adulthood results in the te rmination of criminal offending. Sampson and Laub (1993; 2003) emphasize the quality and char acter of the social bonds, principally in the family and employ ment arenas, as opposed to the specific timing of the social bonds.10 Hence, it is more important to marry someone that has the qualities of a good spouse at any time in adulthood than it is to marry someone without these qualities at a particular age. They maintain that the pathways to crime and conformity are mediated by social bonds to key institutions of social control (Sampson & Laub, 1993, emphasis in original ). Then, their age-graded social bonding theory acknowledges that change in behavior is a tw o-way street; just as the acquisition of strong social bonds to conventional society during adulthood is hypothesized to be the driving force behind criminal desistence, the erosi on of ties to these social institutions is thought to increase the likelihood of partici pation in criminal behavior. One unique aspect of Sampson and Laubs (1993; 2003) theor y is this ability to not only explain criminal desistence but also explain the onset of off ending in adulthood. In 2003, Sampson and Laub extended their age-graded social control theory to include notions of criminal opportunity and incorporate indi vidual choice as important factors in influencing crime (Laub & Sampson, 2003). They now note that social control, routine activities and human agency, directly and inte ractively, influence offending across the life-course (Laub & Sampson, 2003). This inclusion of human agency is particularly noteworthy and relevant for the current study that seeks to place sanction effects within a developmental and life-course perspective. 10 Some developmental theories, however, place heavy emphasis on timing (see Thornberry, 1997). 77

PAGE 78

Sampson and Laubs (1993; 2003) theoretic al approach to explaining the agecrime curve does not see it as the merging of two distinct curves (cf. Moffitt, 1993); rather, the aggregate agecrime curve is viewed as a conglomeration of many unique, individual criminal trajectories over time. All offenders terminate their criminal activity at one time or another (Sampson & Laub, 2005a; 2005b). In this view, the tail of the aggregate age-crime curve is a reflection of di fferences in the timing of essentially inevitable commitments to conventional ways in adulthood. Thus, most individuals are strongly bonded to social institutions in ear ly adulthood and, by late adulthood, almost everyone has claimed a stake in conformity. Central to Moffitts (1993) dual taxonomy and Sampson and Laubs (1993; 2003) age-graded social control theories as well as other developmental and life-course theories are the idea s of trajectories, transitions and turning points. In preparation for placing sanction effects in a developmental perspective, it is these im portant building bloc ks of the perspective which are now discussed. Trajectories, Transitions, and Turning Points A trajectory is a pathway or line of devel opment over the life span such as work life, marriage, parenthood, self-esteem, and criminal behavior (Sampson & Laub, 1992, p. 64). In a full life-course view, a complete offending tr ajectory would include ones criminal involvement over childhood, adol escence, early and even late adulthood. Thornberry (1997) maintains that trajectories have several important dimensions, two of which include entrance and success. Entrance re fers to the fact that some individuals embark on a trajectory whereas other individuals do not. With respect to criminal behavior, a small minority of individuals refrain from engaging in criminal behavior at any age (Moffitt, 1993). These individuals do not offend and therefor e do not enter into 78

PAGE 79

a criminal trajectory. Everyone else, however enters into a criminal trajectory which has some shape or another over the life-course (Sampson & Laub, 2005a). While entrance distinguishes who participates in a given trajectory, the dimension of success acknowledges there are important differences in the quality and character of a trajectory for different peopl e. For example, most indi viduals enter into a work trajectory but only some have highly fulfill ing careers across the life-course. With respect to crime, most indivi duals will enter into a criminal trajectory at some point and there will be varying degrees of success or involvement in crime across the life-course (Thornberry, 1997). Recalling the idea of between-individual st ability, this differential success between individuals will be largely preserved, though not perfectly, across the life-course. Irrespective of this between-individual continuity in behavior, however, it is expected that there will be changes in success experiences within-indi viduals over time. How these changes are thought to occur is now explored. The trajectory concept is often taken to indicate ones behavioral fate is essentially sealed (Sampson & Laub, 2005a). However, an individuals cr iminal trajectory is not etched in stone and can potentiall y be altered in meaningful ways. Termed transitions, important life events that occur over the short-term such as marriage and justice system contact are embedded within tr ajectories (Elder, 1985). In some cases, these transitions can serve as turning points in ones trajectory (Elder, 1985). Several different life transitions have been considered as capable of potentially turning a criminal trajectory for better or worse; t hese include marriage (Laub, Nagin, & Sampson, 1998; Sampson & Laub, 1993; 2003; Laub & Sampson, 2003; Sampson, Laub, & Weimer, 2006), employment/work (Uggen, 2000; Uggen & Staff, 2001; Sampson & 79

PAGE 80

Laub, 1993; 2003; Laub & Sampson, 2003; cf. Wright & Cullen, 2004) military involvement (Elder, 1986; Sampson & Laub, 1993; 1996; Laub & Sampson, 2003) and, importantly, justice system contact (Moffi tt, 1993; 2006; Sampson & Laub, 1997). Aside from the type of specif ic transition that is experienced (e.g., justice system contact, marriage), there is some debate as to what makes a life transition a turning point in a criminal trajectory. Thornberry (1997) discusses the importance of timing, which refers to the idea t hat transitions which occur at specific times in the developmental cycle are likely to alter ones traj ectory. For instance the experience of off-time or precocious transitions to adulthood, which are increased by gang membership in adolescence, can have lastin g adverse effects on criminal behavior in adulthood (Thornberry, Krohn, Lizotte, Smith, & Tobin, 2003; Krohn, Ward, Thornberry, Lizotte, & Chu, 2011). The timing dimension of a trajectory emphasizes the fact that the exact same event such as employment ma y have different outcomes on ones trajectory at different points in time. Recall that so me place stronger em phasis on the quality or character of the event (e.g ., see Sampson & Laub, 1993).11 Hence, if one were formally arrested, as opposed to merely being questioned by the police, the event would hold more salience. As a consequence, the ex perienced transition might be more likely to serve as a turning point in ones offending trajectory. Then, both the timing and the quality of the transition are im portant factors which may infl uence the likelihood of a life transition becoming a turning point. 11 Interestingly, though, their work sometimes neglects this argument. For instance, they explore the causal effect of marriage in generalnot wedlock to a prosocial or resourceful spouse (e.g., see Sampson, Laub, & Wimer, 2006). 80

PAGE 81

There is also differing opinions as to what constitutes evidence of a turning point. In one view, turning points have been conceptualized as being abrupt shifts in ones trajectory (Elder et al., 1991). That is, the experience of a key life event causes a sharp turn in ones behavior, knifing off the past and providing the individual with a fresh outlook. In another view, turni ng points are more akin to in itiation points that set in motion a gradual change in behavior over ti me (Pickles & Rutters, 1991; Sampson & Laub, 1997). In this case, evidence of behavioral change may not be immediately apparent but would reach detect able levels as time passes. In either case, the individuals likely course has, for better or wo rse, been altered. A turning point, of some variety, has occurred if a detectable change in behavior can be traced back to the experience of a specific transition and non-spuriousness can be established. As a final point of consideration, the c oncept of a trajectory need not refer to the pathway over the entire life-course. There are many instances of scholars usefully employing the idea of trajecto ries to study some behavior ov er time, often just a few years or more (Ward, Stogner, Gibson, & Akers, 2010), over a single developmental period (Broidy et al., 2003; Tremblay et al ., 2004; Shaw, Gilliom, Ingoldsby, & Nagin, 2003; Shaw, Lacourse, & Nagin, 2005), or over just a couple of developmental periods such as childhood and adolescence (Nagin et al ., 2003; Broidy et al., 2003). This is not entirely surprising given the lack of data sets that are readily av ailable with measures across the entire life-course. Typically in these cases when the trajectory concept is employed, the objective of looking at behavio r over the long-run remains the same (see, however, Horney, Osgood, & Mars hall, 1995). Also similar, then, is the causal goal of discovering whether certain life transitions serve as turning points in a behavioral 81

PAGE 82

trajectory and whether these transitions may serve as turning points for one type of trajectory but not for another as some criminological theories would claim (e.g., Moffitt, 1993). This latter possibility necessitates a discussion on the existence of distinct criminal trajectory grou ps. Distinct Criminal Trajectory Groups? The possibility that qualitatively distinct groups of offenders exist is among the most disputed issues in all of developmental and life-course criminology (Farrington, 2006). There are several theor ies that maintain there are one or more distinct subpopulations of offenders which are respons ible for the shape of the age-crime curve (Moffitt, 1993; Thornberry & Krohn, 2005; LeBlanc, 2005; Lahey & Waldman; 2005). The most notable of them is Moffitts ( 1993) dual-taxonomy theor y, which explicitly predicts two qualitatively distinct types of criminals, namely adolescence-limited and lifecourse-persistent offenders. Her hypot heses have sparked numerous empirical investigations aimed at uncov ering the validity of the dual taxonomy and the existence of distinct groups of offenders in general. These efforts have employ ed the latent class growth analysis methodology (LCGA), which is also known as semi-parametric groupbased modeling (SPGM) (Nagin, 2005). This technique, which will be discussed in considerable detail in Chapt er 7, empirically determines the number of trajectory groups, their shapes, and the proportion of i ndividuals belonging to each group (Nagin, 2005). In the first application of the latent class growth analysis technique, Nagin and Land (1993) used conviction records from a cohort of over 400 males (see West & Farrington, 1973; 1977) and found four trajectory groups: non-offenders, adolescencelimited offenders, and two variations of chroni c-offenders. Employing data on a sample 82

PAGE 83

of males from the National Youth Survey McDermott and Nagin (1998) found a three group model to be supported by the data. As anticipated by theory (see Moffitt, 1993), the three groups included nonoffenders, adolescence-peaked, and chronic offenders (McDermott & Nagin, 1998). Using data from multiple sources including several Racine Birth Cohorts, the Second Philadelphia Birth Cohort, and the Cambridge Study in Delinquent Development, DUnger and colleagues (1998) uncovered evidence of four to five offending groups in each analysis. While adolescence-limited offenders and chronic-offenders were consistently found ac ross studies, the proportion of individuals belonging to these groups was unanticipated by theory such that there were more chronic offenders. Additionally, they found t hat a group of low-level chronic offenders emerged from the data on a fairly consistent basis. Moffitts (1993) theory anticipates onl y a single group of chronic offenders. However, evidence using LCGA suggests that two groups may be inadequate to characterize all offenders (Moffitt, 2006). Moffitt and her colleagues (1996) found evidence for a child-limited aggressive children group, who exhibit some similarities to the life-course-persistent offender group during childhood but undergo a recovery of sorts leading to the avoidance of life-long problems, which are characteristic of the lifecourse-persistent offenders. Other schol ars have found evidence for a low level chronic group which has been found to c onsistently offend from childhood through adolescence (Fergusson et al., 2000) or from adolescence through adulthood (DUnger, Land, McCall, & Nagin, 1998; Nagin et al., 1995) These individuals persist in offending as do life-course-persistent of fenders but do so at consider ably lower offending rates. 83

PAGE 84

There is also some theoretical discussion of the possibility of an adult-onset group which begins its offending in adulthood (Patterson, 1997). Uncovering the number and character of traj ectory groups has been a very difficult empirical endeavor. The trajectory groups that are uncovered may depend on the type and quality of the data being em ployed (Moffitt, 2006). Eggleston, Laub and Sampson (2004) found that individuals may be classified in different trajectory groups when using longer periods of data, raising questions about group classifications. Some contend that LCGA is not suited to te st whether there are a specific number of trajectory groups (Skardhamar, 2010). While simula tion studies show that LCGA can identify real groups and correctly classify individuals into thes e groups (Brame, Nagin, & Wasserman, 2006; Nagin, 2005), this accuracy is dependent upon the existence of true subpopulations (Skardhamar, 2010). If the method cannot test for the existence of these subpopulations and their existence is not k nown a priori, then the usefulness of the technique may be lessened (s ee Skardhamar, 2010). In focusing on the time periods of child hood and adolescence, Piquero (2008, p. 43) states that it is impressive that across a ll the various studies in different parts of the world using different methodologies to meas ure criminal activity over a similar age range, there have been a consist ent number of trajectories i dentified in these studies. Typically, three to four trajectories are identified, namely low, medium and high groups. While LCGA will likely find a small number of groups whether they truly exist or not (Skardhamer, 2010), it is indeed remarkable that different studies find important similarities in offending behav ior over time when these groups empirically emerge from 84

PAGE 85

85 different samples. But, t he jury is still out on the dual taxonomy as well as whether other qualitatively distinct subpopulat ions of offenders indeed exist. In sum, consideration of life-course theories and notions has flooded scholarly thinking in many disciplines, influencing an assortment of substantive research topics across the behavioral and social sciences (Featherman, 1983). Criminology has undoubtedly been strongly influenced by the popular trend to ta ke long-views of social phenomenon. What is clear is that the idea of distinct criminal trajectories and the LCGA methodology provide exciting hypot heses and ways to assess whether the experience of a certain life transition, such as justice system contact, serves as a turning point in a criminal traj ectory. If hypotheses regardi ng risk factors that distinguish group membership and those related to different ial effects across trajectory groups pan out as expected, evidence for th e existence of distinct crimi nal trajectory groups may be bolstered. For now, the consideration of tr ajectories and trajecto ry groups may provide important insights into how the effects of official intervention on subsequent offending behavior operate. This may impart some under standing for the mixed bag of empirical evidence on the effects of sanctionssee Chapt er 5as well as provide policy-relevant findings for the control of crime and violence.

PAGE 86

CHAPTER 4 SANCTION EFFECTS IN A DEVELOPMENTAL AND LIFE-COURSE PERSPECTIVE With a fundamental understanding of t he contradictory pr edictions of the deterrence and labeling perspectives as well as an appreciation of the theories and issues raised by the life-course criminologi cal paradigm, Chapter 4 places theories of sanction effects within the developmental and life-course perspective. First, a discussion of whether official intervention is a life transition substantial enough to cause a turning point in a criminal trajectory in general is offered. Focusing on taxonomic ideas and theories, the possibility that offici al intervention may be just an experienced life transition for some but ma y be more than that for othe rsmarking a critical turning point in their offending trajectoriesis ill uminated. Next, the possibility that the immediate effects of official intervention may differ from the l onger term effects is thoroughly explored. Finally, potential differences in the role that specific types of official intervention play in bringing about dete rrent or labeling effects are discussed. Official Intervention: Life Transition or Turning Point? The interlocking nature of trajectories and life transitions such as justice system contact can, in some cases, create turning points in ones criminal trajectory (Elder, 1985). Theories limited to time-stable fact ors are thus incapa ble of unpacking the zigzagging and temporally variable patterns of offending (Samps on, 2000, p. 712). From the standpoint of both deterrence and labeling theories, the experience of official intervention is hypothesized to be a life transi tion worthy of turning point status in a criminal trajectory. With re spect to altering ones trajecto ry toward unconformity, state sanctions are viewed as an important institut ion of social control which is linked to cumulative disadvantage (Sampson & Laub, 1997). [Labeling] theory specifically 86

PAGE 87

suggests a snowball effectthat adol escent delinquency and its negative consequences (e.g., arrest, offi cial labeling, incarceration) increasingly mortgage ones future, especially later life chances mo lded by schooling and employment (Sampson & Laub, 1997, p. 15). The application of a deviant label has been likened to a status degradation ceremony in which an individuals total ident ity is transformed and his or her social rank is sent plummeting (Garfinkel, 1956). The criminal label can become an individuals master status that trumps ones former i dentity and ties to conventionality (Becker, 1963). Experiencing an official label often kn ifes off conventional opportunities leaving individuals with few choices but to o ffend (Sampson & Laub, 1997). Then, the life transition of experiencing a sanction is much more than an event that simply comes and goes. Should the label stick to the individual, it seems possible it may have lasting effects on behavior across the life-course. Among the major theories of crime, labeling theory has been argued to be the only true developmental theory given its focus on the intervening mech anisms that unfold over time that cause primary deviance to develop into secondary deviance (Loeber & LeBlanc, 1990; Sampson & Laub, 1997). The in tegration of labelin g theory within the developmental and life-course perspective is not forced but rat her quite natural. Unfortunately, however, despite its obvious affinity to a life course, developmental framework, labeling theory has rarely been vi ewed from this perspective (Sampson & Laub, 1997, p. 9). This omission has resulted in the failure of many prior research efforts to consider a labeling event as a potential turning point in a crim inal trajectory. 87

PAGE 88

Deterrence has served as the backbone of the criminal justice system for hundreds of years and specific deterrence theory suggests that direct sanction experiences play an important role in curb ing ones offending behavior (Beccaria, 1963; Gibbs, 1975). However, the blending of det errence theory with the developmental and life-course perspective is not quite as s eamless. Unlike labeling theory, deterrence theory does not focus on a particular event (e .g., the stamping of one as criminal) marking the definitive mom ent which will determine whet her one will continue their problematic behavior over the long-term. Fo r instance, Stafford and Warr (1993) note that direct and indirect experiences wit h punishment and punishment avoidance, all actively play a role in influencing behavior. When an active offender is not experiencing official intervention, they are experiencing punishment avoidance by definition. This immediate source of information factors into ones decision to continue offending. It is true that individuals who do not have much experience with crime tend to rely more heavily on indirect experiences with punishment and punishment avoidance by necessity (Stafford & Warr, 1993). For most, justice system contact is no everyday event and what happens to oneself is much more important than what happens to ones peers or neighbors. When one does experience official interv ention, this personal and relatively infrequent experience may play an impor tant role in altering the likely course of a criminal trajectory. Mor eover, a meaningful deflection in a criminal trajectory toward conformity need not lead to absolute deterre nce. When coupling these ideas with the possibility of differential sancti on effects, it becomes clear t hat official intervention is an important life transition that may constitute a turning poin t, for better or worse, in criminal trajectories. 88

PAGE 89

Turning Points within Trajectory Groups Gottfredson (2005) notes that current typological research hinges on the assumption that adding age or time to aid in o ffender classification e fforts leads to better predictive ability and a greater understanding of treatment effe cts. While some disagree with this line of research (e.g., see Gottfredson & Hirschi, 1990; Gottfredson, 2005), most scholars recognize that the effects of sanctions will likely differ across divergent types of offenders. This argument has been advanced in theoretical discussions concerning labeling effects as well as those concerning deterrent effe cts. With respect to the former, Braithwaite (1989) states T he first step to produc tive theorizing about crime is to think about the c ontention that labeling offenders makes things worse. The contention is both right and wrong.12 Similarly, Paternoster and Iovanni (1989, p. 381) note that We should not expect labeling effects to be invariant across subgroups. With respect to the latter, Pogarsky (2002, p. 432) claims that punishment threats are consequential only for a subgroup of the gener al population. At the extremes are acute conformists who comply with the law for reasons related to morality and incorrigibles who are committe d to violating the law and are uninfluenced by sanctions. Finally, the rest of society is deterrable meaning they are subject to the effects of sanction experiences (Pogarsky, 2002). Count less other statements on the contingent nature of labeling and deterrent effects can be gathered from studies and theoretical discussions over the years. 12 Braithwaite (1989) is specifically referring to the di fferent outcomes of labels that are tied to differences in how officials respond to criminals. The current st udy, however, deals with differences in outcomes that are the result of offenders differential responses to criminal sanctions. 89

PAGE 90

Perhaps, Sherman (1993, p. 445) puts it best: Does punishment control crime? This question provokes fierce debates in crim inology and public policy. Yet there is ample evidence that it is the wrong question. Widely varying results across a range of sanction studies suggest a far more useful question: under what conditions does each type of criminal sanction reduce, increase, or have no effect on future crimes? Answering that question is central to the futu re of research in crime and delinquency.13 The salience of this question for criminol ogy and public policy still rings true today (Huizinga & Henry, 2008; Bhati & Piquero, 2008). While there is generally an agreement that t he effects of sanctions will likely differ across conditions and types of individuals, specific hypotheses specifying how labeling and deterrent effects are conditional upon offending subpopulations are not well developed (Chiricos et al., 2007). Compar ed to more experienced offenders, novice offenders have been argued to be more likely to experience deterrent effects (Thorsel & Klemke, 1972; Packer, 1968; Matsueda, Kreager & Huizinga, 2006), but also more likely to experience labeling effects (H orwitz & Wasserman, 1979; Paternoster & Iovanni, 1989). In one view, Paternoster and Iovanni (1989) contend that labeling effects may level off as an individual becom es committed to deviance, which indicates that novice offenders would be more likely to experience them. For individuals offending at high levels over time, the labeling process may have already run its course. In another view, experienced offenders have committed more crime before an arrest than novice offenders have, which may make novice offenders more susceptible to specific deterrence due to their greater perceived certainty of punishment following 13 Shermans (1993) theory of defiance draws on notions of legitimacy and fairness to explain why there may be differences in the effects of sancti on experiences on subsequent behavior. 90

PAGE 91

arrest (Matsueda et al., 2006). Given there are hypotheses suggesting that official intervention may serve as a turning point in either direction for certain criminal trajectories, how the effects of official intervention vary across offending subpopulations is truly an empirical question. Still, tax onomic approaches do provi de strong theoretical rationale for anticipating how the effects of sanctions will differ across divergent offending subpopulations. Among the major approaches that explic itly address stability and change in behavior over time, the dual taxonomy approac h unambiguously suggests that different types of individuals may react differently to certain life transitions (cf., state-dependence approach, Sampson & Laub, 1993; 2003). Specif ically, Moffitt (1993; 2006) contends that adolescence-limited off enders will naturally desist from criminal behavior once the rewards of conventionality in early adul thood become apparent. Individuals who are caught for their violations of the law may experience problems with this transition to conformity. That is, they become snared by the criminal justice system causing them to forgo the natural desistence process. Another group, lifecourse-persistent offenders, should be largely unaffected by punishment experiences since their proactive and reactive interactions will steer them to ward adverse consequences and persistent offending throughout the life-course. Thus, Moffitt (1993) clearly anticipates the experience of life transitions may serve as a turning point for one type of individual but will have little influence on anot her type of individual. Moffitt (1993) is not alone in her categoriz ation of offenders; half of the major developmental theories employ the idea of tr ajectory groups in one form or another (Moffitt, 1993; Thornberry & Krohn, 2005; Lahey & Waldman, 2005; LeBlanc, 2005). 91

PAGE 92

LeBlanc (1997; 2005) proposes a three grou p typology which involves persistent, transitory, and common offenders. Persistent offenders are primarily influenced by their lack of control, common offenders are largel y influenced by criminal opportunities, and transitory offenders are influenced by mixtur e of lack of control and opportunities. LeBlanc (2005) argues that individuals are unique at birth and they remain unique throughout the life span. As a consequence, t he level of control is different for each individual and they do not exper ience the same set of situationsIn addition, there are short-term (over time and plac e) within-individual differences in offending because of adaptation, the way an organism adapts thr ough the integration of new external elements or through structural change to fit with the environm ent (pp.156-157). Blending these ideas, it follows that the experience of official intervention may differ amongst individuals who remain uni que throughout the life span. Thornberry and Krohn (2005) do not explicitly hypothesize distinct trajectory groups per se, but they do suggest the causes of offending may vary across time and there are differences in those who tend to start early versus those who tend to start late. For example, later starters (ages 18 to 25) will display more continuity than earlier starters (ages 12 to 18) because they have mo re cognitive deficits. Importantly, they note that justice system sanc tioning, among such things as family bonds and school success, can influence behavior trajectories (Thornberry & Krohn, 2001; 2005). Their theory tends to focus on the consequences of antisocial behavior in leading to additional problematic behavior (i.e., labeling effects) rat her than on deterrent ef fects. Still, given their emphasis on offending trajectories and key differences between individuals who 92

PAGE 93

start offending at separate phases of the life-course, there ma y be some dissimilarities in the ways in which criminal sanctions infl uence the behavior of early and late starters. Finally, Lahey and Waldman (2005) think it is useful to categorize individuals into subpopulations but they reject the idea that there are only two groups. Rather, they suggest something akin to a continuum of of fenders and highlight four main trajectory groups. We posit that the same set of chil d characteristics and social factors influence conduct problems in children who follow all dev elopmental trajectories, but differences in developmental trajectories result from different combinations of the same set of causal influences (Lahey & Waldman, 2005, p. 27). While the author s do not explicitly address labeling or deterrent effects, these subpopulati ons may have differential responsiveness to key life transitions such as justice system contact even though the etiology of offending is simila r. Following logic similar to Moffitt (1993), it would stand to reason that chronic offenders, who have very high levels of antisocial propensity, would be highly unlikely to experienc e deterrent effects. There are some noteworthy variations in the ways in which these developmental and life-course theories cat egorize offenders. Some s uggest there are qualitatively distinct groups with different etiologies, another suggests some differences in those who start offending at different ti mes, and another suggests differ ent offending groups reflect varying degrees of the same underlying constr uct such as antisocial propensity. Putting these differences aside for the moment, th ese theories have an important commonality. They all suggest there is utility in thinking th at types of offenders may differ in one or more key respects. By acknowledging that tr ajectory groups may be indicative of some real differences between off enders, it follows that t hose following along distinct 93

PAGE 94

offending trajectories may be differentially res ponsive to life transitions such as justice system contact experiences. A number of interesting ideas and possib ilities emerge when considering sanction effects across different criminal trajectories To name a few, t he falsely accused (Becker, 1963) cannot, by definition, experi ence a deterrent effect but they can be subject to labeling effects. Incorrigibles (see Pogarsky, 2002) or those on a chronic offending trajectory such as life-coursepersistent offenders (s ee Moffitt, 1993) would be likely unresponsive to deterrent effects. And, the labeling process may also be irrelevant for these individuals or have al ready run its course. Individuals who have decreasing levels of offending over time, may have their desistance process hastened by justice system contact (i.e., the formal costs of crime may give the individual one more reason to stop offending) or reversed (i.e., the labeling process may be invoked, snaring individuals as Moffitt (1993; 2006) might anticipate). The point here is that placing sanc tion effects within a developmental and lifecourse perspective and focusing on trajectory groups reveals exciti ng possibilities. The life transition of official inte rvention may serve as a turning point (deterrent or labeling) for individuals following along one particular trajectory but may have no influence on offending behavior for individuals followi ng along another trajectory. Then, both deterrence and labeling theories may be relevant explanations of crime but for different types of individuals. Inte restingly, the consideration of sanction effects within a developmental and life-course perspective also adds the possibility that deterrent and labeling effects may both be relevant for ex plaining the behavior of the same individual, 94

PAGE 95

both in the shortand long-runs (see Sampson & Laub, 1997). It is this possibility that is now explored. Turning Points in the Shortand Long-runs Representing a fundamental departure from the predi ctions of taxonomic theorizing, Sampson and Laubs (1993; 2003) age-graded social bonding theory rejects the idea that there are fundam entally distinct groups of offenders who may respond to sanctions in different ways. While thei r theory does not categorize offenders into distinct types, it is certainly not devoid of important developmental implications. Most relevant here is that their theory leads one to consider the long-run consequences of official intervention, which may very well di ffer from the more immediate ones (Sampson & Laub, 1997). For the most part, research on labeling has consisted of cross-sectional studies or panel studies entailing modest follow-up periods within rather than across developmental phases (Sampson & Laub, 1997, p. 9, emphasis in original ) and they claim that looking only at the direct effects of official sanc tions is misleading (Sampson & Laub, 1997, p. 18). Recall, labelin g theory has been argued to be a true developmental theory because it focuses on dev iance amplification as a process that unfolds over time (Loeber & LeBlanc, 1990; Sampson & Laub, 1997). A run-in with the law may constitute a turning point in t he life of an individual (Sampson & Laub, 1997) but the labeling process does not happen over night. Labeling involves a sequence of stages including blocked access to conventional opportunities, alteration of self-identity and the acquisition of delinquent peers (Paternoster & Iovanni, 1989). Then, despite what occurs as an immediate outcome of justice system cont act such as a deterrent or 95

PAGE 96

null effect, the consequences of official intervention may lead to subsequent involvement in crime in the l ong-run. The importance of Sampson and Laubs revision to their age-graded social bonding theory is critical for the theoretical possibility that both deterrent and labeling effects may be experienced by the same individuals. Sampson and Laubs (2003) revised theory suggests that criminal behavio r is a function of ones experiences with key life events that can fost er or erode bonds to conventi onal society as well as a product of ones routines and rati onally-calculated decisions. T heir explicit incorporation of rationality suggests that individuals we igh the costs of commi tting crime and are thereby subject to the laws of deterrence. In the short-run an individual may lower their offending behavior if they experience justice system contact since their perceptions of certainty were boosted. Sampson and Laubs (1997) acknowledgement that official intervention can also have longer-term adv erse consequences leads to the possibility that people may be deterred in th e short-run but pulled toward crime in the long-run. One might wonder why official intervention is not anticipated to result in a shortterm labeling effect and a long-term deterrent effect (see Figure 1-3). Why this is certainly an empirical possibilit y, the theoretical rationale is lacking which would predict this sequencing of events. On the one hand, Nagin ( 1998) draws attention to differences in deterrent effects across the shortand long-runs; he claims that while large amounts of evidence have been amassed on short-term deterrent effects, little is known about long-term effects. Evidence from perceptions-based deterrence studies on the interconnection of formal and informal sources of social control point to a possibly substantial divergence between longand short-term effects (p. 4). In this 96

PAGE 97

case, however, Nagin (1998) is referring to the fact that a very effective policy in the short-term, which subjects a large proportion of the populat ion to the rule of law, becomes less effective in the long-term. When a substantial number of individuals are subjected to punishment, this draws att ention to the prevalence of the offending behavior. As a consequence, in dividuals disregard the law which is thought to be outof-tune with typical behavior. Get tough policies centered on deterring marijuana smoking serve as a prime historical exam ple of this process. While one should be aware that immediate deterrent effects may vanish if sanctions in the long-run become commonplace, this is unlikely for serious cr imes especially those involving violence. Regardless, this process does not suggest deterrent effects to develop following labeling or null effects. If anything, it ant icipates that any immediate deterrent effect would erode in the long-run. Types of Official Interventions Understanding how official intervention e ffects may differ in the shortand longruns and determining the extent to which sa nction effects may va ry across offending subpopulations are important steps toward more fully evaluating the efficacy of deterrence and labeling theorie s. However, examinin g contingent labeling and deterrent effects of a single official interv ention may overly simplify a complex problem. Kobrin and colleagues (1972) were among the first to introduce the idea of a sanction pattern in which there are diffe rent levels of sanctions within the various stages of the criminal justice system. This idea, which was originated in the general deterrence framework, can be applied within the specific deterrence fr amework (Huizinga & Henry, 2008). It is reasonable to expec t that official intervention effects may vary based upon the severity of the sanction (Huizinga & Henr y, 2008) and some evidence finds that this 97

PAGE 98

is indeed the case (e.g., see McAra & McVi e, 2007). Thus, a developmentally informed investigation into the effects of sanctions would be better off if it involved some consideration of the type of official in tervention being experienced by an offender. According to specific deterrence t heory, punishments mu st be applied with certainty and celerity and should have a certai n degree of severity in order to deter criminal behavior effectively; these three factors are thought to interact to produce criminal deterrence (see Stafford et al., 1986) Research, however, clearly shows that it is certainty of punishment t hat is most important in t he deterrence process (Nagin, 1998), a fact which may have important implic ations for determining whether official intervention serves as a turning point in a cr iminal trajectory. If certainty of getting caught really is the bulk of how deterrence oper ates, it could be th e case that a minor form of official intervention like a police cont act may be sufficient to influence sanction perceptions and thereby initiate criminal det errence. Recall, however, the idea of tipping points in which sanction effects must r each a certain level to trigger deterrence. From this logic, more severe types of of ficial intervention may be needed to invoke criminal deterrence. To the extent that severity of punishment is relatively unimportant, more severe sanctions including arrest and incarceration may serve to increase the likelihood of a labeling effect. The type of sanction could also hold diffe rential importance across divergent types of offending subpopulations. As an example, let us consi der offenders belonging to a chronic offending trajectory. These i ndividuals may already have had the adverse consequence of official intervention happen to them. With little chance of making things worse for this group, having a more serious o fficial intervention such as incarceration 98

PAGE 99

99 may serve to increase the like lihood that the off ender will be deterred. Now consider a novice offender, a minor form of official intervention such as a police contact may activate deterrence without risk of setting in motion the labeling process. A more serious form of official interv ention such as an incarceration may activate the deterrence process but also expose one to deviant peers, cause an alteration of ones self-identity, and limit life chances in the employment and educational realms. Thus, a stern sanction for individuals following along a very low offending trajectory may do more harm than good in the long-run. Determining how the effects of offici al intervention operate is no easy task and requires the consideration and blending of mu ltiple ideas, theories and perspectives. From this discussion, it appears that empiri cal studies should seek to examine the effects of different gradations of official in tervention (e.g., police contact, arrest) across offending subpopulations and across time. While this is an ambitious undertaking, it may be necessary for one to fully grasp the efficacy of deterrence and labeling theories as well as to yield policy-relevant information concerning the c ontrol of crime and violence with traditional law enforcement me chanisms, which will remain a staple of crime reduction efforts for t he foreseeable future.

PAGE 100

CHAPTER 5 REVIEW OF SANCTION EFFECTS LITERATURE Chapter 5 summarizes what is known empirically about the effects of sanctions on subsequent criminal offending. Studies t hat have examined the link between official interventions and later offending in aggregat ed samples are first reviewed. This literature serves to highlight differences in empirical findings across studies, which may stem from the fact that most empirical tests ignore the po ssibility that sanction effects may vary depending on characteristics of the offender. Next, recent research that has begun to explore contingent labeling and det errent effects across disaggregated samples (i.e., different types of offenders) is reviewed. Particular emphasis is given to empirical research focused on examining differential sanction effects for fundamentally different types of offenders such as novice and chronic criminals. Finally, the strengths and weaknesses of this limited body of empirica l research are fully discussed setting the stage for the current studys investigation into the effects of official intervention across divergent violent offending subpopulations. Sanction Effects in the Aggregate Research that is cast as either deterrence or labeling studies are equally relevant to the analysis of the effects of official intervention on behavior, which can usefully be categorized into studies that examine the ef fects of arrest and studies that examine the effects of sanctions that follow an arrest (Huizinga & Henry, 2008). With respect to the former, some of the most re levant research on the label ing and deterrent effects of arrest on violent behavior st ems from the Minneapolis Dome stic Violence Experiment (MDVE) (Sherman & Berk, 1984a) and the six repl ication studies known collectively as the Spouse Assault Replication Program (SARP) (see Maxwell, Garner, & Fagan, 100

PAGE 101

2002). With funding from the National Inst itute of Justice and the Police foundation, Sherman and Berk (1984a) conducted a random ized experiment to examine the effects of mandatory arrest policies on domestic vi olence recidivism. Consistent with the options police typically have in handling domesti c violence cases, police were instructed to arrest the offender, separate the two par ties, or advise the couple on how to handle its problems. Officers were given a color coded pad that in structed them to administer one of the three treatments to an eligible domestic viol ence offender at random. Delivered treatments were nearly perfect fo r arrests but only about 75% successful for the advise and separate treatment conditions. Unsuccessful delivered-designed matches typically comprised of individuals who were randomly assigned to receive advise or separate treatments but ended up rece iving the arrest tr eatment. Two data sources were used for recidivism including victim interviews and official reports. With respect to victim interview data, individual s who were arrested were less likely than those who were advised to recidi vate. With respect to offici al report data, individuals who were arrested were less likely than those who were separated to recidivate (see Sherman & Berk, 1984a). While this seminal work was used as evidence of a specific deterrent effect, it wa s not without conceptual and methodological flaws. Binder and Meeker (1988) not e that the MDVE study was more akin to a pilot study than a rigorous scientific study that was worthy of suggesting sweeping policy changes. They claimed that the Sherman and Berk (1984b) policy summary report that was released was essentially propaganda and not e there were important problems with the randomization, sample attrition, operat ionalization of recidivism, methodological technique and interpretation of weak effects. For example, comparison groups made by 101

PAGE 102

Sherman and Berk (1984a) were not a priori choices and they interpreted coefficients when the overall model was statistically in significant. The problems surrounding this initial study led to six replic ations studies with some minor variations in their research protocols. The SARP was funded to determine the genera lizability of the em pirical findings from the MDVE (see Maxwell et al., 2002). In the Charlotte replication, Hirschel and Hutchinson (1992) found no evidence for either a labeling or a deterrent effect for victim or offender data across multiple measures of recidivism, in cluding prevalence, incidence and time to failure. The resu lts from the Omaha replication revealed, for the most part, similar findings. However, a notable except ion was that individuals who were issued an arrest warrant (i.e., they we re absent at the scene) had s ubstantially lower recidivism across all measures (Dunford, Huizinga, & Elliot, 1990). Perhaps, these individuals knew they could not violate the law or dr aw attention to themselves given their outstanding warrants. Unlike the Omaha and Charlotte studies that showed by-andlarge null findings, the Dade County replic ation study (Pate & Hamilton, 1992) and the Colorado Springs replication study (Berk, Campbell, Klap, & Western, 1992) both found evidence for a specific deterrent effect with victim data but not with official report data. Intriguingly, the Milwaukee study revealed an initial deterrent effect followed by a longer term labeling effect (Sherman et al., 1991). Compared to those not arrested, those receiving a short arrest of approximately 2. 8 hours showed an initial deterrent effect for the first 30 days following the official in tervention. Howeve r, when following the recidivism of these individuals over a longe r period of time, those arrested experienced a labeling effect that persisted fo r a year (Sherman et al., 1991). 102

PAGE 103

In short, the initial MDVE revealed dete rrent effects and the SARP studies yielded a mix of findings ranging from null effects to both deterrent and labeling effects. Moreover, many efforts at synthesis vary al most as much as do the published reports from the individual sites (Ma xwell et al., 2002, p. 55) so, in many respects, it has been difficult to draw any firm conclusions about s pecific deterrent effects of mandatory arrest policies. Conducting the most sophisticated meta -analyses to date, Maxwell and colleagues (2002) took on the task of synthesis us ing data from five of the MDVE/SARP experimental sites. They ex amined recidivism using both victim and official report measures. The researchers deal with comple xities associated with multi-site analysis and also take into account the variability in the existence, number, and timing of victim interviews as well as differences in the collection methods of official data. The researchers employ 4,032 incidents which invo lve assaults by males of their female partners, comparing the arrest treatment group to a lumped, non-arrest treatment group. Official data revealed that approximately 23% of the sample had recidivated where victimization data regi stered that 43% of the sample had recidivated (Maxwell et al., 2002). Across five outcome measures and controlling for site and suspects characteristics (age, use of intoxicants, race, married, prior arre st, and employment), there were consistently smaller rates of recidivism among individuals in the arrest treatment group compared to th ose grouped together in the nonarrest group. Overall, then, the best evidence that exists from studies that have anal yzed aggregated samples of MDVE/SARP data is that ther e is a very small specific deterrent effect of arrest among the general population (Maxwell et al., 2002). 103

PAGE 104

Aside from these classic fi eld experiments, there is a growing body of literature employing various methodologies that has examined the effect of arrest on subsequent offending behavior. There are numerous studies that have found support, or partial support, for labeling theory (Bernburg & Krohn, 2003; Farrington, 1977; Gold & Williams, 1970; Huizinga, Elliott, & Dunf ord, 1996; Huizinga & Esbensen, 1992; Huizinga, Weiher, Espiritu, & Esbensen, 2003 ; Kaplan & Damphouse, 1997; Paternoster, 1978; Paternoster & Piquero, 1995; Smith, 2006). Fo r example, Bernburg and Krohn (2003) found that having police cont act or being arrested as a juvenile significantly increased serious crime and general delinquency in early adulthood and this effect was significant while controllin g for adolescent delinquency, family poverty, educational capabilities, and race. In direct contrast to st udies that fail to support traditional criminal justice approaches to curb ing crime, a couple of studies have found support for deterrence theory (Cameron, 1964; Smith & Gart in, 1989). For example, Smith and Gartin (1989) reporte d that arrest led to a reduc tion in the number of future police contacts and that arresting an individu al during a police contact also led to a reduction in future police contacts compared to not arresting an individual. However, this latter finding only held for the first few po lice contacts; in fact, after five contacts official intervention effects ar e more supportive of labeling. Several studies report null findings, failing to yield support for either deterrence or labeling theory (Hemphill, Toumbourou, Herrenkohl, McMorris, & Catalano, 2006; Huizinga et al., 1986; Klein, 1986; McAra & McVie, 2007; Thomas, 1977). Unlike the majority of studies examining the effect of arrest on behavior, McAra and McVie (2007) notably employed propensity score matching which is a rigorous methodological 104

PAGE 105

technique aimed at uncovering causal effect s with observational data. They found that once individuals were adequately matched there we re no significant official intervention effects on either the prev alence or frequency of serious delinquency. When considering studies that examine ot her sanctions that may follow an arrest, a similar picture emerges. That is, a numbe r of studies offer some support for labeling theory ( Berg, Consterdine, Hullin McGuire, & Tyrer, 1978 ; Bernburg, Kr ohn, & Rivera, 2006; Chiricos et al., 2007; Fagan, Kupchi ck, & Liberman, 2003; Hagan & Palloni, 1990; Huizinga et al., 2003; Klein, 1986; McAra & McVie, 2007; Thornberry, 1971; Schneider, 1982; Shannon, 1985; Wolfgang, Fi glio, & Sellin, 1972) but only a few offer any support for deterrence theory (Murray & Cox, 1979; Thistlewaite Wooldredge, & Gibbs, 1998; Ventura & Davis, 2005). Additi onally, there are several studies that report essentially no effect of official intervention on s ubsequent behavior (Brennan & Mednick, 1994; Gottfredson, 1999; Huizinga & Espiritu, 1999; Rose & Hamilton, 1970; Smith & Akers, 1993; Spohn & Holleran, 2002; Ta xman & Piquero, 1998). Some studies have more explicitly examined t he extent to which the type of official intervention matters in producin g deterrent or labeling effect s. For example, McAra and McVie (2007) found that appearing in juvenile court leads to increases in serious delinquent behavior whereas simply being arrested had no influence on subsequent behavior. For the sample as a whole, an arrest alone may not co nstitute a status degradation ceremony (see Garfinkel, 1956) thereby failing to trigger the labeling process. Thus, this study in particular dr aws attention to the potential importance of considering different types of official intervention as possible turning points in individuals criminal trajectories. 105

