JUVENILE RISK ASSESSMENT: AN EXAMINATION OF GENDER
By
Valerie R. Anderson

A THESIS
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
MASTER OF ARTS
Psychology
2012

ABSTRACT
JUVENILE RISK ASSESSMENT: AN EXAMINATION OF GENDER
By
Valerie R. Anderson
Female offenders comprise the fastest growing proportion of juvenile court caseloads.
However, most risk assessments are developed for boys, empirically validated on samples of
boys, and subsequently applied to girls with little regard to the appropriateness of the instrument
for young women. The current study examined the differential performance of the Youth Level
of Service/Case Management Inventory (YLS/CMI) by gender through assessing the moderating
effect of gender on the risk-recidivism relationship, the criminogenic predictive validity of the
instrument, and gendered patterns of risk among juvenile probationers (n =1,100) from a midsize
Midwestern community. Findings revealed slight gender differences in both the predictive
validity of the YLS/CMI and the moderating effect of gender on the risk-recidivism relationship.
Unique patterns of risk and need also emerged when samples of boys and girls were analyzed
separately. Directions for future research as well as gender-responsive programming and policy
implications are discussed.

TABLE OF CONTENTS

LIST OF TABLES .........................................................................................................................v
LIST OF FIGURES ..................................................................................................................... vi
INTRODUCTION .........................................................................................................................1
Female Juvenile Delinquency: Definition, Scope, and Impact ............................................4
Historical Perspectives and Theories of Gender Differences in Delinquency .....................7
Girls’ Risk Factors Across Levels of Analysis ..................................................................10
Utility of Risk Assessment Instruments.............................................................................16
Gender and Risk Assessment .............................................................................................19
Female Juvenile Risk Assessment .....................................................................................23
Youth Level of Service/Case Management Inventory (YLS/CMI) and Gender ...............25
Juvenile Delinquency Taxonomies ....................................................................................27
Current Study .....................................................................................................................30
Research Questions ............................................................................................................32
METHODS ...................................................................................................................................33
Sample................................................................................................................................33
Training and Procedures ....................................................................................................34
Measures: Independent Variables ......................................................................................34
Measures: Dependent Variable ..........................................................................................36
RESULTS .....................................................................................................................................38
Research Question 1 ..........................................................................................................38
Research Question 2 ..........................................................................................................40
Research Question 3 ..........................................................................................................41
Research Question 4 ..........................................................................................................43
Research Question 5 ..........................................................................................................44
Research Question 6 ..........................................................................................................45
Research Question 7 ..........................................................................................................47
DISCUSSION ...............................................................................................................................55
Differential Predictive Validity of the YLS/CMI by Gender ............................................55
Gendered Risk Profiles and Cluster Types ........................................................................60
Study Limitations ...............................................................................................................63
Policy Implications and Directions for Future Research ...................................................64
APPENDICES ..............................................................................................................................69
Appendix A: Youth Level of Service/Case Management Inventory (YLS/CMI) Items ...70
Appendix B: YLS Pattern of Need Descriptive Guide and Service Manual .....................72

iii

REFERENCES.............................................................................................................................79

iv

LIST OF TABLES
Table 1.1. Sample Demographics ..................................................................................................33
Table 1.2. Descriptive Statistics for Variables in Analysis ...........................................................35
Table 1.3. Subscale Correlation Matrix .........................................................................................36
Table 1.4. Two-Year Recidivism Rates by Gender and Risk Level .............................................37
Table 2.1. Moderated Logistic Regression for Predicting Recidivism with Total Score ..............39
Table 2.2. Moderated Logistic Regression for Predicting Recidivism by Risk Level ..................39
Table 2.3. Moderated Logistic Regression for Predicting Recidivism by Subscale ......................40
Table 3.1. Predictive Validity of the YLS/CMI: Overall Sample..................................................42
Table 3.2. Predictive Validity of the YLS/CMI: Boy ....................................................................42
Table 3.3. Predictive Validity of the YLS/CMI: Girls...................................................................43
Table 3.4. Gender Differences in the Predictive Validity of the YLS/CMI ..................................44
Table 4.1. Gender Differences in Delinquency Cluster Types ......................................................46
Table 4.2. Corrected Item-Total Correlations and Reliability Scores ...........................................47
Table 4.3. Girls’ Patterns of Risk and Recidivism Rates ...............................................................49
Table 4.4. Boys’ Patterns of Risk and Recidivism Rate ................................................................52

v

LIST OF FIGURES
Figure A.1. Radar Graph of Subscale Peaks in Girls’ Clusters .....................................................50
Figure A.2. Radar Graph of Subscale Peaks in Boys’ Clusters .....................................................54

vi

JUVENILE RISK ASSESSMENT: AN EXAMINATION OF GENDER
INTRODUCTION
Juvenile delinquency is a serious social problem and can act as a pathway to adult
offending (Zahn, Hawkins, Chiancone, & Whitworth, 2008). In the United Sates, girls are the
fastest growing subpopulation of offenders in the juvenile justice system (Chesney-Lind &
Shelden, 2004). By 2008, girls accounted for nearly 30 percent of all delinquency cases. Between
1985 and 2008, the female delinquency caseload grew three percent per year on average and the
male delinquency caseload grew one percent per year on average (Puzzanchera, Adams, &
Sickmund, 2011). During this period of time the female caseload increased 101 percent versus a
54 percent increase in growth among the male caseload; as well, females outpaced males in
average annual growth for all offense categories (Puzzanchera, Adams, & Sickmund, 2011;
Synder & Sickmund, 2006). Given both this increase in girls’ delinquency and the fact that little
research has focused specifically on female juvenile delinquency (Zahn et al., 2008), a better
understanding of girls’ offending has become increasingly of interest to researchers and
practitioners. A large amount of research has identified risk factors related to male delinquency,
but whether the same factors are related to girls’ offending serves as a debate in the literature
(Zahn et al., 2008). Researchers calling for a more nuanced understanding of female juvenile
delinquency question whether there are unique, gender-specific risk factors that contributes to
girls’ system involvement (Bloom, Owen, Covington, 2003; Chesney-Lind, 1997).
One way in which the juvenile justice system responds to delinquency is through the use
of standardized risk assessments. When a youth comes into contact with the system, courts
administer an assessment to identify the level of risk that the youth poses and areas of need for
intervention. Risk assessments may be used for decisions related to processing (e.g., detention)

1

as well as the individualized treatment needs. However, most assessment and correctional
classification systems were developed for boys, empirically validated on samples of all or
majority boys, and subsequently applied to girls with little regard to the appropriateness or
validity of the instrument for this population (Bloom, Owen, Covington, 2003; Chesney-Lind,
1997; Salisbury, Van Voorhis, Spiropoulos, 2009). In the literature on gender-responsive
assessment, this has largely been attributed to the fact that the current generation of assessments
was created from samples of male offenders (Blanchette & Brown, 2006; Reisig, Holtfreter, &
Morash, 2006). In turn, these risk measures and classification systems have been critiqued for an
omission of gender-relevant factors such as mental health issues and history of abuse. Thus,
researchers and feminist scholars have called to question the use of these instruments with girls;
the differential predictive validity based on gender, the appropriateness of the methodology, and
the theories that underpin this assessment method have all been questioned.
In addition to predicting criminogenic risk and need, researchers have begun to use these
assessments to identify patterns of risk and need by creating taxonomies of offenders. These
taxonomies have been based on risk factors associated with delinquency, also referred to as
subgroups or clusters in the literature (Mulder, Brand, Bullens, & van Marle, 2010; Onifade et
al., 2008; Simourd, Hoge, Andrews, & Leschied, 1994). Researchers have suggested that by
understanding the types of clusters youth belong to, practitioners can improve their
understanding of offender etiologies and design appropriate interventions to address theses
unique patterns of risk and need (Onifade et al., 2008). However, the literature in this area has
also focused solely on male offenders or combined samples of youth mostly comprised of boys.
This limited focus calls to question whether there exist specific cluster types for girls’ patterns of
criminogenic risk and need.

2

The study reported here will add to our understanding of gender and risk assessment
through the use of the Youth Level of Service/Case Management Inventory (YLS/CMI), a
widely-used juvenile risk assessment instrument (Hoge, Andrews, & Leschied, 2002). Gender
differences were tested using the YLS/CMI to predict two-year recidivism in a representative
sample of juvenile probationers. The gender-based performance of the YLS/CMI was also
examined in two ways. First, whether or not gender moderates the risk-recidivism relationship
across overall YLS/CMI score, risk levels, and its subscales is examined. Second, the differential
predictive validity of the YLS/CMI based on gender is assessed. In addition to the analysis of the
psychometric properties of the YLS/CMI, a second set of research questions addresses gendered
patterns and differences in a cluster solution developed to understand patterns of juvenile risk
and need (Onifade et al., 2008). Finally, this study cluster analyzed the sample of juvenile
probationers separately by gender to detect if unique patterns of risk and need exist for girls.
The current study adds to the existing literature, albeit conflicting and limited, on juvenile
delinquency, gender, and risk assessment. This study informs theory around gender and crime
recurrence, juvenile justice practice and policy, as well as the current research methodologies
utilized in risk assessment. The broader gender and risk assessment discussion is also informed.
Juvenile delinquency is a serious social problem that has well-documented adverse
effects at both individual and societal levels (Cottle, Lee, Heilbrun, 2001). When aiming to
understand female juvenile delinquency as a population, one must identify the demographics,
patterns of behavior, areas of need and risk, contextual factors related to offending, as well as the
social and juvenile justice systems response to girls’ crime. The first section of the literature
review will provide an overview of female juvenile delinquency (including the scope of female
delinquency, key definitions, and the impact of delinquency), theories that explain gender

3

differences in delinquency, and girls’ risk factors from a multiple levels of analysis perspective.
The second section will cover the utility of risk assessment instruments for youth in the juvenile
justice system, an overview of the gendered nature of risk assessments with both women
offenders and girls, and a review of what is currently known about gender-based performance of
the YLS/CMI. The final section will provide a review of juvenile offending taxonomies and their
application to risk assessment.
Female Juvenile Delinquency: Definition, Scope, and Impact
The Office of Juvenile Justice and Delinquency Prevention (OJJDP) reported that in 2008
juvenile courts across the U.S. handled 1.6 million delinquency cases (Puzzanchera, Adams, &
Sickmund, 2011). Over the course of the past two decades, female juvenile delinquency has
increased compared to male juvenile delinquency and by 2008 girls comprised nearly 30 percent
of juvenile justice involved youth (Puzzanchera, Adams, & Sickmund, 2011). The overall total
number of juvenile arrests in the United States dropped 29 percent for males, however arrests for
females only dropped 14 percent; this pattern is consistent across all crimes, however was most
significant for violent crimes (Brumbaugh, Hardison, & Winterfield, 2010). These arrest trends
indicate that patterns of offending for girls are increasing more or decreasing less than boys
(Zahn et al., 2010). Whereas the overall rate of juvenile arrests decreased, these rates decreased
more for boys than for girls; however, simple assault arrests increased 19 percent for girls and
decreased 4 percent for boys (Snyder, 2008). Although girls account for nearly one third of
juvenile arrests, there is debate as to whether this is a reflection of an increase in deviant
behavior among females or a reflection of a change in the systems response to female offenders
(MacDonald & Chesney-Lind, 2001; Zahn et al., 2010). For example, higher levels of policing of
young women may contribute to the higher rates of girls involved in the juvenile justice system,

4

rather than an actual increase in their deviant behavior. Further, given that the rate of arresting
young men has decreased, this may “free up” enforcement resources to focus on young women.
It should also be noted that the prevalence of juvenile delinquency might shift depending on the
local ecology of the community (e.g., urban versus suburban). As well, disparities in local
policies and differential cultural practices or social responses to deviant behavior among youth
may affect the prevalence of juvenile delinquency within any given community.
Juvenile delinquency is most frequently defined as the involvement of a person younger
than the age of 18 violating the criminal law through behavior such as violent crime, burglary,
property crime, and status offenses (Hoge et al., 1996; Jung & Rawana, 1999). Additionally,
status offenses are youth-specific behaviors such as truancy, running away, or possessing of
alcohol (Zahn et al., 2008). These behaviors are only defined as crimes because of the offender’s
status as a minor. However, it should be noted that definitions of juvenile delinquency are not
consistent across studies and researchers nor are they identical across states or nations. For
example, some researchers have defined juvenile delinquency outside of the legal definition by
studying youth involved in gangs or drug abuse (assessed by self-report measures), and then
using these at-risk behaviors as proxies for juvenile delinquency. Whereas other studies produce
information about juvenile offending from official court records without consulting with or
receiving information from the actual juvenile offenders. As well, other investigations have
included young adults (up to the age of 21) in their studies (Cottle, Lee, Heilburn, 2001). These
discrepancies in defining juvenile delinquency have implications for how data can be interpreted
and the policies and interventions that can be derived from the research. Thus, properly defining
as well as identifying the causes, correlates, and pathways of female offending has been a

5

longstanding quest for juvenile justice researchers, practitioners, and policymakers across
disciplines.
The impact of juvenile delinquency includes both individual-level effects (e.g.,
psychological impact) as well as societal effects (e.g., economic impact). In addition to the
individual and social impact of juvenile delinquency, there is a tremendous economic impact.
States spent nearly $6 billion each year on youth in residential facilities (Justice Policy Institute,
2009; Sickmund, Sladky, & Kang, 2008). The American Correctional Association (2008)
estimated that it costs states approximately $240 per day, roughly $88,000 annually, for each
youth in a detention facility. Incarcerating youth who may not need high levels of supervision is
a costly alternative to other community-based interventions that may be more effective
(American Correctional Association, 2008). In particular, resources allocated to detention
facilities may be better utilized in educational or community services. Cohen (1998) estimated
the external costs of a juvenile offender to be $1.3 million to $1.5 million; higher-risk offenders
were estimated to impose costs to society as high as $36 million. Cohen’s cost-benefit analysis
suggests that providing more appropriate interventions for high-risk youth can save 1.7 to 2.3
million dollars per youth. In addition to the economic impact of incarcerating youth, other
iatrogenic effects of juvenile incarceration include increasing recidivism rates among detained
youth (Benda & Tollet, 1999; Snyder, 2008), incarceration may reinforce delinquent behavior
(Dishion, McCord, & Poulin, 1999; Holman & Ziedenberg , 2006; Snyder & Sickmund, 2006), it
does not meet the mental health needs of youth (Burrell & Bussiere, 2005; Grisso, 2008), and
detention negatively impacts juveniles education and occupational attainment (Freeman, 1991;
Harlow, 2003).
The failure to provide youth with appropriate interventions can increase offense rates and

6

lead to long-term needs of community-level programming. Unfortunately, research on and
evaluation of gender-specific programming for youth in the juvenile justice system is sparse and
methodologically weak. The American Bar Association and National Bar Association (2001)
emphasized the lack of prevention, diversion, and alternative programming for girls. A
systematic review of gender-responsive programs for juvenile justice involved youth indicated
that, to date, few programs have been empirically validated as promising prevention techniques
or interventions for girls (Zahn et al., 2010).
Historical Perspectives and Theories of Gender Differences in Delinquency
As stated above, most theories of crime were developed for male offenders. Therefore,
scholars have questioned whether general theories of crime can adequately explain patterns of
female offending. For example, Funk (1999) identified that both males and females experience
strain, have delinquent associations, and weak social bonds. However, while girls and boys may
experience the same risk factors, they may have differential exposure (e.g., frequency or
severity) to these risk factors. While a history of victimization is considered a risk factor, gender
may moderate this relationship – producing a stronger risk for girls. Chesney-Lind (1997) argued
that abuse in the home especially contributed to female delinquency. As well, behavior labeled
delinquent, such as running away from home and school truancy, often stems from abusive
(physical, sexual, emotional) experiences that young women face in their homes and schools
(Chesney-Lind, 1997).
Other theorists contend that social bonds are more important for females and those
stronger predictors of offending. This position is also embedded in a feminist perspective, where
gender inequality exists in the socialization process as well as expectations in gender roles, and
gender differences rest in girls’ greater need for relationships, and interpersonal connections