PAGE 106

In a comprehensive review of the deterrenc e literature, Nagin (1998) finds that cross-sectional and scenario-based empirica l research provides evidence that perceptions do indeed influence cr iminal offending and claims there is much stronger evidence for the existence of a substantial deterrent effect than what existed previously. However, the empirical studies that are of the highest methodological quality tend to show considerably weaker results compared to those which are conducted with crosssectional data or those which employ few or no control variables (Paternoster, 1987). A recent meta-analysis finds that perceptions of certainty of arrest have a statistically significant but relatively weak effect on crim inal behavior (Pratt et al., 2006). Taking all of the experimental and non-expe rimental studies into account, there is no clear support for labeling or deterrence theor y in the aggregate. This is true for different types of sanctions such as arrest and t hose that follow arrest. The onl y thing that is fairly clear from prior research is that the effects of official intervention, if they exist, are considerably more nuanced requiring investigations into samples that are meaningfully disaggregated and follow individuals ov er the shortand long-terms. The Contingencies of Sanction Effects Of the studies that have exam ined contingent official intervention effects across different groups, subpopulations have typically been defined with respect to sex, race, prior record, and stakes in conformity (Chi ricos et al., 2007). Several studies have examined contingent labeling an d deterrent effects across se x (Baumer, 1997; Chiricos et al., 2007; Gendreau, Little, & Goggin, 1996) and race (Ageton & Elliott, 1974; Berk et al., 1992; Bernburg & Krohn, 2003; Chiricos et al., 2007; Harris, 1975). While it is of academic interest to understand how official intervention effects may vary across race or sex, it is of less practi cal import to isolate these cont ingent effects due to policy 106

PAGE 107

considerations. That is, differential treat ment based on demographic characteristics can result in charges of racial and gender bias in the administration of justice. To a lesser extent, subpopulations that are based on stakes in conformity are subject to the same problem. Sherman, Smith, Schmidt, and R ogan (1992) find that individuals with high stakes in conformity (e.g., employed, high school graduate, married) had lower recidivism rates (see also DeJong, 1997). However, it may also be unethical to differentially intervene based upon whether an offender, for example, has a job or is married. Determining the effects of official in tervention across offending subpopulations (i.e., those with a prior reco rd) appears to be freer from the ethical problems of divergent treatment of offenders. In reality, selective intervention already occurs since police officers and prosecutors have consid erable discretionary power. Moreover, selective intervention policies informed by differential effects for offending subpopulations would guide police officers and prosecutors in a more systematic fashion thereby helping to ensure that different treatment is t he result of behavioral tendencies and not innate characteristics su ch as race or gender. Thus, for these reasons, the most pressing need for research may be on the extent to which the contingency of prior record mo derates the effect of offici al intervention on subsequent offending behavior. In one of the earliest studies examining deterrent and labeling effects across offending subpopulations, Came ron (1964) found that arre sting novice shoplifters typically resulted in the cessati on of their delinquent behavior. The results of this study echoed the notion that official intervention ef fects for novice offenders would be stronger 107

PAGE 108

than for more experienced offenders (Packer, 1968; Thorsell & Klemke, 1972). In a study of drunk drivers, Taxman & Piquer o (1998) used Cox proportion hazard models and found that punishment sentences were mo re likely to increase recidivism than rehabilitation sentences. Moreover, this effe ct was particularly pr onounced for first-time offenders. In a study conducted in New York City, DeJong (1997) found evidence consistent with labeling theory but only for no vice offenders. More specifically, she found that firsttime offenders who were incarcerated for misdemeanors were more likely to be rearrested than individuals who were not incarcerated for their offenses. Similar findings were reported by Horwitz and Wasserman (1979) who found that the positive relationship between severity of official intervention and the lik elihood of recidivism held only for first-time juveniles. Klein (1974) collected two years worth of recidivism data on 100 juveniles from each of 13 police agencies that he categorized as either high or low diversion agencies. He found that arrest and system processi ng decreased the rate of subsequent offending; however, this was found to be the case only for more experienced offenders and not for first-time offenders. In examining the effects of arrest within a criminal career paradigm, Smith and Gartin (1989) found t hat the experience of being arrested is more likely to terminate the criminal career of novice offenders whereas it may only reduce offending rates among more experienced offenders. Interest ingly while another scholar found labeling effects for incarcerat ion using the same data set (see Shannon, 1980; 1985), Smith and Gartin (1989) found deterrent effects for their outcomes. This 108

PAGE 109

point draws attention to the need to consider di fferent types of offici al intervention within the same study. Matsueda and colleagues (2006) linked subsequent offending behavior to experiential risk of apprehensi on, which they defined as t he number of arrests divided by the number of self-reported offenses. They found that as experiential risk of apprehension increased, the perceived risk of apprehension also increased causing a reduction in subsequent offendi ng behavior. These results ar e consistent with specific deterrence theory. What is important is that experi enced offenders have lower experiential risk of apprehens ion values than novice offenders, meaning that experienced offenders are likely less deterred by arrest. A recent felony convictions study whic h aimed to redirect attention to the contingent nature of labeling effects found evidence for labeling theory in general but did not find the observed labeli ng effect to be conditioned by prior record (Chiricos et al., 2007). Specifically, those who had adj udication withheld were less likely to recidivate compared to those who had been adjudi cated. In a rare but highly notable attempt to place sanction effects in a develop mental framework, Bhati (2007; see also Bhati & Piquero, 2008) classi fied individuals based upon dive rgence between their postrelease micro trajectories and counterfactual trajectories to expl ore whether labeling, deterrent, or null (incapacitation) effect s occurred. Her analysis concluded that approximately 56% were unaffected by incarcer ation (i.e., only incapacitation effect), 40% experienced a deterrent effect and onl y 4% experience a labeling effect. Importantly, those with a great er number of arrests were less likely to experience 109

PAGE 110

deterrent effects but those who were sanctioned closer to those prior arrest clusters were more likely to experience deterrent effects. With a mixed bag of empirical evidence, Smith and Gartin (1989) have noted that perhaps the only conclusion we can draw at this point is that t he relationship between punishment and future offending among both novice and experienced offenders is problematic (p. 97). Unfortunat ely, some twenty years later this observation still largely appears to be true. With recent studies c ontinually providing mixed results, key limitations of prior research and only a few explicit attempts to place sanction effects in a developmental and life-cour se framework, much remain s unknown about what types of official interventions result in turning points for what types of criminal trajectories and for how long. Before attemp ting to address this question em pirically, it will be useful here to outline the key problems with prior research that are, in part, responsible for the inability to make any firm conclusions regarding deterrent and labeling effects. Key Limitations of Prior Research Prior research has revealed important insights into the relationship between criminal sanctions and subsequent offending in the aggregate, across different types of offenders, and in both the shortrun and long-run. However, there are several important limitations of prior research that temper the theoretical conclusions about labeling and deterrent effects. The pr incipal issues include: data, offender categorization, and selection effects. Data In a critique of Sampson and Laubs (2005a) empirical work, Robins (2005) reminds us that official a rrest records may su ggest termination of a criminal career because of data problems. For instance, i ndividuals who committed crimes while in 110

PAGE 111

prison, or serving in the m ilitary, or whose crimes were simply undetected do not get counted (Robins, 2005). This is a major problem when assessing whether official intervention serves as a turning point, for better or worse, in a criminal trajectory. Using official data to estimate o ffending trajectories can be pr oblematic for another reason. Arrest records and true offending behavior ma y be correlated but they are certainly not the same thing. Hence, while it is us eful to know whether an arrest reduces the number of future arrests or results in never being arrested again, this unfortunately does not necessarily imply the reduction or ce ssation of offending behavior (Huizinga & Henry, 2008, p. 223). Therefore, empirical efforts aime d at uncovering the effects of official intervention on subsequent off ending are probably on be tter ground if they employ self-report data. Specifically, it is important to employ se lf-report data on actual offending behavior and not self -report data on arrests, since the latter suffers from the same problems as official re cords. Importantly, self-repo rt data has been shown to be a reliable and valid source of information for many empirical investigations (Thornberry & Krohn, 2000). Another key data problem is the use of l ongitudinal data sets with just a couple of waves. While two-wave panel data can help with the establishment of time-order, it does not provide enough time points to examine the effects of official intervention in a developmental perspective. Simply estima ting offending trajectories requires, at minimum, three waves of data. Modeling quadratic or cubic changes in offending trajectories requires four or five waves, respectively. Factoring in the need to follow-up offenders both in the shortand long-runs, it becomes clear that a multi-wave data set designed to follow individuals over an ex tended period of time is necessary to 111

PAGE 112

investigate whether official intervention serves as a turning point in a criminal trajectory. While there is certainly not an abundance of such data sets, they do indeed exist (Huizinga & Henry, 2008). However, these mu lti-wave panel studies have yet to be fully exploited to address the key theoretical questions, which have been illuminated herein by placing official intervention effects re search in a developmental and life-course framework, in a methodological ly rigorous fashion. Offender Categorization To assess whether official intervention serves as a turning point in the criminal trajectories of offenders, a classification procedure for diffe rent types of offenders is necessary. While individuals prior record s have been used to classify persons into offending subpopulations in previous research there is a critical problem with this strategy. For example, tw o individuals who have been arrest ed once could be classified similarly despite the fact that one ma y be a chronic offender and the other an exceptionally infrequent offender Then, classifying individuals based on their prior records as opposed to their offending behavior is somewhat imprecise if the goal is to assess contingent effects based on ones offending trajectory. As an alternative to classification bas ed on prior record, Huizinga and colleagues (1986) classified individuals according to a delinquency typology (e.g., non-delinquent, exploratory delinquent, non-seri ous delinquent, serious delinquent ). However, there are problems with subjective classification procedures as well. First, a subpopulation can be subjectively classified as distinct but only represent random va riation. Second, unusual subpopulations can fail to be classified at all. Third, statis tical tests examining differences across subpopulations may be undermined by uncertainties about the reliability of subpopulation assignment (s ee Nagin, 2005). Finally, subjective 112

PAGE 113

classification schemes may make it difficult to capture trajectories of offending behavior over time, which can limit their ability to as sess whether official intervention serves as a turning point in a criminal trajectory. Nagin (2005; see also Nagin & Land, 1993) proposes a unique solution to this problem that empirically classifies individuals into distinct trajectory groups, which can be used to assess whether the effects of a particular event differ across these subpopulations. Selection Effects In addition to the problems associated with subjective classification, methodological limitations of previous studi es have resulted in the inability to discern between labeling and deterrent effects for offending subpopulations with any definitiveness. A key problem with prior empirical work is the use of methodological designs that are not well-suited for disentangling selection arti facts from real official intervention effects (Smith & Paternoster, 1990).14 With some exceptions (e.g., Chiricos et al., 2007; McAra & McVie, 2007; Smith & Paternoster, 1990), research designs have failed to employ strong methodology that addresses the problems of making causal inferences with non-experimental data. All is not lost. The data, classificati on, and selection artifact problems can be addressed given the availability of high-qualit y longitudinal data for official intervention effects research (Huizinga & Henry, 2008, p. 248) and the advancement of integrated methodological techniques capable of esti mating treatment effects across different 14 Selection artifacts occur when the relationship betwe en official intervention and subsequent behavior is biased due to a correlation of the official intervent ion variable and the error term in a regression model (Smith & Paternoster, 1990). It would be ideal for scientific purposes to randomly assign individuals to treatment groups; however, ethical and practical concer ns frequently make experimental designs in social science impractical. For these reasons, the implem entation of experimental designs in criminology is a rare occurrence (for some classic examples, see Klein, 1986; Sherman & Berk, 1984a). 113

PAGE 114

trajectory groups with observational data (Haviland & Nagin, 2005; Haviland et al., 2007; 2008). Given this synthesis of what is known, what is unknown, and why certain things remain unknown, in regards to the effect s of official intervention on subsequent offending, I am now in a position to out line the data and the me thodological approach used for the current analysis that addresses each of the aforement ioned limitations of prior research. CHAPTER 6 DATA Chapter 6 describes the data and measures explains the research methodology and outlines the analytic strategy. First, the particulars of the Rochester Youth Development Study (RYDS) are discussed, wh ich serves to demonstrate that these data are particularly well-suited to address the empirical aims of the current study. Following a general discussion of the RYDS, the procedures for handling missing data and arriving at the final analysis sample are detailed. Next, the measurement of violent offending behavior, official intervention, and the forty covariates is discussed. Chapter 6 then concludes detailing descriptive statisti cs for the analysis sample on the whole. Overview of Data The current research employs data from the RYDS, which is an ongoing longitudinal study aimed at understanding the causes of crime and delinquency in a developmental and life-course perspective. The RYDS features a total of fourteen waves of data, collected in three phases. Phase one consists of nine waves of data collected at six month intervals beginning in 1988 and ending in 1992. Phase two consists of three waves of data gathered at annual intervals beginning in 1995 and 114

PAGE 115

ending in 1997. Finally, phase three consists of two waves of data collected with a two year interval beginning in 2003 and ending in 2005. The principal respondents (Generation 2 or G2) were approximately 13. 5 years of age at wave one and 31 years of age at wave fourteen.15 The RYDS is a disproportiona te stratified sample of 1,000 individuals in which at-risk youth (i.e., ma les and individuals living in high crime areas) were over-sampled. The Generation 2 sa mple is comprised of 73% males and 27% females. With respect to race, the sample is 68% Afri can American, 17% Hispanic, and 15% White (the male subsample is appr oximately 63% African American, 18% Hispanic, and 19% White). The overall retention rate is good. For instance, 78.5% of the 1,000 individuals in the original sample participated in the survey during wave thirteen, which is some fifteen years fr om the data collection time point. Research subjects were interviewed in sessions lasting approximately an hour and were asked to answer questions on a wide va riety of subjects. These topical areas include but are certainly not limited to such things as neighborhood characteristics, peer influences, gang involvement, social supports, educational aspirations and commitments, psychological well-being, fam ily structure and relationships, sexual activity, and ones behaviors such as self-reported delinquency, violence and drug use. These self-report surveys are supplemented by official records from Rochester Public Schools, police records, social services data, and United States Census data. In addition, the RYDS also follow ed the primary caregivers (Generation 1 or G1) of the main research subjects (G2) for the firs t eight waves in phase one and all three waves 15 The RYDS distinguishes between the principal respondents, who were adolescents during phase one of the study, and their primary caregivers who were adults during this time and were also interviewed as part of the RYDS. The former are referred to as G eneration 2 (G2) whereas the latter are referred to as Generation 1 (G1). 115

PAGE 116

in phase two. This provides information about things such as the parents behaviors and experiences, their perceptions of G2 and their peers, their relationships and living environments and, in many cases, similar in formation that was collected from G2 which can be used to validate and/or compare perceptions between the two generations.16 The RYDS is one of a few data sets that is ideal for official intervention effects research. Non-experimental studies necessi tate the use of data from longitudinal studies with probability sampling designs and high-quality measures (Huizinga & Henry, 2008). One of the key features of the RYDS data is the breadth of data in terms of variables and constructs included in the data set, which is particularly important to the current research endeavor that attempts to isolate official intervention effects with counterfactual methods (i.e., propensity score matching). These methods create covariate balance only on measured constructs used for treatment model estimation. Thus, having measures of relevant covariates is essential. Another important feature of the RYDS data is the longitudinal, multiphase component which ensures there are sufficient waves of data to estimate offending trajectories and still assess the effects of official intervention across the shortand long-runs. Moreover, the large number of waves also permits the estimation of trajecto ries that are not simply linear over time; directional changes can be model ed with, for exampl e, quadratic terms (Nagin, 2005). Finally, the RYDS contains measures of self-reported offending across waves that will be used to estimate the violent offending subpopulations. This is a substantial 16 While not the focus here, it is worth mentioning t hat the RYDS is now interlaced with the Rochester Intergeneration Study (RIS) which was funded to follow the main research subjects children (Generation 3 or G3). Needless to say, this provides a unique o pportunity to look into t he transmission of crime and life circumstances across three generations. 116

PAGE 117

improvement over studies with two-wave panel designs and those that estimate contingent labeling and deterrent effects with official data as the measure of behavior. Analysis Sample The current investigation into the cons equences of official intervention for subsequent violent behavior across violent offending trajectory groups employs data from phases one and two (waves two through ten) from the male s from the RYDS.17 Individuals included in the analysis sample had to be interviewed during every wave between waves six and ten and had to have valid police intervention records during waves seven and eight. In addition, indivi duals must have been interviewed at least twice between waves two through five to be included in the analysis sample. The analysis sample was selected with the rationale that the subject had to be interviewed such that confoundi ng variables used to predict the probability of receiving official intervention were generally available (wave six), data on officially recorded arrests and self-reported police contacts we re available (waves seven and eight), and the violent offending outcomes in the shor t-run (wave nine) and long-run (wave ten) were observed. Given the ability of LCGA to accommodate missing data in its estimation of behavioral trajectories, some missing data was permitted from waves two through five. Nevertheless, missing interviews were not a serious concern during this time period; 93.4% of the final analysis samp le was interviewed during each of waves two through five and approximately 5.9% wa s surveyed during all but one of these waves. Just less than 0.7% of individuals we re interviewed in only two out of four of 17 Data from wave one is not used because one of the si x items that is part of the violence measure is not assessed during this wave. Rape is also not asked in waves eleven and twelve in phase two. Therefore, wave ten is used to assess the long-term effects of official intervention. 117

PAGE 118

these waves (waves two through five). Therefore, no individ ual in the analysis sample had his violent offending trajectory estimated from less than three out of five of the time points since individuals had to be interviewed in wave six to be included in the analysis sample. Moreover, very few trajectories were estimated with any missing data at all (only 6.6%). The final analysis sample that fit the above criteria consists of approximately 82% of t he original RYDS males (N=595) Across several important demographic factors as well as violent criminal behavior, there were no significant differences between the analysis sample and those individuals who did not make the analysis sample but were interviewed during wave one (see Table 6-1).18 As with most data sets, not all of t he 595 male respondents comprising the analysis sample contained valid data on ever y covariate of interest. Even under the ideal case of data from valid respondents being missing completely at random, using a listwise deletion procedure her e would be especially problem atic given the importance of maintaining a sufficient sample size to es timate the effects of official intervention on subsequent violent offending behavior acro ss each of several trajectory groups. Several options for addressing missing data were considered. Haviland and colleagues (2005; 2007; 2008) devel oped the LCGA-PSM methodology and recommended dealing with missing data by creating missing indicator dummy variables and replacing missing values with any score such as the covariate mean given its ease of implementation and appropriateness within the propensity score matching framework. The logic behind this approach is that even though an individuals 18 Means are reported for interpretation purposes but the p-values for the significance testing are associated with the chi-square stat istic for dummy variables (Black, Hispanic, family poverty status) and the Wilcoxon-Mann-Whitney st atistic for the lone count variable (violent crime). 118

PAGE 119

response to a particular item t hat he or she did not respond to is not observable, the fact that he or she did not respond to the item is observable and may reveal important information about the individuals propensity to receive some treatment (see Rosenbaum & Rubin, 1984; see also Haviland et al., 2007). Includi ng missing indicator variables with arbitrary replacement of missing scores (e.g., m ean replacement) to address the problem of missing data results in the potential for the propensity scores to balance both observed scores on the item or scale as well as the frequency of missing scores. In this way, the fa ct that the individual did not respond to an item is used as information helpful in predicting the propens ity to receive treatment and, as a consequence, preserves the case in the analysis. For the present study, however, this appr oach to dealing with missing data within a propensity score matching framework turned out not to be viable for two primary reasons. First, many of the covariates that have missing dat a are reported by G1 (i.e., parental reports). Moreover, these vari ables tended to be the ones with the highest level of missing information such as t he neighborhood disorganization measure, the Achenbach aggression scale, and the PCs educational expectations for the child variable (see Table 6-2). While one can probabl y safely argue that G2 not responding to a question may reveal important informa tion about G2s likelihood to experience official intervention, it is not quite as cl ear how non-response by G1 would be useful in predicting G2s treatment likelih ood. Second, Haviland et al. (2007) argue it is best to estimate propensity scores and perform matchi ng within trajectory groups. When using missing indicator variables in the models fo r predicting treatment within trajectory groups, the problem of complete/quasi-co mplete separation of data points was 119

PAGE 120

exacerbated for certain latent classes. Ultimately, then, missing data was handled with multiple imputation (see Alison, 2001). Multiple imputation uses information from variables in the model and others which are available that may predi ct missing values. Unlike imputation through regression, random error is added to each imputed valu e. Addressing the list of problems associated with single imput ation approaches to dealing with missing data, multiple imputation is a process whereby the imputation procedure is repeated multiple times to produce a single stacked data set containing n copies of the original data set with different imputed values. That is, the n imputed values are different because a random component is added to the value that is the best prediction given the variables included in the imputation model (Alison, 2001). With re latively low levels of missing data, as is the case in the present study five data sets are sufficient to achieve a high level of efficiency. The single stacked data set can then be separated into five individual data sets and analysis can be performed as usual on each of the imputed data sets. When data analysis is completed on each of the n imputed data sets, the results are then averaged and a special standard error is used yiel ding an overall estima te of the effect. SASs Proc MIANALYZE is utilized for this purpose. Measures This section outlines the measures used in the current study. First, the observed variable used to track violent cr ime trajectories prior to t he intervention and assess the violent crime outcomes is discussed. Second, the variables used to define the two types of official intervention experiences are examined. Finally, the covariates used to estimate the propensity scores that are employed in the matching procedure are detailed. 120

PAGE 121

Behavioral Trajectory and Outcome: Violent Crime Youth violence has been identified as a se rious social problem and one that is likely to continue to be responded to with tr aditional approaches to justice. As a consequence, it has been suggested in this st udy that there is an important practical need to understand how official interventio n experiences may influence offending trajectories for different types of violent o ffending subpopulations. In addition to these practical considerations, studying violence provi des for a fairly conservative test of ideas underlying deterrence and l abeling theories. That is, given the etiologies of crimes against persons and crimes against property ma y differ such that the former could be argued to be less influenced by rational choice, official intervention effects found for one or more violent offending subpopulations are noteworthy (especially if a strong methodological design is employed). Violent crime. The outcome variable of interest is self-reported violent crime which is measured similarly as the variables employed to estimate the trajectories of violent crime prior to the exper ience of official intervention. Violent crime is a variety index measuring the total num ber of six different acts of violence an individual committed in a given wave. The six acts include: attacking someone with a weapon, participating in a gang fight, throwing things at people, committing robbery, raping someone, and engaging in other assault.19 Possible scores on this measure range from zero to six. With respect to the trajecto ries, waves two through six are used to estimate the latent class growth analysis models. Wi th respect to the outcome variable, waves 19 Variety indexes are among the most reliable forms of self-reported measures of delinquency (Belson, 1986; Hindelang, Hirschi, & Weis, 1981). 121

PAGE 122

nine and ten are used to examine the short-term and long-term effects of official intervention, respectively. Treatment Conditions: Official Intervention Earlier the idea was advanced that it may be impor tant to consider different types of official intervention experiences when in vestigating how justice system contact may serve as a turning point in a criminal traj ectory. For instance, deterrent or labeling effects may not fully be set in motion without a certain grade or type of sanction. There is a clear difference between an individual who has been arrested with no subsequent action taken by the criminal justice system and one who has been arrested and incarcerated for his offense. Similarly, a different experience occurs when an individual has been contacted by the police and when one has been arrested. Ideally, each of these three types of official intervention wo uld be examined in the same study. In the current study, however, incarceration was too infrequent of an official intervention experience among this sample of juveniles to include in the anal ysis. I discuss the implications of this omission in Chapter 10 of the study. Police contact and arrest. Using both self-report data from the RYDS and official records data from the Rochester Police Depa rtment and County Wide Registration, two types of official interventions are available and used as the treatment conditions. First, police contact is a self-report measure of whether an individual experienced one or more police contacts during waves seven or eight. Second, arrest is a dichotomous official records measure of whether an i ndividual experienced one or more arrests during waves seven or eight.20 Two waves are employed to ensure a sufficient number 20 Official records data on police contacts was not us ed since there was little variance in this measure across waves seven and eight. Self-reported arrest was also considered but descriptive analysis 122

PAGE 123

of individuals actually experi enced an official intervention so that the effects of the treatment can be estimated across different violent crime trajectories. Covariates The treatment model consists of a lar ge number of covariates from several different waves of data. Some time-stable measures were captured at wave one (e.g., race) whereas other measures that vary across time are available at various waves. In most cases, these measures are taken from wave six which is the wave immediately prior to the designated period for treatment. Wave six is selected as these measures are temporally close, yet precede in time, the official intervention experience. The following potentially relevant confounder va riables are organized into the following subcategories: demographics, neighborhood characteristics, family dynamics, school factors, peer associations, values and m ental states, prior delinquent behavior, and prior justice system contact. In describi ng these measures, the general theoretical significance of controlling for each subcategory of variables is offered. It is worth noting that all of the scales were summed and di vided by the total number of items that comprised the scale. This returns the multi-it em scales to their orig inal metrics so that the response options to the individual items can be used to aid in the interpretation of basic descriptive statistics. Demographics Some of the more important control variables in sanction effects research include those identified from the sta tus characteristics hypothesis, which states that certain revealed there were fewer self-reported arrests as co mpared to official records. Nevertheless, results from self-reported arrests largely conform to thos e reported herein and, consequently, the focus will be placed on the presentation of results from official records data for arrest. 123

PAGE 124

types of individuals such as minorities are mo re likely to experience official intervention irrespective of their crimi nal involvement. Put differently, Given the occurrence of a deviant action (delinquency) the decision of organizational agents to sanction officially (to label) an actor is in part determined by the so cial attributes (race, sex, social class) of the offender (Paternoster & Iovanni, 1989, p. 364). Since minorities have been found to commit a greater proportion of personal crimes t han their demographic counterparts (Hindelang, 1978), for example, controlling for basic demographics is essential when assessing the extent to wh ich deviant labels influence subsequent violent crime. In addition, it is critical to control for the respondents age given the established relationship between age and crime (s ee Hirschi & Gottfredson, 1983). Race. Race is a nominal variable that has been recoded into a series of dummy variables. Specifically, Blacks and Hispanics are coded as 1 on their respective dichotomous variables. Whites are left out of the treatment mode l and serve as the reference group. Age. This common demographic factor is a continuous measure of the respondents age during wave six. Family poverty. Family poverty is a dichotomous variable that serves to distinguish between respondents who come from families who are above the federally defined poverty level from those who come fr om families who are below it. To create this measure, the number of individuals in G1s house was determined and the before tax income was calculated. These two values were used and compared against the U.S. Governments federal poverty chart. Respondents with family incomes below the 124

PAGE 125

federal poverty line were coded as 1 whereas those above the poverty line were coded as 0. Neighborhood characteristics Neighborhood characteristics may play a crit ical role in determining the rate of justice system contact and involvement in de linquency. Research suggests that certain segments of society are subj ect to higher levels of pol icing, making the risk of experiencing an official intervention greater in these areas. Impor tantly, the social disorganization perspective leads one to believe that certain neighborhoods such as those with concentrated disadvantage are more likely to have higher crime rates (Shaw & McKay, 1942). While social disorganization and collective efficacy are characteristics of the neighborhood and are c onceptualized to operate as macro-level phenomenon, problematic neighborhood environments can be a breeding ground for criminal opportunities at the individual-level maki ng controlling for neighborhood factors highly important here. Proportion African American. This variable is a meas ure of the proportion of African Americans residing in the youths neighborhood. The measure comes from U.S. Census data for Monroe County during the 1990 data collection period. The proportion of Blacks in the neighborhood is calculated by taking the number of Blacks (not of Hispanic origin) and dividing them by the total number of individuals residing in the corresponding census tract. Thus, higher score s reflect a greater proportion of African Americans in the neighborhood. Values on this measure can theoretically range from 0 to 1. Proportion in poverty. Proportion in poverty is a measure of the proportion of families in poverty residing in the yout hs neighborhood. As for the proportion African 125

PAGE 126

American measure, the measure comes from 1990 U.S. Census data. The proportion of families living in poverty was calculated by taking the number of families in poverty and dividing this value by the number of families residing in the corresponding census tract. Thus, higher scores signify a greater proportion of families living in poverty. The possible scores on this measure range from a minimum of 0 to a maximu m of 1. Neighborhood disorganization. Tapping the level of conflict and violence, drug use, and general disorder in the adolescents neighborhood as perceived by the primary caregiver (G1), this variable is a m easured with a seventeen item scale. The adolescents primary caregiver was asked to report the extent to which a number of things were a big problem, s ort of a problem, or not a problem. Specifically, they were asked about how much of a problem the following things were in their neighborhood: different racial or cultural groups who do not get along with each other; vandalism, building, and personal belongings broken and torn up; little respect for rules, law and authority; high unemployment; winos and junkies; prostitution; abandoned houses or buildings; sexual assaults or r apes; gambling; burglaries and thefts; run down and poorly kept buildings and yards; assaults and muggings; street gangs or delinquent gangs; syndicate, mafia or organized crime; buy ing or selling stolen goods; drug use or drug dealing in the open; and, homeless street people. Higher scores indicate greater neighborhood disorganization. Neighborhood integration. Neighborhood integration is a seven item measure tapping the extent to which neighbors k now one another and whether the people who live in the neighborhood can be relied upon. For example, on a four point Likert type scale, the adolescents primary caregiver (G1) was asked: how many people live in your 126

PAGE 127

neighborhood do you know by sight?; how many people live in your neighborhood do you talk to on a regular basis?; how o ften do you and other people who live in your neighborhood borrow things like tools or re cipes from each other?; how often do you and other people who live in y our neighborhood ask each other to drive or take your children somewhere? Response options incl uded never, seldom, sometimes, and often. Higher scores on this measure are indicative of higher neighborhood integration. Neighborhood satisfaction. This variable is a three item measure tapping into the primary caregivers satisf action with the neighborhood. On a four-point Likert type scale, respondents were asked to state how sati sfied they were with: your relationship with people in this neighborhood; the way people in this neighborhood take care of where they live; the neighbor hood as a good place to bring up children. Response options ranged from very dissatisfied to very satisfied and higher scores on this measure are indicative of higher neighborhood satisfaction. Neighborhood arrest rate. Neighborhood arrest rate is a measure of the resident arrest rate for the respondents census tract. Using official records, the total number of individuals arrested within the youths census tract was divided by the population and standardized by 1, 000. Higher values indicate a greater neighborhood arrest rate. Family dynamics The structure and dynamics of the family c an play an important role in creating opportunities for crime and delinquency. Specifically, traditional family structure and primary caregivers that play an active ro le in supervising their children can reduce opportunities for offending. Individuals who are strongly bonded to their primary 127

PAGE 128

caregivers may have more to lose from hav ing justice system contact experiences and engaging in offending in general. Family structure. Family structure is a dichotomous variable indicating whether a respondent lives at home with both parents. This measure is mostly reported from the respondents primary caregiver (G1) but some adolescents (G2) did provide data on the family structure when G1 interviews were unavailable. Responses were coded such that 0=does not live with both parent s and 1=lives with both parents. Parental supervision. Parental supervision is a four item scale measuring the quality of supervision that the youth self-reports. On a four point Likert type scale, respondents were asked to rate the extent to which the primary caregiver knows where the child is and knows who they are with when they are away from home as well as how important these two supervisory tasks are to the primary caregiver. Higher values designate greater parental supervision of the respondent by the primary caregiver whether this is one of the biological parents or some other male or female. Parental attachment. This variable is an eleven item scale tapping into how attached the respondent is to t he primary caregiver (PC) (e.g., mom, dad, other male, or other female). Respondents were asked a number of questions whic h include: you get along well with your PC; you feel that you ca n really trust your PC; your PC does not understand you (reverse coded); your PC is too demanding (reverse coded); you really enjoy your PC; your PC interferes with your activities (reverse coded); you think your PC is terrific; you feel very angry toward your PC (reverse coded); you have a lot of respect for your PC; you feel very proud of your PC; and, you feel violent toward your 128

PAGE 129

PC (reverse coded). Higher scores indicate a greater level of parental attachment to the primary caregiver. PCs educational expectations for child. This variable is a two-item measure tapping the primary caregivers (G1) expec tancies regarding whether the youth will succeed in educational pursuits. The primary caregiver was asked whether they believed G2 would graduate from high school and whether he would go on to college. Response options to these two questions were no, depends, and yes. Higher scores indicate greater expectations for the ch ild to further his or her education. Attachment to child. Attachment to child is an eleven item scale measuring how attached the primary caregiver (G1) (e.g., mo m, dad, other male, or other female) is to the adolescent (G2). This is a parallel scale to the parental attachment scale above meaning the items are simila r but this measure instead asks about how attached the primary caregiver is to the child. Higher score s indicate a greater level of attachment to the adolescent. Consistency in discipline. Consistency in discipline measures the youths perception of their primary ca regivers (G1) patterns of di scipline. Res pondents were asked the following questions: how often do you get away with things; once PC decides a punishment, how often can you get out of it; how often do you get punished sometimes, but not other time s, for doing the same thing; how often does PC have to ask you to do the same thing more than once; when you get punished, how much does the kind of punishment you get depend on PCs mood. Response options were often, sometimes, seldom, and nev er. Higher scores are coded such that they indicate lower consistency in discipline. 129

PAGE 130

School factors Individuals with greater a ttachments and commitments to school tend to be less involved in delinquent activity. Moreover, these individuals have more to lose if they were to get into trouble with the law and have less time to do so because of their participation in scholastic activities. Si nce school factors are related to both the participation in delinquency and contact with the justice syst em, these constructs are important control variables in sanction effects research. Commitment to school. Commitment to school is a ten item scale measuring ones level of agreement on a four point Likert scale to statements pertaining to school commitment. For instances, respondents were a sked: whether school is boring to you (reverse coded); you like school a lot; you don t really belong at school (reverse coded); you usually finish your homework; you try hard at school; and, getting good grades is very important to you. Response options ranged from strongly agree to strongly disagree. Higher scores on this measure des ignate greater commitme nt to school. Aspirations for college. Aspirations for college is a two item scale indicating how important it is for one to, first, graduate from high school and, second, go to college. On a four point Likert type scale, response options ranged from not important at all to very important. Higher scores indicate great er aspirations for college. Prosocial activities. This variable is a five item scale measuring ones involvement in prosocial activities many of which take place in the school environment. On a four point Likert type scale with re sponse options of never, seldom, sometimes, and often, respondents were a sked how often they: took part in school sports such as intramurals and varsity sports; took part in school activities like clubs or special events like a school play; took part in organized sports or teams outside of 130

PAGE 131

school; took part in any organized musical or singing group, includ ing in school; and, took part in other organized groups like the Y, Boys and Gi rls Clubs, or scouts. Higher scores on this measure indicate greater involv ement in prosocial activities. Educational expectations. Educational expectations is a parallel measure to the parents education expectations of child sca le. While the items are the same, this measure taps the youths (G2) own expecta ncies regarding whether he will succeed in his educational pursuits. As with the other measure, higher scores indicate greater expectations for the individual to further his education. School clubs with friends The last of the school factor variables taps the extent to which individuals engage in school activities such as clubs or special events like a school play with their friends. Respons e options on this measure include never, seldom, sometimes, and o ften. Higher scores on this measure indicate greater involvement in school clubs with friends. Peer associations One of the more robust predi ctors of crime and violence in the literature is peer delinquency. There is also a long li ne of evidence also suggesting that gang involvement influences violence above and beyond the contributions of peer delinquency (e.g., see Thornberry et al., 2003). In addition to the fa cilitating effects of peers and gangs on delinquency, being involved wit h delinquent peers or affiliated with gangs may also make one more likely to ex perience justice system contact (e.g., see Tapia, in press). Thus, controlling for t he effects of peers is important in sanction effects research and six measures are used here to tap into several distinct aspects of peer influence. 131

PAGE 132

Peer delinquency. Peer delinquency is an eight item scale measuring the proportion of ones peers that engage in deli nquent acts. Respondents were asked to indicate whether none, a few, some, or most of their fr iends have done delinquent things which, for example, include: stol e something worth more than $100; attacked someone with a weapon with the idea of se riously hurting them; took a car or motorcycle for a ride or drive without t he owners permission; and, damaged or destroyed someone elses property on purpose Higher scores indicate greater peer delinquency. Peer drug use Peer drug use is a four item sca le measuring the proportion of ones peers that engage in drug use. Respondent s were asked to indicate whether none, a few, some, or m ost of their friends have us ed marijuana, crack, alcohol, or other drugs (e.g., LSD, heroin, acid). Higher scores indicate greater pe er drug use. Peer delinquent values. This variable is a six item scale that measures whether ones peer group would say it was wrong, not say anything, or say it was okay if the respondent engaged in certain delinquent acts incl uding: using a weapon or force to get money or things from people; hitting somebody with the idea of hurting them; stealing something worth $50; damaging or destroyi ng someone elses property on purpose, taking a car or motorcycle for a ride or drive without the ow ners permission, and skipping classes without an excuse. Higher sco res on this measure reflect higher peer delinquent values. Dating. Dating is a single item dichotomous variable measuring whether an individual has a special boyfr iend or girlfriend. Respons es were coded such that 0=no boyfriend/girlfriend or 1=boyfriend/girlfriend. 132

PAGE 133

Sexual activity under fifteen. This variable is a dichotomous indicator measuring whether the respondent had engaged in sexual in tercourse prior to the age of fifteen. Responses were coded such that 0=no sexual activity under fifteen and 1=sexual activity under fifteen. Gang involvement. Gang involvement is a dichot omous measure of whether an individual was involved in a gang at any time between waves two through six. Beginning in wave two, respondents were a sked during the time period since the last interview if they were a member of a street gang or po sse (the term used by Rochester adolescents). Responses to t he questions of gang participation during the five waves on interest were recoded such that 1=gang involvement (one or more waves) and 0=no gang involvement. Values and mental states Individuals with delin quent values have less respect for the law and/or do not personally subscribe to the norms of soci ally acceptable behavior. Those with less belief in conventional society may be more likely to engage in crime and violence (Hirschi, 1969). Moreover, i ndividuals who have delinquent values may have more runins with the law as they fail to respect the authority of the legal system. In addition to ones values, aggressive behavior has been relat ed to certain mental states such as depression. And, given their current ment al state, depressed individuals may be less concerned than they other wise would be with the ramifications of justice system contact. Given these potential theoretical ties to both the treatment and the outcome of interest, these measures also represent variables to include in the treatment model. Delinquent values. Delinquent values is an elev en item scale measuring how wrong one thinks it is to engage in a variety of delinquent behaviors. For example, 133

PAGE 134

respondents were asked about acts of delinquency such as: using hard drugs, drinking alcohol, stealing something worth $50, taki ng a car or motorcycle for a ride without the owners permission, damaging or destroying so meones property on purpose, or hitting someone with the idea of hurting them. Response options included not wrong at all, a little bit wrong, wrong, and very wrong. Hi gher scores indicate greater approval of delinquent behaviors. Depression. This variable is a fourteen item scale tapping how frequently individuals felt depression symptoms such as feeling hopeful about the future; feeling depressed or very sad; not f eeling like eating because you felt upset about something; or, thinking seriously about suicide. On a four point scale, response options ranged from never to often; t hus, higher scores indicate greater depression symptoms. Self-esteem. Self-esteem is a nine item scale measuring the extent to which an individual agrees or disagrees, on a four point Likert scale, with statements about oneself. Examples of statements tapping self -esteem include: at times you think you are no good at all (reverse coded ); you feel that you hav e a number of good qualities; you feel you do not have much to be proud of (reverse coded); or, you feel you are at least as good as other people. Higher scores indicate greater self-esteem. Prior delinquent behavior The importance of controlling for prior behavior in sanction effects research should be relatively obvious. The best predictor of future behavior remains past behavior (White et al., 1990) and deviant behavior is na turally correlated with official intervention experiences. While the LCGA assigns like i ndividuals to the same latent class, prior delinquent behavior is explicitly included to assess the balance on these key covariates at waves two through six. In addition to contro lling violent crime specif ically, it is a good 134

PAGE 135

idea to control for delinquency more generally. Drug use and general delinquency are naturally related with justice system contac t and they are also related to violent behavior. Indeed, ingestion of certain subst ances such as alcohol or pcp can lead to violence and committing general delinquency would lead to more opportunities for violent encounters. Drug use and general de linquency are measured at wave six as described below. Violent crime. Violence is a self-reported variet y index tapping into the total number of six distinct acts of violence an individual commi tted in a given wave. The violent acts include: attacking som eone with a weapon; engaging in a gang fight; throwing things at people; committing robbery; raping someone; and, engaging in other assault. Scores on this measure range from zero to six and higher scores indicate greater involvement in violent crime. This is the same measure that is used for the behavioral trajectories and outcomes. To be clear, a violence variable for each wave is included in the treatment model. In other words, five diff erent violent crime measures (waves 2 though 6) are employed as covariates. General delinquency. General delinquency is a se lf-reported variety index tapping into the total number of thirty-two distinct acts of general delinquency committed in wave six. The general delinquency acts cover a wide range of criminal and delinquent behavior and include such things as: truancy; carrying a weapon; property damage; public rowdiness; thef t of several different amount s (e.g., less than $5, $5-50, $50-100, $100+); auto-theft; assault; robbery; obscene phone calls; fraud; marijuana sales; hard drug sales; etc. Scores on this measures range from zero to thirty-two and higher values reflect greater involvement in various acts of general delinquency. 135