7

(Mazerolle, 1998). Moreover, girls’ risk for delinquency increases when others in their
relationships engage in delinquent activity. Funk (1999) has suggested that gender differences lie
in the structural forces that may impact the types of risk factors researchers identify as critical to
female and male offending as well as the systems response to those risk factors.
A feminist perspective of crime argues that the differences between girls’ and boys’
experiences must be understood with patriarchy as the central focus when examining
delinquency trajectories (Holsinger, 2000). This perspective advocates that structural oppression,
traditional gender roles, and abuse from males contribute to women’s and girls’ offending.
Therefore, gender differences may be located in developmental processes, the deviant behavior
itself, as well as the formal and informal responses to delinquency (Belknap & Holsinger, 2006).
In addition, the pathways perspective emphasizes childhood trauma (e.g., abuses, neglect) and
girls’ responses to trauma (e.g., running away) as central to understanding the causes of female
offending (Belknap & Holsinger, 2006). In particular, Belknap and Holsinger identified that
these abuses start earlier and last longer for girls than boys. Daly (1992) identified that women’s
pathways into crime included specific subgroups of offenders (e.g., battered women,
economically-motivated women, drug-connected women). Within this framework, research
efforts on women offenders should be characterized by recognizing women’s social location, life
experiences, and the broader context of their offending (Reisig, Holtfreter, & Morash, 2006).
From a historical perspective, the juvenile justice system has acted in ways to control girls’
sexuality, has been more concerned with girls’ virginity, has more frequently institutionalized
girls for immoral behavior (e.g., premarital sex, reproduction) and has worked to control girls’
sexuality through status offenses such as incorrigibility, running away, and truancy (Morash,
2006; Pasko, 2010). The pathways perspective examines systemic connections among the

8

marginalization, victimization, and girls’ experiences within the justice system by situating girls’
and women’s crimes in the context of their environment as well as their relationships (Pasko,
2008).
In addition, intersectionality theory explains and examines the matrix of intersecting
hierarchies related to race, class, and gender (Collins, 1990, 1998). It can examine structural
intersectionalities by additionally focusing on legal, economic, and social barriers (Erez et al.,
2009). Other scholars have examined the impact of intersections of race, class, and gender in
shaping women and girls’ life experiences and their social connection to violence (Simpson &
Elias, 1995). In particular, feminist criminologists have used Collins’ concept of intersectionality
to understand how race, class, and gender shape girls’ and women’s exposure to different life
experiences, which in turn influences their relationship with crime and delinquency (Jones, 2008;
Potter, 2006).
Additional research examining intersectionality and the scope of female offending
highlight changes in patterns of arrest over time (Stevens, Morash, & Chesney-Lind, 2011) as
well as examining the gender convergence hypothesis by including race as well (Goodkind et al.,
2009). Stevens, Morash, & Chesney-Lind (2011) highlighted shifts in policies (e.g., zerotolerance) and practices, not necessarily changes in girls’ behavior, that fuel increasing violent
arrests among girls. Stevens et al. (2011), described how these arrest trends were
disproportionately impacting African-American girls and that the system is “moving girls deeper
into the justice system [which] can have clear and racialized negative consequences” (p. 740).
Additionally, Goodkind et al. (2009) found diminutive support for the gender convergence
hypothesis given that data over time do not show increases in girls’ drug use or violent behavior.
Goodkind and colleagues reported that the greatest gender divergence occurs in measures of

9

violence (e.g., reporting engaging in fighting; reporting injuring someone) and rates of substance
abuse, both of which are greater for boys with slight differences by race/ethnicity. For example,
African American youth have the greatest gender convergence in measures of violence; however,
they have the greatest divergence in the substance abuse measures (Goodkind et al., 2009).
In sum, both historical and contemporary feminist perspectives and theories on crime
suggest that women and girls have been underrepresented in criminological theory and research.
This underrepresentation stems from the broader, structural inequality, discrimination, and
disadvantage that women and girls face. Moreover, gender differences exist in both the
seriousness and extent of criminal behavior as well as differential exposure to problems such as
abuse, emotional distress, and poverty (Blanchette & Brown, 2006; Hannah-Moffat, 2009;
Holtfreter & Cupp, 2007).
Girls’ Risk Factors Across Levels of Analysis
At the individual level, there is documentation that biological, psychopathological,
familial, peer, and school relationships all impact the risk-levels of girls. Biologically, early
puberty increases girls’ risk for juvenile delinquency, which is explained in the research by a gap
between biological and emotional maturation and can be moderated by living in a disadvantaged
neighborhood or dysfunctional family (Moffitt, 1993; Obeidallah et al., 2004). Sexual and
physical abuse as well as neglect are risk factors for both genders, however, girls experience
more victimization overall (e.g., sexual assault and harassment), which may also have a
moderating effect of delinquency (Acoca, 1998). Psychologically, depression and anxiety
disorders have been associated with juvenile delinquency and girls have tended to receive these
diagnoses more frequently than boys (Teplin et al., 2002). In the family context, research has
shown multiple risk and protective factors including level of parental supervision (e.g., higher

10

levels of supervision correlate negatively with delinquency), the family’s history of criminal
justice system involvement, overall instability in family structure and functioning, as well as
neglect and maltreatment (Moffitt et al., 2001; Gaarder & Belknap, 2002). As well, female
delinquents are more likely than boys to have criminally involved parents, parents with substance
abuse problems, fragmented families, and to have suffered from more physical and sexual abuse
by a family member or in the home (Acoca, 1998; Pasko 2008). In turn, the experiences of these
abuses also fuel the mental health needs of young women. For example, the experience of
physical and/or sexual abuse increases the likelihood of substance abuse, depression and anxiety
disorders, self-harm, eating disorders, and suicide attempts (Bailey and McCloskey, 2005;
Chesney-Lind and Sheldon, 2004; Pasko, 2008; Ullman, 2004).
Finally, other proximal factors such as peer relationships and involvement in school are
also documented to impact delinquency risk level (see Morash, 1986 and Haynie, 2001). Girl
offenders have higher rates of truancy, lower school attachment, higher rates of suspension than
non-delinquent girls. Further, many schools are ill-equipped to handle and work with girls
properly through their multitude of needs (Gaarder and Belknap, 2002; Somers and Gizzi, 2001).
Neighborhood-level factors that also serve as risk factors for girls’ offending include living in
disadvantaged neighborhoods, experiencing economic hardships, inadequate supervision, high
rates of gang involvement, attending under-resourced schools (Acoca, 1998; Gaarder and
Belknap, 2002; Kellam et al., 1998; Kroneman et al., 2004).
Wong et al. (2010) provided an overview of studies on adolescent females in Europe that
examined risk factors for offending as compared to adolescent males. Wong and colleagues
separated their findings by individual, family, school, and peer factors that influenced
delinquency. While males and females shared many common risk factors, it was noted that girls

11

were more impacted by negative life events, abuse by parents, and internalizing disorders. In
turn, boys were more impacted by birth complications, low self-control, and school factors
(Wong et al., 2010). Additional reviews and meta-analyses of individual-level risk factors for
girls’ offending have been conducted in the United States. The most commonly examined factors
across these studies include physical and sexual abuse, mental health needs, substance abuse,
criminality within the family, and neighborhood disadvantage (Hoyt and Scherer, 1998; Hubbard
and Pratt, 2002; Simourd and Andrews, 1994). Hubbard and Pratt’s (2002) meta-analysis of the
predictors of female offending revealed that the main predictors of male delinquency, specified
by past studies, such as history of system involvement, delinquent peer associations, antisocial
attitudes and personality were also strong predictors of female delinquency with mean effect
sizes ranging from .18 to .53 with prior history and delinquent peers as the strongest variables.
As well, school relationships and physical and/or sexual abuse, both of which have been
excluded from many previous studies on adolescent offending, have strong effect sizes and need
to be considered in future research and policy (Hubbard and Pratt, 2002).
Consistent with a levels of analysis perspective on juvenile delinquency, the context of
gender also shifts based on the level at which we approach the analysis. Gender at the societal,
macro-level impacts socioeconomic status and opportunities related to work, education, social
expectations, and access to resources. These gender archetypes have implications for the
interrelationship between how the construction of gender at multiple levels impacts juvenile
justice levels of analysis. In many respects, these various levels of gender fit together to form
mutually reinforcing perspectives on delinquent behavior such that gender ideologies and norms
are pervasive at a structural level, within a local community, in families and peer groups, as well
as at the individual level.

12

In a study examining adult female recidivism, poverty was the most significant predictor
of future offenses, whereas obtaining public assistance was the strongest predictor of criminal
desistence (Reisig, Holtfreter, & Morash, 2002). This raises important questions about how
influential receiving public assistance (e.g., welfare, food stamps) can be from a public policy
perspective. Another example is the removal of Supplemental Security Income (SSI) for
individuals with addictions to alcohol or drugs. The effect of this change appears to be an
increase in drug and property offenses after the loss of benefits (Swartz, Martinovich, &
Goldstein, 2003). These types of policies may have implications for juvenile justice involved
youth as a proximal risk factor at the family or community level. For example, a female-headed
household—an indicator of concentrated poverty—is a proximal risk factor for juvenile
delinquency. Given this knowledge, the feminization of poverty will inadvertently affect the
outcomes of youth in those households. This has been documented in the neighborhood effects
literature that supports the connection between the concentration of poverty (among women and
youth) and criminal behavior (Morash, 2006). This has further implications for constraints posed
at the neighborhood-level. Women have less legitimate opportunities to make money resulting in
increased strain on educational attainment and overall public health. To date, little research has
identified higher-order contextual influences of juvenile delinquency (Onifade et al., 2011) and
even less regarding how gender is implicated in these relationships.
Despite the increases in arrest rates of girls, much of the literature suggests that girls’
behavior has not changed in recent years, thus other factors are likely contributing to these
increases (Zahn et al., 2008). Certain policies may also differentially impact girls’ involvement
with the juvenile justice system. Two key policies that have been indicated to disproportionately
impact girls’ arrests include the zero-tolerance policies in schools and the “mandatory arrest”

13

laws in domestic violence cases (Feld, 2009; Zahn et al., 2008). For example, schools’ zero
tolerance policy in regards to youth violence may impact the amount of police referrals girls
receive in school. This policy is particularly important when assessing the context in which
females utilize violence—girls tend to commit more violent offenses with/against intimates (e.g.,
family members, close peers, romantic partners).
A girl may also be arrested for simple assault in the case of domestic dispute between the
girl and a family member. This increase in the arrests of girls appears to be a residual effect of
pro-arrest or mandatory arrest policies set in place for domestic violence incidents (Zahn et al.,
2008). Strom et al. (2010) noted that several previous studies found these policies extend beyond
intimate partners to other domestic relationships such as family members (parents, siblings,
children). In particular, juveniles were more likely to be arrested than adults when fighting with
parents or other family members, especially in mutually combative situations (Buzawa &
Hotaling, 2006). Strom et al. (2010) used data from the National Incident Based Reporting
System (NIBRS) to analyze the influence of these dual-arrest policies on the likelihood of arrest
in family disputes by gender. Strom and colleagues found both male and female juveniles were
significantly more likely to be arrested in a state with a pro-arrest policy (versus discretionary
arrest policies) and these arrest policies had a greater influence of girls’ arrests. In terms of zero
tolerance policies in the school context, the American Psychological Association (APA) Zero
Tolerance Task Force (2008) reviewed a 20-year history of implementation and found that zero
tolerance policies conflict with the knowledge related to adolescent development (e.g.,
exacerbate normative adolescent challenges) and may negatively impact the relationship of
education with juvenile justice system. For example, the reliance on more severe consequences
to student misbehavior has resulted in an increase in utilization of the juvenile justice system for

14

infractions not previously considered threatening or dangerous and were once handled within the
school system (APA Zero Tolerance Task Force, 2008).
Other institutionalized factors that researchers identified as contributing to a differential
systems response to girls’ offending include the relabeling of status offenses, net-widening
practices, as well as court processing and sentencing decisions (Javdani, Sadeh, & Verona, 2011;
Onifade et al., 2009). Such policy changes and differential systems response to girls’ offending
unwittingly play a role in the increasing numbers of female system involvement (see also
Chesney-Lind & Okamoto, 2001, MacDonald & Chesney-Lind, 2001, Steffensmeier et al.,
2005).
In sum, a large body of research documents girls’ risk factors for offending across levels
of analysis. However, the majority of this research has focused on individual-level factors such
as biological, psychological, and environmental proxies of risk. Additional research has indicated
the importance of assessing ecological neighborhood-level factors (Onifade et al, 2011) as well
as how the justice systems response may differentially impact girls’ involvement (Javdani,
Sadeh, & Verona, 2011). In particular, the relabeling of status offenses and the unintended
consequences of dual-arrest policies have been identified as disproportionately affecting girls’
contact with the system. Given the variety of risk factors associated with offending, the next
section will focus on the development of criminogenic risk assessment tools used to predict the
likelihood of juveniles reoffending.
Utility of Risk Assessment Instruments
Given these perspectives on offending and the context of risk factors associated with
female juvenile delinquency, an understanding of the history and utility of risk assessment
instruments with delinquent youth is needed. Many courts respond to youth by administering risk

15

assessments at each point in time they come into contact with the juvenile justice system. There
is evidence that as few as six to eight percent of repeat juvenile offenders commit the majority of
offenses in a community during any given year (Schumacher & Kurz-Gwen, 2000; Snyder &
Sickmund, 2006). Juveniles at highest risk to offend are those who offended in the past (Cottle,
Lee, Heilbrun, 2001). These repeat offenders typically share a common set of risk factors
(Moffitt, 1993). The main advantages of using risk assessment instruments include the effective
allocation of resources and provide juvenile justice practitioners information about the needs of
youth on their caseloads (Funk, 1999). Thus, properly defining and examining risk level of youth
is paramount.
Risk assessment instruments include factors that are associated with delinquency such as
offense history and substance abuse (Cottle, Lee, & Heilbrun, 2001). These measures provide
court personnel with empirically-based tools to aid in decision making regarding placement
decisions and treatment planning by using data to predict the likelihood of future crime (Funk,
1999). Three generations of risk assessment measures have been identified (Bonta, 1996). First
generation measures consisted of professional opinions, which were not strong predictors of
recidivism. Second generation measures represented numerical sums of static risk factors such as
criminal history and current offense (Bonta, 1996). The latest generation of assessment tools
includes both static and dynamic risk factors and serves the dual function of assessing both
criminogenic risks as well as needs that can serve as points of intervention (Schwalbe, 2008).
Static risk factors, variables do not change naturally or cannot be changed through
intervention, that have been identified as the most predictive of recidivism include demographic
characteristics such as male gender and lower socioeconomic status, early age of offending
onset, a greater quantity of arrests, more serious offenses, and longer incarcerations (Cottle, Lee,

16

& Heilbrun, 2001; Schwalbe, 2006). Dynamic risk factors include variables that are malleable to
intervention. Dynamic risk factors identified as significant predictors of recidivism include
family instability, association with delinquent peers, substance abuse, and inadequate use of
leisure time (Cottle, Lee, & Heilbrun, 2001). Both types of risk factors contribute to the RiskNeeds-Responsivity (RNR) model, which was develop from personality and cognitive social
learning theories of behavior (Andrews et al., 2011). There is a substantial amount of evidence in
support of the RNR model for crime prevention (Andrews & Bonta, 2010; Lowenkamp, Latessa,
& Smith, 2006).
Researchers have recently pushed for a better understanding of risk assessment measures
to predict recidivism among juveniles in order to uncover areas of need for court-involved youth
(Onifade et al., 2008). Juvenile justice practitioners use standardized risk assessment instruments
to identify areas of risk (e.g., recidivism) and/or need (e.g., mental health). These are critical
tools used by the court in determining treatment decisions and programming for youth
(Brumbraugh et al., 2010). A more nuanced understanding of risk levels and crime patterns is
important for predicting which types of youth reoffend. Practitioners may utilize findings from
the assessments to uncover specific areas of need and risk that may act as contributing factors to
future delinquent behavior (Hoge & Andrews, 1996). The traditional approach in juvenile justice
practice is for practitioners to simply use their clinical, professional judgment to classify youth
and assess areas of need most amendable to intervention (Onifade et al., 2009). Thus, researchers
have advocated for risk assessment protocol and actuarial prediction methods over the use of
potentially biased informal assessment (Bonta, 1996; Grove & Meehl, 1996). In terms of policy,
the risk principle tends to drive correctional budgets and programming, therefore utilizing factors

17

related to recidivism are given priority (Cullen, Fisher, & Applegate, 2000; Salisbury, Van
Voorhis, & Spiropoulos, 2009).
Cottle, Lee, & Heilbrun (2001) identified age at first arrest, offense history, nonsevere
pathology, and family and social factors as the strongest risk factors associated with recidivism.
Several risk-needs models have emerged from prior reviews of the literature. The risk-needsresponsivity model was developed from the meta-analytic literature on risk assessments. The risk
principle emerged from findings that the most effective programming producing reduction in
recidivism is aimed at intensive services for medium- and high-risk offenders (Lipsey, 1992;
Lipsey & Wilson, 1998; Lowenkamp & Latessa, 2002; Lowenkamp, Latessa, & Holsinger,
2006). Whereas the needs principle targets dynamic factors that are correlated with recidivism
(Andrews, Bonta, & Hoge, 1990). However, very few of these meta-analyses contain studies on
women or girl offenders – which assumes that programming for females should mirror
programming designed for males (Chesney-Lind, 2000). For example, Andrews et al., (2011)
reviewed five studies examining the LS/CMI and YLS/CMI, two of which included samples of
juvenile offenders, and of those two studies focusing on youth one had a sample of 27 females
compared to 113 males, and the other had a sample of 81 females as compared to 327 males.
Across all five studies, the total sample size of was 354 females and 2,069 males.
The literature examining gender and risk assessment is mixed regarding providing
support to the propositions that there are unique factors contributing to female offending.
Numerous studies have reported that dynamic risk assessments are valid for females (e.g.,
Andrews et al., 2011; Blanchette & Brown, 2006; Holsinger, Lowenkamp, & Latessa, 2003),
whereas other researchers disagree with the gender equity finding in risk assessments (e.g.,
Blanchette, 2004; Reisig et al., 2006). Previous meta-analyses of risk assessments (Loeber &