PAGE 136

Drug use. This variable is a self-reported va riety index tapping into the total number of ten different drugs t hat were used in wave six. The drugs include: marijuana; inhalants; hallucinogens; coca ine; crack; heroin; pcp; tranquilizers; downers; and, uppers. Scores on this measur e range from zero to ten with higher scores indicative of higher drug use variety during wave six. Aggression. Aggression is measured with elev en items from the Child Behavior Checklist which was administered to the parents during wave six (Achenbach, 1991). This measure assesses general aggressive tendenc ies in youth but also includes some items that reflect hyperactivity and impulsive ness. Primary caregivers of subjects reported whether G2 had done an y of the following: been unable to sit still or been restless or hyperactive; been cruel to animals ; been cruel, bullying, or mean to others; demanded a lot of attention; felt others were out to get hi m; been impulsive or acted without thinking; screamed a lo t; been stubborn, sullen or i rritable; had sudden changes in mood or feelings; had te mper tantrums or a hot temper; and, threatened people. Response options included never, sometim es, and often. Higher scores on this measure indicate greater aggression. Prior official intervention Individuals with a history of justice system contact will be more likely than those without it to experience it again. This is, in part, because justice system contact is an official measure of delinquency and reflects in volvement in crime. Additionally, when an individual is arrested or an official po lice contact has been made, the individual may be integrated into offender databases either formally or informally and may be suspected of committing other crimes in wh ich officers do not have immediate leads. For these reasons, it is imperative to contro l for prior official intervention experiences. 136

PAGE 137

Previous contact or arrest. The variable is measured using official records data from the Rochester Police D epartment and County Wide Regist ration. The measure is a dichotomous indicator of whether an indi vidual experienced one or more arrests or police contacts during waves one through six. That is, the variable was coded such that 0=no previous contact or arrest and 1=pr evious contact or arrest. Descriptive Statistics Table 6-2 contains descriptive statistics of the violent crime and official intervention measures as well as the covariates used to predict the probability of receiving an official intervention treatment. For each variabl e measured at the interval level of measurement, the mean, standar d deviation, minimum and maximum are reported. For dichotomous variables, the standard deviation is not reported. In this case, of course, the mean can be interpreted as the percentage of individuals in the analysis sample who scored on the dummy variable. In a ddition to the standard descriptive statistics that are reported, the last two columns show the number of valid cases for variables with missing information and the wave(s) in which the variable was measured, respectively. When a specific value of N is not given for a variable, it has no missing information in the analysis sample (N= 595). The discussion below focuses on describing the typical or average male in the sample as well as highlights the prevalence of dichotomous measures. While measures of dispersi on are not explicitly discussed in most cases, it should be kept in mind that there tended to be substantial variation in the sample acro ss most of these measures as indicated by the standard deviation as well as the minimum and maximum scores (see Table 6-2). With respect to basic demographics, t he analysis sample is comprised of 63% Blacks, 18% Hispanics, and 19% Whites, which is very similar to the original RYDS 137

PAGE 138

male subsample. Nearly one-third (30%) of respondents families are below the federal poverty line. And, the typical youth is roughl y 16 years of age at the time of the wave six interview. The average respondent resides in an area with heavy concentrations of minorities and relatively high levels of poverty. Specifically, the residents in a typical census tract are 53% African American and 33 % of a typical tracts residents are in poverty. The neighborhood arrest rate ranges from approximately no arre sts to about eight arrests with an average of 4.12 arrests. The av erage neighborhood is not especially wellintegrated with an average neighborhood integration score equal to 2.19. On average, neighbors tend to interact and do favors fo r one another more or less seldom. Nevertheless, the 2.88 average neighborhood satisfaction score reported by the primary caregiver is close to the scale anchor of satisfied. The mean neighborhood disorganization score is 1.64. Thus, the va rious physical and social disorders in a runof-the-mill neighborhood are clos er to being somewhat of a problem than being not a problem at all. What is clear is th at the typical neighbor hood is not plagued with neighborhood disorganization as being a large problem. Approximately 33% of yout h respondents live with both par ents. Despite the fact that many of these adolescent s live in a fractured home environment, the typical young male experiences fairly high attachment to his primary caregiver and reports high levels of parental supervision. On average, the primary caregive r reports relatively similar levels of attachment to the youth as the average youth does. The individual most responsible for raising G2 tends to be somewhat consistent in their discipline of G2. On 138

PAGE 139

average, he or she deviates from consistent punishment patterns closer to seldom than often. The typical respondent has fairly high aspi rations for college; a mean score of 3.34 is a fair amount above the scale anchor corre sponding to important. For the sample as a whole, these high hopes for college are coupled with moderately-strong school commitment (mean=3.08). Howe ver, the average male in the sample engages in prosocial activities and school clubs with friends closer to seldom than often. Interestingly, G2 reports higher levels of educational expectations for themselves than G1 does of them. Over one-quarter of the males in the samp le (27%) were involved in a gang prior to the treatment period. Ov erall, 46% percent of the males are dating during wave six and 34% had engaged in sexual intercourse prio r to the age of fifteen. On average, respondents have a small proportion of thei r friends engaging in delinquency and drug use. Likewise, the typical respondents peer group has relatively low delinquent values. Turning attention to values and mental st ates, the average male seldom experienced depressive symptoms and had a pretty good self-i mage overall. Similar to ones peer group, male respondents had relatively low le vels of delinquent values on average. With respect to violent criminal behavior prior to wave seven, the average violent crime score was 0.56 during wave two and this value steadily decreased to a low of 0.36 by wave six. In wave six, the mean general delinquency score was 1.38 and the average drug use score was only 0.15. The most delinquent indi vidual engaged in 19 distinct acts of general delinquency and the individual with the greatest drug use variety used five different substances. On average, primary caregivers repor ted that their youth 139

PAGE 140

had relatively low levels of aggression. Justice system contact during waves one through six was not uncommon. Specifically, 34% of the males had at least one police contact or arrest during this time per iod as measured by official statistics. Table 6-3 takes a closer look at the percentage of individuals committing zero through six acts of violence prior to and following the official intervention treatment period. In wave two, almost 65% of t he sample committed zero acts of violence, approximately 21% committed one type of vi olent act, and approximately 14% of the sample committed two or more di stinct types of violent acts. In wave six, about 75% of the sample failed to commit any acts of violence, less than 17% committed one type of violent act, and about 8% of the sample engaged in two or more types of violence. In general, the sample as a whole was becoming less violent between waves two and six. More specifically, the proportion of individuals in each violent category (i.e., one to five distinct acts of violence) declined fairly systematically during the pre-treatment period whereas the proportion of individ uals in the non-violent category (i.e., zero distinct acts of violence) increased. With respect to t he outcomes, at wave nine approximately 85% of the sample did not commit any acts of violence and at wave ten about 81% of the sample refrained from engaging in violenc e. Those committing violence committed between one and five types of violent acts in wave nine and between one and four types of violence in wave ten. Despite the better half of the sample being non-violent in a given wave both prior to the official in tervention period and dur ing the short-run and long-run follow up periods, a sizeable proporti on of the sample engaged in one type of violence and, across all waves, some indi viduals committed several distinct acts of violence. As a final point of considerat ion regarding the samples violent offending 140

PAGE 141

behavior, the maximum score was at least four across all wave s indicating that at least one male committed four or more distinct types of violence at each point of observation. Regarding the violent behaviora l outcomes, the mean val ues for violent crime in the shortand long-runs are 0. 21 and 0.26, respectively (see Table 6-2). With respect to the official intervention treatments, appr oximately 22% of the sample experienced one or more arrests as measured by offici al records whereas about 32% experienced one or more police contacts as measured by self-reports. That is, of the 595 males in the analysis sample, 131 persons had been arrested between waves seven and eight and 188 individuals reported havi ng had a police contact during this same time. On the whole, justice system contact is not a rare event which, of course, is a necessary precondition of any data used to estimate treat ment effects of official intervention across the entire sample (Chapter 8 explores whet her there also is a nontrivial proportion of individuals who have experienced justice system contact in each violent offending trajectory group specifically). Supplementing Table 6-2, T able 6-4 reports the descripti ve statistics for variables that had missing data following t he multiple imputation proc edure. The means reported in the table are the average of the five mu ltiple imputed data sets. It should be noted that the means and standard devia tions here closely mirror those reported for the valid cases (cf. Tables 6-2 and 6-3). Solving the problem of missing data, the multiple imputation procedure has had little to no effe ct on the measures of central tendency and dispersion for the various covariates. With an understanding of the RYDS data, the analysis sample used in the current study, th e measurement of the key constructs as well as a sense of the descriptive statistics, the research methodology used to assess 141

PAGE 142

the effects of official in tervention experiences on subsequent violent offending across different trajectory groups can now be discussed in detail. 142

PAGE 143

Table 6-1. Comparison between analysis sample and individuals excluded from the analysis sample across valid cases fo r key demographic factors and violent behavior at wave one. Mean Analysis Sample Mean Excluded Group p-value Race Black* 0.625 0.686 0.182 Hispanic* 0.180 0.179 0.984 Family poverty level* 0.290 0.295 0.933 Proportion African Am erican 0.531 0.544 0.587 Proportion in poverty 0.333 0.326 0.614 Violent crime# 0.443 0.577 0.240 Notes: Chi-square significance test. # Wilcoxon-Mann-Whitney significance test. 143

PAGE 144

Table 6-2. Descriptive statistics for valid cases in the analysis sample (N=595). Mean SD Min Max N Wave Violent crime outcomes Short-run 0.21 0.59 0.00 1.00 --9 Long-run 0.26 0.63 0.00 1.00 --10 Official Intervention Officially recorded arrest 0.22 --0.00 1.00 --7-8 Self-reported police contact 0.32 --0.00 1.00 --7-8 Covariates Demographics Race Black 0.63 --0.00 1.00 --1 Hispanic 0.18 --0.00 1.00 --1 Age 16.47 0.81 13.90 18.20 --6 Family poverty 0.30 --0.00 1.00 550 4 Neighborhood characteristics Proportion African Amer. 0.53 0.27 0.00 0.99 --1990C Proportion in poverty 0.33 0.14 0.00 0.56 --1990C Neighborhood disorgani zation 1.64 0.64 1.00 3.00 502 6 Neighborhood integr ation 2.19 0.68 1.00 4.00 557 5 Neighborhood satisf action 2.88 0.66 1.00 4.00 558 5 Neighborhood arrest rate 4.01 1.98 0.12 7.87 --A Family Family structure 0.33 --0.00 1.00 --6 Parental supervi sion 3.52 0.46 1.25 4.00 576 6 Attachment to parent 3.39 0.43 1.36 4.00 572 6 PCs edu. expectation of child 2.33 0.83 1.00 3.00 515 6 Attachment to child 3.48 0.46 1.73 4.00 535 6 Consistency in discipline 2.36 0.52 1.00 4.00 562 6 School Factors Commitment to school 3.08 0.38 1.10 4.00 529 6 Aspiration for college 3.34 0.86 1.00 4.00 --6 Prosocial activi ties 1.68 0.66 1.00 3.80 --6 Educational expectations 2.60 0.73 1.00 3.00 591 6 School clubs with friends 1.23 0.57 1.00 4.00 --6 Peer Associations Peer delinquen cy 1.32 0.51 1.00 4.00 558 6 Peer drug use 1.41 0.53 1.00 3.50 571 6 Peer delinquent va lues 1.31 0.42 1.00 3.00 591 6 Dating 0.46 --0.00 1.00 --6 Sexual activity under 15 0.34 --0.00 1.00 ----Risky time with friends 2.16 0.71 1.00 4.67 559 6 Gang involvement 0.27 --0.00 1.00 --2-6 Values and Mental States Self-image 3.21 0.42 2.22 4.00 --6 Depression 1.98 0.47 1.00 3.57 --6 144

PAGE 145

Table 6-2. Continued Mean SD Min Max N Wave Delinquent val ues 1.40 0.43 1.00 3.20 593 6 Prior criminal behavior Violent crime (Tim e 1) 0.56 0.91 0.00 4.00 579 2 Violent crime (T ime 2) 0.48 0.87 0.00 5.00 583 3 Violent crime (T ime 3) 0.47 0.85 0.00 4.00 588 4 Violent crime (T ime 4) 0.37 0.72 0.00 4.00 587 5 Violent crime (T ime 5) 0.36 0.73 0.00 5.00 --6 General delinq uency 1.38 2.32 0.00 19.00 --6 Drug use 0.15 0.49 0.00 5.00 --6 Aggression 0.38 0.35 0.00 1.83 533 6 Prior Justice System Contact Previous contact or arrest 0.34 --0.00 1.00 --1-6 Notes: N values only reported for variables that hav e missing cases. If no N value is reported N=595. 1990C=1990 United States Census for Monroe County. Table 6-3. Number of males committing zero through six distinct acts of violent crime before and after the treatment period. Number of distinct acts of violent crime 01234 56 Violent crime trajectory Time 1 (wave 2) 375 (64.77) 121 (20.90) 49 (8.46) 29 (5.01) 5 (0.86) 0 (0.00) 0 (0.00) Time 2 (wave 3) 407 (69.81) 107 (18.35) 41 (7.03) 22 (3.77) 5 (0.86) 1 (0.17) 0 (0.00) Time 3 (wave 4) 412 (70.07) 114 (19.39) 30 (5.10) 27 (4.59) 5 (0.85) 0 (0.00) 0 (0.00) Time 4 (wave 5) 435 (74.11) 106 (18.06) 32 (5.45) 11 (1.87) 3 (0.51) 0 (0.00) 0 (0.00) Time 5 (wave 6) 448 (75.29) 100 (16.81) 30 (5.04) 15 (2.52) 1 (0.17) 1 (0.17) 0 (0.00) Violent crime outcomes Short-run (wave 9) 506 (85.04) 66 (11.09) 16 (2.69) 3 (0.50) 3 (0.50) 1 (0.17) 0 (0.00) Long-run (wave 10) 483 (81.18) 81 (13.61) 20 (3.36) 9 (1.51) 2 (0.34) 0 (0.00) 0 (0.00) Notes: Percentages (%) reported in parentheses. 145

PAGE 146

Table 6-4. Descriptive statistics fo r variables with imputed data (N=595). Mean SD Min Max Demographics Family pover ty 0.30 0.46 0.00 1.00 Neighborhood characteristics Neighborhood disor ganization 1.67 0.64 1.00 3.00 Neighborhood inte gration 2.20 0.67 1.00 4.00 Neighborhood sati sfaction 2.88 0.66 1.00 4.00 Family Parental super vision 3.52 0.46 1.25 4.00 Attachment to parent 3.38 0.44 1.00 4.00 PCs edu. expect. of child 2.29 0.82 1.00 3.00 Attachment to child 3.47 0.46 1.73 4.00 Consistency in discipline 2.35 0.52 1.00 4.00 School Factors Commitment to school 3.05 0.39 1.10 4.00 Educational expectations 2.60 0.74 1.00 3.07 Peer Associations Peer delinque ncy 1.32 0.51 1.00 4.00 Peer drug use 1.41 0.52 1.00 3.50 Peer delinquent values 1.31 0.42 1.00 3.00 Risky time with friends 2.17 0.71 1.00 4.67 Values Delinquent va lues 1.40 0.43 0.82 3.20 Prior criminal behavior Violent crime (T ime 1) 0.57 0.91 0.00 4.00 Violent crime (Time 2) 0.48 0.87 0.00 5.00 Violent crime (Time 3) 0.47 0.85 0.00 4.00 Violent crime (Time 4) 0.37 0.72 0.00 4.00 Aggression 0.38 0.34 0.00 1.83 Notes: Mean and SD are averages of the five multiple imputed data sets. Min and Max are minimums and maximums across all five of the data sets. 146

PAGE 147

CHAPTER 7 METHODOLOGY Chapter 7 lays out the integrated met hodology employed to determine whether there are differential sanction effects for violent offending subpopulations. Given the advanced nature of the integrat ed methodology, a separate treatm ent of each of the two statistical methods is offered. The latent class growth analysis approach to classifying individuals into distinct latent offending su bpopulations is first reviewed. Subsequently, the propensity score matching approach to dealing with the issue of selection artifacts is explained. The two methodologies are then discussed together to illustrate how this integration is a useful approach for assessing o fficial intervention as a turning point in a criminal trajectory for divergent violent offending subpopulations. Finally, the analytic strategy followed in the current study is summarized. Method Explicitly addressing the issues of offender classification and se lection effects that were identified as key problems with prior re search, this section details the Latent Class Growth AnalysisPropensity Score Matc hing (LCGA-PSM) integrated methodology employed to test the research hypotheses The latent class growth analysis and propensity score matching analysis sections are each subdivided into three topics: conceptual overview, technical details, and m odel estimation. In this way, pertinent information regarding the motivation for the te chnique and the basic ideas that underlie it are first articulated, which can be us ed to facilitate comprehension of the key equations and the model estimati on procedures. 147

PAGE 148

Latent Class Growth Analysis Conceptual overview There has been a growing interest in studying criminal behavior in a longitudinal framework that stretches well beyond using l ongitudinal data to establish one of the three classical requirements for causal in ferencetime-order. Longitudinal growth modeling permits an assessment of intra-individual change in behavior or the modeling of individual-level heterogeneity in behavioral trajectories over an extended period of time. Several methodological approaches to modeling behavior over time have been advanced, two of which include hierarchic al linear modeling (Bryk & Raudenbush, 1987; Goldstein, 1995; Raudenbush & Bryk, 2002) and latent gr owth analysis (McArdle & Epstein, 1987; Meredith & Tisak, 1990; Muthn, 1989; Willett & Sayer, 1994).21 Unlike wave to wave regression, which yields esti mates of relative change in a sample over time (i.e., similarity of the rank ordering of subjects from one wave to another), these approaches capture absolute change in offending behavior within the individual. As a result, they have been extremely influential in helping to study the initial status and trends of criminal trajectories as well as the relationships between initial status and trends and factors that infl uence these parameters. While informative, these approaches are not without specific limitations. Most notably, both make the assumption that a continuous distributi on function underlies the data. In other words, hier archical linear modeling and la tent growth analysis assume everyones trajectory follows the same general path through time albeit there is variation around the populations average trajectory which is typically in the form of a 21 One key difference between HLM and latent growth analysis is that the former assumes observed variables and the later assumes latent variables. 148

PAGE 149

multivariate normal distribution (Nagin, 2005, p. 25). However, a variety of disciplines including criminology have theories which suggest that distinct developmental groups may exist and, some claim, that the effects of certai n life transitions embedded within trajectories may result in different outcomes across di vergent developmental groups (Caspi, 1998; Cloninger, 1986; Kandel, 1975; Loeber, 1991; Moffitt, 1993; Patterson, DeBaryshe & Ramsey, 1989). The assumption of a continuous distribution function made in hierarchical linear m odeling and latent growth analysis is quite restrictive and shuts the door to empirical investigations into phenomena that are of important substantive interest to some devel opmental theories of crime. In response to this limitation of traditional growth modeling approaches, a third methodological technique aimed at modeling individual-level heterogeneity in behavioral trajectories has been gaining traction in recent years. Known both as semi-parametric group-based modeling (Nagin & Land, 1993; N agin, 2005) and latent class growth analysis (Muthn & Muthn 1998-2010), this technique assumes the presence of a latent categorical variable that differentiates subgroups of individuals, which may have distinct etiologies as specified by tax onomic theories of offending (e.g., see Moffitt, 1993).22 When relaxing the assumption of a c ontinuous distribution function, there may be natural groupings of like trajectories that readily emerge from the data. That is, the variation around the population s average trajectory that is apparent with traditional growth modeling methods may constitute meaningfu l clusters of individual trajectories. Given the idea that groups of individuals may be following along distinct offending trajectories, a problem arises regarding how to classify individuals into developmental 22 Other common terms used to refer to the technique include trajectory analysis, finite mixture modeling and group-based modeling. 149

PAGE 150

groups. As previously discussed, subjective classification procedures may identify groups as distinct but represent only random va riation, may fail to identify meaningful groups, and may yield unreliable statistical test s as a result of uncertainties of group assignments (Nagin, 2005). Additionally, with subjective classification, one cannot determine whether developmentally distinct trajectory groups may exist outside of their assumed existence (Nagin, 2005).23 To complicate matters, la tent trajectory groups are, by definition, not directly observable. If they were known, standard methods would be sufficient for estimating parameters across the different classes of offenders. To overcome these problems, Nagin and Land (1993) pioneered an innovative technique that empirically classifies individuals re trospectively based on existing longitudinal observations of behavior (see also Nagin, 2005).24 In short, while the mathematics behind the methodology are fairly involved, the principal idea is that the number of trajectory groups as well as the propor tion of individuals within each group and the growth characteristics of the distinct devel opmental trajectories can be empirically determined using longitudinal data (Nagin, 2005). The technique permits the best of both worlds; it provides a way to assess withi n-individual changes in offending over time 23 As a counterpoint, some argue that latent clas s growth analysis or semi-parametric group-based modeling cannot test for the existence of groups (see Skardhamar, 2010). 24 An alternative approach to latent class growth analysis is generalized growth mixture modeling. Some discussion of why this approach is not employed here is warranted. Unlike latent class growth analysis, generalized growth mixture modeling allows variation around the means of the trajectory groups (Muthn, 2001). Then, growth mixture modeling is a hybrid of the group-based approach and traditional growth modeling approaches and typically results in the extr action of fewer groups to adequately model the data. However, Nagin (2005) notes this approach has a number of important drawbacks that limit its utility. First, modeling both between-group and within-group variat ion provides for a highly complicated statistical model which limits its applicability to model count and other types of data. Second, modeling within-group variation leads to an important conceptual dilemma, since it may become unclear what exactly constitutes a group. This is perhaps the most important limitation. Several developmental theories suggest there are fundamentally different types of offenders and generalized growth mixture modeling may only muddy, not clear, the waters. Third, the layering of heterogenei ty which is characteristic of the general growth mixture modeling approach creates problems with model identification. Fourth, generalized growth mixture modeling may create false groups due to th ese model identification issues (see also Bauer & Curran, 2003; 2004). 150

PAGE 151

while simultaneously capturing important betw een-individual differences in the intercept and growth factors of differ ent trajectories. Technical details While group membership is not directly observable, the num ber of trajectory groups can be ascertained from the data and distinct trajectories can be captured with unique polynomial functions of time (Nagi n, 2005). I begin with an explanation of the general base model and focus on acts of viol ent crime as the behavioral trajectory of interest. The base model has the following formula: )( )(i J j j j iYPYP, where is ones sequence of vi olent crime over time, iYj is the probabilit y of a random individual in the population bel onging to trajectory group j is the probability of given ones group membership j The summation indicates that the J conditional likelihood functions,, require aggregation to result in the unconditional probability of the data. Thus, on t he left side of the equation, is the unconditional probability of observing individual i s sequence of violent crime ov er time (Nagin, 2005, p. 25).)(i jYPiY )(i jYP )iY ( P25 The general likelihood of the entire sample (N) can be found as the product of the individual likelihood functions as shown: 25 For a given group, there is the assumption of co nditional independence meaning that ones behavior at time t is independent of ones prior behavior. In other words, for individuals in the same group, their behaviors are assumed not to be serially correlated (i.e. individual level trends from the group trend are uncorrelated). However, there will be serial correlat ion elsewhere such that the prior behaviors will be correlated with future behaviors in the general population (Nagin, 2005, p. 26). The conditional independence assumption is not unique to LCGA. HLM makes the same assumption of conditional independence but does so at the individual level. LCGA makes this assumption at the group level (Nagin, 2005, pp. 26-27). 151

PAGE 152

)(i NYPL The base model can be easily adapted fo r censored normal, binary, and count data. I skip ahead and specify the equations relevant to c ount data since the current study is measuring violent crime trajectories of the count variety. The formula for the adapted base model for count data (Poiss on distribution) is shown below: )(it it jt it jy e ypjit ,...)2,1,0( itywhere, is the probability of one having committed any non-negative integer number of violent acts at a given time, is the mean event rate of violent crime for all individuals belonging to group j at time t is the expected number of violent crimes for all individuals belonging to group j at time t Thus, the model assigns a probability to each possible outcome such that there is a probability of committing zero acts, a probability of committing one act, a probability of committing two acts, so on and so forth. Importantly, the pr obability assigned to each spec ific outcome possibility ( ) is dependent upon (Nagin, 2005, p. 32).)(it jypjtji)(it jypjt26 26 For example, if is equal to one, then there will be a relative ly high probability for the occurrences of zero or one acts of violence, lo wer probability of the occurrences of two acts, still lower probability of occurrence of three acts, so on and so forth. If is higher and equal to say five, there will be a relatively low probability of occurrence for zero acts of violence, an increasing probability up through five acts of violence (with the probability peaking at five), and then decreasing probabilities past five acts of violence. Thus, when for each group becomes sufficiently large, results from the Poisson model and the censored normal model will converge. 152

PAGE 153

In the basic Poisson model, the link f unction connects the event rate for all individuals with time by assuming varies with time as suggested by the following formula: jt3 3 2 2 10)ln(it j it j it jj jtTTT, where and are time, squared time, and cubed time and ,, and are the four estimated parameters th at indicate the level and trend in offending behavior for each group over time.itT2itT3itTj0j1j2j327 In other words, these param eters determine the shape of the trajectories. One can also estimate linear or quadratic forms of the model by not including and modeling squared and cubed time or just cubed time, respectively. The full model is discussed here for the sake of completeness but one should keep in mind that an all-quadratic model is used to identify the latent classes of violent offenders in this study. The quadratic model is more parsimonious than the general cubic model and is a logical choice to capture changes change in trajectories over time in this study; more specifically, the trajectories ar e estimated only over a few years during adolescence and the quadratic term is sufficient to model meaningful changes in violent crime among the offending su bpopulations during this ti me period. To those with experience modeling within-individual change in a hierarchical linear modeling framework, t he above equation should share many similarities to the analysis of population average traj ectories of count data. A key difference, however, is 27 It is worth noting that the modeling of as opposed to is a matter of convenience as it ensures that the algorithm used to estimate maxi mum likelihood estimates cannot result in negative values of Specifically, taking the exponential of log of t he event rate for individuals in group j at time t (i.e., e) will always yield a value greater than zero. )ln(jtjtjt)jtln( 153

PAGE 154

that there are J sets of the parameters estimat ed for the model. Thus, for each trajectory group there are four parameters that are es timated that correspond not to overall population averages but to group-specific averages. To understand the meaning of these parameters, it is helpful to consi der the parameter estima tes for a model with a two class solution ( J=2). For simplicity, I assume these two classes are non-offenders and adolescence-limited offenders. First, a group of non-offender s would have four parameter estimates that all failed to be significantly different from zero ( j =1). This is because non-offenders, by definition, have a violent crime trajectory which has an intercept equal to zero ( ), a slope that is not increasing or decreasing ( ), and no directional changes in behavior over time ( ; ). Second, another group akin to Moffitts (1993) adolescenc e-limited offenders would have several statistically significant parameter estimates ( j =2). Assuming the first point of data collection was after onset of offending, thes e offenders would have a significant positive initial status estimate ( ), a significant positive linear term ( ), but a significant negative quadratic term ( ).01 002 001 101 202 302 102 228 Importantly, the diffe rent parameters that are estimated for each group permit assessments of differences across groups in both the amount of violent crime and the development of this behavior over time (Nagin, 2005). For example, the level difference between a group of nonoffenders and a group of steady offenders would be important and so would the difference between a group of steady offenders and another group of desisting offenders. 28 The cubic term may or may not be significant depending on how bell shaped the offending behavior was over time. 154

PAGE 155

Model estimation Longitudinal data is necessary in order to employ LCGA procedures and, at minimum, three time points are necessary to es timate linear trajectories of violent crime. Additional time points are necessary if a re searcher seeks to measure changes in the direction of a trajectory (modeled by quadratic and cubic terms). Thus, a minimum of five waves of data are necessary to fully model levels and changes in offending trajectories. Given the current study is fo cusing on the effects of sanctions as turning points in the lives of individuals following al ong different trajectories (e.g., chronic offenders), modeling directional changes in offending traj ectories is important. However, given there are only five waves of data and the trajectories span just a couple of years, intra-individual change in viol ent offending can be adequately captured using linear and quadratic terms. The model search described below begins with a quadratic model rather than the more comp lex cubic model. Once the base model has been appropria tely adapted for ones data type (e.g., Poisson) and the number of parameters per group to be estimated have been determined (e.g., quadratic model), one esti mates several models each with a different number of latent classes specified (e.g., c=2, 3, 4). One must select among the different models estimated and various statis tics are employed for this purpose. The most common statistic to determine the number of trajectory groups is the Bayesian Information Criteria (BIC), which can be used to determine the number of groups in both nested and unnested models (D'Unger, Land, Mc Call, & Nagin, 1998; Nagin, 1999). BIC scores are calculated wit h the following equation: BIC = log (L) + log (n) k. 155

PAGE 156

where L denotes the ma ximum likelihood, n denotes the sample size, and k denotes the number of parameters. BICs exact a penalty for adding parameters thereby favoring parsimony (Jones, Nagin, & Roeder, 2001; Nagin, 2005).29 Following the selection of the model with the appropriate number of groups, the chosen model can be further refined in an optional second stage that elimi nates non-significant growth parameters. For instance, the violent offending pattern s of abstainers, by definition, can be adequately modeled with only an intercept since they have neither growth nor change in growth over time. Since BIC scores favor parsimony, models that successfully capture the intra-individual changes in the sample with the fewest number of parameters will be preferred. Upon settling on a final LCGA model, posterior probabilities of group assignment can be used as a way of assessing the relative precision of the best fitting model. When the mean posterior probabilities of group assignment are greater than the standard cutoff of 0.7, it indi cates that the model is ade quately precise in its group assignments (see Nagin, 2005). In addition, t he odds of correct classification (OCC) can be used to assess model precision. Val ues greater than five for each trajectory group indicate adequate class assignment (Nagi n, 2005). Additional details regarding assessing model precision are discussed during the presentati on of the results (Chapter 8). There are several important descriptive outputs immediately worthy of consideration once it has been determined t hat the best fitting model is adequately 29 Mplus BIC scores have positive values and the choice of the best model is given by the smallest BIC score. Mplus BIC scores = log (L ) + log ( n ) k Nagins (2005) BIC scores differ by a factor of negative two; thus, BIC = -2 log ( L ) + log (n) k 156

PAGE 157

precise in its group assignments. The number of distinct developmental trajectories that are extracted from t he data is obviously important. A nd, the proportion of individuals belonging to each identified subpopulation is al so very useful information. Finally, the statistical significance of t he parameter estimates and the shapes of each trajectory group should be explored. Visual representations of the distin ct trajectories can greatly facilitate group comparisons. These trajecto ry groups can then be used for various purposes, one of which is to estimate t reatment effects across the different subpopulations. Obtaining treatment effect s with observational data in general is the subject of the next sect ion. Propensity Score Matching Analysis Conceptual overview Centuries ago John Locke ( 1975 [1690]) stated: that which produces any simple or complex idea, we denote by the general na me cause, and that which is produced, effect (p. 324). While the meaning of caus e and effect is quite clear, establishing a causal relationship in social science has prov en to be most challenging. At the heart of methodological approaches to uncover causal relationships is an attempt to observe the counterfactual (i.e., what would have occu rred within an individual if he or she would have experienced a different treatment c ondition). Although the counterfactual framework is largely contributed to wor ks of Neyman (1923) and Rubin (1974; 1978; 1980; 1986), the importanc e of knowing what would have happened in the counterfactual case has a long history which c an be traced, at least, as far back as the time of Aristotle (Holland, 1986). In the counterfactual fr amework, each individual is thought to have both a possible outcome of re ceiving treatment and a possible outcome of not receiving treatment (Guo & Fraser, 2010). Naturall y, the measured outcome is 157

PAGE 158

always the possible outcome that correspond s to the treatment condition that was actually experienced. The c ounterfactual, or the possible outcome that corresponds to the treatment condition that was not exper ienced, can never truly be measured in actuality (Guo & Fraser, 2010). This consti tutes the fundamental problem of causal inference (Holland, 1986). T hen, the central task for all c ause-probing research is to create reasonable approximations to this physically impossible counterfactual (Shadish, Cook, & Campbe ll, 2002, p. 5). The scientific approach that is regarded as the gold-standard for causal investigations is the experimental design, which was first formally developed and implemented in the field of agriculture dur ing the 1920s (see Fisher 1925; 1926). The experimental design involves randomly assign ing individuals to either an experimental condition or a control condition. In theory, this reduces all differences between the groups with the exception of the treatment, which is left to the experimenter to manipulate. Hence, true random ization is the hallmark of experimental designs. While the counterfactual can never truly be observed and, consequentl y, a single individuals treatment effect will never be known, the counterfactual framew ork holds that the average treatment effect can be obtained by taking the difference in means between the experimental and control gr oups (Guo & Fraser, 2010). If the two groups are truly equal across all factors other than the tr eatment following randomization procedures, then the mean difference is the typical causal effect of the treatment on the outcome of interest. While experimental designs drive research efforts in the natural sciences, ethical and practical considerations c onstrain opportunities for experimental designs in social 158

PAGE 159

science. As a consequence, observational studies tend to dominate many social science fields including criminology. A num ber of methodological techniques (e.g., ordinary least squares regression, matching, stratification) have been advanced over the years to establish causal relationshi ps when experimental data is unattainable, impractical or, simply, unavailable (Guo & Fraser, 2010). Matching in observational studies is a popular methodological techni que because it closely mimics experimental designs; matching creates group equivalency on variables that are used explicitly in the matching procedure. For example, matched -pairs methodology employed to assess the effects of official intervention would involve taking one individual with certain characteristics (e.g., white, male, singleparent home) that was arrested and matching him or her to another indivi dual who has the same characteristics but did not get arrested. In this way, group equivalency is achieved on key confounders leaving the researcher to estimate the average treatment e ffect of arrest. Of course, the researcher is forced to make the assump tion that there ar e no additional variables that were not used in the matching procedure which would subs tantially bias the findings. The more relevant control variables one includes in t he matching procedure, the closer it becomes to a true experimental design and the more likely that the empirical findings are not subject to substantial amounts of hidden bias. The interaction of two key problems wit h traditional matching approaches limits their utility for establishing cause. First, the number of potentia l confounders that can be included in traditional matching procedures is limited. Taking the above example with only three binary covariates, there are eight (23) possible combinations of individuals; in this case, matching those arrested with those not arrested across these 159

PAGE 160

variables would probably be easily accomplished. However, things get considerably more complex if say there are just ten pot entially relevant control variables. The number of possible combinations of those variables jumps to 1,024 (210). Finding matches for individuals when only a handful of co variates are relevant can be difficult, if not impossible, and the complications gr ow exponentially wit h each added covariate under the matched-pairs method (see Rosenbaum, 2002). Second, when using observational data, the assumption of ignorable treatment assignment is likely to be violated. The assumption holds that, condi tional on covariates, the assignment of an individual to a treatment or control group is independent of the outco me (Guo & Fraser, 2010). If this assumption is violated, it m eans that some covariate is significantly related to treatment assignment. Taken toge ther, these two issues create a difficult dilemma. While it would appear wise for resear chers to account for as many covariates as possible to meet the assumption of ignorable treatment assignment, it becomes increasingly difficult to fi nd matched pairs as additional covariates are considered.30 Acknowledging the benefits of matched met hods but noting the practical limitations of such approaches, Rosenbaum and Rubin (1983) developed a matching technique which seeks to create covariate balance across a large number of variables by matching individuals on only one variable their propensity for treatment. They demonstrated that, in theory, individuals can be matched based on a single propensity score which can create covariate balance on a ll variables that are used to make up the 30 An additional assumption of propensity score analys is is the stable unit treatment value assumption (SUTVA), which is the a priori supposition that an outcome value for an individual assigned to a specific treatment is similar regardless of the mechanism used to assign treatment and irrespective of the type of treatments received by other individu als (Guo & Fraser, 2010). There is no reason to suspect that these assumptions have been grossly violat ed. Specifically, an individuals tr eatment status should not influence outcomes of other individuals via changes in social interactio ns or environmental conditions. 160

PAGE 161

propensity score (Rosenbaum & Rubin, 1983). True propensities for treatment cannot be directly known but they can be estimat ed using logistic regression methods. Interestingly, using propensity scores that are estimated has been shown to actually remove more bias than would be the case if true propensity scores could be used since the former removes both systematic and r andom biases whereas the latter only removes systematic imbalance (s ee Joffe & Rosenbaum, 1999). Technical details With an understanding of the logic behind the counterfactual framework and the problem propensity score matching attemp ts to solve, key equations and model estimation procedures are now discussed. Rosenbaum and Rubin (1983) defined the propensity score for individual i as the conditional probabi lity of assignment to a treatment ( ) versus non-treatment (1 Wi 0 Wi) given the vector of observed covariates ( ), which is represented in mathematical form as follows: ix )|1()(ii ixXWiprxe While the complex derivations and proofs are omitted here, is a balancing measure that summarizes the information available from the observed covariates. Rosenbaum and Rubin (1983; see also Rosenbaum, 2002) demonstrated that the propensity score itself can balance the differences across th e vector of observed covariates thereby creating equivalent treatm ent and control groups, across observed variables, with a single score. When individuals have the same propensity score, it does not automatically guarantee in prac tice that they will have the exact same set of observed covariates (e.g., male, White, single-parent). In fact, there will be instances in which two individuals with the same propensity scores do not have the same xi. Importantly, )(ixe 161

PAGE 162

however, any differences in xi between individuals with t he same propensity scores are chance differences as opposed to systematic differences (Rosenbaum, 2002). Guo and Fraser (2010) show that the assu mption of ignorable treatment can be linked to the propensity score with the following equation: )(|iiixewx This equation indicates that a treatment assignment ( ) is independent of the vector of observed covariates ( ), conditional on the propensity score ( ).iwix )(ixe31 There are two important implications of this equation. First, in dividuals with similar propensity scores should have a similar vector of observed covariates (less any chance differences), meaning that the distribution of the observed covariates will be similar across the treatment and control groups. Second, every individual has the same probability of treatment assignment c onditional on the propensity, mimicking randomization of treatment and control groups as in true experiments (Guo & Fraser, 2010). To assess the success of the propensity score matching procedure for creating covariate balance, standardized bias statisti cs are utilized. First, for the unmatched sample, the standardized bias statistic is defined as the absolute difference in covariate means (or proportions for dichotomous va riables) divided by a special standard deviation. Its formula is given by: xXp Xt XsMM d /| |*100 31 Notation: independent (); conditional (|) 162

PAGE 163

where denotes the standardized bias st atistic for some covariate X before matching, is the mean of the treatment group, and is the mean of t he group of potential controls before matching. is the special standard devia tion for continuous measures and is calculated as follows: XdXtMXpMXs 2/)(22Xp Xt Xsss, and, is calculated as follows for dichotomous measures that are coded in binary format: 2/))1()1((Xp Xp Xt Xt XMMMMs Thus, a standard bias statistic in this form t hat is equal to 30 signifies that the covariate means or proportions are out of balance by 30% of a standard deviation. The standard cutoff value for a covariate to be considered to be balanced is 20, which equates to a difference in covariate means or proportions less than or equal to 1/5 of a standard deviation (Haviland et al., 2007). In the ideal case, covariates would have values equal to 0 indicating perfect balance. In practi ce, however, excellent covariate balance is present when standardized bias statistics are below a value of 10. With matching procedures other than 1:1 pair matching, the formula for the postmatching standardized bias statistic is modified slightly to account for the fact that a treated individual is matched to one or more controls and/or a control individual is matched to one or more treated individuals. Specifically, while the general formula for the standardized bias statistic remains simila r, the means for the treated and/or control groups are not simple arithmetic means. Extending Hav iland and colleagues (2007; 2008) formula for 1:K variable matching to full matching and eliminating the stratification 163

PAGE 164

summation (as no stratification is conducted in this analysis) leads to the following formula: x Xcm Xtm XmsMM d /| |*100 where denotes the standardized bias st atistic for some covariate X after matching, is the unweighted mean of means for the treated matched to the controls, and is the unweighted mean of means of the controls matched to the treated. For the following notation, subscript i refers to treated participants and subscript j represents control participants. To ultimately obtain is first calculated which represents is the average of for all j controls matched to participant i The formula for is as follows: XmdXcm XtmM MXcmMciMcijXciM im j cij i ciX m M11 where is the value of covariate X post-matching for the jth control that matches to treated participant i These mean values, are then averaged across nt total treated individuals as the following formula illustrates: cijXciM tn i ci t XcmM n M11, The value for is calculated similarly for matc hing procedures, such as full optimal matching, that also have the pot ential for a variable number of treated individuals to be matched to a single control subject (e.g., see Figure 7-1). Analogous to the previous calculations, the formula for is given by: XtmMtjM 164

PAGE 165

im i tji j tjX m M11, where subscript i still refers to treated participants and subscript j still represents the control participants. is the value of covariate X post-matching for the ith treated that matches to control participant j Thus, is the average of for all i treated individuals matched to each control participant j is then averaged across nc total control individuals as shown: tjiXtjMtjiXtjM cn j cj c XtmM n M11. In the case of full matchi ng with variable treatment-c ontrol and control-treatment possibilities, the post-matching means and do not represent simple arithmetic means as do the pre-matching means and Still, they are easily calculated and directly comparable as they use the same special standard deviation ( ). XtmMXtMXcmMXpMXsPercentage bias reduction statistics are often calculated in conjunction to assess the extent to which the matching proc edure reduced the pre-existing covariate imbalances. With the standardized bias statistics calculated both preand postmatching, the percentage bias reduction ( %BR ) is simply calculated as follows: )/(1%XXmddBR Collectively, the covariate balance statis tics permit one to asse ss whether covariate balance has been achieved through propensity score matching for each covariate as measured by a standard metric (i.e., 0.20) as well as to specify exactly how much bias has been reduced. These analyses are noteworthy for their transparency and that Xmd 165

PAGE 166

they are conducted independent of, and prior to the stage of estimating the treatment effect. Recall that an individuals counterfactual can never truly be observed but that an average treatment effect can be obtained. When )(|iiixewx is true and balances the observed covariates as indicated by acceptable standardized bias statistics, the formula for the unbiased aver age treatment effect (ATE) at a given propensity score is: )(ixe )(ixe )](|([)]0),(|,([)1),(|,([01 0 1 i ii iixeYYEEWxeYEWxeYEE In other words, for all indi viduals with the same propensity score, the mean difference in outcome between the treatment and control groups serves as an unbiased estimate of ATE at that particular propensity score. And, in the case of propensity score matching, the mean of these propensity-specific matched differences is the unbiased estimate for the ATE as shown below: ATE =)0|()1|()](|([0 1 01 WYEWYExeYYEEi. With the technical details laid out, attention is now turned to model estimation procedures. Model estimation The first step in model estimation is the s pecification of the tr eatment model to be used to obtain the best estimates of the true propensity scores. Correct specification of a treatment model is necessary because e ffect estimates are dependent upon initial model specification (Rubin, 1997). Guo and Fraser (201 0) note that the correct specification of a treatment model involves obtaining propens ity scores that balance the two groups on the observed covariates. Includi ng theoretically relevant covariates and 166

PAGE 167

specifying the appropriate functional form of them, including polynomial and interaction terms as necessary, are the keys to succe ss (Guo & Fraser, 2010). Some researchers have casted a wide net identifying as many covariates as possible in the data (see Bingenheimer, Brennan, & Earls, 2005). However, having more covariates in the model than necessary can actually worsen probl ems with common support (Smith & Todd, 2005) and result in wider ranges for estimate d treatment effects (see Hirano & Imbens, 2001). Therefore, instead of including all pos sible variables in the data set, potentially relevant covariates with theoretical links to treatments and outcomes are included (see Chapter 6). Interactions terms are added to the treatment model if balance is not successfully achieved with a strictly additive treatment model. With estimated propensity scores, the matc hing procedure is now executed. The current study employs optimal matching procedures (see Haviland et al., 2007). Unlike greedy matching approaches such as nearest neighbor matching, optimal matching is not subject to the same pr oblems including common suppor t region sensitivity to the specifications of the treatment model and large case loss due to insufficient common support overlap (Guo & Fraser, 2010). With greedy matching, propensity score matched pairs are built by selecting a treated subject and finding the best available control subject that has not yet been used, under matching without replacement. Once the treated subject has been ma tched to the control subjec t that pair is eternally wedded. This type of matching is greedy in the sense that a matched pair being formed is the only concern at the time. T he procedure pays no mind t hat the ability to form other good matched pairs later may be gr eatly hampered by the current pairing. 167

PAGE 168

While greedy matching approaches pick the best pair at any given time, optimal matching is a smart process whereby matc hed pairs are built, and sometimes rebuilt, with the end goal of obtaining the lowest total distance between the treatment and control propensity scores. Pair s that are initially wedded can later be divorced from one another should the treated and control subjects be better paired with other individuals to best solve the optimization problem. T hus, optimal matching approaches are not greedy since they are concer ned with the overall gr oup of matched pairs rather than the current pairing. Within optimal matching, full matching is employed which matches each treated individual to one or more controls and each c ontrol individual to one or more treated. For the current study, full matching is critical as the small sample size across trajectory groups will need to be retained. An optimal full matching approach retains all individuals in the sample and matches treat ed and control subjects into matched sets which results in the best solution to the overall optimization pr oblem. It is not uncommon to have a single treated matched to a large number of c ontrol subjects or vice versa. While this procedure is a departure from 1:1 matching, it tends to result in substantial improvement in overall bias reduction especially when compared to 1:K or variable matching approaches (e.g., see G uo & Fraser, 2010). The matching procedure is executed in R using optmatch (for a detailed review, see Hansen, 2004; see also Guo & Fraser, 2010). Given full matching is a relatively new procedure, Figure 7-1 provides an illustration of how full matching differs from other more common approaches to matching (i.e., 1:1, 1:K). In 1:1 matchi ng, each treated subject is matched to exactly 168

PAGE 169

one control subject. While this approach under an optimal matching scheme tends to reduce substantial amounts of bi as, it does not use all of the available information. In 1:K matching, each treated subject is matched to a specific number of control subjects. In this example, 1:K matching happens to us e most of the available information given there are just slightly more t han twice the number of controls to treated subjects but it does so at a huge cost. Several of the matched pairs do not have similar propensity scores and bias reduction in this case will be less than ideal. While not illustrated, an alternative to 1:K matching is achieved by permitting K to vary (with or without restrictions). This is known as variable matching and is a matching strategy that does result in better bias reduction as compared to 1:K matching where K is fixed at some value. Full matching is an extension of th is 1:K variable matching scheme which also allows K:1 variable matching, typically resulting in the greatest level of bias reduction (see Guo & Fraser, 2010). As illustrated in Fi gure 7-1, full matching uses all available information and matches each treated subject to one or more cont rol subjects and/or matches each control subject to one or more treated subjects. In short, the example demonstrates that full matching results in s ubstantial bias reduction which was not the case with 1:K fixed matching and will use all available information which was a limitation of the 1:1 matching strategy in this example. The next step involves a reassessment of the differences between the treated and control groups along each of the covariates us ed to estimate the propensity scores. In the context of propensity score matching, Ha viland and colleagues (2008) note that it is a mistake to see the t statistic as a formal test of balance when applied to estimated quantities, such as the propensity score, wh ich were estimated from the data to 169

PAGE 170

discriminate treated and control groups (p. 431; see also Ho, Imai, King, & Stuart, 2007; Imai, King, & Stuart, 2006). A main reason t statistics do not constitute formal tests of covariate balance is because bal ance is a property not of the population but rather of the observed sample itself (Ho et al., 2007; Imai et al., 20 06). T-statistics are not directly comparable preand post-matching since these two samples are often of different sizes (Haviland et al., 2007). More importantly, t -tests are functions of remaining observations, variance of tr eated and control groups, and the ratio of remaining treated and control subjects. As such, a t -statistic can actually lead to a conclusion that balance has been achieved even though balance has really worsened in reality (Ho et al., 2007; Imai et al., 2006). In addition to these key problems with using t tests, the nature of hypothesis testing is such that it leaves researchers with the impression that balance has been achieved and that any remaining imbalance is simply ignorable; however, less bias is always better and this is especially so for any covariate that may be strongly related to the outcome of interest (Ho et al., 2007; Imai et al., 2006). As opposed to using t -statistics, the absolute standar dized difference in covariate means (dx) is directly compared to their corre sponding pre-matching values. In this way, one is able to determine how successful the optimal full matching procedure was in removing differences between the treated and control groups across the observed covariates. If this process is unsuccessf ul (i.e., there are s ubstantively meaningful differences between the treated and control groups across one or more observed covariates after matching), polynomial and in teraction terms are added to the treatment model and the process of optim al matching and reassessment will continue. Ultimately, 170

PAGE 171

the goal is to obtain two groups that lack differences with respect to all observed covariates. Once all differences have been adequately removed across all covariates, the final step in the model estimation procedure involv es the estimation and interpretation of the ATE. For this purpose, the Hodges-Lehmann aligned rank test is employed to estimate the ATE of the optimally and fully matched sa mple. In addition to the statistical significance of the Hodges-Lehmann aligned rank test, the absolute standardized difference in means is used which yields an es timate of effect si ze (Haviland et al., 2007). An LCGA-PSM Integrated Method Latent class growth analysis provides resear chers with an important tool to classify individuals into different developmental tr ajectories and model intra-individual change in behavior over time. Propensity score matchi ng analysis permits re searchers to create covariate balance on observed variables, to expl icitly assess the relative success of the matching procedures, and to estimate average treatment effects. Recently, Haviland and colleagues (Haviland & Nagin, 2005; Hav iland et al., 2007; 2008) noted the unique benefits of each of these two techniques and suggested that they c ould be integrated to address important questions that emerge from the life-course cr iminological literature. As has been thoroughly discussed in this paper, many life-course theories anticipate that the effects of treatments may depend on ones behavioral trajectory (e.g., Moffitt, 1993). The method is well-suited to answer w hether there are trajectory group-specific effects of certain treatments such as official intervention. Moreover, the use of latent class growth analysis and propensity score matching in conjunction helps to create balance both on prior outcome histor y and on other measured confounders, 171

PAGE 172

respectively, providing for a good estimation of the treatment effe ct with observational data (Haviland et al., 2007; 2008). The LCGA-PSM integrated method is ess entially the sequentia l employment of latent class growth analysis followed by propensity score matching within trajectory groups. Discussing the procedure, Hav iland and colleagues (2007) write: The integration of group-based trajec tory modeling and propensity scores is composed of a three-stage analysis. The first stage involves estimating a group-based trajectory model fo r the outcome and participants of interestIn the second stage, each tr eated individual is matched with one or more untreated individualsWe t hen check the degree of success of the matching strategy in achieving balance In the third stage of the analysis, the treatment effect of the event of interestis analyzedwithin and across trajectory groups. (p. 249). In short, two highly useful techniques in their own right are brought to bear on longitudinal data in a systematic and straightforward manner, which ultimately provides an ideal way to determine whether an offici al intervention experience serves as a turning point, in either direction, for one or more distinct violen t offending trajectory groups. Analytic Strategy Group based trajectory models are esti mated using a self-reported violent delinquency scale composed of count s of violent criminal acts. Specifically, a Poisson model is estimated and the number of violent offending trajectory groups is determined using the BIC score and other relevant selection criteria. Next, posteri or probabilities of group assignment and other common post-esti mation statistics are examined to determine the relative precision of the best fitting model. The number of trajectory groups, the proportion of individuals within each trajectory group, and the shapes of the offending trajectories are explored. To suppl ement this descriptive analysis, trajectory 172

PAGE 173

group-specific descriptive statistics ar e reported and one-way ANOVA models are estimated. This serves to determine whether any covariates are significantly related to trajectory group membership. With the best fitting LCGA model, indivi duals within each subpopulation are split into two groups: 1) those receiving official in tervention and 2) those not receiving official intervention. Covariate balance is asse ssed prior to matching using the absolute standardized difference in covariate means. Next, logistic regres sion models are used to estimate propensity scores for each indivi dual. Optimal full matching procedures are implemented and then covariate balance is r eassessed for each subpopulation to test the successfulness of the matching procedures. Changes to the treatment model are done as necessary and this cyclical proce ss is repeated until the best possible model has been estimated or no substantial bias remains. In additi on to carrying out propensity score matching within each trajectory group, it is also done for the sample as a whole. The integration of the two advanced quantit ative techniques culminates with the estimation of the treatment effe ct. More specifically, with tw o statistically similar groups within each subpopulation, statisti cal significance of the effect of official intervention on violent outcome behavior is calculated for each of the violent offending subpopulations using the Hodges and Lehmann (1962) aligned rank test. A dditionally, the absolute standardized difference in outcome means is reported as a measure of effect size. Similarly, the treatment effects are esti mated for longer-term out comes to determine whether there are differential effects over time This trajectory group-specific analysis is 173

PAGE 174

repeated for the other official intervention (e.g., arrest). Fo r general comparative purposes, the analysis is carried out on t he sample as a whole as well. 174

PAGE 175

175 Treated sub j ect Control sub j ect Matched p air 1:1 1:K Full matching ( K=2 ) Figure 7-1. Illustration of three alternative matching strategies.

PAGE 176

CHAPTER 8 RESULTS: VIOLENT CRIME TRAJECTORIES OF OFFENDING Chapter 8 presents the results of the Poisson LCGA model search for the appropriate solution and describes in detail characteristics of the trajectory groups that emerge from the selected model. First, standard model sele ction criteria are used to select the model with the appropr iate number of latent classes from five alternative Poisson quadratic LCGA models that have one through five latent classes, respectively. With the number of violent crime trajectory groups determined, the model is refined to determine the order of the polynomials nece ssary to capture the within-individual change for each violent offending trajectory group. Next, t he precision of trajectory group assignments is assessed using standard criteria including mean posterior probabilities, odds of correct classificati on, and correspondence between the proportion of individuals in each latent class bas ed upon the estimated model and the proportion based on maximum probability assignment. The parameter estimates of the selected model are then discussed and the shapes of the trajectories an d the proportion of individuals belonging to each are reviewed. To supplement this descriptive analysis, trajectory group-specific descriptive st atistics and results of one-way analyses of variance are reported. Finally, the trajecto ry group-specific treatment experiences and violent crime outcomes in the shor tand long-runs are explored. Model Selection Table 8-1 contains information on key model selection criteria for five different quadratic latent class growth models with lat ent classes ranging from one to five. As evidenced by the high Bayesian Information Criter ia (BIC), the one class model is by far the worst fitting model suggesting the populati on as a whole appears to be comprised of 176

PAGE 177

some meaningful subpopulations. The th ree and four class models have BIC scores that are relatively similar to one another yet substantivel y lower than the one, two, and five class models. The three class model has the lowest BIC and the Schwarz (1978) and Kass & Asswerman (1995) metric for model comparison suggest there is nearly a 72% chance that the three class model is t he correct model. There is approximately a 28% chance that the four class model is the correct model and essentially no chance that the one, two, or five class m odel accurately represents the data. It is worth noting that in addition to using m odel selection criteria such as the BIC, Nagin (2005, p. 75) states that selection must balanc e the sometimes competing objectives of model parsimony and capturing the distinct ive features of t he data. In the present study, parsimony is favored since t he identification of a large number of latent classes is of no utility if the effects of of ficial intervention cannot be ascertained within the trajectory groups. If th is is the case, the analysis moves from one of causal discovery to one that is quite descriptive in nature. The four class Poisson LCGA quadratic model that was esti mated essentially breaks the low classthat is in the three class model and will be discussed shortlyi nto two different latent classes. One of these classes is slightly trending upward a nd the other is slightly trending downward but, importantly, both are offendi ng at levels that are between the other two trajectories identified. Estimates of the precision of class assignments of the four class model are not all above conventional cutoffs though they are nearly so. In sum, the three class model is selected as the best fitting model given the BIC provides the strongest support for it and most of the key feat ures of the data are captured by it while keeping trajectory 177

PAGE 178

groups as large as possible.32 This latter point is important as studies have had difficulty matching within small trajec tory groups (Haviland et al., 2007). The selection of the number of latent classes is of pr imary concern for any latent class growth analysis but an opti onal second stage of model se lection can further refine the model by adjusting the num ber of free parameters necessary to model changes in offending over time across each of the latent classes. For instance, some trajectory groups which are stable across time (e.g., non-offenders) may adequately be captured with an intercept only and, as a consequence, estimated growth pa rameters for this class would be statistically insignificant. Simila rly, the trajectory of certain latent classes may be adequately captured with just a linear term Recall the BIC favors parsimony so the lowest values on these metrics woul d be expected for models with the fewest growth parameters that adequatel y represent all relevant within-individual change in violent crime over time. Indeed, when higher order pol ynomials are estimated and lower order polynomials are sufficient to descr ibe the level and trends in the trajectories, standard errors of the growth parameters and intercepts can be elevated potentially rendering certain parameter estimates insignifi cant (Nagin, 2005). Table 8-2 contains model selection criteria fo r a parsimonious three class Poisson LCGA model with the highest significant polynomials which are ne cessary to model the change exhibited in each specific violent crime trajectory (i.e., 2,0,2) and the all-quadratic trajectory model that was estimated for selection of the number of latent cla sses (i.e., 2,2,2). There is strong evidence in favor of the parsimonious three class model which restricts the order 32 While model selection criteria as we ll as practical considerations lead to the selection of the three class model, the differences in the two aforementioned gr oups in the four class model may be substantively important from a theoretical standpoint, a point which is discussed in some detail in the final chapter of the study. 178

PAGE 179

of polynomials to only those above the intercept that are significant at the highest order (p<0.05). Thus, the simpler model is sele cted to investigate the research hypotheses regarding differential sanction effects across violent offending subpopulations.33 Assessment of Model Adequacy Before interpreting the parameters and analyz ing the trajectories themselves, an important procedure in LCGA is assessing t he extent to which the selected model has accurately assigned individuals to specific latent classes. The LCGA model estimates a probability of belonging to each of the latent classes for ever y individual. That is, every individual receives three different estimat ed probabilities of group membership in which the total sums to one and individuals are assi gned to a particular viol ent crime trajectory based upon maximum probability assignment. For example, consider a hypothetical individual with probabilities of belonging to latent classes one through three equal to 0.8, 0.125, 0.075, respectively. This individ ual would be assigned to be a group member in latent class one based on maximum probability assignment since he or she had an 80% chance of belonging to that gr oup. Group mean posterior pr obabilities can be used as a measure of model adequacy; Table 8-3 cont ains the mean posterior probabilities for group assignment to the three violent crime trajectories (column) by the assigned group based on maximum probability assignment (row). For individuals who were actually assigned to class one, their mean posterior probabilities of belonging to classes one through three are 0.891, 0.000, and 0.109, respectively (see Table 8-3). It is to be expected that the highest average posterior probabilities for group assignment are for the groups to which individuals were actually assigned. This is the case here. In an 33 While this optional model refinement stage was underta ken, it is important to note that it did not have any influence on the proportions of individuals falling within each latent class. 179

PAGE 180

ideal world, values of one would grace the diagonal of the table signifying that individuals could be perfectly classified into di stinct trajectories. In practice, however, the ideal will not be achieved; the conventio nal cutoff for average posterior probabilities for group assignment across all groups has been set to 0.7 (Nagin, 2005). Focusing on the diagonal of Table 8-3, a ll of the three violent crime trajectory groups have values that exceed 0.7. In fact each of these values is at least above 0.8 indicating good model precisi on. Other statistics also suggest that individuals have indeed been assigned to violent crime trajectory groups with adequate precision. Table 8-4 contains values of the quantity j minus Pj, which provide information on the concordance of the proportion of individuals belonging to the latent classes based on estimated model ( j) and the proportion of individuals belonging to them based on maximum posterior probability assignment rule (Pj). This is a commonsensical comparison of the adequacy of a LCGA model If individuals were assigned to groups with absolute certainty, t he difference between these values would be zero. Importantly, this number will be small when ther e is low assignment error. Overall, there appears to be acceptable d egrees of concordance between j and Pj for each of the three latent classes. Specifically, t he proportion assigned to classes one and two is slightly smaller whereas the proportion assi gned to latent class three is somewhat larger. Table 8-4 also reports the odds of correct classification (OCCj), which offers another way of assessing the adequacy of trajectory group assignments. OCCj values that are larger signify better accuracy in class assignment. As a conventional cutoff, values exceeding five have been suggested to indicate high assignment accuracy (see 180

PAGE 181

Nagin, 2005, p. 89). For all violent crime trajectory groups, the OCCj exceeds this conventional cutoff also indicating good model adequacy.34 With the number of latent classes determined, the order of polynomials necessary to capture within-individual change in violence across the trajectory groups specified, and t he demonstration that the parsimonious three class Poisson LC GA model is adequately precise in its classification of individuals into subpopulations, attention is now turned to describing the relevant features of the esti mated violent crime trajectori es for the three distinct gr oups. Trajectory Group Characteristics Figure 8-1 plots the model estimated expect ed counts of distinct violent acts for the total sample. The total sample average trajectory is slightly declining in a linear fashion from waves two through six. While this graphic of a simple growth curve model informs the average trend of the aggregated sample, it wa s noted earlier that several model selection criteria suggested meaningful subpopulat ions exist in the data; specifically, the strongest ev idence supported three distinct violent crime trajectory groups. Figure 8-2 plots the model estimated expect ed counts of distinct violent acts for the trajectory groups across waves two th rough six, permitting a concise graphical overview of how violent crime unfolds over ti me for the different trajectory groups and aiding in the assignment of m eaningful group names. From th is point forward, latent classes one through three will be referred to as high offenders, non-offenders, and low offenders, respectively. An initial observati on is that there is evidence of betweenindividual stability in offending behavior ov er time. Across all of the waves, non34 The OCCj is calculated as follows: [ pp / (1 pp)] / [j / (1 j)]. 181

PAGE 182

offenders have the lowest expected counts and high-chronic offenders have the highest expected counts. The violent cr ime trajectory of the low offenders is also sandwiched between high offenders and non-offenders across all waves of data. With respect to within-individual stability in violent crime over time, the trajectory groups are marked with a mixtur e of within-individual stab ility and within-individual change. One class has strong stability in t he lack of violent criminal behavior. Nonoffenders consistently have essentially zero expected counts of offending across all waves. The low trajectory group has an ex pected count of approxim ately 0.8 at wave two but by wave six the expected count drops to about 0.4. This decline does not occur in a linear fashion. Specifically, the expe cted counts for the low offending group drop during the first couple of time points and then stabilize. The high offending group exhibits the highest expected counts across all waves but its offending trajectory follows an inverted parabolic shape. That is, the leve l of violence of this group rises slightly from wave two to wave three, peaks at appr oximately 1.8 acts of violence during wave three, and then decreases to approximat ely 1.2 acts by wave six. Table 8-5 summarizes the total number of individuals and proportion of individuals in each violent crime trajectory group. Just under half of the sample is classified as low offenders and nearly forty percent are classified as non-offenders. The high trajectory group that emerged from the data is smaller, comprising approximately eleven percent of the sample. Importantly, each subpopula tion identified by the LCGA procedure contains a fairly sizeable proportion of the sample which is a nece ssary prerequisite to investigating treatment effect s of official intervention ac ross distinct violent crime trajectories. At the same ti me, the high trajectory group only has 68 males in it leading 182

PAGE 183

to some concern over the size of this gr oup. Recall, that matching within trajectory groups becomes more difficult for smaller subpopulations (Hav iland et al., 2007). Supplementing the graphical di splay of the violent crime trajectories (see Figure 82), Table 8-6 contains the parameter esti mates, standard errors, t-statistics, and pvalues for the intercept and growth factors fo r the three latent cl asses. Non-offenders (class two) have their trajectory characte rized by an intercept only. High and low offenders (classes one and three) have their tr ajectories captured with an intercept as well as both linear and quadratic terms. For all groups, the intercepts are statistically significant. This parameter estimate is not especially important as it only indicates whether the expected counts of violent acts during wave two is significantly different from one. More meaningful are the size and direction of statistically significant parameter estimates for t he linear and quadratic terms which were needed to describe within-individual change in violent offendi ng behavior for the high and low offending groups. High offenders have a statistically signifi cant linear term that is positive which indicates that the log of the expected c ounts of violent crime increases by 0.175 (p<0.01). Of course, this growth is not li near given the significant quadratic term which is negative (p<0.01). The low group has a statistically significant linear term that is negative indicating the log of the expected counts of violent cr ime decreases 0.415 (p<0.01). This decline is al so not linear, however, as indica ted by the positive quadratic term for low offenders in this ca se (p<0.01). In short, the sizes and directions of the linear and quadratic terms for the high and low o ffending trajectories are responsible for their observed inverse parabolic and j-shaped curves (see Figure 8-2). 183

PAGE 184

In sum, a parsimonious three class Poi sson LCGA model adequ ately captures the data. The trajectories that emerged from the data provide some important differences with respect to between-individual and withinindividual stability and change in violence over time. And, these distinct patterns of offending that were identified through empirical means will prove useful in investigat ing the extent to which official sanctions make matters better or worse for indi viduals following along distinct offending trajectories. Before asse ssing official intervention effects on subsequent violent behavior, however, trajectory group-specific descriptive statistics are reported. Trajectory Group-specific Descriptive Statistics Chapter 6 reported the descriptive stat istics for the aggregated sample. This section reports descriptive statistics for each of the three violent off ending trajectories. In addition, one-way ANOVAs are estimated wh ich assess whether t here are significant differences between the means of the forty covariates across the distinct trajectory groups. It is important to point out here t hat many of the covari ates are measured at wave six, which is during the final wave of the violent crime trajec tory. Therefore, any significant differences in covariate means across the trajectory groups should not be construed as risk factors for trajectory grou p membership. This is especially the case with respect to time-varying variables, which comprise the overwhelming majority of the covariates. Moreover, it is probably best not to think of thes e covariates as outcomes of ones violent crime trajectory. Rather than focusing on a caus al connection, it might be preferable to view these associations as supplementing general descriptions of the trajectory groups. Table 8-7 contains the means for the covari ates measured at the interval level of measurement and the means for binary covariates across the three trajectory groups. 184

PAGE 185

In addition, the table contains results of an overall F-test for mean differences and Tukeys statistical tests for specific group comparisons. There are a couple of notable differenc es between the groups with respect to demographics. The high offenders (class one) have a greater proportion of Blacks and are older on average than the other two classes. High offenders are comprised of 75% Blacks are have an average age of 16.76. T he three trajectory groups have similar levels of Hispanics and there ar e also no statistically signif icant differences in family poverty levels. There is only one neighborhood characteristic that is significantly related to trajectory group membership. Those individuals in the low offending group have a significantly higher percentage of African Americans residing in their neighborhood (55%) than non-offenders (49%). However, the differ ence between high offenders (57%) and non-offenders does not reach stat istical significanceeven though high offenders have the highest mean percentage of African Americans residing in the neighborhood. The rest of t he covariates are not signifi cantly different across the trajectory groups but many ar e, at least, in a direction that would be consistent with theoretical expectations. Unlike neighborhood characteristics, all six family factors are significantly related to trajectory group membership Non-offenders ( 40%) are significantly more likely than both low (30%) and high (24%) offenders to live with both par ents. Non-offenders also tend to experience greater levels of parental supervision and their primary caregivers have higher levels of attachment to t hem than both low and high offenders. Nonoffenders are more likely to be attached to their primary caregiver and to receive 185

PAGE 186

consistent discipline than low offenders only. Finally, primary caregivers educational expectations for their high-offending child ar e significantly lower than those of both nonoffending and low-offending juveniles. Four out of five school factors have signi ficant F-statistics indicating that these variables too are associated with trajectory group membership on the whole. However, only three of the five Tukey tests reveal si gnificant mean differences between specific groups. Of these three factors, non-offenders exhibit significantly more commitment to school compared to low and high offenders and they tend to engage in more school clubs with friends than do low offenders. N on-offenders also have higher aspirations for college than do low and high offenders. Relat edly, low offenders also have significantly higher aspirations for co llege than do high offenders. All of the peer influence covariates are significantly and strongly related to trajectory group membership. Wi th the exception of the vari ables in the prior criminal behavior subcategory, upon which the trajectory groups are based, the peer influence variables tend to have the largest F-stat istics and have means t hat are the most noticeably different across the trajectory groups. Six of t he seven variables are such that all group means are significantly diffe rent from one another. High offenders have greater peer delinquency, higher peer drug use, greater pee r delinquent values, more involvement with dating, have higher amounts of risky time with friends, and more gang involvement than low and non-o ffenders; and, low offenders have greater levels of these variables as compared to non-offender s too. Compared to only 25% of nonoffenders, 43% of high offending males, and 40% and low offending males have 186

PAGE 187

engaged in sexual intercourse before the age of fifteen. These differences between the two offending groups and the one non-offending group are statistically significant. Consistent with literature on the relationship between de pression and delinquent values and offending behavior, trajectory group membership is related to these variables. High offenders have greater delinquent values than both low and nonoffenders. Relatedly, low offenders have hi gher delinquent values than non-offenders. Non-offenders exhibit fewer depressive symptoms than both low an d high offenders. There were no significant differences in t he self-image means across the three latent classes. It goes without saying that there are large and significant differences in the means of violent offending behavior across the three di stinct trajectory groups. More relevant to our discussion is the fact that general delinquency and drug use exhibit a similar relationship to trajectory gr oup membership. That is, hi gh violent offenders engage in more general delinquency and more drug use than low and non-offenders and, also, low violent offenders significantly commit more of these type s of behaviors compared to non-offenders. On average, low and high violent offenders have similar levels of aggression and these two groups both have signi ficantly higher levels of aggression than do non-offenders. Similarly, these tw o groups have experienced higher levels of prior justice system contact in the form of an arrest or police contact as compared to non-offenders, which is not entirely surprising. Trajectory Group-specific Interv ention Experiences and Outcomes Recall, about 22% of the samp le experienced official inte rvention in the form of an actual arrest during either or both waves si x and seven as measured by official police records in Rochester (n=131). This is a fairly sizeable proportion of the sample 187

PAGE 188

permitting a general assessment of the effects of official intervention on subsequent violent offending behavior for the male sample as a whole. What is more pertinent to the current discussion is whet her there is a sufficient number of individuals who experienced official interventi on across each of the three violent crime trajectories. If a trajectory group has only a few individuals w ho experienced an official intervention then it may be impossible to estimate a treatment effect of official intervention on subsequent violent offending for that group with the current data. Fortunately, this does not appear to be the case. Figure 8-3 displays the number of individuals arrested and not arrested across each of the violent offending trajectory groups. All trajectory groups had a number of individuals who exper ienced an arrest during the period of interest. Not surprisingly, the high offenders had the greatest proportion of individuals having experienced an official intervention (41.2%) and the non-offender s had the lowest (14.5%). The proportion of low offenders experiencing official intervention lies in the middle at 23.5%. Whereas approximately 22% of the sample experienced an arrest as measured by official data, 32% of the sa mple experienced a police co ntact during either or both waves six and seven as measured by self-r eport data (n=188). Figure 8-4 displays the self-reported official intervent ion experiences of individu als across each of the three violent crime trajectories. Mirroring the pattern of results of arrest, the self-report data on police contact reveals that non-offenders have the lowest proportion of individual experiencing official intervention (19.2%) followed in order by low offenders (37.2%) and high offenders (50.0%) Not surprisingly, t he number of individuals experiencing official intervention in the form of a police contac t is higher compared to arrest across all 188

PAGE 189

trajectory groups; hence, arrest s follow police contacts but not all police contacts result in arrests. In sum, there appears to be a sufficient number of individuals who experienced the treatment whether that is an arrest as recor ded by official records or a police contact as documented through self-r eport (see Figures 83 and 8-4). The proportion of individuals experiencing o fficial intervention across the trajectory groups makes logical sense; that is, the mo re problematic the trajectory group, the greater proportion of individual s experiencing official interv ention. Focusing on the efficacy of policing, it is reassuring to note that individuals belonging to more problematic violent crime trajectory groups are experiencing o fficial interventions at the highest rates. However, focusing on the ra mifications of official intervention for subsequent behavior across distinct violent offending trajectory groups may or may not support the justice system contact pa tterns that are observed herein. Table 8-8 contains the violent crime means for three different trajectory groups. A general examination of the wave nine and wa ve ten mean outcome violent crime scores shows that between-individual stability in offending is lar gely preserved over both the shortand long-runs. That is, high offenders still have the highest violent crime mean in both outcome waves, followed in order by low offenders and non-offenders. Results from one-way ANOVA and Tukeys B tests reveal that these mean differences are all statistically significant in the short-r un (wave nine), however, high offenders and low offenders are not significantly different fr om one another in the l ong-run (wave ten). These two groups do have significantly higher violent crime means than non-offenders in the long-run. With a very thorough descrip tion of the three violent offending trajectory groups, I now turn to investigating whether se lf-reported police contacts and/or officially 189

PAGE 190

recorded arrests alleviate, exacerbate, or have no affect on t he violent criminal behavior in the shortand long-runs fo r the aggregated sample and, most importantly, for each of the distinct offending traj ectories. 190

PAGE 191

Table 8-1. Model selection cr iteria for five different Po isson LCGA models estimating violent crime trajectories. Latent Class Log Likelihood BIC Prob. model correct 1 -2730.044 5479.254 0.000 2 -2387.141 4819.002 0.000 *3 -2342.381 4755.037 0.720 4 -2330.550 4756.928 0.280 5 -2328.991 4779,364 0.000 indicates selected model. Table 8-2. Model selection cr iteria for two alternative th ree class Poisson LCGA models with different orders of polynomials across the trajecto ry groups. Order of Polynomials Loglikelihood BIC Probability model correct (2,2,2) -2328.991 4779.364 0.000 *(2,0,2) 2342.480 4742.457 1.000 indicates selected model. Table 8-3. Mean posterior probabilities for tr ajectory group assignment for the selected three-group Poisson LCGA model. Group Assigned Group 1 2 3 1 0.891 0.000 0.109 2 0.000 0.899 0.101 3 0.058 0.126 0.816 Table 8-4. Additional diagnos tics for the selected threegroup Poisson LCGA model. Assigned Group j Pj j Pj OCCj 1 0.131 0.114 0.017 54.383 2 0.415 0.393 0.022 12.535 3 0.454 0.492 -0.038 5.331 Table 8-5. Total number and proportion of individuals within each violent crime trajectory group (N=595). Assigned Group Nj %j High (C1) 68 11.4 Non-offenders (C2) 234 39.3 Low (C3) 293 49.2 191

PAGE 192

192 Table 8-6. Parameter estimates for the selected three class Poisson LCGA model. Assigned Group Parameter Estimate S.E. t p-value High (C1) Intercept 0.507 0.119 4.255 0.000 Linear 0.175 0.083 2.114 0.035 Quadratic -0.064 0.021 -3.086 0.002 Non-offenders (C2) Interc ept -3.472 0.422 -8.129 0.000 Low (C3) Intercept -0.296 0.102 -2.900 0.004 Linear -0.415 0,114 -3.646 0.000 Quadratic 0.065 0.028 2.335 0.020

PAGE 193

Table 8-7. Trajectory gr oup-specific descriptive statistics and ANOVA results. High (C1) Nonoffenders (C2) Low (C3) F-test Tukeys B Mean Mean Mean Demographics Race Black 0.75 0.55 0.66 6.11** C1>C2; C3>C2 Hispani c 0.16 0.20 0.17 0.38 None Age 16.76 16. 35 16.50 7.13** C1>C2; C1>C3 Family poverty 0.24 0.28 0.33 1.48 None Neighborhood characteristics Proportion African Amer. 0.57 0.49 0.55 4.94** C3>C2 Proportion in poverty 0.33 0.31 0.34 1.50 None Neighborhood disorgani zation 1.68 1.60 1.66 0.51 None Neighborhood integration 2.02 2.21 2.22 2.37 None Neighborhood satisfac tion 2.81 2.93 2.86 1.07 None Neighborhood arrest rate 4.25 3.78 4.13 2.69 None Family Family structure 0. 24 0.40 0.30 4.40* C2>C3; C2>C1 Parental supervision 3. 41 3.60 3.48 6.11** C2>C3; C2>C1 Attachment to paren t 3.46 3.46 3.32 7.92** C2>C3 PCs edu. expectations of ch ild 1.93 2.47 2.30 9.86** C2>C1; C3>C1 Attachment to child 3.36 3.61 3.41 14.05** C2>C3; C2>C1 Consistency in discip line 2.32 2.44 2.29 5.67** C2>C3 School Factors Commitment to school 2. 98 3.16 3.03 9.13** C2>C3; C2>C1 Aspiration for college 2.97 3.49 3.30 10.53** C2>C3>C1 Prosocial activities 1.60 1.75 1.63 2.72 None Educational expectations 2.47 2.69 2.56 3.28* None School clubs with friends 1.16 1.32 1.17 5.08** C2>C3 Peer Associations Peer delinquency 1.82 1.11 1.37 61.30** C1>C3>C2 193

PAGE 194

Table 8-7. Continued. High (C1) Nonoffenders (C2) Low (C3) F-test Tukeys B Mean Mean Mean Peer drug use 1.96 1.19 1.45 69.64** C1>C3>C2 Peer delinquent values 1.64 1.17 1.34 38.56** C1>C3>C2 Dating 0.75 0.33 0.50 22.01** C1>C3>C2 Sexual activity under 15 0. 43 0.25 0.40 7.74** C1>C2; C3>C2 Risky time with friends 2.71 1.87 2.26 47.82** C1>C3>C2 Gang involvement 0.88 0.04 0.31 144.99** C1>C3>C2 Values Self-image 3.14 3.23 3.21 1.16 None Depression 2.10 1.90 2.02 6.73** C1>C2; C3>C2 Delinquent values 1. 71 1.25 1.45 38.85** C1>C3>C2 Prior criminal behavior Violent crime (Time 1) 1.74 0.00 0.75 164.75** C1>C3>C2 Violent crime (Time 2) 2.09 0.00 0.49 302.30** C1>C3>C2 Violent crime (Time 3) 2.06 0.00 0.47 310.45** C1>C3>C2 Violent crime (Time 4) 1.61 0.00 0.38 228.24** C1>C3>C2 Violent crime (Time 5) 1.38 0.00 0.41 139.35** C1>C3>C2 General delinquency 4. 07 0.32 1.60 93.74** C1>C3>C2 Drug use 0.50 0.03 0.17 26.33** C1>C3>C2 Aggression 0.47 0.26 0.46 22.36** C1>C2; C3>C2 Prior Justice System Contact Previous contact or arrest 0.53 0.20 0.41 20.88** C1>C2; C3>C2 p 0.05; **p 0.01 194

PAGE 195

Table 8-8. Trajectory groupspecific mean outcomes in the short-run and long-run. High (C1) Nonoffenders (C2) Low (C3) F-test Tukeys B Mean Mean Mean Short-run (wave 9) 0.47 0.04 0.28 19.46** C1>C3>C2 Long-run (wave 10) 0.40 0.16 0.31 5.50** C1>C2; C3>C2 p 0.05; **p 0.01 195

PAGE 196

23456 Wave 0 0.1 0.2 0.3 0.4 0.5 Violent Crime Total Sample, 100.0% Figure 8-1. Estimated trajectory of violent off ending during adolescence for the total sample (c=1). 196

PAGE 197

23456 Wave 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 Violent Crime High, 13.1% Non-offenders, 41.5% Low, 45.4% Figure 8-2. Estimated trajectories of violent offending during adolescence fo r the three latent classes (c=3). 197

PAGE 198

40 220 224 28 34 69 0 50 100 150 200 250 300 350 High offenders Non-offenders Low offendersTotal Number of Individuals in Each Latent Cla s Not Arrested Arrested Figure 8-3. Number of individua ls experiencing arrest in waves seven and/or ei ght across the three violent crime trajectory groups. 198

PAGE 199

199 34 189 184 34 45 109 0 50 100 150 200 250 300 350 High offenders Non-offenders Low offendersTotal Number of Individuals in Each Latent Cla s No Police Contact Police Contact Figure 8-4. Number of indivi duals experiencing police contact in waves sev en and/or eight across the three violent crime trajectory groups.