18

Dishion, 1983; Simourd & Andrews, 1994) found no gender differences between males and in
terms of risk factors that are predictive of future offending. The leading argument by supporters
of the later position focuses on the is whether gender-specific factors such as history of trauma,
self-esteem, family issues, and relationships are important to future offending or if they represent
prevalent areas of need in the lives of girls and women that are unrelated their system
involvement. Few studies, however, have addressed the recommendations of gender-responsive
literature (Salisbury, Van Voorhis, & Spiropoulos, 2009). Specifically, prior research has not
included the very variables that the gender disparity literature has argued are relatively more
important for young women. There is a history of gender-symmetry arguments in juvenile risk
assessment, that is, the same risk factors can be applied equally to boys and girls (see Ilacqua,
Coulson, Lombardo, & Nutbrown, 1999; Jung & Rawana, 1999; Simourd & Andrews, 1994).
Until recently, few scholars have empirically tested gender differences in variables related to
juvenile offending.
Gender and Risk Assessment
Typically risk assessment instruments either ignore gender (e.g., there is no discussion of
gender or only use male samples) or only use male gender as a predictive criminogenic risk
factor for boys. Combining male and female samples in data analysis may reduce its
effectiveness for girls because of the smaller percentages (typically one quarter to one fifth) of
delinquent girls compared to boys; thus, particular risk factors that may be more salient for girls
do not carry as much weight. Moreover, female risk factors may be “hidden,” thereby reducing
the benefits of using assessment with girls. Chesney-Lind (1986) suggested this was a problem
with her “add women and stir” metaphor: a small sample of girls within a larger sample of boys

19

will likely result in a failure to detect gender-specific risks and needs. Funk (1999) argued that
female risk factors should be analyzed separately from male risk factors in assessment research.
Van Voorhis and Presser (2001) surveyed state correctional classification directors and
found that 36 states had not validated their systems on female offenders, women were being
over-classified based on the assessments, and were being held at higher custody levels. In
addition, these assessments fail to identify the unique needs of women offenders (Van Voorhis &
Presser, 2001). Other researchers have argued that gender-responsive factors are not necessary to
incorporate in criminogenic risk assessments, because few studies have detected gender
differences in terms of the predictive validity of the. Andrews et al. (2011) aggregated youth and
adult data on with the LS/CMI and YLS/CMI to assess if the predictive validity across the eight
subscales are gender-neutral and found that recidivism was strongly associated with YLS score
for both females (AUC = .827) and males (AUC = .746). Andrews and colleagues detected one
difference in the substance abuse female in that it is a stronger predictor for females than for
males. In addition, Andrews et al. (2011) argued that there was not evidence to support the need
for gender-responsive considerations in risk assessments since there is an absence of empirical
evidence for this position as well as existing evidence that supports gender-neutrality in the
predictive validity of the major risk and need factors.
Caulfield (2010) reviewed various adult assessment measures and evaluated the genderspecificity of criminogenic risks and needs. In particular, Caulfield argued that the mere
presence of a risk or need does not presume the etiology is the same across genders. For
example, much of the literature on violent female offending asserts that women’s use of violence
is qualitatively different from men’s use of violence (Blanchette & Brown, 2006). Caulfield
further argued that not only are assessments developed on male offenders less reliable with

20

female offenders, there is a need to focus future research gender-specific criminogenic needs of
women as well as research that addresses the context of women’s lives and their pathways into
and out of the justice system. Salisbury, Van Voorhis, & Spiropoulos (2009) assessed whether
gender-responsive needs were factors related to recidivism and identified that child abuse,
relationships, adult victimization, parental stress, and limited self-efficacy were all differential
risk factors for women. They also argued that the treatment of gender-responsive needs, in
addition to generic risk factors will also reduce recidivism.
Research on the Level of Service Inventory (LSI) has produced conflicting results. Some
researchers have found the validity of the LSI to be robust with female offenders; others have
identified significant differences between women and men on the subscales of the LSI
(Holsinger, Lowenkamp, & Latessa, 2003). Holtfreter and Cupp (2007) reviewed empirical
evidence for the validity of the LSI with women and found that only 25 percent of the literature
even reported statistics for female offenders and typically utilized non-representative samples.
Two critiques of the LSI that Hannah-Moffat (2009) presented were: 1) the current literature’s
sole focus on testing the predictive validity of generic risk factors developed from large samples
of men, and 2) the resulting lack any understanding of female crime etiology, or gendered
patterns of desistence. As well, Hannah-Moffat critiqued the criminological literature for
conceptualizing gender as a binary “sex variable,” which virtually ignores non-aggregate
differences between women and men as well as within-group differences. Taylor and
Blanchette’s (2009) review suggested that when gender-specific items were included in risk
assessments, they fared better than so called gender-neutral assessments. For example,
augmenting traditional “gender-neutral” assessments with gendered items may provide evidence
for the incremental validity in terms of predictive rigor (Taylor & Blanchette, 2009)

21

It is also the case that many of the factors on the LSI such as family issues or antisocial
influences are constrained by the instrument’s theoretical basis, social learning theory. The
argument by non-supporters of this method indicate the lack of gender-specific risk factors (e.g.,
trauma history, dysfunctional relationships, mental health) may limit the effect of other factors
that are more salient for women offenders such as support, conflict, and safety (Salisbury, Van
Voorhis, & Spiropoulos, 2009). That is, more gender-related factors may not properly map onto
the dimensions of family or relational issues as specified on the LSI (Salisbury et al., 2009),
which is largely a critique of the content validity of the assessment for women (Davidson &
Chesney-Lind, 2009). From a policy standpoint, if these gender-specific factors are not
predictive of recidivism, they may be seen as irrelevant even if they are highly prevalent in the
lives of women (Blanchette & Brown, 2006). Morash (2009) reflected that the paradigm (e.g.,
positivism, modernism) that supports the assumptions underlying actuarial risk prediction
methods is in conflict with feminist and pathways explanations of crime. Morash suggested that
research within one paradigm can inform research within another paradigm; therefore, within
this framework, a complete understanding of gendered risk factors in offending cannot be fully
documented by current risk assessment tools (Salisbury, Van Voorhis, & Spiropoulos, 2009).
In sum, the main discussion around gender and actuarial assessment tools is developed in
the adult corrections literature on risk and recidivism. Very little is known about general versus
gender-specific factors in risk assessment among juvenile offenders. What is known about
gender and juvenile assessment is outlined in the next section. However, the theories that
underpin the argument that gender matters—that is, the causes of crime or pathways into crime
are gendered and largely stem from structural gender inequality—can be translated in attempt to

22

understanding female juvenile risk assessment as well (Smith, Cullen, & Latessa, 2009; HannahMoffat, 2009).
Female Juvenile Risk Assessment
Current research has suggested a need to understand some of the dynamic processes
involved in girls’ delinquency as well as their risk and protective factors (Zahn et al., 2008).
Little is known about how many girls comprise the high-risk, repeat offender demographic.
Furthermore, there are very few valid and reliable gender-specific risk assessment instruments in
existence (Brumbraugh et al., 2010). For example, Brumbraugh and colleagues reported that
mental health instruments were most sensitive to gender differences in juvenile populations;
however, out of the 35 risk assessment tools that the researchers reviewed, only five of them
showed promise as gender-appropriate instruments across multiple jurisdictions. Multiple factors
need to be considered when assessing the gender-based analysis or the gender-based
development of the instrument.
Gavazzi, Yarcheck, & Chesney-Lind (2006) analyzed the gender-sensitivity of global risk
indicators on a sample of adjudicated boys and girls and found that girls scored significantly
higher on items relating to family, mental health, traumatic events, and peer relationships.
Gavazzi and colleagues identified that these risks fall into “gender-sensitive” domains. In
particular, boys scored higher on prior offense and differed on the type of offense for which they
were referred (e.g., boys were more likely to be detained for person or property-related offenses
and girls were more likely to be detained for incorrigibility or domestic violence).
Belknap & Holsinger (2006) applied the feminist pathways explanation to better
understand risk factors associated with female delinquency and found that girls report higher
rates of abuse and victimization, although boys also reported high rates of abuse compared to the

23

general population. Girls also reported higher levels of mental health issues (e.g., high rates of
harming themselves and suicide attempts), lower self-esteem, and girls reported higher levels of
abandonment by a parent (Belknap & Holsinger, 2006).
Emeka & Sorensen (2009) examined a generic risk assessment instrument and found that
the instrument worked well for male recidivism, but only provided marginal predictive validity
for females. Emeka and Sorensen reviewed a series of gendered risk factors that may
differentially predict girls’ offending. This set of risk factors includes offense type (e.g., girls are
more likely than boys to referred to court for status offenses), family problems, mental health
needs, and traumatic experiences (Chesney-Lind & Shelden, 2004; Funk, 1999; Gavazzi,
Yarcheck, & Lim, 2005; Gavazzi et al., 2006). Emeka and Sorensen’s (2009) analysis identified
that recidivism rates for girls were lower across risk categories than boys and the AUC for male
offenders was .706, but .611 for female offenders.
Funk (1999) found that instruments developed from combined samples of males and
females fall short in two ways: they explain less variance in girls’ reoffending (17 percent with
the combined model as compared to 31 percent with the female predictors) and do not capture
female risk factors such as child abuse and running away. The female-specific instrument
predicted female recidivism twice as well as the combined instrument. Therefore, Funk urged
juvenile justice researchers and practitioners to develop separate assessment instruments for girls
and boys.
Schwalbe (2008) offered insight into gender differences offense history—girls are
referred more for status offenses and males are more likely to be engaged in person-related
offenses. Schwalbe found that eight out of nine risk factors measured in the North Carolina

24

Assessment of Risk (NCAR) predicated recidivism for males, but only five out of the nine
predicted recidivism for females, with much less accurate predictions for white females.
While meta-analyses of risk assessment instruments have reported strong effect sizes and
adequate predictive validity, much less research has focused on the performance of these
instruments with marginal groups. Schwalbe’s (2008) meta-analysis of 19 studies on juvenile
risk assessment instruments yielded 25 effect sizes for males (ranging from r = .13 to .44) and 24
effect sizes for females (ranging from r = .03 to .57). Schwalbe concluded that only large
differences in base rates of offending (greater than 70 percent) pronounced meaningful gender
differences and risk assessments actually are more effective with female offenders than male
offenders.
Youth Level of Service/Case Management Inventory (YLS/CMI) and Gender
The YLS/CMI, a formal multi-domain risk assessment instrument, is comprised of both
static and dynamic risk factors that are predictive of recidivism and identifies areas of need that
are malleable to intervention efforts (Hoge & Andrews, 1996; Onifade et al., 2009). The
YLS/CMI is an adaptation of the LSI, a version for adults (Andrews & Bonta, 1995). Several
studies have assessed the predictive validity and reliability of the YLS/CMI, however notable
flaws with these studies have been documented including small sample sizes (Jung & Rawana,
1999; Schmidt, Hoge, & Gomes, 2005), retrospective scoring of the instrument (Marczyk,
Heilbrun, Lander, & DeMatteo, 2005), limited generalizability of the setting (Schmidt et al.,
2005), and brief recidivism follow-up (Catchpole & Gretton, 2003). The YLS/CMI is intended
for use by juvenile justice practitioners when developing treatment plans for youth on their
caseloads as well as detained youth (Andrews & Bonta, 1995). To date, research on risk
assessment and classification of juveniles is very limited compared to that of adult offenders.

25

In a study of youth with the largest sample size to date, Bechtel, Lowenkamp, and
Latessa (2007) reported an area under the curve (AUC) of .60, indicating moderate to large
predictive power, with an AUC for an institutionalized sample of .56 and an AUC for a
community-based sample of .64. They also found that the YLS/CMI did not have strong
predictive validity for females, but especially white females; however, females only comprised 7
percent of the study sample. The YLS/CMI is a stronger predictor for both white and non-white
males in either community or institutional settings (Bechtel et al., 2007). In another study on the
predictive validity of the YLS/CMI, Ilacqua et al. (1999) used a matched sample of boys and
girls based on risk score and found no differences in risks; however, matching on scores would
cancel out any real gender differences.
Schwalbe’s (2006) meta-analysis of third-generation risk assessment instruments found
that the YLS/CMI was the most widely used and studied by researchers (k = 11). The predictive
validity of the YLS/CMI has been reported an AUC of .64, the weighted mean effect size across
studies (Schwalbe, 2006). In addition to the larger debate on the differential predictive validity
of risk assessments by gender and the psychometric properties of assessments, Flores et al.
(2004) indicated that the YLS/CMI has strong predictive validity across age, gender, and race.
Onifade et al. (2009) found that the YLS/CMI is an adequate predictor, yet did detect slight
gender differences at two-year follow-up, where the YLS/CMI correctly classified 75 percent of
white females, 69 percent of African American females, 62 percent of white males, and 60
percent of African American males, with the only significant difference between white females
and African American males. Given this limited, and conflicting, body of research, further
studies should assess the differential predictive validity of the YLS/CMI by gender. Finally,

26

Andrews and colleagues (2011) noted that the gender-neutrality of the subscales of both the
LS/CMI and YLS/CMI has not been deeply explored.
Juvenile Delinquency Taxonomies
Researchers have raised the question as to whether juvenile offenders should be treated as
a homogenous group based on risk level or if unique patterns of offenders exist beyond basic risk
profiles (Onifade et al., 2008). Moffitt (1993) created one of the earliest taxonomies of offending
and suggested that there are two different types of antisocial juveniles—those whose behavior is
adolescent-limited and those whose behavior is life-course persistent. Moffitt identified that
adolescent-limited offenders are strongly influenced by contextual factors and typically begin in
mid-adolescence and desist in early adulthood. Whereas with the other group of adolescents,
antisocial behavior begins earlier and persists throughout the lift time, progressively becoming
more serious as they get older (Moffitt, 1993). Previous to Moffitt’s taxonomy, a few young
offender typologies were created. Hewitt and Jenkins (1946) looked at correlations among
demographic, family, and education variables on youth from schools or child welfare agencies
and developed three types of offenders: the “socialized delinquent” (e.g., delinquent peers, gang
involvement, school truancy), the “unsocialized aggressive” (e.g., violent behavior, pro-criminal
attitudes), and “overinhibited” (e.g., submissive, worrisome).
Quay (1964) built upon this work using a homogenous sample of young male offenders
by looking at factors related to behavior and delinquency. Based on the items utilized, Quay
found four factors: the “socialized-substructural” delinquent (e.g., normal psychological profile),
the “unsocialized psychopathic” (e.g., aggressive, irritable), “disturbed-neurotic” (e.g., anxious),
and “inadequacy-immaturity” (e.g., underdeveloped coping skills). However, these early studies

27

were very limited in their statistical sophistication and ability to classify offenders into mutually
exclusive categories.
Mulder, Brand, Bullens, and van Marle (2010) developed a classification of serious
juvenile offenders using 70 static and dynamic risk factors from the Juvenile Forensic Profile
(Brand & van Heerde, 2004) across seven domains: offense history, family and environment,
substance use, psychological factors, interpersonal relationships, and conduct while in the
institution. Other classification work identified youth based on offense type (e.g., violent,
property, sex offenses), however Mulder et al. (2010) argued that developing a classification
based on a set of risk factors instead of offense type provides more information about the
potential underlying problematic behavior. Mulder and colleagues found six distinct subgroups:
antisocial juvenile offenders (e.g., lack of empathy, substance abuse), frequent offenders with
substance abuse, a “flat profile” group (e.g., consistently low scores), juvenile offenders with
family problems (e.g., neglect, domestic violence, poor parenting), and two sex offender
groups—one group lacking cognitive and social skills and the other with sexual problems only.
Mulder and colleagues also examined differential recidivism rates across the subgroups and
found that each subgroup consisted of its own set of risk factors that significantly predicted
recidivism after treatment. However, this study was limited by the use of case file information
and many of the subgroups were rather small (Mulder et al., 2010).
Other researchers developed an empirically-based taxonomy of young male offenders
using the YLS/CMI with cluster analytic techniques (Simourd, Hoge, Andrews, & Leschied,
1994). Their solution yielded five distinguishable types: “generalized high risk/need” (e.g.,
offense history, spikes in family and attitudes domain), “low risk” (e.g., uniformly low scores
across domains), “difficulties in community” (e.g., delinquent peers, problems in school,