PAGE 200

CHAPTER 9 RESULTS: SHORT-TERM AND LONG-TERM SANCTIONS EFFECTS Chapter 9 reports the main findings of the study, which include the extent to which official intervention effects vary across dist inct violent crime trajectories and whether any deterrent or labeling effects differ in the shortand long-runs. First, bivariate estimates of the relationship between o fficial intervention and subsequent violent offending are reported for the aggregated sample and the three violent crime trajectory groups. Next, the results of the propensit y score matching procedures are reported demonstrating the successfulness of the technique for creating covariate balance on potentially important confounder variables for the aggregated sample and, of course, for each of the trajectory groups. These result s are reported for both official intervention treatments. Finally, the effects of the tw o treatments on subseque nt violent offending are estimated with the propensity score matc hed samples to assess whether deterrent or labeling effects exist for the aggregated sample and for one or more violent offending trajectory groups specifically. These m odels are estimated with both short-run and long-run outcomes to ascertain whether any total sample effects and/or trajectory group-specific effects that are free fr om bias differ over time. Bivariate Relationships between Of ficial Intervention and Violence This section examines the extent to which official intervention experiences are associated with violent behavio r without controlling for any covariates. Should a statistical relationship between official inte rvention and violent behavior fail to exist at all, it might be unnecessary to employ propensity score matching to attempt to create covariate balance across forty different covariates with the ultimate goal of estimating 200

PAGE 201

an unbiased treatment effect.35 In other words, a firstorder issue is whether the independent and dependent variables of intere st are related to one another at the bivariate level. Bivariate Relationships in the Short-run The bivariate effects of official intervention on violent behavior are estimated using the Wilcoxon-Mann-Whitney test, which assesses differences between two groups when the dependent variable is not both measured at the interval level and normally distributed. Recall, the violent crime outcome variable is measured at the interval level but is a count of the variety of violent crime acts that an individual engages in during a specific time period. Common of count out comes employed in criminological research, the variable is positively skewed and hence not no rmally distributed. With that said, the bivariate relationship between arrest and vi olent behavior in the short-run is first examined. Figure 9-1 illustrates the mean differences between those experiencing an arrest and those not experiencing an arrest for the total sample and across each of the three trajectory groups. In addition, the figure contai ns the main conclusions (p-values) of the Wilcoxon-Mann-Whitney tests for statistical significance (see Table 9-1 for details). Focus first on the total sample which is si tuated at the left side of the table. The average violent crime score in the short-run for the males that were not arrested during the treatment period is 0.162 whereas the mean for those who were arrested is over twice as high, with an average value of 0.374. This total sample mean difference in violent crime during wave nine is 0.212 and is statistically significant (p<0.001). 35 This, of course, is under the assumption that t here is no reason to suspect a suppression effect. 201

PAGE 202

Before reporting trajectory group-specific re sults, it is worth mentioning that while the statistical test reported herein is referred to as a bivariate statistical test it is not exactly a bivariate test. Recall, the latent class grow th analysis disaggregated the sample into violent offending trajectory groups and the LCGA statistical model, by definition, does not permit within-class variati on in the intercepts or growth factors. Therefore, the bivariate re lationship between arrest and violent behavior within a trajectory group is better conceptualized as a partial relationship between these two variables with prior violent behavior as the lone control variable. Given prior violent offending behavior is likely among the most important control variables, if not the most important, it may not be surprising to find non-significance for some of these bivariate relationshipseven without controlling fo r a host of other relevant variables. With that said, focus now on the three traj ectory group-specific bivariate findings. As an initial observation, the relations hips between arrest and subsequent violent offending behavior in the short-run across the th ree trajectory groups are in a direction consistent with the labeling perspective. Ho wever, only the mean difference for the low offending trajectory group is statistically signifi cant (see also Table 9-1). The wave nine violent crime mean difference between those a rrested and those not arrested for the low offending trajectory group equals 0.165 (p 0.05). While the m ean difference between high offenders who experienced an arrest and t hose who did not is 0.293, which is the highest of any group, there is no evidence of a statistically significant bivariate relationship between arrest and violent crime in the short-run (p>0.10). The mean difference for the non-offending trajectory group is non-significant and equals 0.053, which was the lowest of the three groups (p >0.10). Interestingly, the average violent 202

PAGE 203

crime score for those individuals who were on a low offending trajectory is actually higher than it is for those individual s who were following along a high offending trajectory but were not subject to the official interv ention of arrest. Figure 9-2 reports the shor t-run bivariate results for the self-reported police contact official intervention. Unlike t he arrest results for wave nine, the mean differences are all noticeably larg er and, importantly, statistically significant in this case (see also Table 9-2). For the sample on the whole, the mean for individuals experiencing a police contact during the treatm ent period is about 3 times higher than for individuals who did not experience a polic e contact during this same time period. That is, the mean difference is 0.302 and is statistically significant (p 0.01). Focusing on the trajectory group-specific analyses, polic e contacts are significantly related to violent crime in the short-run for all three of the latent classes. On average, high offenders who experience a polic e contact commit 0.412 time s more types of violent crimes than their counterparts (p 0.05). While the mean difference for non-offenders is again the smallest of the three groups, it is equal to 0.084 and is statistically significant (p 0.05). For the low offending trajectory group, the average outcome difference for males who were contacted by the polic e and those who were not equals 0.285 (p 0.01). In sum, the total sample analysis indicate s there are statistically significant mean differences in wave nine violent crime for bot h the arrest and police contact treatments. In the short-run, there is consistent empirical support fo r labeling theory across the different violent offending trajectories specif ically for the police contact treatment. The arrest analyses showed trends that were in a direction consistent with labeling theory, but only the low offending trajectory group had a short-run outcome that was 203

PAGE 204

significantly different from zero. A very interesting observation that occurred across the trajectory groups for both official interventio n treatments that has perplexing implications is worth noting here. Specif ically, the mean differences were consistently largest for high offenders, second largest for low o ffenders, and smallest for non-offenders. Putting aside for the moment that these fi ndings are bivariate relationships, these patterns raise the possibility that labeling may be somewhat more pronounced precisely for those individuals who should, by def inition, have the mo st justice contact experiences. These issues ar e further explored in the discussion. The focus now is on whether there are bivariate relationships between official intervention experiences and violent crime several years after the justice syst em contact experiences. Bivariate Relationships in the Long-run Sampson and Laub (1997) raise the possibili ty that short-term sanction effects may vary from longer-term ones. Somewhat unexpectedl y, only low offenders who were arrested had worse outcomes compared to their counterparts but individuals who experienced a police contact were more likel y to have higher violent criminal behavior than their counterparts at wave nine regardl ess of which behavioral trajectory they belonged to. Examination of outcomes at wave ten permit an assessment of whether those who experienced labeling effects at wave nine continue to experience these effects at wave ten or whet her they are even further pronounced once labeling has had additional time to run its course. In addition, it permits one to assess whether delayed labeling effects may emerge for the arrest treatment and, while not hypothesized, whether evidence of deterrence exists in the long-run. Figure 9-3 illustrates the mean differences between individuals experiencing an arrest and those not experiencing an arrest for the total sample as we ll as the distinct 204

PAGE 205

violent offending trajectory group s. None of these bivariate re lationships are statistically significant and all are quite small (see also Table 9-3). Mean differences for the total sample as well as for high offenders and nonoffenders are in a direction consistent with labeling but are not significant (p>0.10). The mean difference for the low offending group is in a direction consistent with deterrence but is not significant (p>0.10). Thus, the significant bivariate relationships found at wave nine for the total sample and the low offending group disappear by wave ten. Figure 9-4 contains a summary of the relationship between police contact during waves seven and eight and violent crime at wave ten. Overall, i ndividuals experiencing a police contact commit more types of viol ence compared to those not experiencing a police contact (p 0.01). With respect to the trajectory group-specific analyses, only low offenders experiencing a police contact had si gnificantly worse outcomes than their trajectory group counterparts; the significant mean difference for this group equals 0.192 (p 0.01). The mean differences for the high offending and non-offending groups are 0.147 and 0.047, respectively, but these are not statistica lly significant (p>0.10). The wave ten mean differences for the high offending, non-offending, and low offending trajectory groups correspond to approximately 47%, 56%, and 52% of their wave ten values, respectively. In short, while the short-run violence outcome patterns for the police contact treatment are similar, labeling effects are less strong in wave ten than they are in wave nine. Moreover, signific ance tests suggest that short-term labeling effects may persist over the long haul only for low offenders. Covariate Balance Assessments of Treatment Models While the previous analysis was informative it did not account for a number of variables (other than, to a ce rtain extent, prior behavior) that could potentially confound 205

PAGE 206

the relationship between official interventio n and violent behavior in the shortand longruns. This section takes on the task of creating equivalent groups on measured covariates through propensity score matching fo r the sample as a whole and for each of the three violent offending trajectories across both treatments of official intervention. To begin, covariate means are estimated for t hose experiencing official intervention and those not experiencing it and st andardized bias statistics ar e computed. This analysis is carried out across each of the five MI data sets for the aggregat ed sample as well as for the three violent offendi ng subpopulations. MI aver ages of covariate means and standardized bias statistics are highli ghted in the Tables 9-5 through 9-12.36 This initial assessment of the unmatched samples shows wh ich covariates are out of balance and demonstrates the need fo r a statistical remedy to the problem, which in this case is propensity score matching analysi s. Using a logistic regression model, conditiona l probabilities (i.e., propensity scores) of receiving official interventi on were estimated using the fort y covariates as predictors. This was repeated for each of the MI data sets for the total sample and for each trajectory group. As outlined in the methodology section, the propensity scores are then used in the optimal full matching with restrictions routine in R (see Hansen, 2004). Mean differences and standardized bias statis tics are recalculated after matching for each MI data set. The right side of Tabl es 9-5 through 9-12 cont ains the MI average 36 Pre-matching covariate means and standardized bias statistics for variables with no missing information are similar across the imputed data sets. Only thos e variables that have imputed data have differences across the five MI data sets. Post-matching, all cova riates may have differences across the MI data sets (even those with no missing information) since the es timated propensity scores w ill differ by some amount given they are estimated with slightly different data sets due to the random component that is added to imputed values during the multiple imputation proced ure. Overall, any pre-matching and all postmatching differences are relatively small across t he different imputed data sets and the averages of the MI data sets well represent the covariate means and st andardized bias statistics. Details on covariate balance for each of the individual MI data sets for all analyses reported herein are available upon request. 206

PAGE 207

standardized bias statistics after matching and percentage of bias r eduction statistics. With this information, the ability of the estimated pr opensity scores to create quasiequivalent treatment and cont rol groups is discussed. Collectively, these steps demonstrate the extent to which the matching procedure can produce unbiased treatment effect esti mates, under both the strongly ignorable treatment assignment and stable unit treatment value assumptions. Unfortunately, there is no guarantee that pr opensity score methods will be able to create covariate balance on all variables in general or for every trajectory group specifically (Haviland & Nagin, 2005; Haviland et al., 2007; 2008). Bu t, the benefit of knowing this information upfront is a boon as the researcher can stat e for which trajectory groups a particular data set can support causal inference and for which it cannot. Traditional statistical approaches such as regression analysis do not yield this information in such a transparent fashion.37 Treatment Model for Arrest Total sample Table 9-5 contains the covariate means and standardized bias statistics for the aggregated sample both preand post-matching for the arre st treatment condition. Before matching on propensity scores, 28 of the original covariates are significantly out of balance (i.e., dX>0.20) and five other covariates have standardized bias statistics that are greater than ten. This degree of imbalance pre -matching should come as no 37 Regression modeling requires due diligence in en suring the appropriatenes s of employing certain control variables. As an example, consider a data se t in which all of the individuals who experience an arrest cluster very high on a variable measuring risky time with friends whereas all those not experiencing an arrest cluster on the very low end of this measure. Regression analysis will blindly control for risky time spent with friends even though this variable in this case could not really be held constant in a meaningful sense. 207

PAGE 208

surprise given our reliance upon observationa l data. Approximatel y 74% of the males that were arrested were Black compared to 59% of those individuals who were not arrested. However, the treatment and non-treatment groups were 19% and 18% Hispanic, respectively. Correspondingly, one notices that the standardized bias statistics indicate that Blacks ar e significantly out of balance ( dBlack=31.74) while Hispanics have very good balance ( dHispanic=3.65). Age is the co variate that has the second highest standardized bias statistic ( dAge=77.71) indicating that the covariate means are out of balance pre-matching by nearly 80% of a standard deviation. At the same time, age is a covariate that provides a good example of the need to consider practical differences in means in addition to balance statistics. For instance, even though this covariate is highly out of balance, the differences in the average ages are not especially large in a practical sense. Those experiencing an arrest tend to be just under 17 whereas those not experiencing an arrest are about 16 and 4 months on average. Individuals who are arrest ed during waves seven and/or eight tend to reside in neighborhoods that have a great er portion of their residents in poverty and of African American racial background. Not surprisi ngly, the arrest group has a higher average neighborhood arrest rate. Wh ile these covariates are out of balance according to conventional criteria, the neighborhood integr ation, satisfaction, and disorganization measures have acceptable levels of pre-matchi ng balance. Four of th e six family factors are significantly out of balance pre-matching. There are non-trivial differences in the covariate means of the parental supervi sion, primary caregivers educational 208

PAGE 209

expectations of child, and attachment to child covariates as well as important differences in the proportion of indi viduals living with both parents. Four out of five of the school factors are significantly out of balance and the commitment to school covariate is just shy of also exceeding the c onventional cutoff. The means for aspirations for college, involv ement in prosocial activities, educational expectations and involvement in school club s with friends are higher by an amount that suggests these covariates are substantially out of balance, which could bias any estimate of the effect of arrest on subseque nt violent crime. The peer association cluster of variables also has all but one of t he covariates significantly out of balance. The mean differences in peer delinquency, peer drug use, peer delinquent values, and risky time spent with friends are such that these covariate means are larger for the arrest group than the group of potential controls. In addition, the proportion of individuals who are actively dating and hav e been involved in a gang is substantially greater for the arrest group. With respect to values and mental states only the delinquent values covariate is significantly imbalanced. The covariate m eans for self-image and depression have very good balance pre-matching. As is to be expected, prior delinquent behavior is substantially imbalanced across the two conditi ons such that those who were arrested have higher violent crime covariate means across all time points than those who were not arrested. In addition, individuals who were arrested tend to have higher levels of general delinquency and aggression. They also tend to have higher levels of drug use but this covariate just missed the conventional cutoff ( dDrugUse=19.54); still, balance could be substantially improved for this covari ate too. Consistent with the idea that the 209

PAGE 210

best predictor of future behavior is past behav ior, the covariate tapping prior justice system contact is out of balance by over 93% of a standard deviation. This was the variable that was most out of balance pre-matching. The average covariate is out of balance by nearly 30% of a standard deviation and the logit of the propensity sco re is imbalanced by nearly 1.5 standard deviations, which is quite large. There were covariates in each sub-area that were significantly out of balance suggesting it is important for sanction effects researchers to account for various things such as demographics, prior behavior, and school factors. In short, the treatment and potential control groups for the sample as a whole are different in key respects meaning that the bivariate estimates r eported earlier are biased estimates of the influence of arrest on subsequent violent behav ior. In practice, it would be considered a gr eat success if all of the standardized bias statistics were less than 10 post-matching. At the least, one hopes to see each of these values less than the standard cutoff for acceptabl e covariate balance, which is set to 20 by convention. Of course, the closer the statistic is to zero the better. After matching individuals on their propensity scores, all of the covariates have acceptable levels of balance. Moreover, the matchi ng procedure has resulted in excellent balance for all but one of the covariates. Peer delinquency is t he covariate that is most out of balance post-matching but this variable not only has an acceptable level of balance but fairly good balance ( dPeerDelinquency-m=10.65). In all but four co variates (family poverty, neighborhood disorganization, consistency in discipline, and depression), bias has been reduced and often by subs tantial amounts. For a typical co variate, 70% of the bias has been removed by matching and the median percent age bias reduction is equal to 87%. 210

PAGE 211

Percentage bias reductions in the covariat es topped out at approximately 98%. While bias actually worsened for four variables as noted by their negative %BR scores, the post-matching standardized bias statistics for t hese variables are still all below seven. The average covariate differs by less than 4.5% of a standard deviation post-matching. Moreover, the small mean differences that remain are far less systematically in a direction consistent with expectations. Given these facts, it is not surprising that the logit of the propensity score is essentiall y perfectly balanced following the optimal full matching procedure. Thus, the matching pr ocedure for the arrest treatment for the aggregated sample was highly successful. I now assess whether this good fortune in post-matching covariate balance also occu rs for the propensity score matching procedure within each trajectory group. Trajectory groups Tables 9-6 through 9-8 repor t the covariate means and standardized bias statistics for high offenders, non-offenders, and low off enders, respectively. Focusing first on high offenders, 22 covariates have standardized bias statistics greater than 20 and another 10 variables have values that exceed 10 (see Table 9-6). Family poverty is out of balance by nearly 33% of a standard dev iation but the other demographic variables are not substantially out of balance. Five of the six neighborhood factors are imbalanced prior to matching. Those experiencing an arre st live in neighborhoods with a greater percentage of Afric an Americans and families in poverty. These individuals also reside in neighborhoods that tend to have lower neighborhood satisfaction scores, higher arrest rates, and higher integration. With respect to family factors, there are nontrivial differences in the covariate means for the parental supervision, attachment to parent, primary caregivers educational ex pectations for child, and constancy in 211

PAGE 212

discipline covariates. Those individuals w ho experienced an arrest were involved in substantially lower prosocial activities and school clubs with friends. The differences between the two groups for the other school fact ors are not substantially out of balance but they could be improved. Two of the seven peer asso ciation factors are out of balance according to the conventional cutoff. Individuals who were arrested during the treatment period also had high er delinquent values than those not arrested by a difference of nearly 3/10 a standard deviation. Most of the prior criminal behavior covariates are imbalanced but not necessarily in directions that would be expected. The covariate with the greatest pre-matching mean difference is prior justice system contact. On average, the typical covariate is out of balance by approximately 25% of a standard deviation and the logit of the propensity score is imbalanced by a sizeable 3.15 standard deviations. The right half of Table 9-6 demonstrates that a large number of covariates remain out of balance post-matching. Specifically 15 covariate means differ by more than twenty percent of a standard deviation. Wh ile the average covariate is now out of balance by less 19% and the median covariate is now imbalanced by approximately 9%, the means of the logit of the propensity score are still di fferent by more than three standard deviations. In short, the propens ity score matching procedure has not produced two groups similar enough to support causal inference.38 While this is 38 Using a logistic regression model, conditional probabi lities (i.e., propensity scores) of receiving official intervention were estimated using the forty covariates as predictors. It is important here to mention an unanticipated point of departure from the original analys is plan. The problem of complete separation of data points was detected. Complete separation of data points occurs when an outcome variable is perfectly predicted by a single variable or set of predictors. Quasi-complete separation of data points occurs when an outcome variable is nearly perfectly predicted by a variable or set of variables. The difference between the two is that in the latter case there remains at least one instance across a predictor variable or set of predictor variables for which both ou tcomes are observed. The practical implications of these problems for propensity score matching ar e substantial. When an individuals treatment 212

PAGE 213

unfortunate, it is not entirely surprising given the prior resear ch of Haviland and colleagues. The non-offending trajectory group has a total of 17 covariates that are substantially imbalanced and another 9 that have less than optimal levels of bias before matching (see Table 9-7). The proportion of Blacks is substantially higher and the proportion of Hispanics is lower for low offenders who were arrested compared to those who were not arrested. Two of the six ne ighborhood factors includi ng proportion African American and neighborhood arrest rate are im balanced. Three of the seven family factors are imbalanced with the primary caregivers educat ional expectations for child covariate being the most so. This variable is out of balance by slightly more than 0.5 standard deviations. The childs own educationa l expectations for himself is the only school factor that is not adequately balanced pre-matching. The covariate means for peer delinquency, peer delinquent values, and ri sky time spent with friends are larger by a non-trivial amount for low offenders who were arrested during the treatment period. Those not arrested exhibited higher levels of depressive symptoms and lower levels of delinquent values. Not surprisingly, the prio r violent criminal behavior of non-offenders is not significantly different across the fi ve time points and these means are essentially zero in all cases. The prior justic e system contact variable has the greatest standardized bias statistic and is out of balance by nearly an entire standard deviation. assignment is perfectly predicted (i.e., they have a predicted probability of zero or one), a major assumption of the distribution of propensity score is violated. Without being able to obtain useable propensity scores, advanced quant itative matching methods are rendered useless. The issue of complete or quasi-complete separa tion of data points is relatively uncommon in large samples; however, this problem can occur at a nontrivial rate when a large number of variables are used to predict a binary variable with a small sample si ze. In this study, forty covariates are being used to predict a binary official intervention treatment vari able within trajectory groups of which some are relatively small. To address this problem, Firths regression was used. Nevertheless, successful balance was unable to be obtained. Matching for this group using propensity scores estimated for the total sample also resulted in unsatisfactory balance achievement. 213

PAGE 214

On average, covariate means and proportions are different by nearly 24% of a standard deviation and the logit of the propensit y score is imbalanced by 1.83 standard deviations. Following the optimal full matching procedure, all but one of the covariates has acceptable levels of bias. That is, only drug use has a standardized bias statistic that is greater than 20.39 It is worth noting that the matching procedur e actually worsened the bias of eleven of the forty covariates. Howe ver, this trade-off proved to be beneficial overall given a median bias reduction of 58% and a 96% bias reduction in the logit of the propensity score. A typica l covariate is now imbalanc ed by less than 10% whereas it was imbalanced by nearly 24% prior to matching. While acceptable levels of balance were not achieved for the drug use variabl e, covariate balance was substantially improved for the non-offending group overall to a level that supports reasonably strong causal inference. The low offending trajectory group is no exception to pre-matching covariate imbalances. Specifically, 20 of the covari ates have standardized bias statistics that exceed the standard cutoff and another 10 vari ables have mean differences greater than 1/10 a standard deviation. The proporti on of Hispanics and mean age differences between those arrested and not arrested are different by non-triv ial amounts. With respect to out of balance neighborhood ch aracteristics, the arrested group has a substantially higher percentage of its residents in poverty and a higher area arrest rate. Family structure, primary caregivers educational expectations for child, and PCs 39 Including drug use and the violent crime measures created quasi-complete separation of data points. To address this, these variables were excluded from the propensity score model. Firths regression was attempted with these variables included but this resu lted in other covariates being out of balance postmatching. 214

PAGE 215

attachment to child are imbalanced by amounts that could confound the results linking violence and official intervention for this trajectory group. Three of the five school factors are out of balance wit h aspirations for college being the most problematic covariate in this subgroup. Five of the seve n peer association covariates differ by more than 1/5 of a standard deviation While the covariates of self-image, depression, and delinquent values are balanced pre-matching, most of the variables capturing prior delinquent behavior are out of balance. Fi nally, the proportion of low offenders experiencing prior justice system contact who were arrested is over twice the proportion of those not arrested, a va lue that separates the means by approximately 75% of a standard deviation. On average, covariat es differ by more than one-quarter of a standard deviation and the logit of the propensity score is out of balance by over one and two-thirds of a standard deviation. As the right part of Table 9-8 illustrate s, the matching procedure has resulted in substantial reductions in bias. On average, covariates are out of balance by less than 6% of a standard deviation and no covariate is out of balance by more than 13% of a standard deviation. Only one covariate (consis tency in discipline) experienced a bias increase. In addition to no variables exceeding the conventional cutoff, only two covariates (aspirations for college and educational expectations) exceeded a standardized bias statistic of 10. The median and mean bi as reductions were 75% and 60%, respectively. The percentage bias reducti on in the logit of the propensity score is 99%; these means now differ by only approxim ately 2% of a standard deviation. In short, the treatment and control groups have been matched with a high degree of success, which ultimately means that strong caus al inference for the effects of arrest on 215

PAGE 216

subsequent violence can be made from post-matching analysis conducted within the low offending trajec tory group. Across all trajectory groups, there are subs tantial differences in covariate means between those experiencing an arrest and thos e not experiencing it during waves seven and/or eight. This is not to say that all of the same variables are in balance or out of balance across the three trajectory groups. In fact, this is clearly not the case. Only neighborhood arrest rate, prim ary caregivers educational expectations for child, drug use, and prior justice system contact are substantially imbalanced across all groups ( dx 20). And, only neighborhood disorganization, commitment to school, and self-image are reasonably balanced pre-matching ac ross the trajectory groups ( dx< 20). Of the remaining covariates, 17 are imbalanced ac ross one violent offending trajectory group and 16 are imbalanced for two groups. Thes e pre-matching differences in covariate balance are informative since they suggest potentially different selection mechanisms across divergent offending trajectories. Attention to pre-matching imbalances across latent classes can also identify variables that may be robustly related to official intervention across trajectory groups such as neighborhood arrest rate and drug use. In sum, it has been shown that, prior to matching, some covariates have consistent imbalance across the trajectory groups and m any others are imbalanced in some trajectory groups but not in others. Despite some important differences in which covariates are imbalanced across the latent classes, it can be safely concluded that each trajectory group has a larg e number of covariates that have substantial bias and a number of others that have less than optimal amounts of bias. Without removing these key differences in covariate means or pr oportions between those experiencing an arrest 216

PAGE 217

and those not experiencing one ac ross each of the trajectory groups, any estimates of the trajectory group-specific treatment effe cts would be untrustworthyas is the case with the bivariate associations between o fficial intervention and subsequent violent behavior. For the arrest treatment, the propensity score matching procedure created excellent bias for the total sample and for two latent classe s permitting trajectory groupspecific treatment effect estimates for non-offenders and low offender s. Unfortunately, however, the data do not support causal inference for high offenders. Treatment Model fo r Police Contact Total sample Table 9-9 reports treatment and cont rol group means and standardized bias statistics before and after matching for the polic e contact treatment for the total sample. Focusing on the pre-matching statistics, well over half of t he variables exhibit substantial imbalance. In addition to thes e 24 imbalanced covariates, there are eight additional covariates that have standardized bias statistics greater t han 10. Overall, the typical covariate is out of balance by approx imately 26% of a standard deviation and the logit of the propensity score is imbalanced by more than 90% of a st andard deviation. There are some noticeable trends in cova riate imbalances pre-matching among the subcategories of variables which give some insight into the types of covariates that pose the greatest validity threats. For instance, none of the means for the treated and potential control groups for t he neighborhood variables are substantially out of balance (though covariate balance could be improved) Contrast this with prior behavior variables which all show substantial imbal ances between the two groups. A number of the covariates in the family, school, peer, and values and mental states sub-areas are imbalanced in directions consistent with expectations suggesti ng the importance of 217

PAGE 218

accounting for these variables in sanction effe cts research as well. Not surprisingly, those with a police contact dur ing the treatment period experienced more prior justice system contact. In sum, the two groups are not equivalent with respect to a number of important variables that could confound the observed relationship between police contact and subsequent violent crime. Before discussing the post-matching balance stat istics, it is important to point out that there is a large degree of consistency in the factors that are imbalanced for the police contact and arrest treatment variables prior to matching (cf. Ta bles 9-5 and 9-9). There are, however, a coupl e of notable differences between the pre-matching imbalances. For one, Blacks are overrepres ented among the arrest treatment group but this is clearly not the case for the police cont act variable. There are also differences in the imbalances of race at t he macro-level (proportion Afri can American). Self-image seems to be imbalanced for the police contact treatment but not for the arrest treatment and there are a few other more minor diffe rences throughout. By and large, however, there is a good deal of similarity between the variables that tend to be imbalanced between the two different official inte rvention measures. Once the treatment and control groups are matched based upon their propensities to experience a police contact, every covariat e exhibits not only acceptable balance but excellent balance. Specifically, the means of any covariate do not differ by more than 7.25% of a standard deviation. The bias reducti on statistics indicate that all except four of the variables underwent a bias reduction as denoted by the positive %BR values. The average percentage bias reduction was 46% and the median percentage bias reduction was 86%. The average covariate is now imbalanced by less than 4% of a 218

PAGE 219

standard deviation and, moreover the group means of the logit of the propensity scores are essentially identical. The full optimal matching procedure has created two groups that are very comparable fo r the aggregate sample. Trajectory groups Tables 9-10 through 9-12 contain the preand post-matching covariate means and standardized bias statistics along with perc entage bias reduction values for the high offenders, non-offenders, and lo w offenders, respectively. The high offenders have 25 covariates that have standardized bias st atistics greater than 20 and an additional 10 covariates that have standardized bias stat istics greater than 10 (see Table 9-10). Three of the four demographi c factors are imbalanced and two of the six neighborhood factors are also out of balance prior to matching. Two-thir ds of the family variables exhibit non-trivial differences in covariate means with parental supervision being the covariate most out of balance. Four out of five school factor s are substantially imbalanced. Four of the seven peer influenc e variables are out of balance with the other three variables fairly close to the conventional cutoff. Self-image and delinquent values are two additional variables that hav e non-trivial differences in their covariate means. Most of the prior behavior covariates are imbalanced as is the prior justice system contact variable. The average covariat e is out of balance by approximately 28% of a standard deviation. The means of the l ogit of the propensity score are different by nearly two standard deviations. Following the matching procedure, the treatment and control groups are not sufficiently balanced. While the covariat e imbalances of some variables have been reduced to a large degree or essentially entirely eliminated (e.g., aggression and Blacks), the imbalance for many other variables is far worse post-matching than it was 219

PAGE 220

pre-matching. In several cases, these nega tive bias reductions are quite substantial taking variables from acceptable levels of imbalance pre-matching to unacceptable levels post-matching (e.g., attachment to parent, gang involvement, and depression). In fact, the average bias reduction value is -0.78 indicating a bias increase. The median bias reduction for a covariate is 24%. The differences in the logit of the propensity scores across the treated and control groups was reduced by 85% but this global summary balance measure is still out of bal ance by nearly 29% of a standard deviation. After matching, the average cova riate differs by 22% of a st andard deviation. Similar to the high offending group for the arrest treat ment, the data cannot support good causal inference. Non-offenders have 16 covariates that ar e out of balance but another 14 that have less than optimal levels of imbalance meaning that only ten covariates have excellent balance pre-matching. Age is the only dem ographic variable imbalanced and none of the neighborhood characteristic variables is different between groups. The covariate means for three of the six fam ily covariates, three of the fi ve school factors, and four of the seven peer influence variables differ by more than 1/5 of a standard deviation. General delinquency, aggressi on, and prior justice system contact are imbalanced by this criterion as well. The covariate means for a typical variabl e for the non-offending trajectory group differ by approximately 19% of a standard deviation. And, the means for the logit of the propensity score diverge by 1.33 standard deviations. After matching, all but two covariates have acceptable levels of covariate balance. Violent crime at times two and four have standardized bias statistics that exceed 20.40 40 Violent crime variables were omitted from the propensity score models due to quasi-complete separation of data points. Firths regression was attemp ted but did not yield better bias reduction overall. 220

PAGE 221

It is important to note that any differences in violent crime for the non-offending group were created through the MI procedure. Recall, trajectories were estimated with up to two waves of missing data. Following traj ectory group estimation and prior to the estimation of the logistic regression model s, data were imputed and in a few instances an individual was assigned a non-zero scoreeven though they were classified as a non-offender by the LCGA modeling procedur e. Thus, the high standardized bias statistics for the violent cr ime measures are products of extremely small standard deviations of these prior violent crime meas ures since nearly every one has a value of zero by definition. Thus, even though two va riables are technically out of balance according to conventional cutoffs, this has li ttle influence on our ability to make a causal inference in this case. Indeed, the mean scores on these measures are essentially identical. With that said, the average covariate is now imbalanced by only about 7% of a standard deviation and the logit of the propensity score is out of balance by approximately 5% of a standar d deviation. These values are down from 19% and 132%, respectively. In short, the matchi ng procedure for the non-offending trajectory group for the police contact treatment was successful. Finally, Table 9-12 reports the preand post-matching covariate balance statistics for the low offending trajectory group for t he police contact treatme nt. The covariate means for a total of 15 of the measures are substantially different across the two groups. Another 15 variables have less than ideal covariate balance. None of the demographic or neighborhood variables are out of balance. Ha lf of the family variables, two-thirds of the school fact ors, five-sevenths of the peer association measures, and two-thirds of the values and mental states variables are out of balance. The majority of 221

PAGE 222

prior criminal behavior covariates are imbal anced but the prior justice system contact variable is not. Overall, the average cova riate imbalance is about 19% of a standard deviation and the logit of the propensity score is out of balance by slightly more than one standard deviation. After matching, all forty covariates have standardized bias statistics below 10. The average covariate is now imbalanced by less than 5% of a standard deviation. Three variables had greater bias post-matching but even these variables are still easily within acceptable levels of covariate balance. The average bias reduction was 32% and the median bias reduction for a covariate was 73 %. The logit of the propensity score underwent a 99% bias reduction leaving the co variate means to differ by less than 1% of a standard deviation. The optimal full matching procedure has created a treatment and control group for the low offending trajectory in which the treatment effect of police contact on subsequent violent offending can be estimated with confidence. With respect to covariate imbalance across the three violent criminal offending trajectory groups, there are again some similarities and also some important differences. Eight variables had acceptable levels of covariate balance pre-matching across all trajectory groups (e.g., neighborho od disorganization, sexual activity under fifteen) whereas nine variables had unaccept able levels across all trajectory groups (e.g., risky time with friends, general delinquency). The rema ining 23 variables were out of balance for either one or two of the lat ent classes. While the particulars of the variables tended to vary some across the traj ectory groups, there were a large number of pre-matching differences for all trajec tory groups leading one to question whether the bivariate results will hold once bias has been removed. Unfortunately, the matching 222

PAGE 223

procedure only successfully removed bias in the covariates for the non-offending and low offending trajectory groups. It is for these two groups which reasonable trajectory group-specific treatment effect estimates c an be obtained. I now turn to the primary analysis, which quantifies effects for the arre st and police contacts treatments in both the shortand long-runs. Treatment Effect Estimat es of Official Interventi on on Subsequent Violence Three types of treatment e ffects are generally available for post-matching analysis. Most importantly, the average treatment effect (ATE) is the average effect of a treatment on an outcome for all individuals in the sample (or subsample). In addition to this overall effect, two other treatment effects are often of inte rest to policy researchers. The first of these and the one more often repo rted is the average treatment effect on the treated (ATT), which is the average effect of some treatment on an outcome for individuals who are typically treated. Se cond, the average treatment effect on the untreated (ATU) is the average effect of so me treatment on an outcome for individuals who typically avoid treatment. Each of these treatment effects seek to provide an estimate of the counterfactual; that is, ATE, ATT, and ATU estimate avera ge outcome differences of individuals being treated and not treated. However, they differ with respect to their reference populations. To help understand how these treat ment effect estimates are dissimilar, it is useful to consider their meaning in reference to the counterfactual for a single individual (for a discussion, see Morgan & Winship, 2007). In the present case, ATE is defined as the expected observed difference in violent crim e for a randomly selected individual from the entire sample if the individual could have experienced both the treatment condition of official intervention and the condition of no o fficial intervention. ATT is the expected 223

PAGE 224

observed difference in violent crime for a randomly selected individual who experiences official intervention if the individual coul d have experienced both treatment conditions, whereas ATU is the expected observed diffe rence in violent crime for a randomly selected individual who does not experience offi cial intervention if the individual could have both experienced an official inte rvention and not experienced one. Because the interest here is in the general effects of arrest and police contacts on subsequent violent behavior, I focus on the av erage treatment effect (ATE) of official intervention on subsequent violent offending. All of the build-up involving bivariate assessments and assessments of covariate balanc e culminates with the straightforward estimation of ATE using the Hodges-Lehmann aligned rank test fo r matched sets. I begin with arrest and estimate the treatment effects of this official intervention on subsequent violent offending behavior in t he short-run for the total sample and then across trajectory groups. Similarly, I es timate the average treatment effects for the police contact official intervention in the shor t-run. This is follo wed with an exploration of whether any of these treatm ent effects persist or differ in the long-run. Treatment Effects in the Short-run Recall, the bivariate relationship betw een arrest and subsequent violent offending behavior at wave nine for the total sample was statistically significant (see Table 9-1). With a matched sample, the treatment effect estimate of arrest for the total sample remains statistically significant (p 0.05). Table 9-13 contains the details of the HodgesLehmann test of statistical si gnificance. It should be not ed that the Hodges-Lehmann estimates reported in the table are averages of the five MI data sets and the standard errors that are reported we re calculated using appropriate procedures for handling MI data sets (this also applies to Tables 914 through 9-16). To aid in interpreting the 224

PAGE 225

treatment effects, the column at the far right of the table reports Cohens d statistic. While the short-run treatment effect of arrest for the to tal sample is statistically significant, it is best charac terized as a small effect. Thus, while arresting someone may exacerbate their violent offending behavior in the short-run there are certainly other factors that are probably much more impor tant in explaining subsequent violent behavior. Nevertheless, the fa ct that arrest predicts late r violent crime at all after successfully matching on forty cova riates is important. Table 9-13 also contains the results of the trajectory group-specific analyses. The treatment effects were not estimated for the high offending trajectory group since substantial amounts of covari ate imbalance remained. Even though only the bivariate relationship for low offenders was statistically significant, recall that the expected counts of violence in the short-run were higher for those arrested compared to their counterparts across all trajectory groups in the bivariate analysis (see Table 9-1). Similar to the bivariate analysis, the treatm ent effect estimates for the non-offending group fails to be statistically significant. W hat is more, however, is that the observed effect is now in the opposite direction. Once individuals are successfully matched on the covariates, those experiencing an arre st appear less likely to commit acts of violence during wave nine but, to be clear, this relationship is not statistically significant. Conversely, the treatment effect for the low offending trajectory group is still in a direction that would support labeling theory. Like the non-offending group, however, this treatment effect estimate is also not st atistically significant. Thus, the significant bivariate relationship between arrest and short-run violent crime for the low offending trajectory group does not hold after matching. 225

PAGE 226

Table 9-14 contains the Hodges-Lehmann s hort-run treatment effect estimates and Cohen d values for the police contact official intervention. Beginning again with the full sample, the aligned rank test shows t hat individuals self-reporting a police contact during waves seven and/or eight are likely to engage in more distinct acts of violent crime in wave nine. This treatment effect is statistically signific ant with an alpha level equal to 0.01. Similar to the short-run vi olent outcome results for the treatment of arrest, the treatment effect is relatively small ( d =0.30). Thus, there is consistent evidence of a fairly small labeling effect acro ss both types of official intervention for the sample as a whole. The trajectory group-specific analysis reveals mixed support for the labeling perspective. While the bivariate relations hip between police contact and violent crime during wave nine was statistically significant for the non-offending trajectory group, the treatment effect estimate for this relationshi p fails to achieve stat istical significance. However, the treatment effect for the low of fending trajectory group is statistically significant (p 0.05). It should be noted t hat this effect is the strongest of all of the significant treatment effects across all models. Still, this effect is probably best described as small ( d =0.35). For low offenders, experi encing a police contact results in an observed short-run outcome that is supportive of the basic tenants of labeling theory. This is not the case for non-offenders (p>0 .05) and, unfortunately, the data do not permit us to discern with any definitiveness whether this is true for high offendersthe trajectory group that is probably most important for policy. At best, one can attempt to glean information from the unmatched and matched results for the low offending trajectory group and the preand post-matching covariate bal ance assessments. It was 226

PAGE 227

found that the bivariate findings held for the low offending trajectory group in the matched sample. For both low and high offender s pre-matching, the average covariate was out of balance by approxim ately one-quarter of a standard deviation. And, it was true that the difference in wave nine viol ent crime means was largest for the high offenders. Given this information, it is reas onable to think that the bivariate relationship for high offenders may have also held. Howe ver, this is quite speculative and the author is certainly in no place to draw any sci entifically sound conclusions in this regard. Treatment Effects in the Long-run Table 9-15 contains the treatment effect es timates for the arrest condition for the total sample as well as the non-offending an d low offending trajectory groups. For the sample on the whole, the Hodges-Lehmann aligned rank test is not statistically significant indicting there is no treatment effect of arrest. Likewise, the treatment effect estimates for the non-offending and the lo w offending groups are not statistically significant. This is not especia lly surprising given the fact t hat all of these relationships were not statistically significant at the bivari ate level. Thus, the total sample treatment effect of arrest in the s hort-run is not sustained over the long haul and no trajectory groups suddenly have a labeling or deterrent treatment effect emerge in wave ten. Figure 9-16 reports the long -run treatment effects of t he police contact official intervention. Recall, the bivariate relati onships between violent crime at wave ten and police contact were statistically significant for the total sample as well as for the low offending trajectory group (see Table 9-4) Once individuals are matched based upon their propensities to experienc e a police contact during the treatment period, there are no longer any significant relationships between police contact and subsequent violent offending. In sum, labeling effects on the whole or for any trajectory group specifically 227

PAGE 228

228 are not sustained in the long-run. In addition, labeling effects do not appear to need time to run their course; in other words, there is no evidenc e in this sample that shortrun deterrent or null effects eventually bec ome long-run labeling effects as Sampson and Laubs (1997) discussions of the influences of official intervention across the lifecourse suggest as an empi rical possibility.

PAGE 229

Table 9-1. Bivariate associations between a rrest and violent crime in the short-run for the total sample and by trajectory group. Sum of Scores Expected Under Ho Mean Score Z Total Sample Not Arrested 134209.5 138272.0 289.24 3.77** Arrested 43100.5 39038.0 329.01 --Trajectory Groups High offenders (C1) Not Arrested 1298.0 1380.0 32.45 1.22 Arrested 1048.0 966.0 37.43 --Non-offenders (C2) Not Arrested 23415.5 23500.0 117.07 0.69 Arrested 4079.5 3995.0 119.98 --Low offenders (C3) Not Arrested 32075.5 32928.0 143.19 2.01* Arrested 10995.5 10143.0 159.35 --** p 0.01; p 0.05; p 0.10; Wilcoxon-Mann-Whitney test of statistical significance. Table 9-2. Bivariate associations between police contact and violent crime in the shortrun for the total sample and by trajectory group. Sum of scores Expected under Ho Mean score Z Total Sample No Police Contact 113527.0 121286.0 278.94 6.43** Police Contact 63783.0 56024.0 339.27 --Trajectory Groups High offenders (C1) No Police Contact 1002.5 1173.0 29.48 2.51* Police Contact 1343.5 1173.0 39.51 --Non-offenders (C2) No Police Contact 21939.5 22207.5 116.08 1.97* Police Contact 5555.5 5287.5 123.45 --Low offenders (C3) No Police Contact 25256.5 27048.0 137.26 3.71** Police Contact 17814.5 16023.0 163.44 --** p 0.01; p 0.05; p 0.10; Wilcoxon-Mann-Whitney test of statistical significance. 229

PAGE 230

Table 9-3. Bivariate associations between arre st and violent crime in the long-run for the total sample and by trajectory group. Sum of Scores Expected Under Ho Mean Score Z Total Sample Not Arrested 137778.0 138272.0 296.94 0.42 Arrested 39532.0 39038.0 301.77 --Trajectory Groups High offenders (C1) Not Arrested 965.5 966.0 34.48 0.00 Arrested 1380.5 1380.0 34.51 --Non-offenders (C2) Not Arrested 23510.0 23500.0 117.55 0.04 Arrested 3985.0 3995.0 117.20 --Low offenders (C3) Not Arrested 33055.5 32928.0 147.57 0.29 Arrested 10015.5 10143.0 145.15 --** p 0.01; p 0.05; p 0.10; Wilcoxon-Mann-Whitney test of statistical significance. Table 9-4. Bivariate associations between police contact and violent crime in the longrun for the total sample and by trajectory group. Sum of Scores Expected Under Ho Mean Score Z Total Sample No Police Contact 116284.0 121286.0 285.71 3.77** Police Contact 61026.0 56024.0 324.71 -Trajectory Groups High offenders (C1) No Police Contact 1131.0 1173.0 33.26 0.64 Police Contact 1215.0 1173.0 35.74 --Non-offenders (C2) No Police Contact 22035.0 22207.5 116.56 0.74 Police Contact 5460.0 5287.5 121.33 --Low offenders (C3) No Police Contact 25521.5 27048.0 138.70 3.04** Police Contact 17549.5 16023.0 161.00 --** p 0.01; p 0.05; p 0.10; Wilcoxon-Mann-Whitney test of statistical significance. 230

PAGE 231

Table 9-5. Treatment and control group covariate means and bias statistics before and after PSM fo r the arrest treatment (total sample). Before Matching After Matching M MXt Xp dX MXtm MXcm dXm %BR Demographics Race Black 0.74 0.59 31.74 0.71 0.72 3.65 0.89 Hispanic 0.19 0.18 3.65 0.20 0.20 1.79 0.51 Age 16.92 16.35 77.71 16.83 16.88 7.33 0.91 Family poverty 0. 31 0.30 2.95 0.30 0.33 5.68 -0.93 Neighborhood characteristics Proportion African Americ an 0.59 0.51 27.69 0.56 0.56 2.32 0.92 Proportion in poverty 0.36 0.32 27.99 0.35 0.35 3.08 0.89 Neighborhood disorganizati on 1.69 1.66 4.56 1.69 1.66 5.47 -0.20 Neighborhood integrati on 2.16 2.21 6.86 2.16 2.18 3.31 0.52 Neighborhood satisfacti on 2.82 2.89 10.89 2.86 2.86 3.31 0.70 Neighborhood arrest rate 4.53 3.86 34.40 4.30 4.28 1.92 0.94 Family Family structure 0.24 0.36 24.48 0.28 0.28 3.26 0.87 Parental supervision 3.44 3.54 20.79 3.48 3.49 5.44 0.74 Attachment to paren t 3.43 3.37 13.79 3.40 3.39 5.31 0.61 PCs edu. expect. of ch ild 1.89 2.40 63.43 2.00 2.02 7.50 0.88 Attachment to child 3.32 3.52 43.36 3.38 3.38 3.62 0.92 Consistency in discipli ne 2.34 2.36 4.24 2.32 2.29 6.57 -0.55 School Factors Commitment to school 2.99 3.07 18.96 3.01 3.02 5.15 0.73 Aspiration for college 3.08 3.41 37.19 3.13 3.16 3.73 0.90 Prosocial activities 1.55 1.71 26.60 1.55 1.56 4.96 0.81 Educational expectations 2.45 2.64 24.89 2.44 2.44 2.17 0.91 School clubs with friends 1.10 1.26 32.13 1.12 1.11 2.42 0.92 Peer Associations Peer delinquency 1. 48 1.28 37.11 1.46 1.40 10.25 0.72 Peer drug use 1. 55 1.37 34.31 1.55 1.54 2.27 0.93 231

PAGE 232

Table 9-5. Continued. Before Matching M M After Matching M Xt Xp dX MXtm Xcm dXm %BR Peer delinquent values 1.39 1.29 22.44 1.38 1.36 4.97 0.78 Dating 0.56 0.44 24.57 0.55 0.54 3.07 0.88 Sexual activity under 15 0.28 0.36 16.20 0.31 0.30 3.36 0.79 Risky time with friends 2.41 2.10 42.47 2.38 2.36 3.83 0.91 Gang involvement 0.44 0.22 48.21 0.40 0.40 4.54 0.91 Values Self-image 3.19 3.21 6.39 3.20 3.20 1.59 0.75 Depression 2.00 1. 98 4.78 2.00 2.01 5.16 -0.08 Delinquent values 1.50 1.37 28.24 1.47 1.45 6.99 0.75 Prior criminal behavior Violent crime (Time 1) 0.80 0.51 31.69 0.77 0.78 2.79 0.91 Violent crime (Time 2) 0.76 0.41 38.67 0.64 0.66 3.22 0.92 Violent crime (Time 3) 0.74 0.40 36.84 0.67 0.64 4.38 0.88 Violent crime (Time 4) 0.55 0.32 28.33 0.49 0.48 6.88 0.76 Violent crime (Time 5) 0.57 0.30 33.81 0.55 0.50 6.40 General delinquency 2.26 1.13 45.24 2.15 1.93 8.90 0.80 Drug use 0.23 0. 13 19.54 0.21 0.20 1.83 0.91 Aggression 0.50 0.35 41.73 0.47 0.45 5.95 0.86 Prior Justice System Contact Previous contact or a rrest 0.67 0.25 93.98 0.58 0.58 2.35 0.98 Global Covariate Balance Stats. Mean ----29.32 ----4.42 0.70 Median ----28.11 ----3.78 0.87 Maximum ----93.98 ----10.25 0.98 Logit propensity score -0.27 -2.25 147.72 -0.70 -0.70 0.26 1.00 Notes: The following is relevant for Tables 9-5 through 9-12. MXt = treatment group covariate mean before matching; MXp = potential control group covariate mean before matching; dX = standardized covariate bias before matching (percentage of a standard deviation); MXtm = treatment group covariate mean after matching; MXcm = control group covariate mean after matching; dXm = standardized covariate bias after matching (percentage of a standard deviation); %BR = percentage reduction in standardized bias through propensity score matching. 232

PAGE 233

Table 9-6. Treatment and control group covariate means and bias statistics before and after PSM fo r the arrest treatment (High offenders). Before Matching After Matching M MXt Xp dX MXtm MXcm dXm %BR Demographics Race Black 0.71 0.78 13.96 0.78 0.78 1.55 0.89 Hispanic 0.18 0.15 7.72 0.14 0.13 4.19 0.46 Age 16.82 16.72 11.81 16.77 16.69 9.95 0.16 Family poverty 0.16 0.30 32.64 0.17 0.26 19.87 0.39 Neighborhood characteristics Proportion Afric an American 0.61 0.55 23.71 0.58 0.57 4.07 0.83 Proportion in poverty 0.35 0.31 26.86 0.35 0.34 3.61 0.87 Neighborhood disorganizati on 1.71 1.73 5.79 1.77 1.78 2.22 0.62 Neighborhood in tegration 2.12 1.97 22.12 2.15 1.97 28.05 -0.27 Neighborhood sa tisfaction 2.65 2.93 39.04 2.57 2.83 37.42 0.04 Neighborhood arrest rate 4.91 3.80 59.20 4.46 4.46 0.04 1.00 Family Family structure 0.25 0.23 5.88 0.25 0.25 0.35 0.94 Parental supervision 3.32 3.46 22.30 3.43 3.29 23.70 -0.06 Attachment to parent 3.52 3.32 42.05 3.50 3.22 50.97 -0.21 PCs edu. expect. of child 1.57 2.07 63.86 1.62 1.97 44.37 0.31 Attachment to child 3.31 3.37 11.94 3.38 3.27 22.10 -0.85 Consistency in discipline 2.38 2.27 22.72 2.34 2.10 47.03 -1.07 School Factors Commitment to school 2.86 2.93 16.35 2.85 2.85 1.52 0.91 Aspiration for college 2.89 3.03 13.51 2.87 2.84 2.98 0.78 Prosocial activities 1.39 1.75 58.20 1.39 1.72 53.90 0.07 Educational expectations 2.43 2.50 8.65 2.46 2.48 2.61 0.70 School clubs with friends 1.02 1.25 49.08 1.03 1.27 51.72 -0.05 Peer Associations Peer delinquency 1.88 1.77 15.27 1.77 1.73 4.94 0.68 Peer drug use 1. 96 1.93 5.88 1.89 1.89 0.47 0.92 233

PAGE 234

Table 9-6. Continued. Before Matching M M After Matching M Xt Xp dX MXtm Xcm dXm %BR Peer delinquent values 1.69 1.59 20.66 1.60 1.58 3.46 0.83 Dating 0.71 0.78 13.96 0.67 0.79 27.10 -0.94 Sexual activity under 15 0.39 0.45 11.59 0.44 0.51 15.48 -0.34 Risky time with friends 2.70 2.70 6.43 2.69 2.59 13.11 -1.04 Gang involv ement 0.96 0.83 46.58 0.98 0.78 62.50 -0.34 Values Self-image 3.13 3.14 3.96 3.12 3.11 1.12 0.72 Depression 2.12 2.08 10.10 2.13 2.15 3.39 0.66 Delinquent values 1.80 1.66 28.86 1.72 1.74 4.59 0.84 Prior criminal behavior Violent crime (Time 1) 1.54 1.90 29.97 1.60 2.08 40.35 -0.35 Violent crime (Time 2) 1.94 2.15 17.69 2.02 2.09 5.75 0.68 Violent crime (Time 3) 1.89 2.16 22.02 1.96 2.02 5.25 0.76 Violent crime (Time 4) 1.79 1.48 29.39 1.63 1.57 5.44 0.81 Violent crime (Time 5) 1.54 1.28 22.13 1.33 1.43 8.44 General delinquency 3. 89 4.20 7.91 3.43 5.07 42.28 -4.34 Drug use 0.39 0.58 23.14 0.44 0.67 29.15 -0.26 Aggression 0.54 0.46 23.24 0.50 0.46 8.69 0.63 Prior Justice System Contact Previous contac t or arrest 0.79 0.35 97.94 0.78 0.45 72.07 0.26 Global Covariate Balance Stats. Mean ----24.85 ----19.14 0.17 Median ----22.07 ----8.57 0.46 Maximum ----97.94 ----72.07 1.00 Logit propensity score 1.22 -1.47 314.82 1.24 -1.38 306.27 0.03 234

PAGE 235

Table 9-7. Treatment and control group covariate means and bias statistics before and after PSM fo r the arrest treatment (Non-offenders). Before Matching After Matching M MXt Xp dX MXtm MXcm dXm %BR Demographics Race Black 0. 82 0.50 72.77 0.77 0.72 14.69 0.80 Hispanic 0. 12 0.21 25.15 0.14 0.17 14.09 0.44 Age 16.86 16.27 89.79 16.80 16.82 2.60 0.97 Family poverty 0. 32 0.28 9.45 0.35 0.32 15.36 -0.63 Neighborhood characteristics Proportion African Americ an 0.58 0.47 39.53 0.54 0.54 6.19 0.84 Proportion in poverty 0.34 0.31 19.66 0.33 0.33 8.32 0.58 Neighborhood disorganizati on 1.56 1.64 12.55 1.58 1.57 13.47 -0.07 Neighborhood integration 2.17 2.22 7.09 2.14 2.10 13.59 -0.92 Neighborhood satisfacti on 2.96 2.91 7.90 2.88 2.89 6.72 0.15 Neighborhood arrest rate 4.41 3.67 36.37 4.07 4.18 8.01 0.78 Family Family structure 0.26 0.42 33.18 0.26 0.25 5.64 0.83 Parental supervision 3.58 3.60 4.31 3.62 3.65 8.97 -1.08 Attachment to parent 3.53 3.45 19.32 3.52 3.49 12.97 0.33 PCs edu. expect. of child 2.10 2.52 53.11 2.24 2.19 13.44 0.75 Attachment to child 3.44 3.62 39.86 3.55 3.57 9.16 0.77 Consistency in discipli ne 2.39 2.45 11.72 2.39 2.35 13.34 -0.14 School Factors Commitment to school 3.17 3.15 5.57 3.20 3.17 13.86 -1.49 Aspiration for college 3.38 3.51 15.36 3.38 3.38 16.29 -0.06 Prosocial activities 1.72 1.76 4.70 1.75 1.76 9.20 -0.96 Educational expectations 2.44 2.74 38.82 2.50 2.51 7.38 0.81 School clubs with friends 1.27 1.32 7.35 1.32 1.26 7.89 -0.07 Peer Associations Peer delinquency 1. 27 1.09 40.51 1.14 1.14 5.50 0.86 Peer drug use 1.24 1.19 13.79 1.17 1.20 12.12 0.12 235

PAGE 236

Table 9-7. Continued. Before Matching M M After Matching M Xt Xp dX MXtm Xcm dXm %BR Peer delinquent values 1.26 1.16 24.15 1.19 1.23 12.23 0.49 Dating 0.35 0.33 5.91 0.31 0.34 7.07 -0.20 Sexual activity under 15 0.32 0.24 19.83 0.27 0.27 7.31 0.63 Risky time with friends 2.05 1.84 32.18 2.00 1.98 11.46 0.64 Gang involvement 0. 03 0.04 5.79 0.04 0.04 2.41 0.58 Values Self-image 3.25 3. 22 6.92 3.34 3.29 11.00 -0.59 Depression 1.81 1.92 22.20 1.79 1.80 3.51 0.84 Delinquent values 1.32 1.24 21.82 1.29 1.30 3.83 0.82 Prior criminal behavior Violent crime (Time 1) 0.00 0.01 13.03 0.00 0.00 1.28 0.90 Violent crime (Time 2) 0.01 0.00 13.40 0.00 0.00 3.11 0.77 Violent crime (Time 3) 0.00 0.00 6.53 0.00 0.00 0.68 0.90 Violent crime (Time 4) 0.00 0.00 6.00 0.00 0.00 6.68 -0.11 Violent crime (Time 5) 0.00 0.00 0.00 0.00 0.00 0.00 General delinquency 0. 47 0.30 23.89 0.39 0.47 12.65 0.47 Drug use 0.00 0. 04 25.65 0.00 0.07 41.70 -0.63 Aggression 0.31 0.26 18.95 0.26 0.25 6.35 0.66 Prior Justice System Contact Previous contact or arre st 0.56 0.14 99.44 0.43 0.39 11.45 0.88 Global Covariate Balance Stats. Mean ----23.84 ----9.54 0.27 Median ----19.13 ----8.64 0.58 Maximum ----99.44 ----41.70 0.97 Logit propensity score 0.25 -4.03 183.69 -0.77 -0.95 7.71 0.96 236

PAGE 237

Table 9-8. Treatment and control group covariate means and bias statistics before and after PSM fo r the arrest treatment (Low offenders). Before Matching After Matching M MXt Xp dX MXtm MXcm dXm %BR Demographics Race Black 0.71 0.64 14.42 0.66 0.66 6.21 0.57 Hispanic 0.23 0.15 20.45 0.24 0.23 5.04 0.75 Age 16.99 16.35 84.88 16.89 16.85 5.46 0.94 Family poverty 0. 37 0.32 10.82 0.37 0.37 2.43 0.78 Neighborhood characteristics Proportion African Americ an 0.58 0.55 14.48 0.55 0.55 5.13 0.65 Proportion in poverty 0.37 0.33 32.60 0.35 0.35 5.64 0.83 Neighborhood disorganizati on 1.74 1.67 12.45 1.72 1.69 7.04 0.43 Neighborhood integrati on 2.17 2.23 9.69 2.23 2.23 3.32 0.66 Neighborhood satisfacti on 2.82 2.87 7.06 2.86 2.87 4.01 0.43 Neighborhood arrest rate 4.43 4.04 20.56 4.18 4.14 7.14 0.65 Family Family structure 0.23 0.32 20.12 0.27 0.30 7.68 0.62 Parental supervision 3.42 3.50 18.22 3.48 3.46 7.35 0.60 Attachment to paren t 3.35 3.31 9.10 3.33 3.33 3.71 0.59 PCs edu. expect. of ch ild 1.92 2.35 53.70 2.01 1.96 6.49 0.88 Attachment to child 3.26 3.46 40.96 3.34 3.33 3.80 0.91 Consistency in discipli ne 2.29 2.29 1.52 2.26 2.25 5.66 -2.72 School Factors Commitment to school 2.96 3.02 14.84 2.99 2.98 9.12 0.39 Aspiration for college 3.01 3.39 43.28 3.14 3.08 10.76 0.75 Prosocial activities 1.52 1.67 25.36 1.48 1.46 5.54 0.78 Educational expectations 2.46 2.58 15.52 2.44 2.37 12.40 0.20 School clubs with friends 1.04 1.21 39.61 1.06 1.06 2.23 0.94 Peer Associations Peer delinquency 1. 43 1.35 14.27 1.39 1.38 3.83 0.73 Peer drug use 1. 54 1.43 20.83 1.56 1.55 2.61 0.87 237

PAGE 238

Table 9-8. Continued. Before Matching M M After Matching M Xt Xp dX MXtm Xcm dXm %BR Peer delinquent values 1.32 1.35 6.33 1.34 1.34 3.94 0.38 Dating 0.59 0.47 24.43 0.56 0.55 2.95 0.88 Sexual activity under 15 0.22 0.45 51.09 0.26 0.25 6.05 0.88 Risky time with friends 2.46 2.22 34.38 2.48 2.47 3.32 0.90 Gang involvement 0.43 0.28 33.46 0.39 0.40 5.75 0.83 Values Self-image 3.18 3.22 9.57 3.22 3.20 7.91 0.17 Depression 2.05 2.01 7.00 2.02 2.04 6.05 0.14 Delinquent values 1.47 1.44 5.24 1.44 1.43 3.81 0.27 Prior criminal behavior Violent crime (Time 1) 0.90 0.71 22.59 0.89 0.90 4.86 0.78 Violent crime (Time 2) 0.66 0.45 29.31 0.55 0.54 1.92 0.93 Violent crime (Time 3) 0.63 0.43 28.79 0.52 0.46 8.85 0.69 Violent crime (Time 4) 0.31 0.40 15.68 0.30 0.33 4.66 0.70 Violent crime (Time 5) 0.46 0.39 10.35 0.41 0.43 5.37 General delinquency 2.48 1.33 50.55 2.03 2.03 5.94 0.88 Drug use 0.28 0. 14 26.46 0.25 0.25 6.06 0.77 Aggression 0.57 0.42 42.28 0.50 0.53 8.00 0.81 Prior Justice System Contact Previous contact or a rrest 0.68 0.33 74.93 0.56 0.56 0.99 0.99 Global Covariate Balance Stats. Mean ----25.43 ----5.48 0.60 Median ----20.50 ----5.50 0.75 Maximum ----84.88 ----12.40 0.99 Logit propensity score 0.22 -2.65 167.12 -0.61 -0.65 2.27 0.99 238

PAGE 239

Table 9-9. Treatment and control group covariate means and bias statisti cs before and after PSM for the police contact treatment (total sample). Before Matching After Matching M MXt Xp dX MXtm MXcm dXm %BR Demographics Race Black 0. 63 0.62 0.74 0.63 0.64 4.84 -5.54 Hispanic 0.19 0.17 4.41 0.19 0.19 1.82 0.59 Age 16.60 16.41 23.65 16.55 16.53 4.89 0.79 Family poverty 0. 30 0.31 2.37 0.30 0.32 5.52 -1.33 Neighborhood characteristics Proportion African Americ an 0.54 0.53 4.43 0.53 0.54 5.23 -0.18 Proportion in poverty 0.34 0.32 17.09 0.34 0.34 4.42 0.74 Neighborhood disorganizati on 1.67 1.67 1.27 1.68 1.70 3.70 -1.92 Neighborhood integrati on 2.16 2.21 7.58 2.17 2.15 2.84 0.63 Neighborhood satisfacti on 2.83 2.90 11.62 2.87 2.87 3.21 0.72 Neighborhood arrest rate 4.22 3.91 15.42 4.11 4.13 5.21 0.66 Family Family structure 0.31 0.34 5.38 0.34 0.34 3.05 0.43 Parental supervision 3.44 3.55 23.63 3.49 3.48 2.88 0.88 Attachment to paren t 3.33 3.41 16.12 3.39 3.37 4.07 0.75 PCs edu. expect. of ch ild 2.05 2.40 41.70 2.19 2.16 4.18 0.90 Attachment to child 3.34 3.54 40.83 3.43 3.42 2.83 0.93 Consistency in discip line 2.30 2.38 15.15 2.34 2.31 6.35 0.58 School Factors Commitment to school 2.94 3.10 40.35 3.00 3.01 4.10 0.90 Aspiration for college 3.14 3.43 33.57 3.25 3.23 3.96 0.88 Prosocial activities 1.61 1.71 15.13 1.65 1.63 4.68 0.69 Educational expectations 2.45 2.67 28.85 2.52 2.50 3.19 0.89 School clubs with friends 1.18 1.25 12.74 1.20 1.19 2.18 0.83 Peer Associations Peer delinquency 1. 49 1.25 45.95 1.37 1.38 2.86 0.94 Peer drug use 1. 57 1.34 44.47 1.47 1.47 2.38 0.95 239

PAGE 240

Table 9-9. Continued. Before Matching M M After Matching M Xt Xp dX MXtm Xcm dXm %BR Peer delinquent values 1.40 1.26 32.69 1.36 1.35 2.86 0.91 Dating 0.58 0.41 34.91 0.54 0.54 1.88 0.95 Sexual activity under 15 0.35 0.34 3.04 0.36 0.38 7.25 -1.38 Risky time with friends 2.36 2.08 40.27 2.28 2.28 4.44 0.89 Gang involvement 0.40 0.21 41.62 0.33 0.33 2.61 0.94 Values Self-image 3.13 3.24 27.94 3.17 3.17 4.45 0.84 Depression 2.04 1.95 19.17 2.02 2.02 5.02 0.74 Delinquent values 1.54 1.33 47.50 1.46 1.46 1.22 0.97 Prior criminal behavior Violent crime (Time 1) 0.74 0.50 26.27 0.67 0.66 3.75 0.86 Violent crime (Time 2) 0.63 0.42 24.36 0.55 0.57 4.68 0.81 Violent crime (Time 3) 0.66 0.38 31.67 0.54 0.56 3.13 0.90 Violent crime (Time 4) 0.57 0.28 39.93 0.46 0.46 1.96 0.95 Violent crime (Time 5) 0.56 0.27 37.73 0.42 0.43 1.70 General delinquency 2.41 0.90 61.95 1.67 1.64 2.53 0.96 Drug use 0.29 0. 09 35.11 0.19 0.20 3.78 0.89 Aggression 0.49 0.33 45.46 0.43 0.43 2.16 0.95 Prior Justice System Contact Previous contact or a rrest 0.46 0.29 35.71 0.39 0.39 2.70 0.92 Global Covariate Balance Stats. Mean ----25.94 ----3.61 0.46 Median ----27.10 ----3.46 0.86 Maximum ----61.95 ----7.25 0.97 Logit propensity score -0.25 -1.17 92.37 -0.63 -0.63 0.23 1.00 240

PAGE 241

Table 9-10. Treatment and contro l group covariate means and bias statistics be fore and after PSM for the police contact treatment (High offenders). Before Matching After Matching M MXt Xp dX MXtm MXcm dXm %BR Demographics Race Black 0.65 0.85 48.95 0.73 0.73 0.00 1.00 Hispanic 0.21 0.12 24.14 0.24 0.26 4.62 0.81 Age 16.71 16.81 12.97 16.59 16.78 24.36 -0.88 Family poverty 0. 18 0.29 26.40 0.43 0.20 51.50 -0.95 Neighborhood characteristics Proportion African Americ an 0.59 0.56 11.57 0.66 0.64 6.44 0.44 Proportion in poverty 0.35 0.31 27.61 0.37 0.33 33.13 -0.20 Neighborhood disorganizati on 1.74 1.69 13.71 2.03 1.76 40.93 -1.99 Neighborhood integrati on 2.08 2.00 12.55 1.93 1.87 8.78 0.30 Neighborhood satisfaction 2.86 2.80 8.74 2.57 2.65 11.33 -0.30 Neighborhood arrest rate 4.75 3.76 52.66 4.28 4.45 9.00 0.83 Family Family structure 0. 29 0.18 28.01 0.26 0.31 11.42 0.59 Parental supervision 3.21 3.60 70.74 3.48 3.71 42.04 0.41 Attachment to parent 3.39 3.43 11.89 3.49 3.36 23.59 -0.98 PCs edu. expect. of child 1.73 2.00 33.08 2.31 1.92 48.19 -0.46 Attachment to child 3.29 3.39 21.65 3.36 3.44 16.54 0.24 Consistency in discipli ne 2.31 2.34 7.39 2.32 2.23 17.62 -1.38 School Factors Commitment to school 2.82 3.00 45.89 2.98 2.84 36.52 0.20 Aspiration for college 2.79 3.15 36.47 3.34 2.89 46.53 -0.28 Prosocial activities 1.48 1.72 37.78 1.88 1.85 5.55 0.85 Educational expectations 2.44 2.50 7.13 2.87 2.71 19.92 -1.79 School clubs with friends 1.09 1.23 27.13 1.47 1.48 2.48 0.91 Peer Associations Peer delinquency 1. 96 1.67 40.48 1.89 1.58 44.73 -0.10 Peer drug use 2. 02 1.86 27.27 1.76 1.74 5.01 0.82 241

PAGE 242

Table 9-10. Continued. Before Matching M M After Matching M Xt Xp dX MXtm Xcm dXm %BR Peer delinquent values 1.68 1.59 17.06 1.66 1.52 26.65 -0.56 Dating 0.79 0.71 20.48 0.76 0.81 12.74 0.38 Sexual activity under 15 0.38 0.47 17.91 0.47 0.47 0.12 0.99 Risky time with friends 2.81 2.60 30.40 2.60 2.66 9.06 0.70 Gang involvement 0.85 0.91 18.33 0.77 0.94 48.97 -1.67 Values Self-image 3.08 3.19 25.01 3.34 3.28 13.49 0.46 Depression 2.11 2. 09 3.82 2.11 2.21 20.59 -4.39 Delinquent values 1. 89 1.54 82.16 1.73 1.61 27.07 0.67 Prior criminal behavior Violent crime (Time 1) 1.75 1.76 1.38 1.75 2.24 39.39 -27.47 Violent crime (Time 2) 1.99 2.15 13.85 2.19 2.03 14.11 -0.02 Violent crime (Time 3) 1.85 2.24 33.72 1.97 1.81 13.21 0.61 Violent crime (Time 4) 1.71 1.50 19.84 1.74 1.46 26.56 -0.34 Violent crime (Time 5) 1.56 1.21 30.23 1.71 1.12 50.05 General delinquency 4. 82 3.32 39.08 4.57 3.33 32.37 0.17 Drug use 0.65 0.35 36.22 0.53 0.42 23.59 0.35 Aggression 0.56 0.43 35.81 0.42 0.41 1.28 0.96 Prior Justice System Contact Previous contact or a rrest 0.65 0.41 48.51 0.53 0.49 9.22 0.81 Global Covariate Balance Stats. Mean ----27.70 ----21.97 -0.78 Median ----26.76 ----18.77 0.24 Maximum ----82.16 ----51.50 1.00 Logit propensity score -1.07 1.50 184.93 -0.13 -0.52 28.59 0.85 242

PAGE 243

Table 9-11. Treatment and contro l group covariate means and bias statistics be fore and after PSM for the police contact treatment (Non-offenders). Before Matching After Matching M MXt Xp dX MXtm MXcm dXm %BR Demographics Race Black 0.62 0.53 18.93 0.60 0.58 7.10 0.63 Hispanic 0.16 0.21 13.22 0.16 0.14 9.93 0.25 Age 16.61 16.29 42.04 16.54 16.51 6.43 0.85 Family poverty 0. 27 0.29 5.76 0.28 0.29 7.11 -0.23 Neighborhood characteristics Proportion African Americ an 0.53 0.48 18.08 0.51 0.50 7.88 0.56 Proportion in poverty 0.32 0.31 4.44 0.32 0.32 5.75 -0.30 Neighborhood disorganizati on 1.62 1.63 3.41 1.60 1.59 7.80 -1.28 Neighborhood integrati on 2.15 2.23 11.14 2.14 2.17 7.22 0.35 Neighborhood satisfaction 2.91 2.92 3.07 2.89 2.87 3.44 -0.12 Neighborhood arrest rate 4.00 3.73 13.66 3.88 3.86 5.08 0.63 Family Family structure 0. 38 0.40 4.99 0.40 0.39 5.90 -0.18 Parental supervision 3.63 3.59 8.96 3.62 3.60 10.74 -0.20 Attachment to paren t 3.50 3.45 13.02 3.52 3.52 3.86 0.70 PCs edu. expect. of ch ild 2.21 2.52 39.49 2.28 2.29 5.93 0.85 Attachment to child 3.49 3.62 31.16 3.54 3.55 3.27 0.89 Consistency in discip line 2.35 2.46 20.83 2.37 2.36 6.71 0.68 School Factors Commitment to school 3.01 3.19 48.77 3.05 3.08 10.05 0.79 Aspiration for college 3.24 3.55 35.89 3.34 3.31 5.96 0.83 Prosocial activities 1.71 1.76 6.90 1.74 1.77 5.59 0.19 Educational expectations 2.47 2.75 37.06 2.52 2.57 9.35 0.75 School clubs with friends 1.40 1.30 15.22 1.37 1.40 5.70 0.63 Peer Associations Peer delinquency 1. 22 1.09 31.31 1.17 1.13 9.18 0.71 Peer drug use 1. 26 1.18 20.53 1.23 1.21 7.00 0.66 243

PAGE 244

Peer delinquent values 1.28 1.15 31.50 1.26 1.25 4.40 0.86 Table 9-11. Continued. Before Matching M M After Matching M Xt Xp dX MXtm Xcm dXm %BR Dating 0.38 0. 32 12.69 0.37 0.40 6.18 0.51 Sexual activity under 15 0.31 0.23 17.67 0.31 0.33 5.62 0.68 Risky time with friends 1.98 1.85 22.01 1.96 1.98 9.97 0.55 Gang involvement 0.04 0.04 3.75 0.03 0.03 5.06 -0.35 Values Self-image 3.15 3.24 22.41 3.18 3.20 6.68 0.70 Depression 1.96 1.89 15.88 1.93 1.95 7.23 0.54 Delinquent values 1.30 1.24 16.45 1.29 1.27 5.77 0.65 Prior criminal behavior Violent crime (Time 1) 0.00 0.01 13.41 0.00 0.01 14.19 -0.06 Violent crime (Time 2) 0.01 0.00 11.69 0.00 0.01 22.94 -0.96 Violent crime (Time 3) 0.00 0.00 6.72 0.00 0.00 2.27 0.66 Violent crime (Time 4) 0.00 0.00 6.17 0.00 0.00 21.39 -2.46 Violent crime (Time 5) 0.00 0.00 0.00 0.00 0.00 0.00 General delinquency 0.53 0.27 30.82 0.47 0.50 4.52 0.85 Drug use 0.07 0. 03 17.97 0.07 0.08 5.43 0.70 Aggression 0.33 0.25 30.01 0.30 0.30 5.77 0.81 Prior Justice System Contact Previous contact or a rrest 0.33 0.16 39.94 0.32 0.31 7.12 0.82 Global Covariate Balance Stats. Mean ----18.67 ----7.29 0.34 Median ----16.17 ----6.31 0.63 Maximum ----48.77 ----22.94 0.89 Logit propensity score -0.49 -2.42 132.55 -0.85 -0.93 5.21 0.96 244

PAGE 245

Table 9-12. Treatment and contro l group covariate means and bias statistics be fore and after PSM for the police contact treatment (Low offenders). Before Matching After Matching M MXt Xp dX MXtm MXcm dXm %BR Demographics Race Black 0.62 0.68 11.67 0.65 0.65 3.21 0.73 Hispanic 0.20 0.15 13.04 0.18 0.17 4.11 0.68 Age 16.57 16.46 13.20 16.49 16.50 4.18 0.68 Family poverty 0. 35 0.32 5.69 0.35 0.37 5.69 0.00 Neighborhood characteristics Proportion African Americ an 0.53 0.57 15.37 0.55 0.54 5.55 0.64 Proportion in poverty 0.35 0.33 18.51 0.34 0.34 6.03 0.67 Neighborhood disorganizati on 1.67 1.70 5.18 1.70 1.70 1.75 0.66 Neighborhood integrati on 2.19 2.24 7.38 2.19 2.22 5.82 0.21 Neighborhood satisfacti on 2.79 2.90 17.38 2.85 2.84 2.13 0.88 Neighborhood arrest rate 4.14 4.13 0.40 4.14 4.09 4.59 -10.35 Family Family structure 0. 29 0.30 2.35 0.32 0.31 6.51 -1.77 Parental supervision 3.44 3.51 15.78 3.46 3.44 7.91 0.50 Attachment to paren t 3.25 3.36 22.90 3.31 3.29 5.97 0.74 PCs edu. expect. for ch ild 2.10 2.34 30.02 2.21 2.20 2.23 0.93 Attachment to child 3.30 3.48 35.29 3.38 3.38 3.88 0.89 Consistency in discip line 2.28 2.30 4.17 2.29 2.29 3.30 0.21 School Factors Commitment to school 2.95 3.03 20.29 2.99 2.99 2.90 0.86 Aspiration for college 3.20 3.36 18.65 3.30 3.31 3.38 0.82 Prosocial activities 1.61 1.65 6.69 1.62 1.61 2.97 0.56 Educational expectations 2.44 2.62 22.73 2.53 2.52 4.27 0.81 School clubs with friends 1.11 1.20 18.41 1.12 1.13 3.37 0.82 Peer Associations Peer delinquency 1. 45 1.33 24.90 1.39 1.38 5.42 0.78 Peer drug use 1. 55 1.40 29.91 1.49 1.50 4.79 0.84 245

PAGE 246

246 d Table 9-12. Continued. Before Matching M M After Matching M Xt Xp X MXtm Xcm dXm %BR Peer delinquent values 1.37 1.33 11.19 1.37 1.36 5.23 0.53 Dating 0.60 0.45 30.51 0.54 0.56 5.30 0.83 Sexual activity under 15 0.36 0.42 12.48 0.39 0.39 2.21 0.82 Risky time with friends 2.38 2.22 23.52 2.33 2.33 4.26 0.82 Gang involvement 0.40 0.26 30.67 0.34 0.35 4.49 0.85 Values Self-image 3.13 3.25 29.29 3.18 3.14 8.85 0.70 Depression 2.06 2.00 12.44 2.04 2.07 9.69 0.22 Delinquent values 1.53 1.40 30.91 1.50 1.51 8.48 0.73 Prior criminal behavior Violent crime (Time 1) 0.74 0.77 4.25 0.72 0.75 6.97 -0.64 Violent crime (Time 2) 0.47 0.52 7.80 0.48 0.47 4.20 0.46 Violent crime (Time 3) 0.56 0.43 19.37 0.49 0.46 6.06 0.69 Violent crime (Time 4) 0.46 0.34 20.92 0.41 0.40 2.80 0.87 Violent crime (Time 5) 0.48 0.37 16.42 0.44 0.45 3.74 0.77 General delinquency 2.44 1.11 63.50 1.82 1.76 3.96 0.94 Drug use 0.27 0. 11 27.78 0.22 0.20 4.59 0.83 Aggression 0.54 0.40 37.29 0.48 0.50 5.46 0.85 Prior Justice System Contact Previous contact or a rrest 0.45 0.39 11.82 0.42 0.44 4.56 0.61 Global Covariate Balance Stats. Mean ----18.75 ----4.77 0.32 Median ----17.90 ----4.53 0.73 Maximum ----63.50 ----9.69 0.94 Logit propensity score 0.05 -1.01 101.15 -0.40 -0.41 0.71 0.99

PAGE 247

Table 9-13. Post-matching short-run treatment e ffects of arrest for the total sample and by trajectory group. Est. SE t d Total Sample 3346.28 1692.02 1.98* 0.26 High offenders --------Non-offenders -164.08 190.80 -0.86 -0.12 Low offenders 862.37 568.00 1.52 0.31 ** p 0.01; p 0.05; Table 9-14. Post-matching short-run treatment effects of police contact for the total sample and by trajectory group. Est. SE t d Total Sample 5696.27 2078.11 2.74** 0.30 High offenders --------Non-offenders 378. 50 312.55 1.21 0.18 Low offenders 1722.42 743.64 2.32* 0.35 ** p 0.01; p 0.05; Table 9-15. Post-matching long-run treatment effects of arrest for the total sample and by trajectory group. Est. SE t d Total Sample -503.31 1610.89 -0.31 -0.07 High offenders --------Non-offenders -285.96 269.49 -1.06 -0.33 Low offenders 125.47 475.17 0.26 0.04 ** p 0.01; p 0.05; Table 9-16. Post-matching long-run treatment effects of police contact for the total sample and by trajectory group. Est. SE t d Total Sample 2538.03 1895.17 1.34 0.12 High offenders --------Non-offenders -48.19 329.39 -0.15 -0.07 Low offenders 1455.22 858.37 1.70 0.25 ** p 0.01; p 0.05; 247

PAGE 248

0.162 0.350 0.241 0.374 0.643 0.088 0.406 0.035 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 TOTAL SAMPLE** High offenders(C1) Non-offenders(C2) Low offenders(C3)*Average Violent Crime Not Arrested Arrested **p 0.01; *p 0.05; 0.10 Figure 9-1. Short-run violent crime mean comparisons of those arrested and not arrested for the total sample and by violent crime trajectory group before matching. 248

PAGE 249

0.113 0.265 0.174 0.415 0.677 0.111 0.459 0.027 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 TOTAL SAMPLE** High offenders(C1)* Non-offenders(C2)* Low offenders(C3)**Average Violent Crime No Police Contact Police Contact **p 0.01; *p 0.05; 0.10 Figure 9-2. Short-run violent crime mean comparisons of thos e with and without a police contact for the total sample and by violent crime trajecto ry group before matching. 249

PAGE 250

0.254 0.375 0.160 0.317 0.290 0.429 0.177 0.290 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 TOTAL SAMPLE High offenders(C1) Non-offenders(C2) Low offenders(C3)Average Violent Crime Not Arrested Arrested **p 0.01; *p 0.05; 0.10 Figure 9-3. Long-run violent cr ime mean comparisons of those arrested and not arrested for the total sample and by violent crime trajectory group before matching. 250

PAGE 251

251 0.206 0.324 0.153 0.239 0.383 0.471 0.200 0.431 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 TOTAL SAMPLE** High offenders(C1) Non-offenders(C2) Low offenders(C3)**Average Violent Crime No Police Contact Police Contact **p 0.01; *p 0.05; 0.10 Figure 9-4. Long-run violent cr ime mean comparisons of those with and without a police contact for the total sample and by violent crime trajecto ry group before matching.