28

unproductive use of leisure time), “family and personal distress” (e.g., family conflict, high
scores in the personality domain), and “economically disadvantaged” (e.g., issues with family
finances). These solutions were subsequently validated on external criteria such as offense,
disposition, program compliance, and successive convictions (Simourd et al., 1994). However,
since data were coded from case file information, findings from this study may differ from other
research on this instrument that used structured interviewing.
Onifade and colleagues (2008) used the YLS/CMI to group offenders based on patterns
of criminogenic need and risk beyond the traditional, uni-dimensional low, moderate, and high
risk levels. This study produced five unique patterns of risk that were cross-validated in two
populations of offenders. These five clusters included a low risk group, a moderate risk group
with environmental needs, another moderate risk group with family needs, an two high risk
groups—one with a history of offense and another with first-time offenders (Onifade et al.,
2008). The authors argue that more research needs to be produced in the area of understanding
both the differential predictive validity of assessments as well as identifiable patterns of risk that
may moderate the risk-recidivism relationship across gender and racial groups. Finally, the study
identified that juveniles with minimal system contact are indeed distinguishable, which has
considerable policy (e.g., diverting offenders, reducing caseloads) and treatment implications
(Onifade et al., 2008).
Thus, is critical to explore the role of gender in the development of offending taxonomies
and patterns of risk given the wide gap in this already scant body of literature. To date,
researches have only used male offenders in their classification procedures (Moffitt, 1993;
Mulder et al., 2010; Simourd et al., 1994), or analyzed combined samples of male and female
offenders (Onifade et al., 2008). Standards set by the National Council on Crime and

29

Delinquency (1998) required that assessment tools be analyzed for validity, reliability, and
equity across all types of juvenile offenders. Therefore, the study reported here examined the
gender-based performance of a widely used juvenile risk assessment instrument as well as
develop a gender-specific taxonomy of juvenile offender risk and need profiles.
Current Study
The current study adds to the discussion in the literature around gender and risk
assessment by examining patterns of results using the YLS/CMI, one of the most popular risk
assessment measures. While the current research was limited by using a measure (the YLS/CMI)
historically developed for boys, it provided a beginning step towards determining to what extent
there were unique patterns of risk as a function of gender. Future work will need to also examine
uniquely female risk factors within a competitive methodology (e.g., assess the incremental
validity by adding gender-specific risk factors into the model). Prior reviews have consistently
concluded that there is limited research on the gender specificity of risk assessments in
populations of juvenile probationers. Many risk assessments have been developed on samples
solely comprised of boys or majority boys with a small sample of girls (Blanchette & Brown,
2006; Reisig, Holtfreter, & Morash, 2006). As well, most studies do not conduct analyses
separately by gender, which is considered best practice when analyzing gender differences in
offending. While Andrews et al. (2011) argued that there is a lack of empirical evidence to
support the need for gender-specific items or instruments, the review only examined one study
exclusively using the YLS/CMI with a relatively small sample size of girls (n = 81).
Additionally, this study noted limitations in variation of recidivism follow-up periods, anywhere
from one to four years, and the data used in analysis were overrepresented by Canadian
databases created by the authors of the LSI (Andrews et al., 2011). Finally, this study will also

30

contribute by integrating other statistical examinations of gender (e.g., interaction effects) across
risk scores and domains in additional to the traditional tests of predictive validity.
The study reported here examined how the psychometric properties of the instrument
vary based on gender. First, assessing how gender acts as a moderator can inform researchers
and practitioners of how specific risk factors differentially affect girls versus boys and more
specifically the strength and/or direction of this relationship (Aguinis, 2004). In addition, this
study adds to the literature by assessing gender differences of each domain by comparing the
predictive validity of the YLS/CMI. There are currently no studies that have tested if the
predictive validity across domains is significantly different by gender for adolescent offenders.
Further, this study examined the gendered nature of a juvenile offending typology
(Onifade et al., 2008) and assessed if unique offending typologies exist when analyzed separately
based on gender. No studies to date have been published assessing gender-specific patterns of
offending in this nature. Given that researchers have identified that the risk-recidivism
relationship is different for members of marginal groups (e.g., girls, racial/ethnic minority)
(Onifade et al., 2009), it is of importance to understand which specific factors are implicated in
this relationship. In addition, this study has implications at the level of a local county court and
has the possibility of translating across other juvenile court settings. The results will provide the
court with richer information on gender and risk assessment, gender differences in recidivism
rates, as well as provide a more detailed understanding of gendered typologies of offending.
Court personnel can use this information to make better treatment decisions by sorting boys and
girls into appropriate community-based programs, more intensive services, or diversion
programming based on differences in their risk and need profiles. Finally, this study will be used

31

to develop a discourse in court settings around the importance of gender and gender-specific
considerations in risk assessment, processing, and treatment modalities.
Research Questions
This study aims to assess the validity of the YLS/CMI subscales and cluster types by gender
through the following seven research questions:
1. Does gender moderate the relationship between the YLS/CMI risk score and recidivism?
Does gender act as a moderator across risk levels?
2. Does gender moderate the relationship between the eight YLS/CMI subscales and
recidivism?
3. How well can the YLS/CMI correctly identify juveniles who will reoffend based on
gender?
4. What is the predictive validity of each subscale of the YLS/CMI based on gender?
5. Do gender differences in the predictive validity exist across the domains of the
YLS/CMI?
6. Are there gender differences in the original delinquency unit cluster solutions?
7. Are there unique cluster solutions based on gender? If unique gender clusters exist, what
are the patterns of reoffending across typologies?
In order to answer these research questions, descriptive, assessment and recidivism data were
collected on a sample of juvenile probationers.

32

METHODS
Sample
The sample is drawn from secondary data in the probation division of a medium-sized
Midwestern juvenile court. This is a cross-sectional sample (n=1,110) of both male (n = 832) and
female (n = 278) youth under jurisdiction of the court between 2004 and 2008. There were
neither refusals nor duplicate cases. All youth under supervision of the court were assessed at
intake. The assessment was interview-based and conducted by the youth’s juvenile court officer.
Any identifying information was removed from the assessments prior to being made available to
the researcher by the court. This timeframe in the study allows for a 24-month follow-up period
to assess recidivism. Recidivism is defined as any new court petitions within this two-year period
of time following the administration of the YLS/CMI. All recidivism data were collected through
the court’s data management system. Table 1.1 provides demographic and basic descriptive
information of the probation sample that used in this study. There were not significant
differences in age or race between girls and boys in this sample. However, of notable interest, are
the differences between person and property offenses across gender. Person-related offenses
accounted for 49.3 percent of girls’ arrests, while property offenses accounted for 47.0 percent of
boys’ arrests.
Table 1.1. Sample Demographics

Average Age
Caucasian
Latino/Hispanic
African-American
Multi-Racial
Other Race
Person

Girls
n = 278
M (SD)
14.87 (1.32)
N (%)
110 (39.9)
24 (8.7)
92 (33.3)
46 (16.7)
4 (1.4)
132 (49.3)
33

Boys
n = 832
M (SD)
14.79 (1.51)
N (%)
338 (40.9)
74 (9.0)
295 (35.6)
104 (12.6)
16 (1.9)
209 (26.1)

Table 1.1 (cont’d)
Property
97 (36.2)
377 (47.0)
Weapon
3 (1.1)
21 (2.6)
Drug
4 (1.5)
63 (7.9)
Status
18 (6.7)
24 (3.0)
Public Ordinance
10 (3.7)
29 (3.6)
Sex
0 (0.0)
71 (8.9)
Other
4 (1.5)
8 (1.0)
*Valid Percent of Crime Type (10 girls and 31 boys missing original charge information)
Training and Procedures
Court personnel were trained in administering the YLS/CMI. Each juvenile court officer
received four days (32 hours total) of training prior to using the instrument. The trainings
consisted of providing definitions, explaining item-by-item scoring criteria, practicing
interviews, as well as listening to and coding cases. Inter-rater reliability checks were performed
routinely and consistently reached at least 90 percent exact agreement.
Measures: Independent Variables
This section reviews the measures utilized in the study. This includes an explanation of
the predictor variables as well as how recidivism is used as a dependent variable. The YLS/CMI
is a 42-item instrument that is comprised of eight subscales. The items of each of the subscales
are dichotomously scored (no = 0, yes = 1); thus, scores can range from 0-42. The eight
subscales assess both static and dynamic risk factors for future offending (Hoge et al., 2002):
1. Prior offenses/dispositions. Five items focusing on official offense history (e.g.,
“three or more prior convictions”).
2. Education. Seven items focusing on school performance and behavior (e.g., “low
achievement”).
3. Leisure and recreation. Three items focusing on use of free time (e.g., “lack of
organized activities”).

34

4. Peer relationships. Four items focusing on the characteristics of acquaintances and
friends (e.g., “lack of positive acquaintances”).
5. Substance abuse. Five items focusing on drug and alcohol use/abuse (e.g.,
“occasional drug use”).
6. Family and parenting. Six items focusing on family relationships and parental
behavior (e.g., “inadequate supervision”).
7. Attitudes and orientation. Five items focusing on antisocial tendencies (e.g., “not
seeking help”).
8. Personality and behavior. Seven items reflecting disruptive behavior and personality
characteristics (e.g., “short attention span”).
Items within each of the subscales are computed to create a summated score for each risk
domain with scores ranging from 3-7. Appendix A includes the 42 items of the YLS/CMI.
Table 1.2. Descriptive Statistics for Variables in Analysis
Girls
Boys
n = 278
n = 832
M (SD)
M (SD)
History
.65 (1.04)
.71 (1.14)
Education
3.22 (1.96)
3.01 (1.81)
Leisure & Recreation
1.59 (1.01)
1.52 (1.03)
Peers
2.22 (1.38)
2.21 (1.36)
Family
3.01 (1.78)
2.67 (1.78)
Substance Abuse
1.29 (1.53)
1.37 (1.53)
Attitudes & Orientation
1.02 (1.36)
1.22 (1.45)
Personality
3.15 (1.90)
2.60 (1.89)
Overall YLS/CMI Score
16.21 (7.86)
15.36 (8.01)
N (%)
N (%)
Low Risk
54 (19.4)
189 (22.7)
Moderate Risk
167 (60.1)
473 (56.9)
High Risk
57 (20.5)
170 (20.4)

X2 or t

*
*
*

Table 1.2 provides descriptive statistics of the overall YLS/CMI score, YLS/CMI
subscale scores, and distribution of the sample across risk levels. Differences in mean scores

35

were tested in order to identify significant differences in assessment scores by gender. The
descriptive analysis revealed significant differences among three subscales. Girls scored
significantly higher than boys in the Family and Parenting as well as the Personality and
Behavior subscales, while boys scored significantly higher in the Attitudes and Orientation
domain. There were no mean differences in the YLS/CMI composite score or across risk levels.
The subscale correlation matrix for the eight different domains is presented in Table 1.3.
Table 1.3. Subscale Correlation Matrix
1
2
3
4
5
6
7
1
-2
.192
-3
.177
.421
-4
.266
.417
.434
-5
.272
.316
.275
.400
-6
.271
.331
.374
.468
.324
-7
.116
.485
.563
.368
.148
.278
-8
.285
.431
.433
.441
.285
.384
.501
*Subscale Labels (1- Prior/History, 2- Family/Parenting, 3- Education, 4- Peers, 5- Substance
Abuse, 6- Leisure/Recreation, 7- Personality/Behavior, 8- Attitudes/Orientation)
Measures: Dependent Variable
The outcome of interest in this study is recidivism. Recidivism is coded 1 if there was a
new offense in the 24-month follow-up period or 0 if there was no new petition within that
timeframe. Recidivism data are collected on the same list of youth with an initial YLS/CMI
through the court data management system. If a juvenile aged out of the system during the
follow-up period, adult records will be checked as well. Table 1.4 provides the two-year followup recidivism rates of the sample by gender as well as by risk level. These recidivism rates reveal
that boys (RR = 52.0%) in this sample reoffend at higher rates than girls (RR = 40.3%) two-years
following initial assessment. The largest difference in recidivism rate was found among the
moderate risk group in which girls’ rate was 41.9 percent and boys’ rate was 57.1 percent at twoyears following the assessment date.
36

Table 1.4. Two-Year Recidivism by Gender and Risk Level
Girls
Boys
n = 278
n = 832
Recidivists
112
433
Non-Recidivists
166
399
Recidivism Rate
40.3%
52.0%
Low Risk
27.8%
31.7%
Moderate Risk
41.9%
57.1%
High Risk
47.4%
61.2%

37

RESULTS
There were a total of 1,110 juvenile probationers in this study including 278 girls and 832
boys. Within the sample of girls, 112 reoffended within the two-years of their initial assessment
(40.3 percent of the sample) and 433 boys reoffended (52.0 percent of the sample). Based on
recidivism rates, theses findings indicate significantly greater delinquency among the male
sample as calculated by an independent samples t-test. The results will be presented in order by
research question. First, in order to answer the first and second research questions, a moderated
binary logistic regression was calculated to assess whether gender moderates the relationship
between the overall YLS/CMI risk score and two-year recidivism. First, the YLS/CMI risk score
was grand-mean centered. Boys were coded 0 and girls were coded as 1 in the database. Second,
an interaction variable was computed in SPSS to determine risk score*gender. After these new
variables are created, gender, YLS/CMI risk score, and the interaction variable will be included
as covariates in the regression equation. As well, three logistic regression equations were
calculated for each of the three risk levels by using gender as the independent variable and twoyear recidivism as the dependent variable. Odds ratios are examined to determine the strength of
the moderated risk-recidivism relationship. In order to answer RQ2, a moderated binary logistic
regression was conducted for each of the eight subscales of the YLS/CMI and recidivism. All of
the subscale scores will be grand-mean centered prior to computing the gender*subscale score
interaction. Eight separate regression equations were computed to test whether gender moderates
the relationship between any of the subscales and recidivism.
1. Does gender moderate the relationship between the YLS/CMI risk score and recidivism?
Does gender act as a moderator across risk levels?

38

Table 2.1 presents the results of the logistic regression predicting the dichotomous
outcome at each level of gender. Score, gender, and the product of score and gender were
included as variables in the model. The overall risk score was not moderated by gender with an
odds-ration of .994 (ns). Based on the Pseudo R2, the range of variation explained in future reoffense is 3.9-5.2 percent.
Table 2.1. Moderated Logistic Regression for Predicting Recidivism with Total Score
Variable
B
S.E.
Wald
Sig
Exp(B)
YLS Score
.040
.009
19.667
.000
1.041
Gender
.520
.143
13.207
.000
1.681
YLS Score X Gender
-.006
.009
.408
.523
.994
Constant
-.422
.124
11.511
.001
.656
2
2
2
-2log likelihood=1495.76; X =44.087; Cox and Snell R =.039; Nagelkerke R =.052
Note: p<.05
Next, the categorical variable, risk level, rather than the continuous variable, raw risk
score, was used as the predictor. Table 2.2 presents the results of examining the effect of gender
at different levels of risk. The logistic regression revealed no significant effect of gender among
low risk (Odds-Ratio = .827, ns) and high risk (Odds-Ratio = .571, ns) juveniles. However, the
moderate risk group was moderated by gender given that girls were nearly half as likely than
boys to reoffend within two-years among this group with an odds-ratio of .543 (p < .01).
Table 2.2. Moderated Logistic Regression for Predicting Recidivism by Risk Level
Variable
B
S.E.
Wald
Sig
Exp(B)
Low Risk
-.190
.342
.309
.578
.827
Constant
-.765
.156
23.996
.000
.465
2
2
2
-2log likelihood= 300.04; X =.314; Cox and Snell R =.001; Nagelkerke R =.002
Moderate Risk
-.611
.182
11.252
.001
.543
Constant
.285
.093
9.426
.002
1.330
2
2
2
-2log likelihood= 873.32; X =11.405; Cox and Snell R =.018; Nagelkerke R =.024
High Risk
-.560
.308
3.297
.069
.571
Constant
.455
.157
8.349
.004
1.576
2
2
2
-2log likelihood= 305.965; X =3.306; Cox and Snell R =.014; Nagelkerke R =.019
Note: p<.05

39

2. Does gender moderate the relationship between the eight YLS/CMI subscales and
recidivism?
Another series of moderated logistic regression models were conducted to test the riskrecidivism relationship by risk domain. The results for these eight models, separated by subscale,
are presented in Table 2.3. Findings revealed that gender only moderated the relationship
between one of the eight subscales and recidivism. Gender moderated the family risk domain
with two year recidivism in which the odds of girls’ reoffending was 1.181 times greater given
their family risk score (Odds-Ratio = 1.181, p < .05).
Table 2.3. Moderated Logistic Regression for Predicting Recidivism by Subscale
Variable
B
S.E.
Wald
Sig
Exp(B)
History
.347
.067
26.959
.000
1.415
Gender
-.473
.142
11.169
.001
.623
History X Gender
-.231
.134
2.954
.086
.794
Constant
.084
.071
1.411
.235
1.088
2

2

2

2

2

2

2

2

2

2

2

2

-2log likelihood=1495.06; X =41.944; Cox and Snell R =.037; Nagelkerke R
Family
.050
.039
1.663
.197
Gender
-.549
.145
14.296
.000
Family X Gender
.167
.081
4.192
.041
Constant
.086
.070
1.536
.215

=.049
1.052
.578
1.181
1.090

-2log likelihood=1515.57; X =22.857; Cox and Snell R =.020; Nagelkerke R =.027
Education
.134
.039
11.782
.001
1.143
Gender
-.497
.142
12.299
.000
.608
Education X Gender
-.058
.074
.604
.437
.944
Constant
.090
.070
1.649
.199
1.094
2
2
2
-2log likelihood=1513.41; X =25.012; Cox and Snell R =.022; Nagelkerke R =.030
Peers
.262
.053
24.783
.000
1.300
Gender
-.480
.141
11.506
.001
.619
Peers X Gender
-.168
.104
2.625
.105
.845
Constant
.084
.070
1.426
.232
1.088
-2log likelihood=1500.17; X =38.258; Cox and Snell R =.034; Nagelkerke R =.045
Substance
.184
.047
15.527
.000
1.202
Gender
-.231
.186
1.551
.213
.794
Substance X Gender
-.176
.093
3.616
.057
.839
Constant
.081
.070
1.339
.247
1.084
-2log likelihood=1509.89; X =27.187; Cox and Snell R =.024; Nagelkerke R =.032
Leisure
.209
.068
9.522
.002
1.233
40

Table 2.3 (cont’d)
Gender
Leisure X Gender
Constant

2

-.481
-.187
.086

.141
.139
.070

11.639
1.799
1.526
2

.001
.180
.217

2

-2log likelihood=1517.19; X =21.242; Cox and Snell R =.019; Nagelkerke R
Personality
.122
.037
10.682
.001
Gender
-.551
.145
14.365
.000
Personality X Gender
.005
.075
.005
.944
Constant
.100
.070
2.027
.155
2

2

2

.618
.830
1.090
=.025
1.130
.576
1.005
1.105

-2log likelihood=1512.14; X =26.285; Cox and Snell R =.023; Nagelkerke R =.031
Attitudes
.153
.049
9.686
.002
1.165
Gender
-.454
.142
10.301
.001
.635
Attitudes X Gender
-.037
.102
.130
.719
.964
Constant
.076
.070
1.186
.276
1.079
2
2
2
-2log likelihood=1515.25; X =23.180; Cox and Snell R =.021; Nagelkerke R =.028
3. How well can the YLS/CMI correctly identify juveniles who will reoffend based on
gender?
In order to test the predictive validity of the instrument, a receiver operating characteristic
AUC was calculated on risk scores for the entire sample as well as disaggregated by gender. The
AUC is the ability of the instrument to correctly classify boys and girls who will reoffend. This
calculation is ideal for assessing the predictive validity because it corrects for base rate by
calculating the ratio of true-positives to false-positives. AUCs were calculated separately for the
boy sample, the girl sample, and the overall combined sample. Furthermore, the sample was
separated by gender and AUCs were calculated for each of the subscales for the boy sample, the
girl sample, and the overall combined sample. This calculation was repeated for each of the eight
subscales to assess the predictive ability of each of the subscales. AUC values above .50 indicate
adequate predictive validity (Schmidt et al., 2005), whereas predictive power is considered
strong if the AUC value is above .60 (Rice and Harris, 1995). Results from the overall sample
are presented in Table 3.1. Findings reveal that the ROC resulted in a test statistic of .590 (p <

41

.001) for the overall YLS/CMI score and .591 (p < .001) by risk level prediction. The test
statistics for the subscales ranged from .543-.579, all of which were statistically significant
predictors (p < .05).
Table 3.1. Predictive Validity of the YLS/CMI: Overall Sample (n=1,110)
Area
Under the
Curve
Standard
(AUC)
Error
P value
History
.579
.017
.000
Education
.558
.017
.001
Leisure & Recreation
.543
.017
.013
Peers
.578
.017
.000
Family
.545
.017
.009
Substance Abuse
.562
.017
.000
Attitudes
.562
.017
.000
Personality
.558
.017
.001
Total YLS
.590
.017
.000
Risk Level
.591
.017
.000

95%
Confidence
Interval
.545-.612
.524-.591
.509-.577
.545-.612
.511-.579
.528-.595
.528-.596
.525-.592
.557-.624
.558-.624

While findings indicated that the YLS/CMI has adequate predictive validity for the
overall sample, slight differences emerged for the disaggregated groups. Findings for the sample
of boys are presented in Table 3.2. For the sample of boys, the ROC resulted in a test statistic of
.601 (p < .001) for the overall YLS/CMI score and .602 (p < .001) for risk level, both of which
indicate strong predictive power.
Table 3.2. Predictive Validity of the YLS/CMI: Boys (n=832)
Area
Under the
Curve
Standard
(AUC)
Error
History
.594
.020
Education
.568
.020
Leisure & Recreation
.558
.020
Peers
.594
.020
Family
.533
.020
Substance Abuse
.580
.020
Attitudes
.563
.020
42

P value
.000
.001
.004
.000
.103
.000
.002

Confidence
Interval
.555-.632
.529-.607
.519-.597
.555-.632
.493-.572
.541-.618
.524-.602

Table 3.2 (cont’d)
Personality
Total YLS
Risk Level

.565
.601
.602

.020
.020
.020

.001
.000
.000

.526-.604
.562-.639
.563-.640

Findings for the sample of girls are presented in Table 3.3. Among this sample, the ROC
resulted in a test statistic of .573 (p < .05) for the overall YLS/CMI score and .564 (ns) for risk
level, indicating adequate predictive power for only the overall risk score.
Table 3.3. Predictive Validity of the YLS/CMI: Girls (n=278)
Area
Under the
Curve
Standard
(AUC)
Error
History
.529
.035
Education
.540
.035
Leisure & Recreation
.500
.035
Peers
.534
.035
Family
.607
.034
Substance Abuse
.501
.035
Attitudes
.547
.035
Personality
.568
.035
Total YLS
.573
.035
Risk Level
.564
.035

P value
.412
.257
.990
.337
.002
.988
.183
.053
.039
.068

Confidence
Interval
.460-.598
.472-.608
.432-.569
.465-.603
.541-.673
.431-.570
.478-.616
.501-.636
.505-.641
.496-.633

4. What is the predictive validity of each subscale of the YLS/CMI based on gender?
Table 3.2 presents the AUCs for the sample of boys. Among the subscales, the test
statistics results in AUCs of .594 (p < .001) for the history domain, .568 (p < .01) for the
education domain, .558 for the leisure and recreation domain (p < .01), .594 for the peers domain
(p < .001), .580 for the substance abuse domain (p < .001), .563 for the attitudes and orientation
domain (p < .01), and .565 for the personality domain (p < .01), all of which had statistically
significant predictive power. The family domain is the only subscale that produced weak
predictive power with an AUC of .533 (ns).

43

Table 3.3 presents the AUCs for the sample of girls. Among these subscales, the test
statistic resulted in AUCs of .529 for the history domain (ns), .540 for the education domain (ns),
.500 for the leisure and recreation domain (ns), .534 for the peers domain (ns), .501 for the
substance abuse domain (ns), .547 for the attitudes and orientation domain (ns), and .568 for the
personality and behavior domain (ns). The only subscale that produced statistically significant
predictive validity for the sample of girls was the family domain with an AUC of .607 (p < .01).
5. Do gender differences in the predictive validity exist across the domains of the
YLS/CMI?
Certain subscales appear to be stronger predictors of recidivism for boys than girls.
MedCalc for Windows, version 9.5.0.0 (MedCalc Software, Mariakerke, Belgium) was used to
test for differences between the boy and girl subsamples based on their AUCs. The MedCalc
correction used in the analysis for pairing AUCs is similar to a paired samples t-test; however, it
is considered more sensitive to statistical significance (Hanley & McNeil, 1983).
Results of the gender differences in the predictive validity of the YLS/CMI are presented
in Table 3.4. The only subscale in which significant differences were noted was the substance
abuse scale. The substance abuse AUC for the sample of girls was .501 and the AUC for the
sample of boys was .581 (p < .05). This indicates that the substance abuse domain has
significantly greater predictive ability of recidivism for boys than girls. The remainder of the
subscales as well as the total YLS/CMI score and risk levels were not significantly different by
gender.
Table 3.4. Gender Differences in the Predictive Validity of the YLS/CMI (n=1,110)
Girls
Boys
Differences
Girls AUC
Std. Error
Boys AUC
Std. Error
P value
History
.529
.035
.594
.020
.107
Education
.540
.035
.568
.020
.487
Leisure & Recreation
.500
.035
.558
.020
.150
44

Table 3.4 (cont’d)
Peers
Family
Substance Abuse
Attitudes
Personality
Total YLS
Risk Level

.534
.607
.501
.547
.568
.573
.564

.035
.034
.035
.035
.035
.035
.035

.594
.533
.581
.563
.565
.601
.602

.020
.020
.020
.020
.020
.020
.020

.137
.061
.047
.691
.941
.487
.346

6. Are there gender differences in the original delinquency unit cluster solutions?
In order to examine gender differences in the original delinquency unit cluster solutions
(Onifade et al., 2008), the composition of juveniles within each cluster was examined by gender
as well as the two-year recidivism rates for each cluster. These clusters were created on samples
of intake and probation youth to provide nuance to the standard risk levels (Onifade et al., 2008).
Results from this analysis are presented in Table 4.1. Findings indicate that the negligible risk
cluster includes the largest subpopulation, comprising 28.8 percent of the girls and 30.5 percent
of the boys. Within this subpopulation, girls have an average YLS/CMI score of 6.79, while boys
average score is 6.36. Juveniles within the negligible risk cluster have low risk scores across all
of the domains. Boys have a slightly higher recidivism rate (36.61 percent) than girls (31.25
percent) in the negligible risk cluster. The environmental needs cluster is comprised of 21.6
percent of the girl sample and 23.3 percent of the boy sample. Within the environmental needs
cluster, the average score of girls is 14.12 and the average score of boys is 13.34. This includes
peaks in prior offenses, leisure and recreation, peer relations, and substance abuse. The
recidivism rate of boys (60.82 percent) is higher than the recidivism rate of girls (46.67). The
family needs cluster includes 16.2 percent of the girl population and 16.3 percent of the boy
population. The average overall YLS/CMI score for girls within this cluster is 18.58, while the
average score for boys is 17.63. This includes peaks in the education, family and parenting,
45

attitudes and orientation, as well as personality and behavior domains. Again, the boys’
recidivism rate (58.09 percent) is higher than girls’ recidivism rate (40.00 percent). The high risk
with offense history cluster comprises 24.8 percent of the sample of girls and 19.2 percent of the
sample of boys. This is the only cluster in which girls have a higher proportion of their sample
represented than boys. The average YLS/CMI score for girls in this cluster is 22.86 and 22.69 for
boys. This cluster reflects high risk scores in all eight of the domains. The recidivism rates for
the high risk with offense history are 42.03 percent for girls and 51.25 percent for boys. Finally,
while the high risk first offense cluster is the smallest subpopulation overall, comprising 8.6
percent of the girls and 10.7 percent of the boys, it also has the highest recidivism rates overall
by gender. Girls’ recidivism rates within this cluster are 50.00 percent and boys’ recidivism rates
are 69.66 percent. The average YLS/CMI score of girls in this cluster is 29.25 while the average
score for boys is 28.87. This cluster reflects high risk scores in all eight of the domains. See
Appendix B for the “YLS Pattern of Need Descriptive Guide and Service Manual” (2007), which
further describes cluster composition as well as recommended clinical intervention and services.
Table 4.1. Gender Differences in Delinquency Cluster Types
Girls
Boys
Girls
(n=278)
(n=832)
Recidivism Rate
N (%)
N (%)
%
Negligible Risk 80 (28.8)
254 (30.5)
31.25
Environmental Needs 60 (21.6)
194 (23.3)
46.67
Family Needs 45 (16.2)
136 (16.3)
40.00
High Risk w/ History 69 (24.8)
160 (19.2)
42.03
High Risk First Offense 24 (8.6)
89 (10.7)
50.00

Boys
Recidivism Rate
%
36.61
60.82
58.09
51.25
69.66

An additional cross tabulation of cluster type by gender revealed differences in the
proportion of males and females across each clusters. Since girls account for 25 percent of the
total sample, girls are slightly overrepresented in the high risk with offense history since females
comprise 30.1 percent of this cluster, while girls are underrepresented in the high risk with first
46

offense cluster by only comprising 21.2 percent of the cluster sample total. The other three
cluster types have comparable proportions to the total sample. The negligible risk cluster is
comprised of 24.0 percent girls, the environmental needs cluster is comprised of 23.6 percent
girls, and the family needs cluster is comprised of 24.9 percent girls.
7. Are there unique cluster solutions based on gender? If unique gender clusters exist, what
are the patterns of reoffending across typologies?
In order to examine differential patterns of risk and need by gender, two separate cluster
analyses were conducted to test for boys’ and girls’ unique patterns of risk and needs. First, a
reliability analysis to confirm the internal structure of the total 41-item scale produced a
Cronbach’s Alpha of .808. Table 4.2 presents the corrected item-total correlations for each of the
items on the subscale for the overall sample as well as separately for the sample of girls and the
sample of boys. The alpha coefficients for the overall sample ranged from .74 to .81 across the
subscales. The alpha coefficients for the subscales ranged from .74 to .79 for the sample of girls.
Alpha coefficients ranged from .74 to .81 for the sample of boys. As well, refer back to Table 1.3
for the subscale correlation matrix. These composite subscale variables are used at the input
clustering criteria.
Table 4.2. Corrected Item-Total Correlations and Reliability Scores
Item-Subscale Correlations
Total Sample
Girls
Boys
Offense History
α = .76
α = .75
α = .77
Three or More Prior Convictions
.62
.55
.67
Two or More Failures to Comply
.74
.76
.73
Prior Probation
.76
.70
.78
Prior Custody
.76
.76
.77
Three or More Current Convictions
.37
.38
.37
Education
α = .75
α = .77
α = .75
Low Achievement
.57
.60
.56
Problems with Teachers
.73
.71
.74
Problems with Peers
.66
.68
.65
47

Table 4.2 (cont’d)
Disruptive Classroom Behavior
Disruptive Behavior on School Property
Truancy
Leisure and Recreation
Lack of Organized Activities
Could Make Better Use of Time
No Personal Interests
Peer Relationships
Lack of Positive Peer Acquaintances
Lack of Positive Friends
Some Delinquent Peer Acquaintances
Some Delinquent Friends
Substance Use
Occasional Drug Use
Chronic Drug Use
Chronic Alcohol Use
Substance Abuse Interferes with Life
Substance Use Linked to Offense(s)
Family
Inadequate Supervision
Difficulty in Controlling Behavior
Inappropriate Discipline
Inconsistent Parenting
Poor Relations (Father-Youth)
Poor Relations (Mother-Youth)
Attitude and Orientation
Not Seeking Help
Actively Rejecting Help
Defies Authority
Antisocial/Pro-Criminal Attitudes
Callous, Little Concern for Others
Personality
Short Attention Span
Poor Frustration Tolerance
Verbally Aggressive/Verbally Intimidating
Explosive Episodes
Physically Aggressive
Inadequate Guilt Feelings
Inflated Self-Esteem

.73
.64
.44
α = .81
.80
.79
.68
α = .79
.74
.76
.67
.69
α = .79
.71
.81
.52
.86
.70
α = .75
.60
.64
.66
.73
.48
.57
α = .77
.72
.60
.67
.73
.69
α = .74
.49
.69
.70
.71
.64
.46
.34
48

.78
.73
.51
α = .81
.81
.81
.65
α = .79
.77
.76
.69
.68
α = .79
.69
.79
.66
.87
.68
α = .75
.59
.64
.66
.74
.48
.58
α = .77
.72
.70
.64
.71
.63
α = .74
.56
.72
.63
.70
.60
.51
.39