PAGE 252

CHAPTER 10 CONCLUSIONS Chapter 10, which is the final chapter in this work, provides a summary of the principal research findings and subsequently considers the lim itations of the study. With these qualifications in mi nd, the implications of the st udy that can be safely drawn are discussed; more specifically, the rele vancies of the findings for theories on sanctions effects and policies aimed at contro lling violent crime are considered. Finally, Chapter 10 concludes with recommendations for future research endeavors seeking to examine the effects of official interventi on experiences on subseq uent crime in general and within a life-course and devel opmental framework specific ally. Research Summary Using data from the male s ubsample of the RYDS, the current study examined the extent to which official intervention, such as officially recorded arrests and self-reported police contacts, influenced subsequent violen t crime differently or similarly across violent offending subpopulations in both t he shortand long-runs. To address the empirical aims of the cu rrent research, a three-stage LCGA-PSM integrated methodological approach was employed. First, latent class growth analysis was used to estimate the number and characteristics of violent offending sub populations. Next, propensity score matching was employed to create quasi-equivale nt treatment and control groups in the total sample as well as within each trajectory gr oup specifically. In the current study, full optimal matching procedures were used to attempt to reduce or eliminate pre-matching biases on forty covariates. Finally, total sample treatment effects as well as trajectory group-specific treatment effects were estimated using the Hodges-Lehmann aligned rank test on the matched samples. 252

PAGE 253

LCGA was employed to empirically determi ne the number of traj ectory groups that best characterized the violent offending pattern s in the data. Giv en the use of a count measure of violent crime as well as suffici ent time points to estimate changes in offending over time, a Poisson LCGA was employed with all-quadratic polynomials. The results indicated that ther e was essentially a zero probability that a one, two, or five class model captured the offending pattern s in the data. T here was a non-trivial probability that a four class model was the correct model (28%) but the best evidence favored a three class solution (72%). Following selection of the three class model the order of polynomials was reduced such that two latent classes were characterized by linear and quadratic terms and one latent class was characterized with only an intercept. Empirical evidence strongly favored the more parsimonious three class model (2,0,2) over the more complex allquadratic three class model (2,2,2). The mean posterior probabilities for trajectory group assignment for the parsim onious model were all greater than 0.81, which is well above the conventional cutoff of 0.7. Other model selection criteria also suggested the model was adequately precise in its classification. Of the total analysis sample (N=595), approxim ately 11.4% were classified as high offenders, 39.3% as non-off enders, and 49.2% as low offenders. The offending trajectory of high offenders followed an inverted parabolic shape with offending peaking during wave three. Across all waves, the means for the high offenders were above one and this latent class was consistently the mo st problematic violent offending trajectory group across all time points. The trajecto ry of the non-offending group was flat by definition and the model estimated essentially zero acts of violence for this group across 253

PAGE 254

all waves. Non-offenders ar e the trajectory group whose offending patterns, or lack thereof to be precise, were adequately captured with an intercept only. The low offending trajectory followed a j-shape decline. That is, the expected counts for the low offending trajectory group declined from waves two through four and then largely stabilized thereafter. Overall, there was strong evidence of between-individual stability in offending at the same time there wa s evidence of within-individual change in offending over time for the high and lo w offending trajectory groups. Results of one-way ANOVAs and corres ponding Tukey tests for specific group mean differences revealed that 30 of the 40 co variates were associated with trajectory group membership. There were noteworthy trends in the relationships between the covariates and the latent classes. For instance, there were minimal differences between the three trajectory groups with res pect to the six neighborhood characteristic covariates but all family variables, all peer association measures and, not surprisingly, all prior behavior variables exhibited differenc es among some or all of the groups. The demographic, school factors, and values and mental states subc ategories had a mix of their variables associated with trajectory group membership. Supporting continued between-individual stability in offending, the average shor t-run violent crime outcome was significantly the highest for the high offending group, second highest for the low offending group, and lowest for the non-offending group. This same general pattern of results is maintained in the long-run. A sizeable proportion of the males experienced official in terventions in the form of officially recorded arrests and self-reported police contacts. Specifically, 22% of the total sample was arrested and 32% reporte d a police contact during the treatment 254

PAGE 255

period. High offenders experienced official inte rventions at the highest rate; 41% of this group was arrested and 50% had experienced a po lice contact. For the low offenders, approximately 24% was arrest ed and 37% had a police contact. Not surprisingly, nonoffenders had the lowest proporti on of individuals who had run-ins with the law. The proportion of individuals in this trajectory group that had an arrest or police contact was nearly 15% and 19%, respectively. Bivariate relationships between official intervention experiences and subsequent violent behavior provided split support for label ing and null effects. No evidence was found that supported a s hort-run or a long-run deterrent effect of either measure of official intervention. In the short-run, all mean differences in violent crime for the total sample and across the three trajectory grou ps were in a direction consistent with labeling theory; however, some of these relationships did not reach statistical significance. Arrest was significantly associ ated with short-run violen t crime for the total sample as well as for the low offending trajec tory group. Police contact was significantly related to short-run violent crime for the tota l sample as well as for all three of the violent offending subpopulations. In the long-run, all except one mean diffe rence in violent crime (i.e., arrest treatment for the low offenders) were in a direction cons istent with labeling theory; however, most of these relationships were not statistically significantly. Arrest was not significantly associated with violent crime in the long-run for any trajectory group, nor was it for the total sample. And, police cont act was only associated with violence for the total sample and the low offending trajectory group. Thus, there is stronger evidence of 255

PAGE 256

labeling theory in the short -run as compared to the long -run and no evidence that these types of official interventions curb subsequent violent offending behavior. The optimal full matching procedure was hi ghly successful in creating covariate balance for the aggregated sample for both the arrest and police contact treatments. Pre-matching mean covariate imbalances were approximately 29% and 26% of a standard deviation for the arrest and police c ontact treatments, respectively. The corresponding post-matching mean covariate imbalances dropped to approximately 4% of standard deviation in both cases. No covari ates had variables that were substantially imbalanced following the matching procedure. Matching was less successful in obtaining group equivalences within trajectory groups. On the one hand, very good covariate balance was achieved for the nonoffending and the low offending trajectory gr oups for both the arrest and police contact treatments. For all of these bal ancing models, at least 96% of the bias in the logit of the propensity score was eliminated. And, t he mean and median covariate imbalance was below ten in all cases. On the other hand, the full optimal matching procedure was unable to create treatment and control groups that were simila r on the forty covariates for the high offending trajectory group. This was the case for both the police contact and the arrest treatment. T hus, treatment effect estima tes are only presented for the total sample as well as the non-offending and low offending trajectory groups. The main findings of the current res earch include mixed support for labeling effects in the short-run follow ed by consistent null effects in the long-run. For both the officially recorded arrest and self-report ed police contact treat ments, there are statistically significant short-run effects of official intervention on violent crime for the 256

PAGE 257

total sample. The effect sizes are 0.26 and 0.30, respectively, which are both characterized as relatively small effects. Ne vertheless, the fact that official intervention influences subsequent violent behavior in t he short-run after accounting for forty covariates is noteworthy. The only trajecto ry group-specific effect is that of police contact on short-run violent crime for the low offending tr ajectory group. All other trajectory group-specific effect s in the short-run are not statistically significant and all long-run effects, for the total sample and for all trajectory groups, are also not significant. In sum, when individuals are successfully matched on forty covariates there is only evidence for a short-run labeling effect of the police contact tr eatment for the low offending trajectory group. For the matc hed, aggregated sample (which includes high chronic offenders who could not be matched within their trajectory group) there is evidence for short-run labeling effects for bot h measures of official intervention. Perhaps the clearest conclusion is that t he experience of polic e contact and arrest appears to not have any crime reducing effects in the shortor long-run; this is the case for the total sample generally and each of the distinct violent offending trajectory groups specifically. Limitations While the current study has attempted to address important policy-relevant questions, which were illuminated when placing sanction effects in a developmental and life-course perspective, using both a high-quality longitudin al data set and sophisticated statistical analyses, it is certainly not without limitations. While it was argued that the RYDS data were among the best available data sets to address the empirical aims of the current study, the measurem ent of certain variables c ould have been better. Recall, 257

PAGE 258

for instance, that there were few differenc es in neighborhood characteristics across the violent offending subpopulations, which seems to counter to the basi c idea that social structure should be related to individual criminal behavior (e.g., see Akers, 1998). Several neighborhood factors ar e determined solely by the primary caregivers perceptions of the neighborhood. Specifically, disorganization, satisfaction, and integration are nei ghborhood level phenomena that are being estimated by a single individual. These perceptions probably do not fully represent the neighborhood on the whole. Therefore, the extent to which these variables truly are accounted for in the analysis is unknown. While these neighborhood measures are indeed relatively weak indicators of the neighborhood level constructs they do permit some degree of control over these factors. As such, one is probabl y better off including these variables in the propensity score models than excluding them Other neighbor hood measures are calculated from US Census or other official data so the aforement ioned criticism does not apply to all of the neigh borhood characteristics that were controlled. More important than these more minor i ssues concerning the measurement of certain covariates are the lim itations with the conceptualiz ation, operationalization, and measurement of the two most important variables in the st udy, namely violent crime and official intervention. A variety index was selected as the variable to measure violent crime. This was selected given the fact that variety indexes have been found to be among the most reliable forms of self-r eported measures of crime and deviance (Belson, 1986; Hindelang, Hirschi, & Weis, 1981). Ho wever, there are other conceptualizations of violent crime includ ing prevalence (i.e., whether one commits any acts of violence at all) and frequency (i.e., the total number of violent acts one commits). 258

PAGE 259

There are naturally some interrelationships between these different facets of the criminal career though they are distinct c oncepts. And, some have maintained that these different conceptualizations of off ending behavior may have different causes (Blumstein, Cohen, & Farrington, 1988). Future research might seek to compare the effects of official intervention on preval ence and frequency in addition to variety as it could be the case. In addition, while ther e were practical reasons for focusing the current investigation on violent crime, t here is a need to examine how the effects of arrest and police contact may influence other types of crime and delinquency. In short, additional research should consider both which types of criminal behavior and which specific dimensions of the criminal car eer official intervention experiences may influence. One of the strengt hs of the current study was that it examined two types of official intervention. However, the police contact and arrest treatm ent variables came from two different data sources. To so me extent, this limits the abili ty to compare the differences in the gradation of offending. Ideally, one would want to dire ctly compare the effects of police contacts and arrests on subsequent offending behavior that co me from the same source such as official records. In addi tion, it would have been ideal to have a large enough sample that one could have also examined the effects of stronger sanctions including incarceration. While examining the overall effects of incarceration on crime is often possible with general population samples, it is difficult to estimate trajectory groupspecific effects given the relative infrequencies of these official intervention experiences. This was the case with the current study and therefore it is limited in this regard. Future research should seek to address this problem. 259

PAGE 260

An additional point of consider ation regarding the treatment variable is the fact that justice system contact is often not simply a onetime event. It is im portant to recall that contact with the justice system for this samp le prior to wave seven was considerable (see Table 6-2), which of course means that for some of t hese individuals the official intervention experience during the treatment period was not a novel one. Then, an alternative and probably more realistic concept ualization of official intervention is to consider it as a repeatable treatm ent, rather than a static one. The research questions necessitated the es tablishment of some period that would constitute a treatment period. However, it should be acknowledged that official interventions are repeatable treatments possibly occurring befo re, during, and after the designated treatment period. For some i ndividuals, the ramifications of official intervention such as deviance amplific ation may have already unfolded. While propensity score analysis permits one to ma tch individuals on previous treatment experiences and thereby obtain an unbiased estimate of current treatment experiences, the method does not necessarily capture firs t time treatment effects which might be more, or less, pronounced than those occurring during the selected tr eatment period. This limitation is not unique and is relevant to numerous other studies on sanction effects including those not focused on assessing how prior criminal history may moderate the effects of official intervention on criminal behavior. The current study has addressed a first-order issue: does the effect of official intervention vary across violent offending s ubpopulations and are any significant effects different in the shortand long-runs? Wh ile there is some support for the deviance amplification hypothesis in the short-run, lacking from this re search is an assessment of 260

PAGE 261

the specific mechanisms by which arrests and police contacts may lead to increases in subsequent violent behavior. Labeling theory is explicit in the identif ication of mediating pathways that involve blocked access to c onventional opportunity, alteration of selfidentity, and acquisition of deviant peer s (Paternoster & Iovanni, 1989). Without focusing on these pathways in particular, t he current research can only identify that there is mixed support for the basic idea of deviance amplification and cannot speak to whether the data truly support a labeling process as specified by the theory. In addition, using propensity score matching analysis to address first-order causal issues may complicate secondary efforts to obtain in formation regarding which covariates might explain away the bivariate association bet ween official intervention and subsequent violent offending. For instance, it might be a specific variable that renders the bivariate relationship between a treatment and an outco me non-significant. Regression analyses more easily permit an assessment of the in fluence of adding specific variables on the relationship between the independent and dependent variables of interest. Nevertheless, propensity score matching me thods mimic experimental designs with observational data, providi ng a scientifically sound and transparent approach to identify whether non-spurious relationships exist. And, this is arguably the more important question for policy if not also for theory. While the integrated research methodolog y has supported strong causal inference for the total sample and for tw o of the three trajectory gr oups, treatment effects could not be estimated for arguably the most import ant offending subpopulation, namely high offenders. High offend ers are a small proportion of the population but they are the most problematic with respect to thei r offending and, therefore, t he group that is central to 261

PAGE 262

policy efforts aimed at controlling violent crime. With respect to t hese high offenders, it is known that the treatment(s) and outcome exhibit relationship s at the bivariate level in the short-run that are clearly in a directi on consistent with label ing theory (see Figures 9-1 and 9-2); in one case, this bivariate relationship not only appears large visually but also reaches statistical significance (see Table 9-2). Since the cu rrent study could not create adequate balance across the forty covari ates for the high offending trajectory group, however, it remains unkno wn whether arrest and/or police contact experiences is causally related to subsequent violent behavior. A related point of discussion, which some might argue is a limitation of the current analysis, deals with the procedure of selecting between two LCGA models that have non-trivial probabilitie s of accurately representing the dat a. Recall, the estimated four class model essentially split the low off enders into a rising and declining group. The author is on solid ground for selecting t he three class model for two reasons. Empirically, the three class model was the best choice given the model selection and model adequacy assessment criteria. Practi cally, the three class model was more parsimonious and helped to ensure trajectory gr oups were as large as possible. This latter point proved to be important as treatme nt effects could not even be estimated for the high offending group. A four class m odel may very well have created matching difficulties for individuals classified in either of t hese two alternative groups. Nevertheless, estimating trajectory group-s pecific treatment effects using the three class model rather than the four class model could be seen as running counter to a main goal of the study. For those on a declining trajectory, for example, perhaps official intervention experiences might hasten the desis tance process (deterrence) or cause the 262

PAGE 263

individual to reverse this trend toward confo rmity (labeling). Important policy-relevant knowledge might be obtained if one were to su ccessfully match indi viduals on official intervention experiences in these alternative groups. The limitations of prior research that the study attempted to address were many and varied. In addressing these limitations and by placing official intervention effects research in a developmental and life-course context, the current study has done much to organize theoretical thinking about how sanctions may influence subsequent behavior and has provided an advanced statistical analysis to test these ideas. Despite going to great lengths to overcome limitations of pr ior studies, the current study has a number of issues that lead us to accept the c onclusions with some degree of caution. Unfortunately, many of the aforementioned limitat ions could not be overcome with the data brought to bear on the research ques tion and/or the methodological approach to the topic of study. Fortunately, many of these issues that were raised lead to additional avenues of research, some of which have been discussed here briefly and all of which will be discussed in the final subsection of this work. Discussion and Implications Notwithstanding the limitations of the study, there are impor tant implications of the findings for theories on sanction effects and for policies centered on controlling violent crime. The results of the cu rrent study would seem to be so mewhat damning to specific deterrence theory. There was no empirical evidence, even at the bivariate level, which supported the deterrence doctrine for the total sample or for any trajectory groups in particular. These results were consistent in the shortand long-runs and across two different measures of official intervention including self-report ed police contacts and officially recorded arrests. One limitation of t he current study may shed some light on 263

PAGE 264

the lack of support found for the deterrence doctr ine. The study used arrest and police contact variables as official intervention tr eatments. While certai nty of punishment has been the primary focus of empirical research deterrence theory calls for certainty, severity, and celerity of punishment and t hese elements may interact to influence behavior (Stafford et al., 1986). Perhaps, an offi cial intervention in the form of an arrest or police contact is insufficient by itself to trigger the desired behavioral response. That is, it could be the case that an actual punishment received by a judge might be necessary to invoke specific deterrence bec ause an arrest alone is technically not a punishment. Perhaps, researchers shoul d examine the interactions among the elements of punishment (Stafford et al., 1986) and/or explore t he effects of more serious official interventions. Regarding the latter, sanctioning a youth for his or her violent offense might make a labeling effect more likely. Thus, there is reason to believe that both deterrent and labeling effects may be higher as justice system involvement increases. While the data did not support the main hy pothesis of specific deterrence theory, mixed support was found for ideas underlying labeling theory. That is, individuals who experienced a police contact fared worse than those who did not. As discussed, however, this support was equivocal as well as dependent on the outcome assessment time period and on trajectory group membersh ip. Before discussing the trajectory group-specific effects and differences in the findings in the shortand long-runs, some attention should be given to whether the results generally support t he principal idea of deviance amplification. The deviance amplification hy pothesis maintains that an individuals offending patterns worsen as one goes from primary deviance to secondary 264

PAGE 265

deviance (Lemert, 1951). The mediating pathwaysblocked access to conventional opportunities, alteration of self-identity, and acquisition of delinquent peerssuggest that offending behavior should increase in quality and quantity (Link et al., 1989; Paternoster & Iovanni, 1989). This occurs because an individual now has little choice but to offend, comes to see oneself as a criminal, and is surrounded by individuals who are likely to exposure him or her to techni ques and reinforce their violent behavior. Overall, this suggests official intervention wi ll result in a turning point in a criminal trajectory that results in more problematic offending behavior than at the time of the official intervention (see Figure 1-1). W hen considered in a developmental and lifecourse perspective, however, this implies that there should be significantly more violence than would be expected at some time in the futu re had the individual not experienced justice system cont act. Thus, it is important to consider what the offending trajectory was prior to the intervention such as whether it was trending upward or downward. For the aggregated sample, the mean number of distinct violent crime acts was 0.36 in wave six, which was down from a high of 0.56 in wave two. Hence, the trajectory for the aggregated sample was declining between waves two and six (see Figure 8-1). The average number of distinct acts of violenc e that was committed during wave nine was 0.374 for the arrest group and 0.162 for the non-arrest group (see Figure 9-1). For the police contact treatmen t, these mean numbers were 0.415 and 0.113, respectively (see Figure 9-2). Both of these average values at wave nine for the treatment groups represent modest mean incr eases over the total sample mean at wave six. When viewed in this light, devianc e amplification is not especially large. 265

PAGE 266

When considering the wave nine outcomes, there are statistically and practically significant differences in the mean violence levels for those w ho experienced official intervention and those who did not. For the total sample, these relationships held when accounting for forty covariates in the short-run for both outcome measures (see Tables 9-13 and 9-14). Taken together, th is provides evidence for a c ausal effect of official intervention on subsequent violent offending in t he short-term. Criminal sanctions make matters worse in the short-run for this samp le on the whole. Whether these findings are also indicative of an amplification effect is subject to interpretation. For some, this issue may be paramount to fully evaluating the efficacy of labeling theory and for others this may be largely irrelevant. Future t heoretical discourse should expand upon this discussion. The findings reveal no evidence that official intervention serves as a different type of turning point for individuals following along distinct violent offending trajectories. That is, arrests or police contacts do not exac erbate violent behavior for some offending subpopulations but alleviate probl em behavior for others. Nevertheless, there is support for short-run trajectory group-specific treatment effects such that official intervention serves as a turning point in the criminal trajectories of low offenders but not for nonoffenders (treatment effect estimates c ould not be obtained for high offenders). Why might there be a statisti cally significant labeling effect for a low offender but not a non-offender? Recall, there were nontrivial differences between the low and nonoffending groups for at least one or more of the covariates across the demographic, neighborhood, family, school, peer association, values and ment al states, prior criminal behavior, and justice system contact subcategor ies (see Table 8-7). These statistically 266

PAGE 267

significant mean differences are largely in directions that are consistent with expectations from a variety of criminological theories. To name a few of these differences, non-offenders had more intact fam ily structures, greater attachment to his or her parent, higher aspirations for college, lower delinquent peers, weaker delinquent values, and lower prior justice system contact. Collectively, these factors might indicate that a youth has a greater stake in conformity and/or resources to th wart adverse effects of deviant labels. Those who are low o ffenders are, by definit ion, on an offending trajectory that is sandwiched between high offenders and non-offenders. These two types of individuals may have largely s ubscribed to nonconformity and conformity, respectively. Alternatively, low off enders might be more easily swayed one way or another. The evidence suggests that low offenders violent offending trajectories take a turn toward nonconformity following the expe rience of an official intervention. It is unlucky that treatment effects coul d not be estimated for the high offending group. Some suggest that the effect of a dev iant label will level off once an individual has charted a course of problematic behavior (Paternoster & Iovanni, 1989). The idea is that while chronic offenders may be less suscept ible to official intervention effects in later adolescence, this does not necessar ily mean that these individuals are not responsive to punishments in ear lier developmental stages. Bivariate evidence suggests that if t he independent variable of interest had any causal effects on violent crime it would make the violence of high offenders worse. While the author is unable to discuss treatm ent effect estimates for the high offending group, bivariate results provi de a certain degree of relevant information (see Figures 9-1 and 9-2). While there are statistically sign ificant short-run bivariate effects for low 267

PAGE 268

offenders across both official intervention measures, there is only a statistically significant short-run bivariate effect for high offenders for the police contact treatment although the practical mean difference betwe en the treated and control groups for the arrest treatment is also quite large. Then, labeling effects may still be observable even for those who are fairly committed to violent behavior, but future research is needed that addresses this specific issue. Sampson and Laubs (1997) discussion regarding the shortand long-run effects of criminal sanctions over the life-course is an important theoretical advancement. They suggest that labeling is a process that unfol ds over time and, as a consequence, one might be more likely to observe differences in behavior once sufficient time has passed for ones conventional opportunities to be bl ocked, self-identity to be converted, and peer influence to become problematic. The empirical findings are not consistent with their predictions regarding a long-run labeling effect. Rather, they are supportive of short-run labeling effect followed by long -run null effects for the total sample. Additionally, there is partial support for this same pattern of results fo r the low offending trajectory group. Why do short-run labeling effects vanish in the long-run when theory predicts that labeling effects in the long-run should actua lly be stronger? Aside from the possibility that the theory is simply wr ong, there are some reasons wh y these findings may not be entirely surprising. Police contacts and arre sts are repeatable treatments. This means that approximately three years had passed fr om the treatment per iod to the long-run violent crime observation period. During this time, individuals may or may not have experienced additional arrests and/or police contacts. The treatment and control groups 268

PAGE 269

that were highly balanced across the forty covariates including prior justice system contact and violent offending behavior becom e less comparable as time passes. In essence, the treatment and control groups can be thought to be blending together in a sense, which leads to the washing out of any long-run treatment effects. The issue of repeatable treatments is probably greater for police contacts and arrests relative to criminal s anctions such as incarceration. Hence, incarceration is a less common experience and there would be less mixing of the treatment and control groups long-term. Incarceration is also a more serious type of sanction that can be seen to more easily block ones access to conv entional opportunities (e.g., reporting of felony status on job applic ations), alter ones identity (e.g., one is sentenced for a crime), and expose one to deviant peers (e.g., socialization in jail/prison). In short, it might be expected that long-run sanction effe cts would be more likely to be observed for more serious types of official intervention. In short, w hat seems to be clear from the analysis is that labeling effects are confi ned to the short-term, which is probably a product of the fact that official interventi on, especially arrest and police contact, is a repeatable treatment. Testing t he long-run effects of official intervention on subsequent offending might be best accomplished by empl oying incarceration as the treatment experience and/or running addit ional analyses on subsamples of individuals who did not crossover with respect to their tr eatment status. The analysis raises another poignant ques tion requiring discussion: why is the labeling process triggered by a police contact for low offenders but not an actual arrest? Even at the bivariate level, there were statistically significant differences for all three trajectory groups in the shor t-run for the police contact tr eatment (see Table 9-2) but 269

PAGE 270

only for low offenders for the arrest treatment (see Table 9-1). Thus, there seems to be more evidence for labeling with a treatment c ondition that is conceivably less serious. These findings may be the result of different sources of data. Recall, arrests are officially recorded whereas police contacts are self-reported. There were more individuals who self-reported a police contac t than experienced an officially recorded arrest (cf. Figures 8-3 and 84). The self-reported measure may have captured official intervention experiences that were missed using official data. For instance, if an individual was arrested outside the relevant jurisdictions then this official intervention experience would not be examined and this i ndividual might be classified as not have received official intervention. However, the self-reported police contact measure would register any police contact. Fu ture research that uses eit her official records or selfreports is needed to more directly compare t he effects of different types of official intervention on crime and violenc e. The results suggest that intervening in the lives of youth that are more or less experimenting with violence (i.e., low offenders) may lead to the au gmentation of this form of criminal behavior. While this might be true for the other trajectory groups as well, this study did not support these conc lusions for non-offenders when accounting for forty covariates and did not provide the opportunity to exami ne treatment effect estimates for high offenders. Ironically, cont rolling violent crime for low offenders might be best accomplished by leaving this group to their own devices. The findings of a significant labeling effect echo Moffitts ( 1993) concerns that adolescence-limited offenders might become snared by the criminal justice system causing the normal desistance process that occurs for this group to go awry. 270

PAGE 271

If police contacts are sufficient to set in mo tion the labeling proces s, the result is a catch twenty-two. While police can use their discretion to decide whom or when to make an arrest, contacting an individual who is suspected to have committed a crime is the mark of sound police work. Failing to per form job functions is obviously problematic in many ways but this research suggests that actually performing them might very well be problematic as wellat least from the point of view of controlling subsequent violent crime. There are issues pertaining to justice and fairness that support official intervention in all instances in which a crime has occurred. And, for sanctions beyond those studied here, incarceration might hol d promise for controlling crime, albeit temporarily, via incapacitation. The implications of the cu rrent research for policy ar e such that nonintervention should be considered as a response to the vi olent offending of low offenders if the goal is to minimize, or more precisely fail to worsen, their offending behavior. The findings here suggest that official intervention is t he antithesis of a panacea for the violent crime problem. In other words, e fforts aimed at controlling vi olent crime should look well beyond traditional approaches to cr iminal justice. For instance, taking a public health approach to the study of violence might help to redefine the problem and involve more resources and agencies in the process of devising and implem enting sustainable solutions to the problem at hand. With t hat said, there are a number of avenues for research that address the limit ations of this study and st udies that have been conducted to date that should be explored before offi cial intervention is deemed an ineffective approach to curbing crime. These are now discussed. 271

PAGE 272

Directions for Future Research The current study provides us with some answers to several important research questions pertaining to the effects of official intervention on subsequent violent offending behavior. However, these answers are certainly not the final word on the matter; rather, the study as a whole is probably better seen as a framework around which future research can build. While add itional studies are needed on the moderating role of various factors including race, gender, and stakes in conformity, the potential for policy-relevant information st emming from theoretically dr iven investigations on the moderating role of criminal history helps to ensure this is an area that will continue to receive empirical attention in the future. Assessments of the ideas advanced by various deterrence, labeling, and developmental and life-course scholars might better inform public policies on the control of crime and vi olence and provide str onger tests of theory if they adhere to some basic recommendations These suggestions for future research are primarily geared toward observational studies. Future research is needed that compares different gradations of official intervention to better assess wh ether labeling or deterrent effects are more likely at different stages in the justice system. I deally, a study would ex amine four different official interventions including police contact, arrest without a formal sanction, arrest followed by an alternative sanction, and arrest followed by a stint in a jail or prison. Doing so would help to determine whether labeling and/or deter rent effects were triggered with none, one, some, or all types of justice system contact. To prevent any differences being attributable to methodological artifacts, similar sources of data, such as official records, should be used whenever possible to examine the effects of the different treatments on subs equent offending behavior. 272

PAGE 273

Future research should also compare t he results of analyses that use official records with those that identify official inte rvention experiences with self-report data. Both methods of measuring official intervention have certain limitations. Official records are probably quite valid for arrests and stro nger sanctions that follow an arrest. However, if one has data on official justice system contact only from a single agency then certain criminal sanctions may go unanal yzed. In addition to this problem, the validity of official records ma y be problematic for police contact experiences specifically. These may be documented less consistently and syst ematically by police officers. Selfreported data might be more valid in some instances but a variety of factors could influence a respondents answers to various surv ey items. Then, future studies should use both forms of data when possible. Despite the limitation of self -report information in general, offending behavior is better captur ed with this type of behavioral measure. Official records will substantially underestimate the true number of offenses that are being committed. As a general rule, future re search should seek to use self-report data to measure offending behavior as well as ot her covariates that serve as control variables. Researchers should use official records data for official intervention experiences and, when available, supplemen t this with self-repor t data for comparison purposes. Studies examining sanction effects should also attempt to investigate the effects of an official intervention on prevalence, vari ety, and frequency of offending within the same study. Employing these different behavioral measures allows one to assess to what extent justice system contact terminates a criminal car eer, to what extent it limits the number of distinct acts that an individual engages in, and to what extent it reduces 273

PAGE 274

the total number of offenses an individ ual commits, respectively. This may be considered in conjunction with the aforementioned considerat ions regarding the exploration of different gradations of crimi nal sanctions. For example, deterrence may lessen the frequency or variety of offending only if an individual is actually sanctioned with a fine or community service (not just arrested) but may actually result in the cessation of offending altogether if the individual experie nces a highly undesirable outcome such as incarceration. As anot her possibility, a criminal sanction might differentially influence frequency and variety of offending. Future research should consider these possibilities. Research on sanction effects needs to continue to examine behavior other than violence and compare differences in the effe cts of official intervention across different types of behavior. While certain data sets may not be able to support such specificity, the effects of criminal sanc tions would ideally be examined both on the specific behavior one was sanctioned for and on offending behavio r more generally. That is, if an individual was arrested for the crime of assault, one might seek to determine the deterrent or labeling effect of the official intervention on several behavioral outcomes including assault, violent cr ime, and criminal behavior in general. This speaks to the specificity or the generality in the effects of official intervention on subsequent offending behavior. All of these recommendations thus far have not considered t he developmental and life-course aspects of sanction effects research. Still, there is a healthy set of issues to address already. I conclude this work by making some recomm endations for future studies that seek to explore differential s anction effects for fundamentally different types 274

PAGE 275

of offenders and/or to examine divergent short-run and long-r un effects. Of course, the general considerations discussed to this point should also be kept in mind for investigations into sanction effects that are informed by ideas from developmental and life-course criminology. While the current research did not result in the conclusion that sanction effects operated in opposite directions for distinct violent offending subpopulations, there was some evidence that sanctions mattered more for low offenders re lative to non-offenders. Given the acknowledgment t hat a small proportion of individuals are responsible the bulk of the crime problem (Wol fgang et al., 1972), it is highly important to evaluate the causal relationship between official intervent ion and subsequent offending behavior. As it turned out, the treatment effects for high offenders could not be estimated given adequate covariate balance could not be achiev ed. Obtaining group equivalences on covariates for chronic offender s may be most challenging wit hout a fairly large sample size since they only represent a small propor tion of offenders (see also Haviland et al., 2007; 2008). Therefore, future research should seek to use other data sets that may have larger sample sizes or might be able to create covariate balance for small, high offending trajectory groups. Relatedly, in cases where the data are not stretched beyond capacity, future research might explore trajectory group-specific effects of alternative latent class growth analysis models when these alternative models have a sizeable probability of represent ing the data, albeit one that may no t the largest. For instance, researchers might consider reporti ng the results of analyses from both a three class model and a four class model should there be a non-trivia l probability that each model is the correct one and the more co mplex model yields information may be 275

PAGE 276

relevant for public policy. In short, res earch should continue to consider official intervention experiences as life transitions that may be turning points, for better or worse, in the lives of di fferent types of offenders. Future research should find promise in the integrated me thodological approach used in this study that blends latent class growth and propensity score matching analyses. This methodology is especially well-suited to invest igate whether a life transition serves as a turning point in a crimi nal trajectory for different types of offending subpopulations. However, it wa s acknowledged that official intervention is technically a repeated treatment and the integrated LCGA-PSM analysis does not directly deal with this problem. To address this concern, an alternative methodological approach to the one taken here might be to combine latent class growth analysis with inverse probability-to-treatment weight ing (IPTW). To my knowledge, no study to date in the field of criminology that has employed this me thodological design, but it might be fruitful for sanction effects researchers to consider doing so. IPTW permits a researcher to assess whether there is intraindividual change in offending beh avior that is a function of intra-individual changes in treatment status while we ighting the data to address confounding (Samspon, Laub, & We imer, 2006). At the same time, the question as to whether official intervention serves as a turn ing point in a criminal trajectory may not be as straightforwardly answered with this approach. Thus, considering official intervention as a repeatable treatment will result in the need to carefully consider the implications for testing theories on sanction effects in a dev elopmental and life-cour se perspective, but this is no reason to shy away from this potentially important empirical endeavor. 276

PAGE 277

The issue of repeatable treat ments also has implications for investigating longterm sanction effects. There are difficulties in estimating the effects of criminal sanctions in the long-run. As was previously discussed, indi viduals who did not experience an official interv ention during the treatment period may experience one later and those who did experience one during t he treatment period may not subsequently experience one. The groups that were form ed during the treatment period are probably less comparable the later in time one examines t he effects of official intervention. Still, this doesnt change the fact that investigat ions into long-run sanctions effects are theoretically important (Sampson & Laub, 1997). Future research might conduct additional analyses to determine the extent which these diminishing effects with time are the result of temporally sensitive deterre nt or labeling effects or whether they are simply observed due to treatment experie nces following the designated treatment period. Haviland and colle agues (2007; 2008) provide some guidance for conducting such an analysis. Finally, future research should conti nue to employ strong methodological approaches such as propensity score matching analysis in conjunction with latent class growth analysis to address similar research questions raised herein. While propensity score analyses support strong causal inference in general, little is known about the best methods to use when the technique is integr ated with latent class growth analysis. Propensity score methods give the researc her a multitude of options, which may be overwhelming at times. There is more t han one way to develop a treatment model (e.g., theoretically-driven, empirically-driven), more than one approach to estimate propensity scores (e.g., logistic regre ssion, probit regression), more than one strategy to address 277

PAGE 278

278 confounding (e.g., matching, we ighting, stratification), more than one way to employ a chosen strategy such as matching (e.g., optimal or greedy; 1:1, 1:K, or full), and more than one approach to evaluate the successfulne ss of the propensity score analysis for creating equivalent treatment and control groups (e.g., st andardized bias statistics, ttests). It would appear then that these advanced, integrated methods are in need of additional research and possi bly simulation analysis to help guide future research endeavors. While some simulation analysis has been conducted on propensity score methods generally, it might be helpful to conduc t these informative studies on data that employs LCGA-PSM integrated methods. Ultimately, propens ity score analysis hinges on its ability to create covariate balance on t he covariates used to predict treatment status but some approaches may do this better than others in the context of trajectory group-specific treatment effects and future re search should seek to determine this to make the statistical analysis more acce ssible to researc hers. This discussion has outlined an agenda for sanction effects studies that touches on many concerns of other scholars and intr oduces several new issues to consider. Taken together, there is a demonstrated need for future research on the topic to address certain limitations of the current study as well as weaknesses of sanction effects studies in general. Notwithstanding t he discourse on future research, the current study has taken an important step in the ri ght direction by raising policy-relevant, developmentally sensitive empirical questi ons regarding the ef fects of official intervention on subsequent violent o ffending behavior and addressing them with carefully chosen data and quantitative techniques that are well-suited for the task.

PAGE 279

LIST OF REFERENCES Ageton, S. S., & Elliott, D. (1974). The e ffects of legal processing on delinquent orientations. Social Problems, 22 87-100. Agnew, R. (1992) Foundation for a general st rain theory of crime and delinquency. Criminology, 30 47-87. Akers, R. L. (1968). Problem s in the sociology of devi ance: Social definitions and behavior. Social Forces, 46 455-465. Akers, R. L. (1990). Rational choice, det errence, and social learning theory in criminology: The path not taken. Journal of Criminal Law & Criminology, 81 653676. Akers, R. L. (1998). Social structure and social learni ng: A general theory of crime and deviance. Boston, MA: Northeastern University Press. Akers, R. L., & Sellers, C. S. (2009). Criminological theories: Introduction, evaluation, and application (5th ed.). New York, NY: Oxford University Press. Alison, P. (2001). Missing data. Thousand Oaks, CA: Sage Publications. Andenaes, J. (1974). Punishment and deterrence Ann Arbor, MI: University of Michigan Press. Austin, J. (2003). Why criminology is irrelevant. Criminology & Public Policy, 2 557-564. Bauer, D. J., & Curran, P. (2003). Distributional assumpti on of the growth mixture models: Implications for over-extraction of latent trajectory classes. Psychological Methods, 8 338-363. Bauer, D. J., & Curran, P. (2004). The int egration of continuous and discrete latent variable models: Potential problems and promising opportunities. Psychological Methods, 9 3-29. Baumer, E. (1997). Levels and predictors of recidivism: The Malta experience. Criminology, 35 601-628. Beccaria, C. (1963) [1764]. On Crimes and Punishment. New York, NY: MacMillan Publishing Company. Becker, G. S. (1968). Crime and punishment: An economic approach. The Journal of Political Economy, 76 169-217. Becker, H. S. (1963). Outsiders: Studies in t he sociology of deviance. New York, NY: The Free Press. 279

PAGE 280

Belson, W. A. (1968). The ex tent of stealing by London boys and some of its origins. Advancement of Science, 25 171-184. Bentham, J. (1948) [2000]. An Introduction to the principles of morals and legislation New York, NY: Hafner Publishing. Berg, I., Consterdine, M., Hullin R., McGuire, R., & Tyrer, S. (1978). The effect of two randomly allocated court procedures on truancy. British Journal of Criminology 18, 232-244. Berk, R. A., Campbell, A., Klap, R., & Western, B. (1992). The deterrent effect of arrest in incidents of domestic violence: A Baye sian analysis of four field experiments. American Sociological Review, 57 698-708. Bernburg, J. G. (2002). State reaction, life-course outcomes, and structural disadvantage: A panel study of the impact of formal criminal labeling on the transition to adulthood. Doctoral Dissertation, Un iversity at Albany, State University of New York. UMI no.: 3058971. Bernburg, J. G., & Krohn, M. D. (2003). Labeling, life chances, and adult crime: The direct and indirect effects of official intervention in adolescence on crime in early adulthood. Criminology, 41 1287-1318. Bernburg, J. G., Krohn, M. D., & Rivera, C. J. (2006). Official labeling, criminal embeddedness, and subsequent delinquency: A longitudinal test of labeling theory. Journal of Research in Crime and Delinquency, 43 67-88. Best, J. (2004). Deviance: Career of a concept. Bellmont, CA: Wadsworth. Bhati, A. S. (2007). Studying the effects of incarcerat ion on offending trajectories: An information-theoretic approach Technical Report. Washington, DC: Justice Policy Center, The Ur ban Institute. Bhati, A. S., & Piquero, A. R. (2008). Estimating the impact of incarceration on subsequent offending trajectories: Dete rrent, criminogenic, or null effect? Journal of Criminal Law & Criminology, 98, 207-253. Binder, A., & Meeker J. W. (1988). Exper iments as reforms. Journal of Criminal Justice, 16, 247-358. Bingenheimer, J. B., Br ennan, R. T., & Earls, F. (2005). Firearm violence exposure and serious violent behavior. Science, 27 1323-1326. Blumer, H. (1969). Symbolic interactionism: Perspective and method. Berkeley, CA: University of California Berkeley Press. Blumstein, A., & Cohen, J. (1987). Characterizing criminal careers. Science, 28 985991. 280

PAGE 281

Blumstein, A., & Nagin, D. (1976). The deterrent effect of legal sanctions on draft evasion. Stanford Law Review, 29 241-75. Blumstein, A., Cohen, J., & Farrington, D. P. (1988). Criminal career research: Its value for criminology. Criminology, 26 1-35. Blumstein, A., Cohen, J ., & Nagin, D. (1978). Deterrence and incapacitation: Estimating the effects of criminal sanctions on crime rates Washington, DC: National Academy of Sciences. Blumstein, A., Cohen, J., Roth, J. A., & Visher, C. A. (1986). Criminal careers and career criminals (Vols. 1 and 2). Washington, DC: National Academy Press. Bonger, W. A. (1916) [1905]. Criminality and economic conditions Boston, MA: Little Brown. Bowers, W. J., & Pierce, G. L. (1980). Deterrenc e or brutalization: What is the effect of executions? Crime & Delinquency, 26 453-484. Braithwaite, J. (1989). Crime, shame and reintegration Cambridge, United Kingdom: Cambridge University Press. Brame, B., Nagin, D. S ., & Wasserman, L. (2006). Ex ploring some analytical characteristics of finite mixture models. Journal of Quantitative Criminology, 22 31-59. Brennan, P. A., & Mednick, S. A. (1994). Learning theory appr oach to the deterrence of criminal recidivism. Journal of Abnormal Psychology 103 430-440. Broidy, L. M., Tremblay, R. E., Brame, B., Fergusson, D., Horwood, J. L., Laird, R., Moffitt, T. E., Nagin, D. S ., Bates, J. E., Dodge, K. A., Loeber, R., Lynam., D. R., Pettit, G. S., & Vitaro, F. (2003). Developmental trajectories of childhood disruptive behavior and adolescent delinquency. Developmental Psychology, 39 222-245. Brown, D. W. (1978). Arrest rates and crime rates: When does a tipping effect occur? Social Forces, 57 671-682. Bryk, A. S., & Raudenbush, S. W. (1987). Application of hierarchical linear models to assessing change. Psychological Bulletin, 101 147-158. Burgess, R. L., & Akers, R. L. (1966). A differential association-reinforcement theory of criminal behavior. Social Problems, 14 128-147. Bursik Jr., R. J. (1988). Social disor ganization and theories of crime and delinquency: Problems and prospects. Criminology, 26 519-552. 281

PAGE 282

Bushway, S. D., Thornberry, T. P., & Krohn, M. D. (2003). Desistance as a developmental process: A comparison of static and dynamic approaches. Journal of Quantitative Criminology, 19 129-153. Cameron, M. O. (1964). The Booster and the Snitch New York, NY: Free Press. Caspi, A. (1998). Personality development acro ss the life course. In N. Eisenberg & W. Daom (Eds.), Handbook on Child Psychology (5th ed.). New York, NY: Wiley. Catalano, R. F., & Hawkins, J. D. (1996). The social development model: A theory of antisocial behavior. In J. D. Hawkins (Ed.), Delinquency and crime: Current theories (pp. 149-147 ). Cambridge, United Kingdom: Cambridge University Press. Chambliss, W. J. (1967). Ty pes of deviance and the effectiveness of legal sanctions. Wisconsin Law Review, 1967 703-719. Chiricos, T., Barrick, K., Bales, W., & Bont ranger, S. (2007). The labeling of convicted felons and its consequences for recidivism. Criminology, 45 547-581. Clark, R. D. (1988). Celerity and specific deterrence: A look at the evidence. Canadian Journal of Criminology, 30 109-120. Clarke, R. V., & Cornish, D. B. (1985). M odeling offenders decis ion: A framework for research and policy. Crime & Justice: An Annual Review of Research, 6 147185. Cloninger, R. C. (1986). A unified biosocial theory of personality and its role in the development of anxiety states. Psychiatric Developments, 3 167-225. Cook, P. J. (1980). Research in criminal deterrence: Laying the groundwork for the second decade. Crime and Justice: A Review of Research, 2 221-268. Cooley, C. H. (1902). Human nature and the social order. New York, NY: Scribners Sons. Cornish, D. B., & Clarke, R. V. (1986). The reasoning criminal New York, NY: SpringerVerlag. Dannefer, D. (1984). Adult dev elopment and social theory: A paradigmatic reappraisal. American Sociological Review, 49 100-116. Davis, F. (1961). Deviance disavowal: T he management of strained interaction by the visibly handicapped. Social Problems, 9 120-132. DeJong, C. (1997). Specific deterrence and surv ival analysis: Integrating theoretical and empirical models of recidivism. Criminology, 35 561-76. 282

PAGE 283

Dunford, F. W., Huizinga, D., & Elliot, D. S. (1990). Roles of arrest in domestic assault: The Omaha police experiment. Criminology, 28 183-206. D'Unger, A. V., Land, K. C., McCall, P. L., & Nagin, D. S. (1998). How many latent classes of delinquent/criminal careers? Results from mixed poisson regression analyses. American Journal of Sociology, 103 1593-1630. Eggleston, E. P., Laub, J. H., & Sampson, R. J. (2004). Methodologi cal sensitivities to latent class analysis of longterm criminal trajectories. Journal of Quantitative Criminology, 20 1-26. Eisner, M. (2003). Long-term historical trends in violent crime. Crime and Justice: A Review of Research, 30 83-142. Elder Jr, G. H., Gimbel, C., & Ivie, R. (1991). The case of military service and war. Military Psychology, 3 215-231. Elder, G. H. (1985). Life-course dynamics: Trajectories and transitions, 1968-1980 Ithaca, NY: Cornell University Press. Elder, G. H. (1986). Military times and turning points in mens lives. Developmental Psychology, 22, 233-245. Elias, N. (1939a) [1978]. The civilizing process: The history of manners (Vol. 1) New York, NY: Urizen Press. Elias, N. (1939b) [1982]. The civilizing process: Power and civility (Vol. 2). Oxford, United Kingdom: Blackwell. Huizinga, D. H., & Espiritu, R. (1999). Delinquent behavior of youth in the juvenile justice system before, during, and after different court dispositions. Unpublished report. Pittsburgh, PA: National Center for Juvenile Justice, National Juvenile Court Judges Association. Ezell, M. E., & Cohen, L. E. (2005). Desisting from crime: Continuity and change in long-term crime patterns of serious chronic offenders. New York, NY: Oxford University Press. Fagan, J. A., Kupchick, A., & Liberman, A. (2003). Be careful what you wish for: The comparative impacts of juvenile versus criminal court sanctions on recidivism among adolescent felony offenders. New York, NY: Columbia Law School Public Law and Legal Theory Working Group. Farrington, D. P. (1977). The effects of public labeling. British Journal of Criminology, 17, 112-125. 283