.72
.61
.41
α = .81
.80
.78
.69
α = .78
.73
.76
.66
.70
α = .78
.71
.82
.47
.86
.71
α = .75
.61
.64
.66
.73
.47
.56
α = .77
.71
.58
.68
.73
.71
α = .74
.50
.69
.72
.70
.65
.45
.32

The TwoStep cluster algorithm produced a three cluster solution for the sample of girls
with the eight input cluster variables. The ratio of cluster sizes (largest cluster to smallest cluster)
was 1.46, while structure silhoutte measure of cohesision and separation in the analysis was .3.
These measures assess the closeness between clusters relative to the closeness of the variables
within each cluster. In terms of the silhouette of cohesion and separation, the coefficients range
between -1 (poor model fit) to 1 (excellent model fit) and are categorized as poor model fit (<
.2), fair model fit (.2 to .5), and good model (> .5) based on the partitioning of the data in terms
of how closely cases within cluster are relative to their distance between the other clusters
(Kaufman & Rousseeuw, 1990). Thus, the three cluster solution in this study’s coeffecient of .3
would indicate a fair rating suggesting that there is a slight degree of closeness to cases in
different clusters, there was a susbstantial distance between them to conclude adeqate differences
between the clusters.
Table 4.3 presents the results from the cluster analysis of the girls’ sample including
mean subscale scores, total YLS/CMI score, and recidivism rates within-cluster. Girls’ cluster
solutions appear undifferentiated from low, moderate, and high risk levels. Risk scores (M =
6.97) and recidivism rates (34.62%) among girls in Cluster One (n = 78) are similar in profile to
an uniformily low risk adolescent. Cluster Two (n = 114) included the largest proportion of girls
(41%) with risk scores (M = 15.83) and recidivism rates (37%) in the moderate risk level.
Finally, Cluster Three (n = 86) yeilded the highest average overall risk score (M = 25.07) as well
as the highest recidivism rate (48.84%) compared to the other two clusters as well as the overall
two-year recidivsm rate for girls (40.29%).
Table 4.3. Girls’ Patterns of Risk and Recidivism Rates
Cluster One
Cluster Two
Subscales
(n=78, 28.1%) (n=114, 41.0%)
Offense
.37 (.65)
.51 (.88)
49

Cluster Three
(n=86, 30.9%)
1.08 (1.35)

Subscale Mean
.64 (1.04)

Table 4.3 (cont’d)
Family
Education
Peers
Substance
Leisure
Personality
Attitudes
Total
Recidivism Rate

1.64 (1.39)
1.24 (1.31)
.79 (.92)
.29 (.70)
.47 (.66)
1.82 (1.65)
.23 (.53)
6.97 (3.57)
34.62

2.95 (1.58)
3.39 (1.61)
2.35 (1.03)
1.41 (1.46)
1.88 (.73)
2.96 (1.63)
.32 (.58)
15.83 (3.39)
37.72

4.34 (1.32)
4.78 (1.20)
3.33 (.94)
2.02 (1.70)
2.23 (.73)
4.59 (1.44)
2.67 (1.16)
25.07 (4.10)
48.84

3.01 (1.78)
3.22 (1.96)
2.22 (1.38)
1.29 (1.53)
1.59 (1.01)
3.15 (1.91)
1.02 (1.36)
16.21 (7.86)
40.29

Figure A.1 is a graphical representation of the standardized mean subscale scores for each
of the three cluster solutions. The radar graph indicates peaks in the standardized domain scores
by cluster type.
Figure A.1 Radar Graph of Subscale Peaks in Girls’ Clusters
Prior/Current
Offenses
Attitudes &
Orientation

Family &
Parenting
Cluster 1

Personality &
Behavior

Education

Cluster 2
Cluster 3

Leisure/
Recreation

Peer Relations
Substance Abuse

For interpretation of the references to color in this and all other figures, the reader is referred to
the electronic version of this thesis.
Another TwoStep cluster analysis was conducted on the separate sample of boys. Results
from this analysis are presnted in Table 4.4 including mean subscale scores, total YLS/CMI
50

score, and the boys’ recidivism rates within each cluster. The cluster solutions that emerged from
the sample of boys are different than that of the findings for the sample of girls. The TwoStep
cluster algorithm produced a six cluster solution for the sample of boys, again with the eight
input cluster variables. The ratio of cluster sizes (largest cluster to smallest cluster) was 1.55 with
a structure silhoutte measure of cohesision and separation of .2. The six cluster solution
coeffieicnt of .2 indicates a fair rating of model fit.
Cluster One (n = 140) comprises 16.9 percent of the sample and has an average overall
YLS/CMI score of 18.61. This cluster has the highest two-year recidivism rate at 65 percent.
Boys within the Cluster One typology have the largest peak in the offense history domain (M =
2.68) and have moderate to high risk scores across the other seven domains. Descriptively,
Cluster One looks similar to the high risk with offense history cluster from the original
delinquency unit cluster solutions (see Appendix B for full cluster descriptions).
Cluster Two (n = 172) is the second largest cluster in the solution comprising 20.7
percent of the sample of boys. Boys within this cluster have an average YLS/CMI composite
score of 11.69 and a two-year recidivism rate of 49.42 percent. Juveniles in Cluster Two have
uniformly low risk scores across the domains with slight peaks in the leaisure and recreation (M
= 1.87) subscale. Relative to the original delinquency unit clusters, Cluster Two is most similar
to the neglible risk group.
Cluster Three (n = 112) is the smallest cluster in the boys’ typology comprising 13.5
percent of the sample. Boys in Cluster Three have an average YLS/CMI score of 18.54, and the
second highest two-year recidivism rate among the clusters (59.82 percent). Boys in Cluster
Three have moderate risk scores across all of the domains with peaks in attitudes and orientation
(M = 2.85), leisure and recreation (M = 2.24), peers (M = 2.50), and education (M = 3.33).

51

Cluster Four (n = 118) comprises 14.2 percent of the sample of boys with an average compsite
YLS/CMI score of 15.77. The two-year recidivism rate for Cluster Four juveniles is 50.85
percent. Cluster Four boys are characterized a high risk score in the substance abuse domain (M
= 3.31) and moderate risk scores across all of the other domains except for offense history.
Clusters Three and Four are most similar to the enivornmental needs typology, however, split
into two separate groups: one group higher in substance abuse (Cluster Four) and the other group
with higher scores in leisure and recreation (Cluster Three).
Cluster Five (n = 114) juveniles comprise 13.7 percent of the boys and has the highest
overall average YLS/CMI score of 19.04 across all six of the clusters. While Cluster Five has the
highest mean score, it also has the lowest recidivism rate at 41.23 percent. Cluster Five youth
look most similar to the family needs cluster of the original delinquency unit typology. They are
characterized by moderate risk scores across all of the domains except for prior offense history
with peaks in the family (M = 4.23), education (M = 3.91), peers (M = 2.69), and personality and
behavior (M = 4.31) domains.
Cluster Six (n = 174) is the largest cluster of boys comprising 21 percent of the sample.
Boys within this cluster have the second lowest recidivism rate of all of the clusters at 47.13
percent and the lowest average overall YLS/CMI score at 11.63. Simliar to Cluster Two, youth in
this cluster are most similar to the negligible risk group in the original delinquency typology.
Cluster Two boys are characterized by uniformily low risk scores with slight peaks in the
education (M = 3.17) and personality and behavior (M = 2.80) domains.
Table 4.4. Boys’ Patterns of Risk and Recidivism Rates
Cluster
Cluster
Cluster
Cluster
Subscales
One
Two
Three
Four
(n=140,
(n=172,
(n=112,
(n=118,
16.9%)
20.7%)
13.5%)
14.2%)

52

Cluster
Five
(n=,114
13.7%)

Cluster
Six
(n=174,
21.0%)

Subscale
Mean

Table 4.4 (cont’d)
Offense
2.68
(.98)
Family
2.91
(1.62)
Education
2.96
(1.76)
Peers
2.64
(1.31)
Substance
1.89
(1.61)
Leisure
1.74
(1.04)
Personality
2.41
(1.68)
Attitudes
1.30
(1.36)
Total
18.61
(7.79)
Recidivism
65.00
Rate

.27
(.61)
2.44
(1.79)
1.98
(1.56)
2.13
(1.32)
.84
(1.09)
1.87
(.89)
1.55
(1.53)
.56
(.95)
11.69
(6.71)
49.42

.68 (.88)
2.42
(1.73)
3.33
(1.65)
2.50
(1.23)
1.37
(1.41)
2.24
(.88)
3.19
(1.93)
2.85
(1.49)
18.54
(8.07)
59.82

.17
(.46)
2.71
(1.85)
3.18
(1.78)
2.38
(1.14)
3.31
(1.14)
1.36
(.93)
1.88
(1.65)
.75
(.97)
15.77
(7.17)
50.85

.14
(.46)
4.23
(1.41)
3.91
(1.75)
2.69
(1.31)
.76
(1.09)
1.37
(.89)
4.31
(1.47)
1.66
(1.52)
19.04
(6.93)
41.23

.34
(.68)
1.84
(1.39)
3.17
(1.79)
1.33
(1.28)
.54
(.99)
.76
(.86)
2.80
(1.79)
.82
(1.22)
11.63
(7.51)
47.13

.71 (1.14)
2.67
(1.78)
3.01
(1.81)
2.21
(1.36)
1.37
(1.53)
1.52
(1.03)
2.60
(1.89)
1.22
(1.45)
15.36
(8.03)
52.10

Figure A.2 graphically respresents the six cluster types. The radar graph denotes peaks in
each of the standardized mean subscale scores for each of the six cluster solutions.

53

Figure A.2 Radar Graph of Subscale Peaks in Boys’ Clusters
Prior/Current
Offenses
Attitudes &
Orientation

Family &
Parenting

Cluster 1
Cluster 2

Personality &
Behavior

Education

Cluster 3
Cluster 4
Cluster 5
Cluster 6

Leisure/
Recreation

Peer Relations
Substance Abuse

54

DISCUSSION
The primary purpose of this study was to add to the existing literature on gender and risk
assessment among juvenile probations. This study examined the psychometric properties of the
YLS/CMI by examining interaction effects of risk scores and gender, the differential predictive
validity of the instrument by gender, and developing gender-specific profiles for patterns of risk
and need based on the YLS/CMI criminogenic risk domains.
Differential Predictive Validity of the YLS/CMI by Gender
The first two research questions assessed if gender was a moderating variable in the riskrecidivism relationship. Gender did not moderate the relationship between the total score on the
YLS/CMI and recidivism nor did risk level reveal significant effects at low or high risk levels.
However, gender did act as a moderator in the relationship between the moderate risk level and
recidivism. The results from the series of moderated logistic regression models on subscale
differences only detected gender differences in the family risk domain, in which girls’ risk for
reoffending was 1.181 times greater odds than that of boys’ risk for reoffending. To summarize
risk level differences, gender only moderated the risk-recidivism relationship for moderate risk
youth. Girls’ risk for recidivism within two years was half that of boys within the moderate risk
level group. All of the risk level-gender-recidivism relationships operate in the same direction
(that is, boys are more likely to reoffend at each level of risk than girls) and, while not
statistically significant (p = .069), the effect of gender as a moderator was also strong among
high risk juveniles.
The moderate risk level finding is particularly interesting given that the majority of
juveniles (57.7%) are classified as “moderate risk” based on their composite YLS/CMI score.
These findings are unique given that many previous studies on the YLS/CMI (Catchpole &

55

Gretton, 2003; Jung & Rawana, 1999; Marczyk et al., 2005) did not separate the analysis by risk
level. Researchers who have examined the predictive validity of the YLS/CMI across subgroups
(e.g., by gender, by race/ethnicity) indicated that future work may need to renorm the instrument
to develop more appropriate cut off risk level scores by subgroup (Flores et al., 2004; Onifade et
al., 2009). Thus, prior research on gender and risk may understate this effect by only examining
composite risk scores. Schmidt and colleagues (2005) found a main effect of gender on
recidivism, but did not detect any gender by risk level interactions. While Schmidt et al. (2005)
assessed gender differences risk level, the study was limited due to a small sample size (n = 104).
However, the finding from this study calls into question whether the assessment functions as
well for moderate risk girls as it does for boys. This is especially important given that most case
planning and the provision of services is critical for this group since most low risk adolescents
are diverted from the system or receive minimal court services and the smallest proportion of
juvenile offenders are classified as high risk and receive intensive services and supervision
levels. This calls into question how to develop and sustain appropriate, gender-sensitive case
management for this group of youth.
Research questions three and four asked if there were gender differences in the predictive
validity of the YLS/CMI. The predictive validity of the overall YLS score was significantly
predictive for the entire sample (AUC = .590), the sample of boys (AUC = .601), as well as the
sample of girls (AUC = .573). Risk level was significantly predictive for the overall sample and
separated sample of boys, but not for the sample of girls. This is likely due to the moderating
effect of gender among the moderate risk group as it relates to future offense. Among the sample
of boys, every subscale had a significant AUC value except for the family risk domain. Among
the sample of girls, the only subscale that produced a significant AUC was the family domain.

56

Research question five identified that while certain subscales appear to be stronger predictors of
recidivism for boys than girls, the only significant finding was the substance abuse subscale,
which has stronger predictive ability for the sample of boy. One limitation to this is that these
independent ROC analyses do not correct for the correlated nature of the subscales as predictor
variables.
Based on the findings from the ROC analysis, the YLS/CMI has adequate predictive
validity overall with AUCs above .5, and AUCs over or very close to .6 for the sample of boys,
demonstrating that the predictive power of this instrument for boys is moderate to large (Rice
and Harris, 1995). These results support previous research that the YLS/CMI does not classify
girls as well as boys (Betchel, Lowenkamp, and Latessa, 2007). This, in part, may be due to the
fact that in instrument was developed on combined samples of male and female youth and
subsequently tested to assess the predictive validity. While the findings in this study are similar
to some of previous literature, more research needs to address gender-specific risk factors in
criminogenic risk and needs assessments (Funk, 1999).
In terms of the differential predictive validity of the instrument by gender, only slight
differences were noted. When testing the magnitude of the difference between the AUCs of boys
and the AUCs of girls, the only significant difference that emerged was the substance abuse
subscale, in which substance abuse score was a significantly greater predictor of recidivism for
boys than for girls. While this shows promise at an empirical level (e.g., risk factors that predict
recidivism for both boys and girls survive statistically), this neglects the theoretical perspective
that is the basis for the argument for gender-responsive assessments as well as programming
from the feminist criminological literature. Future work in the area of gender and risk assessment
needs to account for both empirical predictors of re-offense as well as build upon this theoretical

57

framework that substantiates the gender-salient experiences of girls’ involved in the juvenile
justice system.
From a psychometric standpoint, differences across the individual subscales do not
necessarily matter given their correlation with the composite risk score, and especially since
composite score and risk level are the best predictors of recidivism for both boys and girls. These
data infer the risk side of the assessment’s utility. However, understanding subscales and
variation in scores across gender is critical from the needs and responsivity principles of the
RNR model (Andrews et al., 2011; Andrews and Bonta, 2010; Lowenkamp et al., 2006).
Correctly identifying and responding to risk factors within each domain is critical for case
management and planning purposes. While the subscales may appear to have minimal utility
from a statistical perspective in predicting recidivism, the subscales provide an empirical
template for practitioners to develop appropriate services and respond to the needs of youth on
their caseloads. However, research based on developing these risk factors for the original
YLS/CMI RNR model had limited samples of girls. This may influence the current state of
“gender-neutral” programming – that is, programming for girls is essentially identical to that of
programming for boys (Chesney-Lind, 2000). One way in which future researchers can remedy
gender disparities in “gender-neutral” assessments is to add (or remove) additional covariates
into the in order to remove any significant effects of gender. As well, more in line with the
feminist critique, new assessments with potentially more girl-specific risk and needs should be
developed on samples of girls.
Another way in which gender-neutral assessments can be examined is through a more
qualitative and contextual analysis of each item and domain. In terms of the slight differences in
the predictive validity of the instrument, issues of content validity need to be further explored.

58

Davidson and Chesney-Lind (2009) examined this with adult offenders and found that the
meanings of the items were gender-sensitive and varied given that men and women would
endorse the same item but for qualitatively different reasons. In turn this could cause illusory
correlations of the predictor variables. For example, high scores in substance abuse may be
related to coping with past traumas among girls versus related to more recreational use among
boys. If juvenile justice researchers and practitioners are aiming to be more gender-responsive,
there is a need to researchers and practitioners need to integrate gender-responsiveness across all
levels of juvenile justice processing (e.g., assessment, treatment planning, programming that is
sensitive to gender differences). In turn, this would suggest that all of these processes should be
linked. Finally, a further examination of whether these female-specific needs are vastly prevalent
in the lives of young women, but not necessarily behave as criminogenic risk factors (Blanchette
and Brown, 2006).
Overall, it could be argued that the YLS/CMI is gender neutral, based on an overall
psychometric examination using moderated logistic regression and receiver operating
characteristic area under the curve (ROC AUC) to assess the predictive validity of the
instrument. The findings from the first five research questions of the study revealed that the
YLS/CMI, a “gender-neutral” assessment with general risk factors, is predictive for both girls
and boys. However, the strength of the relationship of these factors with recidivism slightly
differs by gender. The only statistically different findings include gender moderating the
moderate risk and recidivism relationship, and the substance abuse subscale was the only
significantly different predictor of recidivism when examining gender differences across
subscales. This supports the literature that suggests that these general risk factors (Andrews and
Bonta, 2006; Andrews et al, 2011; Lowenkamp et al., 2001) can accurately predict recidivism

59

regardless of gender. However, given the limited strength of this relationship in the findings of
the present study, our understanding of juvenile risk assessment and gender can be improved
with a feminist pathways perspective (Chesney-Lind, 1998). This includes the various risk and
needs discussed in the literature such as history of abuse/trauma, interpersonal relationships (e.g.,
family, romantic partners), and mental health needs that are neglected in these general genderneutral assessments. To date, little research has examined these gender-responsive
recommendations among juveniles and adults (Salisbury et al., 2009). Clearly, future work needs
to examine these gender specific risk factors and determine if they contribute to our
understanding of risks and needs.
Gendered Risk Profiles and Cluster Types
Research question six examined the gendered risk profiles and cluster types among a
sample of juvenile probationers. Girls accounted for 25 percent of the sample and were slightly
overrepresented in the high risk with offense history cluster, while boys were slightly
overrepresented in high risk with first offense cluster. The other three delinquency clusters—
family needs, environmental needs, and negligible risk — were all similar in the distribution of
gender proportions. It is of interest that girls have slightly higher average composite scores
across each cluster type, but boys have higher recidivism rates in every delinquency unit cluster.
From this perspective, the instrument may be over-predicting the risk-recidivism relationship for
girls.
Research question seven tested whether there are unique patterns of risk and need based
on gender. First, this portion of the study found adequate level of internal consistency and good
overall reliability by subscale for the samples of boys and girls. The results of the cluster analysis
revealed distinct patterns of risk and need in the separate samples of boys and girls. The cluster

60

analysis revealed a three cluster solution of the sample of girls. These clusters were
undifferentiated from the low, moderate, and high risk levels produced by the YLS/CMI based
on both composite score as well as after a post-hoc examination of their recidivism rates across
clusters. The second cluster analysis revealed a different pattern of risk and need for the sample
of boys through a six cluster solution. Cluster One included high scores overall, the highest
recidivism rate at 65 percent, and a peak in the offense history domain. Clusters Two and Six
were both comprised of uniformly low scores, most similar to the negligible risk delinquency
unit cluster; however, Cluster Six had slight peaks in the education and personality domains.
Clusters Three and Four were predominantly moderate risk scores in which Cluster Three had
peaks in peers and education, while Cluster Four had a peak the substance abuse subscale.
Finally, Cluster Five had the lowest recidivism rate at 41.23 percent among all of the boys’
clusters, yet had the highest mean score with peaks in the family, education, peers, and
personality risk domains. Another way of viewing these findings is that the YLS, having been
developed on boys, provides a higher fidelity picture of boys’ risks and needs than for girls. This
would be consistent with an argument that there are unique risk/need domains for girls, which
are not included in this assessment.
Consistent with previous research, juveniles may have similar composite scores on the
YLS/CMI, but have unique pathways to receiving those scores (e.g., heterogeneity in the
moderate risk group) as well as rates of recidivism (Onifade et al., 2008). For example, certain
clusters, such as Cluster One (RR = 65%), have higher recidivism rates than boys classified as
high risk (RR = 61.2 %). Whereas for the sample of girls, low (27.8%), moderate (41.9%), and
high (47.4%) recidivism rates were nearly identical to the recidivism rates of girls’ Cluster One,
Cluster Two, and Cluster Three. This use of recidivism rates as an external criterion to evaluate

61

the clusters further supports the finding that the gender-specific clusters are not very different
from the low, moderate, and high risk classifications. In sum, these findings provide a more
detailed understanding of the patterns of risk and need for boys than for girls.
Onifade and colleagues (2008) discussed the need to identify subgroups within risk levels
in order to reduce the number of false negatives and false positives in risk prediction. Future
work on gender and risk assessment should also identify within risk level differences based on
gender since these differential patterns emerged in the gender-specific cluster analysis. As well,
gender-specific clusters can be tested for their increase in the incremental predictive validity of
the YLS/CMI based on either cumulative score and/or risk level (e.g., testing how well new
cluster membership in addition to risk level or raw score improves the model fit). Finally, these
findings support the need for the addition of gender-sensitive items to the pool of risk factors in
order to tease apart which factors are the best predictors for boys and girls and differential
patterns of need based on gender.
Since gender differences exist in the original delinquency unit clusters and new clusters
emerged when the samples were analyzed separately, it is important to continue to consider
analyzing samples separately to detect gender effects and to move towards developing genderspecific profiles that address the unique needs of boys and girls. Given that unique patterns of
risk and need did in fact emerge for boys, but not for girls over and above the standard low,
moderate, and high risk levels generated by raw risk scores may call to question the genderbased utility of alternative risk and need profiles developed on samples of boys or combined
samples of boys and girls (Simourd, 1994; Onifade et al., 2008). This also calls to question as to
whether these slight differences have implications for treatment decisions and the provision of
services.

62

Future research should aim to cross-validate these cluster solutions with additional
datasets with large samples of boys and girls. As well, cluster solutions can be evaluated by
additional external variables. For example, instead of using risk level or domain scores as
predictors of recidivism, utilizing cluster types as independent variables may also yield valid risk
prediction.
Study Limitations
One limitation of this study involves the slight correlations between the subscales in the
cluster analysis. Four pairs of subscales had moderate correlations (e.g., peers/leisure and
recreation, family/personality, education/personality, and personality/attitudes). When
conducting a cluster analysis, input variables should be orthogonal or have low correlations. As
well, while both the boys and girls clusters yield adequate measures of similarity and distance,
more in-depth post-hoc external evaluations of clusters is important if the analysis produced a
somewhat low silhouette measure of cohesion and separation (Kaufman & Rousseeuw, 1990).
Another limitation of this study was the use of recidivism as a binary outcome. The yes
or no outcome gives a limited picture of the extent of the reoffense (e.g., seriousness of the
crime) and time to reoffense (e.g., length of time spent out of court supervision). This may be
remedied by obtaining additional data to create a coefficient to determine the length of time
between initial offense, initial assessment, or release date from court jurisdiction to the next
offense. Survival analysis may be employed to analyze group-based differences in time to next
offense. As well, crime types may be calculated into this coefficient to determine the seriousness
and extent of the petition or charge. The performance of the instrument should be evaluated by
multiple outcome measures (Cottle et al., 2001; Schmidt and Hoge, 2005). In the case of gender
differences, utilizing more advanced outcome measures will give researchers a richer

63

understanding of this phenomenon. Finally, since recidivism was only defined by official court
data, there may be cases of unreported offenses, which would bias these findings. This may be
remedied by utilizing more self-report measures in addition to what is captured by official crime
statistics.
As well as the limitations of using recidivism as a binary outcome, since recidivism is
also a distal outcome and researchers may not see drastic changes in rates over time, other
outcomes should be considered in addition to solely measuring future offending. For example,
other more proximal outcomes (e.g., addressing mental health needs, increasing access to
community resources) for these youth may capture and provide context to their experiences
within the court system and changes based on case management and treatment. Future research
should be aimed at leveraging change at both these proximal and distal outcomes. A final
limitation is related to the inclusion of juveniles from only one court jurisdiction in a mid-sized
Midwestern city. This may suggest potential issues with generalizing findings to other
communities or geographic locations as well as across subpopulations (e.g., gender,
race/ethnicity) of juvenile justice samples.
Policy Implications and Directions for Future Research
Despite some of these limitations, the findings from the predictive validity portion of the
study are especially relevant to practitioners, policymakers, and researchers. Particularly given
the current trend to develop more gender-appropriate tools and services in juvenile justice
practice (Belknap and Holsinger, 2006; Bloom et al., 2002), practitioners should be aware of
gender differences in the classification of juvenile offenders across risk levels. While girls and
boys only have slight differences in their composite scores on the YLS/CMI, they have
significant differences in their overall recidivism rates. This would indicate that a moderate or

64

high risk adolescent girl is not as likely as a moderate or high adolescent boy to recidivate.
Researchers and practitioners should either work to develop tools that work in genderappropriate ways to classify youth based on equal risk levels and not over-predict recidivism for
one subsample. Based on this finding, researchers may work towards developing supplemental
gender-sensitive items for juvenile offenders or develop new assessments that are genderspecific and inclusive of risk factors that are both empirically and theoretically salient for female
offenders. In terms of the cluster types, since the cluster analysis produced similar solutions to
risk level for the girls’ sample, future work should address differences within risk levels.
Adequate sample sizes for each risk level are needed in order to power such a study.
While these results still support the use of the YLS/CMI in predicting recidivism among
both male and female juvenile probationers, more gender-responsive assessments may better fit
the needs principle of the RNR model for girls. In turn this would provide a basis for the
development and implementation of more gender-responsive interventions and services that are
specifically tailored to the needs of girls. While only using items from the YLS/CMI limit the
findings of this study, an instrument historically developed for males, it provided a platform for
understanding the unique patterns of risk and need for boys and girls. Future research should
examine potentially unique female factors in theory that the literature has noted. Previous
reviews of the literature have consistently concluded that, in terms of risk assessment, there are
limited findings on gender-specificity in samples of juvenile offenders (Blanchette & Brown,
2006; Reisig, Holtfreter, & Morash, 2006). Thus, while the theoretical basis for gender
specificity exists in the feminist literature, more empirical work needs to ascertain the validity of
these factors (Acoca, 1998; Belknap and Holsinger, 2006; Bloom et al., 2003; Chesney-Lind and
Sheldon, 2004; Funk, 1999).

65

In addition to developing gender-sensitive assessments that incorporate more appropriate
risk factors and needs, continuing to disaggregate by gender as well as race/ethnicity when
testing differences is critical to developing a more contextualized understanding. It is important
to consider these intersections of gender and race when examining the differential predictive
validity as well as tailoring programming and services (Onifade et al., 2009). Furthermore, in
addition to the ways in which race, class, and gender shape the lives of adolescents, social
settings (e.g., neighborhoods, schools, “the corner”) all shape the lives and trajectories of young
women (Jones, 2010). Future work that focuses on examining intersectionality and youth
involved in the juvenile justice system should also explore the impact of these settings (e.g.,
neighborhood effects) and contextualize how girls are socialized to negotiate threats of violence
and escape violence within the communities in which they live. These findings are critical for
examining how girls’ delinquent behavior is constructed because it is a residual effect of both
social policies and it contextualizes their violence because many of these girls live in violent
communities (Jones, 2010; Stevens, Morash, & Chesney-Lind, 2011).
Future research should also examine measurement invariance versus construct validity –
that is, does bias exist across subpopulations (e.g., gender, race/ethnicity, neighborhood types) or
does bias exist in the relationship between the scale and the outcome. Finally, future research
should also assess longitudinal trends by examining gender differences in scores and recidivism
rates overtime (e.g., 1, 2, 3, and 4 year trends). As well, examining reassessments and how scores
of individuals vary over time may elucidate gender-specific processes of girls’ and boys’
experiences while in the system.
Given the goal of the risk and needs assessment enterprise is to utilize the findings from
these assessments to inform treatment decisions and case management planning, there is a need

66

to be gender-responsive at all levels of the system. One way in which this can be accomplished is
through examining the processes by which assessments are utilized and linking how assessments
are used in practice. Many of the major critiques presented in the feminist literature on adult
offenders related to gender-responsive programming and transforming needs into risk can be
extrapolated to girl offenders (Hannah-Moffat; Morash, 2010). Researchers and policymakers
also need to further examine the dimensions of gender, race, and class as they impact developing
culturally appropriate assessments and services for juvenile justice involved youth.
Promising gender-specific programs that have been identified include the Girls Court in
Honolulu, Hawaii and the Female Offender Services Program in Cook County, Illinois (Pasko,
2008). Girls Court focuses on addressing the specific needs of young women while trying to
minimize residential placement. Girls involved in this program attend various services such as
life skills projects, academic and vocational training and counseling, individual and group
counseling, community service placements, HIV/STD education, and drug treatment if necessary
(Pasko, 2008). The Female Offender Services Program address a number of factors that impact
girls’ pathway to delinquency by training court officers in understanding the context of girls’
offending and creating a safe and trusting environment for services. Major areas that the program
targets include sex education, pregnancy, abuse/neglect, parenting skills, confidence and selfesteem, conflict resolution, career and educational training as well as in cultural diversity and
identity (Pasko, 2008).
While these have been highlighted as promising interviews, there is still a lack of genderresponsive programming for adolescent girls. Two main critiques of “gender-responsive”
programming addressed in the literature include (1) the ways in which gender-responsive
intervention inadvertently reinforces traditional gender stereotypes, and (2) gender-responsive

67

intervention and services may also unintentionally increase control and punishment by equating
girls that have a high level of needs also as having a higher level of risk (Hannah-Moffat, 1999,
2001, 2004; Morash, 2010).
Therefore, further examinations of the utility of clusters in programming decisions, test
and incorporate gender-specific risk factors in crimongenic models and in the development and
provision of services that encompass these gender-specific needs documented by the literature.
Additional research may also examine gender differences in the length of time spent under court
supervision, gender-differences in the provision of services, treatment decisions and case
management, as well as an evaluation of the types of programming that are available for female
juvenile offenders. As well, differences in crime type in both initial offense and recidivism
offense(s) should be examined. Larger sample sizes will be necessary in order conduct more
inferential statistics.
Finally, given the multilevel and contextual nature of juvenile offending (Onifade et al.,
2011), future research should examine neighborhood context and community-level factors as it
intersects with gender and race in order to develop a more comprehensive understanding of
juvenile probationers’ risk factors and needs. Further examining the predictive validity of the
YLS/CMI across subgroups of offenders as well as patterns of risk and need can aid in case
management of juvenile offenders, better classify youth in terms of matching programs and
treatment planning, as well as continuing to divert low risk offenders and tailor more intensive
services to high risk youth. In conclusion, additional examinations of the connection between
theoretical and empirical factors for girls’ offending will be critical in developing gendersensitive assessments and gender-responsive services.

68

APPENDICES

69

APPENDIX A

70

APPENDIX A: Youth Level of Service/Case Management Inventory (YLS/CMI) Items
1. Prior/Current Offenses: Three or More Prior Convictions
2. Prior/Current Offenses: Two or More Failures to Comply
3. Prior/Current Offenses: Prior Probation
4. Prior/Current Offenses: Prior Custody
5. Prior/Current Offenses: Three or More Current Convictions
6. Education: Low Achievement
7. Education: Problems with Teachers
8. Education: Problems with Peers
9. Education: Disruptive Classroom Behavior
10. Education: Disruptive Behavior on School Property
11. Education: Truancy
12. Leisure/Recreation: Lack of Organized Activities
13. Leisure/Recreation: Could Make Better Use of Time
14. Leisure/Recreation: No Personal Interests
15. Peer Relations: Lack of Positive Peer Acquaintances
16. Peer Relations: Lack of Positive Friends
17. Peer Relations: Some Delinquent Peer Acquaintances
18. Peer Relations: Some Delinquent Friends
19. Substance Abuse: Occasional Drug Use
20. Substance Abuse: Chronic Drug Use
21. Substance Abuse: Chronic Alcohol Use
22. Substance Abuse: Substance Abuse Interferes with Life
23. Substance Abuse: Substance Use Linked to Offense(s)
24. Family & Parenting: Inadequate Supervision
25. Family & Parenting: Difficulty in Controlling Behavior
26. Family & Parenting: Inappropriate Discipline
27. Family & Parenting: Inconsistent Parenting
28. Family & Parenting: Poor Relations (Father-Youth)
29. Family & Parenting: Poor Relations (Mother-Youth)
30. Attitudes & Orientation: Not Seeking Help
31. Attitudes & Orientation: Actively Rejecting Help
32. Attitudes & Orientation: Defies Authority
33. Attitudes & Orientation: Antisocial/Pro-criminal Attitudes
34. Attitudes & Orientation: Callous, Little Concern for Others
35. Personality & Behavior: Short Attention Span
36. Personality & Behavior: Poor Frustration Tolerance
37. Personality & Behavior: Verbally Aggressive/Verbally Intimidating
38. Personality & Behavior: Explosive Episodes
39. Personality & Behavior: Physically Aggressive
40. Personality & Behavior: Inadequate Guilt Feelings
41. Personality & Behavior: Inflated Self-Esteem
42. *Unemployment/Not Looking for Work
*Note: The variable Unemployment/Not looking for Work was omitted from the measure. This item was
not relevant to this sample due to age and had no variation.