PAGE 284

Farrington, D. P. (1986). A ge and crime. In M. Tonry & N. Morris (Eds.), Crime and Justice (Vol. 7). Chicago, IL: University of Chicago Press. Farrington, D. P. (1989). Self -reported and official offending from adolescence to adulthood. In M. W. Klein (Ed.), Cross-national research in self-reported crime and delinquency (pp. 399-423). Dordrecht, Netherlands: Kluwer. Farrington, D. P. (1992). Criminal car eer research in the United Kingdom. British Journal of Criminology, 3 2, 521-536. Farrington, D. P. (1998). Studying changes with i ndividuals: The causes of offending. In M. Rutter (Ed.), Studies of psychosocial risk: The power of longitudinal data (pp. 158-183). New York, NY: Cambridge University Press. Farrington, D. P. (2005). The in tegrated cognitive antisocial pot ential (ICAP) theory. In D. P. Farrington (Ed.), Integrated Developmental and Life-Course Theories of Offending (pp. 73-92). New Brunswick, NJ: Transaction Publishers. Farrington, D. P. (2006). Build ing developmental and life-cour se theories of offending. In. F. T. Cullen, J. P. Wri ght, & K. R. Blevins (Eds.), Taking stock: The status of criminological theories (pp. 335-366). New Brunswick, NJ: Transaction Publishers. Featherman, D. L. (1983). Li fe-span perspective in social science research. In P. B. Baltes & O. G. Brim, Jr. (Eds.), Life-span development and behavior (Vol. 5, pp. 1-57). New York, NY: Academic Press. Felson, R. B. (2009). Violence, crime, and violent crime. International Journal of Conflict and Violence, 3 23-29. Fergusson, D. M., Horwood, L. J., & Nagin, D. S. (2000). Of fending trajectories in a New Zealand birth cohort. Criminology, 38 525-552. Fisher, R. A. (1925). Statistical methods for research workers. Edinburgh, United Kingdom: Oliver & Boyd. Fisher, R. A. (1926). The arr angement of field experiments. Journal of the Ministry of Agriculture of Great Britain, 33 505-513. Garfinkel, H. (1956). C onditions of successful degradation ceremonies. American Journal of Sociology, 61 420-424. Gendreau, P., Little, T., & Goggi n, C. (1996). A meta-analysis of the predictors of adult offender recidivism: What works! Criminology, 34, 575-607. 284

PAGE 285

Gibbs, J. P. (1972). Issues in defining deviant behavior. In R. A. Sc ott & J. D. Douglas (Eds.), Theoretical perspectives on deviance (pp.39-68). New York, NY: Basic Books. Gibbs, J. P. (1966). Concepts of dev iant behavior: The old and the new. Pacific Sociological Review, 9 9-14. Gibbs, J. P. (1986). Deterrence theory and research. In G. B. Melton (Ed.), The law as a behavior instrument (pp. 87-130). Lincoln, NE: Univ ersity of Nebraska Press. Gibbs, J.P. (1975). Crime, Punishment, and Deterrence. New York, NY: Elsevier. Goffman, E. (1963). Stigma: Notes in the management of a spoiled identity Englewoods Cliffs, NJ: Prentice-Hall. Gold, M., & Williams, J. (1970). National study of the aftermath of apprehension. Prospectus 3 3-12. Goldstein, H. (1995). Multilevel statistical models (2nd ed.). London, United Kingdom: Edward Arnold. Gottfredson, D. M. (1999). Effects of judges sentencing dec isions on criminal careers (National Institute of Justice Research in Brief) Washington, DC: U.S. Department of Justice. Gottfredson, M. & Hirschi, T. (1988). Sci ence, public policy, and the career paradigm. Criminology, 26 37-56. Gottfredson, M., & Hirschi, T. (1990). A general theory of crime Stanford, CA: Stanford University Press. Gottfredson, M. (2005). Offender classifications and treatment effects in developmental criminology: A propensity/ event consideration. Annals of the American Academy of Political and Soci al Science, 602, 46-56. Gove, W. R. (1980). Labeling and mental illness: A crit ique. In W. R. Gove (Ed.), The labeling of deviance: Ev aluating a perspective (2nd ed., pp. 53-99). Beverly Hills, CA: Sage. Gove, W. R. (1985). The effect of age and gender on deviant behavior. A biopsychosocial perspective. In A. S. Rossi (Ed.), Gender and the life course (pp. 115-144). New York, NY: Aldine. Greenberg, D. F. (1981). Methodological issues in survey research on inhibition of crime. Journal of Criminal Law & Criminology, 72 1094-1101. Greenberg, D. F. (1977). Delinquency and the age structure of society. Crime, law and social change, 1 189-223. 285

PAGE 286

Guo, S., & Fraser M. W. (2010). Propensity score analysis: Statistical methods and applications. Thousand Oaks, CA: Sage Publications. Hagan, J., & Palloni, A. (1990). The social reproduction of a criminal class in working class London. American Journal of Sociology, 96, 265-299. Hansen, B. B. (2004). Full matc hing in an observational study of coaching for the SAT. Journal of the American Statistical Association, 99 609-618. Harris, A. (1975). Imprisonment and the ex pected value of criminal choice: A specification and test of aspects of the labeling perspective. American Sociological Review, 40 71-87. Haviland, A., Nagin, D. S., & Rosenbaum, P. R. (2007). Combin ing propensity score matching and group-based trajectory analysis in an observational study. Psychological Methods, 12 247-267. Haviland, A., Nagin, D. S., Rosenbaum, P. R., & Tremblay, R. E. (2008). Combining group-based trajectory modeling and pr opensity score matching for causal inference in nonexperimental longitudinal data. Developmental Psychology, 44 422-436. Heineke, J. M. (1978). Economic models of criminal behaviour: An overview. In J. M. Heineke (Ed.), Economic models of criminal behaviour Amsterdam, Holland: North Holland Publishing Co. Hemphill, S. A., Toumbourou, J. W., Herrenkohl, T. I., McMorris, B. J., & Catalano, R. F. (2006). The effect of school suspensio ns and arrests on subsequent adolescent behavior in Australia and the United States. Journal of Adolescent Health 39, 736-744. Hindelang, M. J. (1978). Race and involv ement in common law personal crimes. American Sociology Review, 43 93-109. Hindelang, M. J., Hirschi, T., & Weis, J. G. (1981). Measuring delinquency. Beverly Hills, CA: Sage Publications. Hirano, K., & Imbens, G. W. (2001). Estimation of causal effects using propensity score weighting: An application to dat a on right heart catheterization. Health Services & Outcomes Research Methodology, 2 259-278. Hirschel, J. D., & Hutchins on, I. W. (1992). Females spouse abuse and the police response: The Charlotte, North Carolina experiment. Journal of Criminal Law & Criminology, 83 73-119. Hirschi, T. (1969). Causes of delinquency. New Brunswick, NJ: Transaction Publishers. 286

PAGE 287

Hirschi, T., & Gottfredson, M. (1983). Age and the explanation of crime. American Journal of Sociology, 89 552-584. Ho, D. E., Imai, K., King, G., & Stuart, E. A. (2007). Matchi ng as nonparametric preprocessing for reducing model dependence in parametric causal inference. Political Analysis, 15, 199-236 Hobbes, T. (1967) [1651]. Leviathan. Oxford, England: Clarendon Press. Hodges, J. L., & Lehmann, E. L. (1962). Rank methods for combination of independent experiments in the analysis of variance. Annals of Mathematical Statistics, 33, 482-497. Holland, P. W. (1986). Statistics and causal inference. Journal of the American Statistical Association, 81, 945-970. Horney, J., Osgood, D. W., & Marshall, I. H. (1995). Criminal career s in the short-term: Intra-individual variability in crime and its relation to local life circumstances. American Sociological Review, 60 655-673. Horwitz, A., & Wasserman, M. (1979). The effect of so cial control on delinquent behavior: A longitudinal test. Sociological Focus, 12, 53-70. Huizinga, D. H., & Esbensen, F. (1992). An arresting view of juvenile justice. School Safety Spring 15-17. Huizinga, D., & Henry, K. L. (2008). The effect of arrest and justice system sanctions on subsequent behavior: Findings from longi tudinal and other studies. In A. M. Liberman (Ed.), The long view of crime: A synt hesis of longitudinal research (pp. 220-254). New York, NY: Springer. Huizinga, D., Elliott, D. S ., & Dunford, F.W. (1986). The effect of arre st on subsequent delinquent behavior. Paper presented at the Annual Meetings of the Academy of Criminal Justice Sciences, Orlando, FL. Huizinga, D., Weiher, A. W., Espiritu, R., & Esbensen, F. (2003). Delinquency and crime: Some highlights from the Denver Y outh Survey. In T. P. Thornberry & M. D. Krohn (Eds.), Taking stock of delinquency: An over view of findings from contemporary longitudinal studies New York, NY: Kluwer Academic/Plenum Publishers. Imai, K., King, G., & Stuart, E. A. (2006) Misunderstanding among ex perimentalists and observationalists about causal inference. Journal of the Royal Statistical Society, Series A 171 481-502. Jewkes, Y. (2004). Media and Crime Thousand Oaks, CA: Sage Publications. 287

PAGE 288

Joffe, M. M., & Rosenbaum, P. R. (1999). Invited commentary: Propensity scores. American Journal of Epidemiology, 150 327-333. Jones, B. L., & Nagin, D. S. (2007). Adv ances in group-based trajectory modeling and an SAS procedure for estimating them. Sociological Methods Research, 35 542571. Jones, B. L., Nagin, D. S., & Roeder, K. (2001). A SAS procedure on mixture models for estimating developmental trajectories. Sociological Methods Research, 29 374393. Kandel, D. B. (1975). Stages in adolescent involvement in drug use. Science, 190 912914. Kaplan, H. B., & Damphouse, K. R. (1997). Negative social sanctions, self-derogation, and deviant behavior: Main and interactive effects in longitudinal perspective. Deviant Behavior 18(1), 1. Kass, R. E., & Wasserman, L. (1995). A refe rence Bayesian test for nested hypotheses and its relationship to the Schwarz Criterion. Journal of the American Statistical Association, 90 928-934. Kennedy, D. M. (2008). Old wine in new bo ttles: Policing and the lessons of pulling levers. In D. Weisburd & A. A. Braga (Eds.), Policing innovation: Contrasting perspectives. New York, NY: Cambridge University Press. Kleck, G., Sever, B., Li, S., & Gertz, M. (2005). The mi ssing link in general deterrence research. Criminology, 43 623-660. Klein, M. W. (1974). Labeling, deterrence, and recidivism: A study of polic e dispositions of juvenile offenders. Social Problems, 22, 292-303. Klein, M. W. (1986). Labeli ng theory and delinquency policy: An experimental test. Criminal Justice and Behavior, 13 47-79. Kobrin, S., Lubeck, S., Hansen, E. W., & Yeaman, R. L. (1972). The deterrent effectiveness of criminal justice strategies (Final report to the National Institute of Law Enforcement and Criminal Justice, Law Enforcement and Assistance Administration) Washington, DC: United States Government Printing Office. Krohn, M.D., Lopes, G., & Ward, J.T. (in press). Effects of police intervention on later offending. In D. P. Farrington & J. Murray (Eds.), Labeling theory: Empirical tests, Advances in Criminological Theory, 18 New Brunswick, NJ: Transaction Publishers. 288

PAGE 289

Krohn, M. D., Ward, J. T., T hornberry, T. P., Lizotte, A. J., & Chu, R. (2011). The cascading effects of gang membership across the life-course. Unpublished Manuscript. Kupchik, A. (2006). Judging juveniles: Prosecuting adolescents in adult and juvenile court. New York, NY: New York University Press. Lahey, B. B., & Waldman, I. D. (2005). A developmental model of the propensity to offend during childhood and adolescence. In D. P. Farrington (Ed.), Integrated developmental and life-course theories of offending (pp. 15-50). New Brunswick, NJ: Transaction Publishers. Land, K., C., & Nagin, D. S. ( 1996). Micro-models of criminal careers: A synthesis of the criminal careers and life course approaches via semiparametric mixed poisson regression model, with em pirical applications. Journal of Quantitative Criminology, 12 163-191. Land, K., C., McCall, P. L., & Nagin, D. S. (1996). A comparison of poisson, negative binomial, and semiparametric mixed poiss on regression models with empirical applications to criminal careers data. Sociological Methods & Research, 24, 387442. Laub, J. H., & Sampson, R. J. (2003). Shared beginnings, divergent lives: Delinquent boys to the age 70. Cambridge, MA: Harvard University Press. Laub, J. H., Nagin, D. S., & Sampson, R. J. (1998). Trajectories of change in criminal offending: Good marriages and the desistance process. American Sociological Review, 63 225-238. Laub, J. H., Sampson, R. J., & Sweeten, G. A. (2006). Assessing Sampson and Laubs life-course theory of crime. In F. T. Cullen, J. P. Wri ght, & K. R. Blevins (Eds.), Taking stock: The status of criminological theory (pp. 313-333). New Brunswick, NJ: Transaction Publishers. LeBlanc, M. (1997). A generic co ntrol theory of the criminal phenomenon: The structural and dynamic statements of an integrated multilayered control theory. In T. P. Thornberry (Ed.), Developmental theories of crime and delinquency (pp. 215-225). New Brunswick, NJ: Transaction Publishers. LeBlanc, M. (2005). An integrative personal control theory of deviant behavior: Answers to contemporary empirical and theo retical developmental criminology issues. In D. P. Farrington (Ed.), Integrated Developmental and LifeCourse Theories of Offending (pp. 125-163). New Brunswick, NJ: Transaction Publishers. Lemert, E. M. (1951). Social pathology: A systematic approach to the theory of sociopathic behavior. New York, NY: McGraw-Hill. 289

PAGE 290

Lemert, E. M. (1967). Human deviance, social problems, and social control Englewoods Cliffs, NJ: Prentice-Hall. Liazos, A. (1972). The poverty of the sociolog y of deviance: Nuts, sluts, and preverts. Social Problems, 20 103-120. Link, B. G. (1982). Mental patient status, work, and income: An examination of the effects of a psychiatric label. American Sociological Review, 47 202-215. Link, B. G. (1987). Understanding labeling effects in the area of mental disorders: An assessment of the effects of expectations of rejection. American Sociological Review, 52, 96-112. Link, B. G., Cullen, F. T., Fr ank, J., & Wozniak, J. F. ( 1987). The social rejection of former mental patients: U nderstanding why labels matter. American Journal of Sociology, 92 1461-1500. Link, B. G., Cullen, F. T., Struening, E., Shrout, P. E., & Dohrenwend, B. P. (1989). A modified labeling theory appr oach to mental disorders: An empirical assessment. American Sociological Review, 54 400-423. Little, R. J. A., & R ubin, D. B. (1987). Statistical analysis with missing data (2nd ed.). New York, NY: Wiley. Locke, J. (1975) [1690]. An essay concerning human understanding Oxford, United Kingdom: Clarendon Press. Locke, J. (1988) [1689]. Two treatises of government New York, NY: Cambridge University Press. Loeber, R. (1991). Questions and advances in t he study of developmental pathways. In D. Cicchetti & S. Toth (Eds.), Models and integrations Rochester Symposium on Developmental Psychopathology. Rochester, NY: University of Rochester Press. Loeber, R., & LeBlanc, M. (1990). Towa rd a developmental criminology. Crime and Justice: A Review of Research, 12 375-437. Lombroso-Ferrero, G. (1911). Criminal man: According to the classification of Cesare Lombroso. New York, NY: Putnams Sons. MacDonald, J., Bluthenthal, R. N. Golinelli, D., Kofner, A., Stokes, R. J., Sehgal, A., Fain, T., & Beletsky, L. (2009). Neighborhood effects on crime and youth violence: The role of business improvement districts in Los Angeles (Technical Report). Santa Monica, CA : RAND Corporation. Males, M., & Macallair, D. ( 1999). Striking out: The failure of Californias three strikes and youre out law. Stanford Law and Policy Review, 11 65-74. 290

PAGE 291

Mankoff, M. (1971). Societal reaction and career devianc e: A critical analysis. Sociological Quarterly, 12 208-218. Marx, K. (1964) [1844]. Economic and philosophi cal manuscripts of 1844. In K. Marx, Early Writings New York, NY: McGraw-Hill. Matsueda, R. L., Kreager, D., & Huizinga, D. (2006). Dete rring delinquents: A rational choice model of theft and violence. American Sociological Review 71, 95-122. Maxwell, C. D., Garner, J. H., & Fagan, J. A. (2002). The preventativ e effects of arrest on intimate partner violence: Research, policy and theory. Criminology & Public Policy, 2 51-80. McAra, L., & McVie, S. ( 2007). Youth justice? The impact of system contact on patterns of desistance from offending. European Journal of Criminology, 4 315-345. McArdle, J. J., & Epstein, D. (1987). Lat ent growth curves within developmental structural equation models. Child Development, 58, 110-113. McCollum, B. (1996). Testimony before the house s ubcommittee on early childhood, youth, and families. Washington, DC: United States Government Printing Office. McDermott, S., & Nagin, D. S. (1998). Same or different over time? In S. O. White (Ed.), Handbook of law and social science: Youth and justice New York, NY: Plenum Press. McLynn, F. (1989). Crime and punishment in eighteenth-century England New York, NY: Routledge. Meredith, W., & Tisak, J. (1990). Latent curve analysis. Psychometrika, 55 107-122. Moffitt, T. (2006). A review of research on the taxonomy of life-course persistent versus adolescence-limited antisocial behavior. In F. T. Cullen, J. P. Wright, & K. R. Blevins (Eds.), Taking stock: The status of criminology theory (pp. 277-311). New Brunswick, NJ: Transaction Publishers. Moffitt, T. E. (1993). Adolescence-limited an d life-course-persistent antisocial behavior: A developmental taxonomy. Psychological Review, 100 674-701. Moffitt, T. E., Caspi, A., Dickson, N., Silva, P., & Staton, W. (1996). Childhood-onset versus adolescent-onset antisocial conduct problems in males: Natural history from ages 3 to 18 years. Development and Psychopathology, 8 399-424. Monnkonen, E. (2001). Murder in New York City Berkeley, CA: University of California Berkeley Press. 291

PAGE 292

Morgan, S. L., & Winship, C. (2007). Causal inference: Methods and principles for social research. New York, NY: Cambridge University Press. Murray, C., & Cox, L. A. (1979). Beyond probation: Juvenile corrections and the chronic juvenile offender. Beverly Hills, CA: Sage Publications. Muthn, B. (1989). Latent variable modeling in heterogeneous populations. Psychometrika, 54 557-585. Muthn, B. (2001). Second-gener ation structural equation mo deling with a combination of categorical and continuous latent va riables: New opportunities for latent class/latent curve modeling. In A. Sayers & L. Collins (Eds.), New methods for analysis of change. Washington, DC: American Psychological Association. Muthn, L., K., & Muth n, B. O. (1998-2010). Mplus users guide (6th ed.). Los Angeles, CA: Muthn & Muthn. Nagin, D. S. (1998). Criminal deterrence research at the outset of the twenty-first century. Crime and Justice: A Review of Research, 23 1-42. Nagin, D. S. (1999). Analyzing developmental trajectories : A semi-parametric, groupbased approach. Psychological Methods, 4 139-157. Nagin, D. S. (2005). Group-based modeling of development. Cambridge, MA: Harvard University Press. Nagin, D. S., & Land, K. C. (1993). Age, criminal career s, and population heterogeneity: Specification and estimation of a nonparametric, mixed poisson model. Criminology, 31 327-362. Nagin, D. S., & Paternoste r, R. (1993). Enduring indivi dual differences and rational choice theories of crime. Law & Society Review, 27 467-496. Nagin, D. S., & Pogarsky, G. (2001). Integrating celeri ty, impulsivity, and extralegal sanction threats into a model of g eneral deterrence: Theory and evidence. Criminology, 39 865-892. Nagin, D. S., & Pogarsky, G. (2003). An experimental in vestigation of deterrence: Cheating, self-serving bias, and impulsivity. Criminology, 41 167-194. Nagin, D. S., & Tremblay ( 2005). Developmental trajectory groups: Fact or a useful statistical fiction. Criminology, 43 873-904. Nagin, D. S., Farrington, D. P., & Moffitt, T. E. (1995). Life-course trajectories of different types of offenders. Criminology, 33 111-139. 292

PAGE 293

Nagin, D. S., Pagani, L., Tremblay, R. E., & Vi taro, F. (2003). Life co urse turning points: The effect of grade retent ion on physical aggression. Development and Psychopathology, 15 343-361. Newman, G. (1978). The punishment response New York, NY: Lippincott. Neyman, J. S. (1923). Statistical pr oblems in agricultural experiments. Journal of the Royal Statistical Society, Series B-2 107-180. Packer, H. L. (1968). The limits of the criminal sanction Stanford, CA: Stanford University Press. Parker, R. N., & Smith, M. (1979). Deterrence, pover ty, and type of homicide. American Journal of Sociology, 85 614-624. Pate, A. M., & Hamilton, E. E. (1992). Formal and informal deterrents to domestic violence: The Dade County spouse assault experiment. American Sociological Review, 57 691-697. Paternoster, R. (1978). The labeling effect of police app rehension: Identit y, exclusion, and secondary deviance. Doctoral Dissertation, Flor ida State University. UMI no.: 7917065. Paternoster, R. (1987). The deterrent effect of the perceived certainty and severity of punishment: A review of the evidence and issues. Justice Quarterly, 4 173-217. Paternoster, R. (1989a). Absolute and rest rictive deterrence in a panel of youth: Explaining the onset, persistence/desis tence, and frequency of delinquent offending. Social Problems, 36 289-309. Paternoster, R. (1989b). Decisions to partici pate in and desist from four types of common delinquency: Deterrence and rational choice perspective. Law & Society Review, 23 7-40. 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. Paternoster, R., & Piquero, A. (1995). Reconc eptualizing deterrence: An empirical test of personal and vicarious experiences. Journal of Research on Crime and Delinquency 32, 251-286. Paternoster, R., Saltzman, L. E., Chiricos, T. G., & Waldo, G. P. (1982). Perceived risk and deterrence: Methodological artifact s in perceptual deterrence research. Journal of Criminal Law and Criminology, 73 1238-1258. 293

PAGE 294

Paternoster, R., Saltzman, L. E., Waldo, G. P., & Chiricos, T. G. (1983a). Perceived risk and social control: Do sanctions really deter? Law & Society Review, 17 457480. Paternoster, R., Saltzman, L. E., Waldo, G. P., & Chiricos, T. G. (1983b). Estimating perceptual stability and deterr ent effects: The role of perceived legal punishment in the inhibition of criminal involvement. Journal of Criminal Law & Criminology, 74, 270-297. Patterson, G. R. (1997). A developmental model for late-onset delinquency. In D. W. Osgood (Ed.), Motivation and delinquency: Nebra ska symposium on motivation (pp. 119-177). Lincoln, NE: University of Nebraska Press. Patterson, G. R., DeBaryshe, B. D., & Rams ey, E. (1989). A developmental perspective on antisocial behavior. American Psychologist, 4 329-335. Perkins, R. M. (1926). Absurd ities in criminal procedure. Iowa Law Review, 11 297335. Perry, S. W. (2008). Justice expenditure and employment extracts, 2006 (Bureau of Justice Statistics Report). Washington, DC: United States Government Printing Office. Pickles, A., & Rutters, M. (1991) Statistical and conceptual m odels of turning points in developmental processes. In D. Magnu sson, L. Bergman, G. Rudinger, & B. Torestad (Eds.), Problems and methods in longitu dinal research: Stability and change (pp. 133-165). New York, NY: Cambridge University Press. Piquero, A. R. (2008). Taking stock of developmental trajec tories of criminal activity over the life course. In A. M. Liberman (Ed.), The long view of crime: A synthesis of longitudinal research (pp. 23-78). New York, NY: Springer. Piquero, A. R., & Pogarsky, G. (2002). Bey ond Stafford and Warrs reconceptualization of deterrence: Personal and vicarious ex periences, impulsivity, and offending behavior. Journal of Research in Crime and Delinquency, 39 153-186. Pogarsky, G. (2002). Identifying deterrable offenders: Imp lications for research on deterrence. Justice Quarterly, 19 431-452. Pogarsky, G., & Piquero, A. R. (2003) Can punishment encourage offending? Investigating the resetting effect. Journal of Research in Crime and Delinquency, 40, 95-120. Pratt, T., Cullen, F. T., Blevins, K., Da igle, L., & Madensen, T. (2006). The empirical status of deterrence theory: A meta-analysis. In F. T. Cull en, J. P. Wright, & K. R. Blevins (Eds.), Taking stock: The status of criminological theory (pp. 367-395). New Brunswick, NJ: Transaction Publishers. 294

PAGE 295

Quinney, R. (1980). Class, state, and crime. New York, NY: Longman. Rand, M. R. (2008). Criminal victimization, 2008 (Bureau of Justic e Statistics Bulletin ). Washington, DC: United States Department of Justice, Office of Justice Programs. Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models: Applications and data analysis methods (2nd ed.). Thousand Oaks, CA: Sage. Reiss, A. J., & Roth, J. A. (1993). Understanding and prevent violence Washington, DC: National Academy Press. Robins, L. N. (1978). Stur dy childhood predictors of adult antisocial behaviour: Replications from longitudinal studies. Psychological Medicine, 8 611-622. Robins, L. N. (2005). Explai ning when arrests end for serious juvenile offenders: Comments on the Sampson and Laub study. Annals of the American Academy of Political and Social Science, 602 57-72. Rose, G., & Hamilton, R. A. (1970). Effects of a juvenile liaison scheme. British Journal of Criminology, 10, 2-20. Rosenbaum, P. R. (2002). Observational studies (2nd ed.). New York, NY: Springer. Rosenbaum, P. R., & Rubin, D. B. (1983). The central role of the propensity score in observational studies for causal effects. Biometrika, 70 41-55. Rosenbaum, P., & Rubin, D. (1984). Reduci ng bias in observational studies using subclassification on the propensity score. Journal of the American Statistical Association, 79 516. Rousseau, J. J. (1968) [1762]. The social contract. Harmondsworth, United Kingdom: Penguin Books. Rubin, D. B. (1974). Estimating causal effects of treatment in randomized and nonrandomized studies. Journal of Educational Psychology, 66 688-701. Rubin, D. B. (1978). Bayesian in ference for causal effects: The role of randomization. Annals of Statistics, 6 34-58. Rubin, D. B. (1980). Comment on randomization analysis of experimental data: The Fisher randomization test. Journal of American Statistical Association, 75 591593. 295

PAGE 296

Rubin, D. B. (1986). Which ifs have causal answers? Journal of the American Statistical Association, 81 961-962. Rubin, D. B. (1997). Estimating causal effe cts from large data sets using propensity scores. Annals of Internal Medicine, 127 757-763. Saltzman, L., Paternoster, R., Waldo, G. P., & Chiricos, T. G. (1982). Deterrent and experimental effects: The problem of causal order in perceptual deterrence research. Journal of Research in Crime and Delinquency, 19 172-189. Sampson, R. J. (2000) Whither the sociolog ical study of crime? Annual Review of Sociology, 26 711-714. Sampson, R. J., & Groves, W. B. (1989). Community structure and crime: Testing social-disorganization theory. American Journal of Sociology, 94 774-802. Sampson, R. J., & Laub, J. H. (1992). Crime and deviance in the life course. Annual Review of Sociology, 18 63-84. Sampson, R. J., & Laub, J. H. (1993). Crime in the making: Pathways and turning points through life Cambridge, MA: Harvard University Press. Sampson, R. J., & Laub, J. H. (1996). Socioe conomic achievement in the life course of disadvantaged men: Militar y service as a turning point, circa 1940-1965. American Sociological Review, 61 347-367. Sampson, R. J., & Laub, J. H. (1997). A life-course theory of cumulative disadvantage and stability in delinquency. In T. P. Thornberry (Ed.), Developmental theories of crime and delinquency (pp. 133-161). New Brunswick, NJ: Transaction Publishers. Sampson, R. J., & Laub, J. H. (2003). Life -course desisters? Trajectories of crime among delinquent boys followed to age 70. Criminology, 41 301-340. Sampson, R. J., & Laub, J. H. (2005a). A life-c ourse view of the development of crime. Annals of the American Academy fo r Political and Social Sciences, 602 12-45. Sampson, R. J., & Laub, J. H. (2005b). Seduc tions of method: Rejoinder to Nagin and Tremblays developmental trajec tory groups: fact or fiction? Criminology, 43 905-913. 296

PAGE 297

Sampson, R. J., Laub, J. H., & Eggleton, E. P. (2004). On the robustness and validity of groups: Response to Daniel Nagin. Journal of Quantitative Criminology, 20 3742. Sampson, R. J., Laub, J. H., & Weimer C. (2006). Does marriage reduce crime? A counterfactual approach to withi n-individual causal effects. Criminology, 4 465508. Scheff, T. J. (1966). Being mentally ill: A sociological theory Chicago, IL: Aldine. Schneider, A. L. (1982). The ef fects of deinstitutionalizati on of status offenders: A review essay. Criminal Justice Abstracts 182. Schrag, C. (1971). Crime and justice: American style Washington, DC: U.S. Government Printing Office. Schur, E. M. (1973). Radical nonintervention: Re thinking the delinquency problem Englewood Cliffs, NJ: Prentice Hall. Schwarz, G. (1978). Estimating the dimension of a model. Annals of Statistics, 6 461464. Sellin, J. T. (1937). Research memorandum on cr ime in the depression. New York, NY: Social Science Research Council. Shadish, W. R., Cook, T. D. & Campbell, D. T. (2002). Experimental and quasiexperimental designs for generalized causal inference. Boston, MA: Houghton Mifflin. Shannon, L. W. (1980). Juvenile delinquency and adult crime, 1948-1977 (Racine, Wisconsin): Three birth cohorts Iowa City, IA: Iowa Urban Community Research Center, University of Iowa. Shannon, L. W. (1985) A more precise evaluation of the effects of sanctions ( Final report to the National Instit ute of Justice). Iowa City, IA: Iowa Urban Community Research Center, University of Iowa. Shaw, C. R. (1930). The jack-roller: A delinquent boys own story Chicago, IL: University of Chicago Press. Shaw, C. R., & McKay, H. D. (1942). Juvenile delinquency in urban areas Chicago, IL: University of Chicago Press. Shaw, D. S., Gilliom, M., Ingolds by, E. M., & Nagin, D. S. ( 2003). Trajectories leading to school-age conduct problems. Developmental Psychology, 39 189-200. 297

PAGE 298

Shaw, D. S., Lacourse, E., & Nagin, D. S. (2 005). Developmental traj ectories of conduct problems and hyperactivity from ages 2 to 10. Journal of Child Psychology and Psychiatry, 46 931-942. Sherman, L. W. (1993). Defiance, deterrence, and irrelevance: A t heory of the criminal sanction. Journal of Research in Crime and Delinquency, 30 445-473. Sherman, L. W., & Berk, R. A. (1984a). The specific deterrent effects of arrest for domestic assault. American Sociological Review, 49, 261-272. Sherman, L. W., & Berk, R. A. (1984b). The Minneapolis Domestic Violence Experiment. (Police Foundation Reports, No. 1), Washington, DC: Police Foundation. Sherman, L. W., Schmidt, J. D., Rogan, D. P., Gartin, P. R., Cohn, E. G., Collins, D. J., & Bacich, A. R. (1991). From initial det errence to long-term escalation: Shortcustody arrest for poverty ghetto domestic violence. Criminology, 29 821-850 Sherman, L. W., Smith, D. A., Schmid t, J. D., & Rogan, D. P. (1992). Crime, punishment, and stake in conformity: Legal and informal control of domestic violence. American Sociological Review, 57, 680-690. Skardhamar, T. (2010). Distinguishing facts and artifacts in group-based modeling. Criminology, 48 295-320. Smith, D. A, & Paternoster, R. (1990) Formal processing and future delinquency: Deviance amplification as selection artifact. Law & Society Review, 24 11091132. Smith, D. A., & Gartin, P. R. (1989). Specifying specific deterrence: The influence of arrest on future criminal activity. American Sociological Review 54, 94-105. Smith, D. J. (2006). Social inclusion and early desistance from crime (Report No. 12 of the Edinburgh Study of Youth Transiti ons and Crime). Edinburgh, Scotland: Center for Law and Society, Univer sity of Edinburgh, Old College. Smith, J., & Todd, P. (2005). Does ma tching overcome LaLondes critique of nonexperimental estimators? Journal of Econometrics, 125 305-353. Smith, L. G., & Akers, R. L. (1993). A comparison of re cidivism of Floridas community control and prison: A five year survival analysis. Journal of Research in Crime and Delinquency 30, 267-292. Spohn, C., & Holleran, D. ( 2002). The effect of imprisonm ent on recidivism rates of felony offenders: A focus on drug offenders. Criminology 40, 329-357. Stafford, M. C., & Warr, M. (1993). A reconceptualization of general and specific deterrence. Journal of Research in Crime and Delinquency, 30, 123-135. 298

PAGE 299

Stafford, M. C., Gray, L. N., Menke, B. A., & Ward, D. A. (1986). Modeling the deterrent effects of punishment. Social Psychology Quarterly, 49 338-347. Sutherland, E. H. (1937). The professional thief. Chicago, IL: University of Chicago Press. Tannenbaum, F. (1922). Wall shadows: A study in American prisons New York, NY: The Knickerbocker Press. Tannenbaum, F. (1938). Crime and the Community New York, NY: Columbia University Press. Tapia, M. (in press). Gang membership and ra ce as risk factors for juvenile arrest. Journal of Research in Crime and Delinquency Published Online First 2011, doi: 10.1177/0022427810393013. Taxman, F. S., & Piquero, A. (1998). On preventing drunk driving recidivism: An examination of rehabilita tion and punishment approaches. Journal of Criminal Justice, 26 129-143. Thistlewaite, A., Wooldredge, J., & Gibbs, D. (1998). Severity of dispositions and domestic violence recidivism. Crime & Delinquency, 44, 388-398. Thoits, P. A. (1985). Self-labe ling processes in mental illness: The role of emotional deviance. The American Journal of Sociology, 91 221-249. Thomas, C. W. (1977). The effect of legal sanctions on juvenile delinquency: A comparison of the labeling and deterrence perspectives Unpublished report. Department of Sociology, Bowling Green State University. Thomas, C. W., & Williams, J. S. (1977). Actors, actions, and deterrence: A reformulation of Chamblisss typology of deterrence. In M. Riedel & P. A. Vales (Eds.), Treating the offender: Pr oblems and issues. New York, NY: Praeger. Thornberry, T. P, Krohn, M. D., Lizotte, A. J., Smith, C. A., & Tobin, K. (2003). Gang and delinquency in developmental perspective. New York, NY: Cambridge University Press. Thornberry, T. P. (1971). Punishment and crime: The effect of legal dispositions on subsequent criminal behavior. Doctoral dissertation, Un iversity of Pennsylvania. NCJ No.: 068811. Thornberry, T. P. (1987). Toward an interactional theory of delinquency. Criminology, 25, 863-891. Thornberry, T. P. (1997). Developmental theories of crime and delinquency. Advances in Criminological Theory (Vol. 7). New Brunswick, NJ: Transaction Publisher. 299

PAGE 300

Thornberry, T. P., & Krohn, M. D. (2000) The self-report method for measuring delinquency and crime. In D. Duffee, R. D. Cruthfield, S. Mastro fski, L. Mazerolle, D. McDowall., & B. Ostrom (Eds.), Innovations in measurement and analysis (pp. 33-83). Washington, DC: National Institute of Justice. Thornberry, T. P., & Krohn, M. D. (2001) The development of delinquency: An interactional perspective. In S. O. White (Ed.), Handbook of Law and Social Science: Youth and Justice (pp. 289-305). New York, NY: Plenum Publishers. Thornberry, T. P., & Krohn, M. D. (2005). Applying Interactional Theory to the explanation of continuity and change in ant isocial behavior. In D. P. Farrington (Ed.), Integrated Developmental and Life -Course Theories of Offending (pp. 183209). New Brunswick, NJ: Transaction Publishers. Thorsell, B. A., & Klemke, L. W. (1972). The labeling pr ocess: Reinforcement and deterrent. Law & Society Review, 6, 393-403. Tittle, C. R., & Rowe, A. R. (1974). Certainty of arrest and crime rate s: A further test of the deterrence hypothesis. Social Forces, 52 455-462. Tonry, M. (2006). Purposes and functions of sentencing. Crime & Justice, 34 1-53. Tracy, P. E., & Kempf-Leonard, K. (1996). Continuity and discontinuity in criminal careers. New York, NY: Plenum. Tremblay, R. E., Nagin, D. S., Sguin, J. R. Zoccolillo, M., Zelazo P. D., Boivin, M., Prusse, D., & Japel, C. (2004). Physi cal aggression during early childhood: Trajectories and predictors. Pediatrics, 114 e43-e50. Turner, R. H. (1972). Deviance avow al as neutralization of commitment. Social Problems, 19 308-321. Uggen, C., & Staff, J. (2001). Work as a turning point for criminal offenders. Corrections Management Quarterly, 5, 1-16. Uggen, C. (2000). Work as a turning point in the life course of criminals: A duration model of age, employment and recidivism. American Sociological Review, 65 529-546. Ventura, L. A., & Davis, G. (2005). Dome stic violence: Court case conviction and recidivism. Violence Against Women, 11 255-77. Verhulst, F. C., Koot, H. M. Berden, G. F. M. G. (1990) Four-year follow-up of an epidemiological sample. Journal of the American Academy of Child and Adolescent Psychiatry, 29 440-428. 300

PAGE 301

Ward, J. T., Stogner, J., Gibson, C. L., & Akers, R. L. (2010). A new trick for an old dog: Applying developmental trajectories to inform drug use progression. Journal of Drug Issues, 40, 755-782. Wellford, C. (1975). Label ing theory and criminology: An assessment. Social Problems, 22, 332-345. Wellford, C. F., & Triplett, R. A. (1993). The future of labeling theory: Foundations and promises. In F. Adler & W. S. Laufer (Eds.), New directions for criminological theory: Advances in criminology theory (pp. 1-22). Edison, NJ: Transaction. West, D. J., & Farrington, D. P. (1 973). Who becomes de linquent? London, United Kingdom: Heinemann. West, D. J., & Farrington, D. P. (1977). The delinquent way of life. London, United Kingdom: Heinemann. White, J. L., Moffitt, T. E., Earls, F., Robins, L., & Silva, P. A. (1990). How early can we tell? Predictors of childhood conduct disorder and adolescent delinquency. Criminology, 28 507-535. Wikstrom, P. H. (2005). The social orig ins of pathways in crime: Towards a developmental ecological action theory of crime involvement and its changes. In D. P. Farrington (Ed.), Integrated developmental and life-course theories of offending, Advances in Criminology Theory (pp. 211-245). New Brunswick, NJ: Transaction Publishers. Willett, J. B., & Sayer, A. G. (1994). Us ing covariance structure analysis to detect correlates and predictors of change. Psychological Bulletin, 116 363-381. Williams, K. R., & Hawkins, R. (1989). The meaning of arrest for wife assault. Criminology, 27 163-182. Wolfgang, M. E., Figlio, R. M., & Sellin, T. (1972). Delinquency in a birth cohort Chicago, IL: University of Chicago Press. Wright, J. P., & Cullen, F. T. (2004). Employment, peers and life-course transitions. Justice Quarterly, 21 183-205. Yu, J., & Liska, A. E. (1993). The certainty of punishment: A refere nce group effect and its functional form. Criminology, 31 447-464. Zimring, F. E. (1998). American youth violence. New York, NY: Oxford University Press. Zimring, F. E., & Hawk ins, G. J. (1973). Deterrence. Chicago, IL: University of Chicago Press. 301

PAGE 302

302 Zimring, F. E., Hawkins, G., & Kamin, S. (2001). Punishment and democracy: Three strikes and youre out in California. New York, NY: Oxford University Press.

PAGE 303

BIOGRAPHICAL SKETCH Jeffrey T. Ward is a Maryland native w ho earned his Bachelor of Science (2005) degree from Rutgers, The State University of Ne w Jersey, where he graduated summa cum laude and double majored in physics and psychology. He earned his Master of Arts (2007) and Doctor of Philosophy (2011) degrees from the University of Florida, where he majored in criminology, law and so ciety. He was also awarded a Social Science Methodology Graduate Cert ificate (2011) from the Univ ersity of Florida. His substantive areas of research include: sanction effects; violence and gangs; developmental and life-course criminology; and space, place, and crime. In addition to these substantive research areas, he is also highly interested in quantitative methodology and psychometrics. He is a re cipient of the prestigious Harry Frank Guggenheim Dissertation Fellowship and his work has been published in some of the leading journals in the field of criminol ogy. He begins his academic career as an assistant professor in the Department of Crimi nal Justice at the University of Texas at San Antonio. 303