71

APPENDIX B

72

YLS PATTERN OF NEED DESCRIPTIVE GUIDE
AND SERVICE MANUAL
January 2007

DELINQUENCY UNIT CLUSTER TYPES

73

DELINQUENCY UNIT

CLUSTER F: “Negligible Risk”

MEAN YLS SCORE: 5.8 (Low)
PERCENT OF UNIT POPULATION: 30%
ELEVATED (PEAK) YLS RISK AREAS: None
RECIDIVISM RATE (12 month): 12%
Demographic Description:
Cluster F is the largest subpopulation within the Delinquency Unit, comprising 30% of the total
unit population. Cluster F has an average age of 14.9 years. This group is nearly 63 % male,
although girls are over-represented in this group when compared with the overall Delinquency
sample. This group displays a very low rate for recidivism consistent with their very low overall
YLS scores.
Clinical Description:
Cluster F youth display low risk behavior across all YLS criminogenic domain areas. The
offense that brought them before the Court is likely to be a relatively minor charge and reflect an
isolated lapse in judgment, a family altercation, or an association with youth who were engaging
in illegal activities. This group’s risk for recidivism is extremely low and they are unlikely to
commit further offenses regardless of whether they are provided with services. Like other low
risk types, there is real danger in placing Cluster F youth on formal probation and exposing them
to the negative influences of established delinquents.
This population is not likely to appear chronically troubled or distressed. They are likely to have
adequate family structure and support, with family members espousing conventional, pro-social
values, attitudes, and outlooks. Cluster F youth typically present a wide range of interests and are
often already involved in various pro-social activities and organizations.
Service and Intervention Implications:
In the overwhelming majority of cases, youth belonging to Cluster Type F should be steered
completely away from longer term Court involvement and long-term stints of formal probation.
Given this group’s low risk profile, they will not benefit from services geared to help more
established delinquent offenders. This group is likely to have adequate resources and
organization necessary to access community programs, treatment providers, and positive
extracurricular activities. Perhaps the most effect intervention the court system could provide
would be to sanction these youth with brief stints of community service and then dismiss/divert
them from further Court involvement.

74

DELINQUENCY UNIT

CLUSTER G: “Environmental Needs”

MEAN YLS SCORE: 15.6 (Moderate)
PERCENT OF UNIT POPULATION: 22%
ELEVATED (PEAK) YLS RISK AREAS: Prior Offenses/Disposition, Leisure &
Recreation, Peer Relations, Substance
Abuse
RECIDIVISM RATE (12 month): 20%
Demographic Description:
Cluster G includes 22% of the Delinquency Unit, with a mean age of 15.3 years. This group is
nearly 75% male, and is roughly proportional to the Delinquency Unit as a whole. This group
displays a moderate rate of recidivism with approximately 20 % of its members re-offending
within a 12-month period.
Clinical Description:
Cluster G youth display higher risk behavior across four of the eight YLS domain areas,
including Prior Offense History, Leisure Activities, Substance Abuse, and Peer Association.
Their risk profile is exactly opposite that of Cluster H. Cluster G youth are typically unstructured
older youth who are avoiding school and home. They are extremely alienated from educational
and family institutions, and spend much of their time with other alienated peers engaging in
substance use and other unproductive activities. The defining characteristics of this group
include poor motivation and a lack of follow through with basic responsibilities. This group is
highly susceptible to negative, antisocial peer influence. The Cluster G group does not have
higher scores in the Education and Family domains because they are withdrawn from both school
and family life and are not long present or involved enough in either setting to be considered
disruptive.
Service and Intervention Implications:
Cluster G youth are extremely susceptible to negative peer associations and a negative
environmental influence. Their families and school systems appear to have long given up on
engaging them and have essentially “let go.” Subsequently, it is extremely difficult to motivate
the youth, family, or school to reinvest and make changes. Treatment needs may need to center
on alternative education and independent living options if they are to remain in the community.
Providing them with an activity oriented positive peer circle and independent life skills training
may be more effective than attempting to motivate them in traditional individual and family
therapy. Residential and day vocational training programs that have independent living, as their
goal may also be more effective with these youth than traditional therapies designed to improve
family adjustment.

75

DELINQUENCY UNIT

CLUSTER H: “Family Needs”

MEAN YLS SCORE: 17.2 (Moderate)
PERCENT OF UNIT POPULATION: 18%
ELEVATED (PEAK) YLS RISK AREAS: Education, Family & Parenting, Attitudes &
Orientation, Personality & Behavior
RECIDIVISM RATE (12 month): 28%
Demographic Description:
Cluster H includes 18% of the Delinquency Unit. With a mean age of 14.3 years, this group is
the youngest of the delinquency cluster types. They are nearly 75% male, which is roughly
proportional to the Delinquency Unit as a whole. The Cluster H group displays a moderate rate
for recidivism with approximately 28 % of its members re-offending within a 12-month period.
Clinical Description:
Cluster H youth display high-risk behavior across four of the eight YLS domain areas, including
Education, Family and Parenting, Attitudes, and Personality and Behavior. The defining
characteristic of this cluster group is the presence of severe family conflict and an angry,
irritable, reactive disposition by the youth toward parents and authority. This population has very
limited frustration tolerance and anger modulation skills, and is quick to act out in the face of
family conflict. Cluster H youth are described as immature, imprudent, emotionally reactive, and
explosive. They come from families with weak parenting skills, inconsistent discipline, and little
control over their children’s behavior. This group has a higher percentage of abuse and neglect in
its history, and may have a higher rate of domestic violence.
Cluster H youth tend to display low levels of risk on YLS subscales assessing Prior Offense
History, Leisure Activity, Peer Associations, and Substance Abuse. Their acting out tends to
primarily occur at home and school in response to stress and frustration, and their Court charges
often tend to reflect this.
Service and Intervention Implications:
Cluster H youth require intensive family treatment and intervention in parental discipline and
youth management practices. Treatment needs to be family specific and family systemic in
orientation and not focus exclusively on the youth’s behavior in isolation from the family
system. Rather, the youth’s behavior should be placed within the larger context of family
dysfunction, with their acting out being a symptom of family conflict and lack of consistency and
structure. Cluster H youth should not be placed into programs that involve established
delinquents or which treat substance abusers. They may require longer term programming
however, to ensure that positive family systemic changes are being sustained and that parental
discipline and behavior management remains appropriate.

76

DELINQUENCY UNIT

CLUSTER J: “High Risk With Offense History”

MEAN YLS SCORE: 28.5 (High)
PERCENT OF UNIT POPULATION: 16%
ELEVATED (PEAK) YLS RISK AREAS: Prior Offenses/Disposition, Education, Leisure
& Recreation, Peer Relations, Substance
Abuse, Family & Parenting, Attitudes &
Orientation, Personality & Behavior
RECIDIVISM RATE (12 month): 35%
Demographic Description:
Cluster J includes 16% of the Delinquency Unit, with a mean age of 15.3 years. This group is
87% male, and is over-represented by both males and African Americans. Cluster J displays a
high recidivism rate, with 35 % of its members re-offending within a 12 month period. Although
high already, this recidivism rate may still underestimate re-offense risk because many Cluster J
youth have highly structured environments due to intensive probation and placement in
residential care. This obviously diminishes their opportunities to re-offend. It is likely that
tracking these youth post probation will yield significantly higher re-offense rates.
Clinical Description:
Cluster J youth display high-risk behavior across all eight YLS domain areas, suggesting severe
problems in coping and adapting to the mainstream society and its institutions. These youth are
described as angry, provocative, and explosive, with difficulties modulating affect. Cluster J
youth are extremely alienated from home, school, and other societal institutions and may distrust
and outright reject others’ attempts to help them. These youth primarily associate with other
delinquent, antisocial youth and engage in illicit substance abuse. They lack empathy for others
and may tend to see the world as a dog-eat-dog place where self-serving behaviors are justified.
Service and Intervention Implications:
Cluster J youth require a very high degree of environmental structure where there is high
accountability and strong, immediate, and consistent reinforcement for engaging in socially
constructive and empathic behavior. Structured day treatment programs, which contain highly
individualized educational instruction, life skills, and vocational training, are likely to be
necessary to maintain these youth in the community. Some members of Cluster J may have such
great intervention needs that they will require consideration for placement in residential
treatment facilities. Youth who progress in these highly structured environments will need long
term commitments by supportive others and community agencies to provide them with a
sophisticated continuum of care once they return to the community. Without such support and
reinforcement, these youth will likely revert to old, ingrained antisocial adaptations.
Cluster J youth are typically refractory to traditional talk and insight oriented therapies. Very
concrete cognitive, behavioral, and need driven multi-systemic therapies that are experiential and
approximate real life situations will likely prove the most successful with this group.

77

DELINQUENCY UNIT

CLUSTER K: “High Risk”

MEAN YLS SCORE: 23.4 (High)
PERCENT OF UNIT POPULATION: 14%
ELEVATED (PEAK) YLS RISK AREAS: Prior Offenses/Disposition, Leisure &
Recreation, Peer Relations, Substance
Abuse, Family & Parenting, Attitudes &
Orientation, Personality & Behavior
RECIDIVISM RATE (12 month): 32%
Demographic Description:
Cluster K includes 16% of the Delinquency Unit, with a mean age of 14.9 years. This group is
74% male, and is over-represented by European American males. Cluster K members display a
high rate for recidivism, with a 32 % re-offense rate within a 12 month period. Although high
already, this recidivism rate may still underestimate re-offense risk because many Cluster K
youth have highly structured environments due to intensive probation and placement in
residential care. This obviously diminishes their opportunities to re-offend. It is suggested that
tracking these youth post probation will yield significantly higher re-offense rates.
Clinical Description:
Cluster K youth display high-risk behavior across all YLS domain areas except Prior Offense
History. They are virtually identical in clinical description and treatment needs to Cluster J.
Cluster K youth display severe problems in coping and adapting to mainstream society and its
institutions. They are described as angry, provocative, and explosive, with difficulties
modulating affect. This group is extremely alienated from home, school, and other societal
institutions and may distrust and outright reject others’ attempts to help them. Like Cluster J,
Cluster K youth primarily associate with other delinquent, antisocial youth and engage in illicit
substance abuse. They lack empathy for others and see the world as a dog-eat-dog place where
self-serving behaviors are justified.
Service and Intervention Implications:
Cluster K youth require a very high degree of environmental structure where there is high
accountability and strong, immediate, and consistent reinforcement for engaging in socially
constructive and empathic behavior. Structured day treatment programs, which contain highly
individualized educational instruction, life skills, and vocational training, will be necessary to
maintain these youth in the community. Some members of Cluster K have such great
intervention needs that they may require residential care. Youth who progress in these highly
structured environments will need long term commitments by supportive others and community
agencies to provide them with sophisticated continuum of care supports once they return to the
community. Without such support and reinforcement, these youth will likely revert to old,
ingrained antisocial adaptations to the world. Like Cluster J, Cluster K youth are typically
refractory to traditional talk and insight-oriented therapies. Very concrete cognitive, behavioral,
and need driven multi-systemic therapies that are experiential and approximate real life situations
will likely prove the most successful with this group.

78

REFERENCES

79

REFERENCES
Acoca, L. (1998). Outside/inside: The violation of American girls at home, on the streets, and in
the juvenile justice system. Crime and Delinquency, 44, 561-589.
Aguinis, H. (2004). Regression analysis for categorical moderators. New York, NY: The
Guilford Press.
American Bar Association and National Bar Association (2001). Justice by Gender: The Lack of
Appropriate Prevention, Diversion, and Treatment Alternatives for Girls in the Justice
System. Washington, DC: American Bar Association and National Bar Association.
American Correctional Association (2008). Adult and Juvenile Correctional Departments,
Institutions, Agencies, and Probation and Parole Authorities. Alexandria, VA: American
Correctional Association.
Andrews, D. A, & Bonta, J. (1995). The Level of Service Inventory-Revised (LSI-R). Toronto,
Ontario, Canada: Multi-Health Systems.
Andrews, D. A., & Bonta, J. (2010). The psychology of criminal conduct (5th ed.). New
Providence, NJ: LexisNexis/Matthew Bender.
Andrews, D. A., Bonta, J., & Hoge, R. D. (1990). Classification for effective rehabilitation:
Rediscovering psychology. Criminal Justice and Behavior, 17, 19-52.
Andrews, D. A., Guzzo, L., Raynor, P., Rowe, R. C., Rettinger, L. J., et al. (2011). Are the major
risk/need factors predictive of both female and male reoffending? A test with the eight
domains of the Level of Service/Case Management Inventory. International Journal of
Offender Therapy and Comparative Criminology, doi: 10.1177/0306624X10395716.
American Psychological Association Zero Tolerance Task Force (2008). Are zero tolerance
policies effective in the schools? An evidentiary review and recommendations. American
Psychologist, 63(9), 852–862.
Archwamety, T., & Katsiyannis, A. (1998). Factors related to recidivism among delinquent
females at a state correctional facility. Journal of Child and Family Studies, 7, 59-67.
Bailey, J. & McCloskey, L. (2005). Pathways to adolescent substance use among sexually
abused girls. Journal of Abnormal Child Psychology, 33, 39-54.
Banfield, J. D. & Raftery, A. E. (1993). Model-based Gaussian and non-Gaussian clustering.
Biometrics, 49, 803-821.

80

Bechtel, K., Lowenkamp, C. T., & Latessa, E. (2007). Assessing the risk of re-offending for
juvenile offenders using the Youth Level of Service/Case Management Inventory.
Journal of Offender Rehabilitation, 45(3/4), 85-108.
Belknap, J. and Holsinger, K. (2006). The gendered nature of risk factors for delinquency.
Feminist Criminology, 1(1), 48-71.
Benda, B. B. & Tollet, C. L. (1999). A study of recidivism of serious and persistent offenders
among adolescents. Journal of Criminal Justice, 27(2), 111-126.
Blanchette, K. (2004). Revisiting effective classification strategies for women offenders in
Canada. Feminism & Psychology, 14, 231-236.
Blanchette, K., & Brown, S. L. (2006). The assessment and treatment of women offenders: An
integrative perspective. Chichester: Wiley.
Bloom, B., Owen., B., & Covington, S. (2003). Gender-responsive strategies: Research practice
and guiding principles for women offenders. Washington, DC: U.S. Department of
Justice, National Institute of Corrections.
Bonta, J. (1996). Risk-needs assessment and treatment. In A. T. Harland (Ed.). Choosing
correctional options that work: Defining the demand and evaluating the supply (pp. 1832). Thousand Oaks, CA: Sage.
Brumbraugh, S, Hardison, J. L., & Winterfield, L. A. (2010). Suitability of Assessment
Instruments for Delinquent Girls. Bulletin. Washington, DC: U.S. Department of Justice,
Office of Justice Programs, Office of Juvenile Justice and Delinquency Prevention.
Burrell, S., & Bussiere, A. (2005). “Difficult to place”: Youth with Mental Health Needs in
California Juvenile Justice. San Francisco, CA: Youth Law Center.
Catchpole, R. E. H., & Gretton, H. M. (2003). The predictive validity of risk assessment with
violent young offenders: A one-year examination of criminal outcome. Criminal Justice
and Behavior, 30(6), 688-708.
Caulfield, L. (2010). Rethinking the assessment of female offenders. The Howard Journal, 49(4),
315-327.
Chesney-Lind, M. (1997). The female offender: Girls, women, and crime. Thousand Oaks, CA:
Sage.
Chesney-Lind, M. (2000). What to do about girls? Thinking about programs for young women.
In M. McMahon (Ed.), Assessment to assistance: Programs for women in community
corrections (pp. 139-170). Lanham, MD: American Correctional Association.
Chesney-Lind M. (1986). Women and crime: The female offender. Signs, 12, 78–96.

81

Chesney-Lind, M., & Okamoto, S. (2001). Gender matters: Patterns in girls’ delinquency and
gender responsive programming. Journal of Forensic Psychology Practice, 1, 1-28.
Chesney-Lind, M., & Shelden, R. G. (2004). Girls, delinquency and juvenile justice. (3rd ed.).
Belmont, CA: Wadsworth.
Chesney-Lind, M. (2000). What to do about girls? Thinking about programs for young women.
In M. McMahon (Ed.), Assessment to assistance: Programs for women in community
corrections (pp. 139-170). Lanham, MD: American Correctional Association.
Cohen, M. A. (1998). The monetary value of saving a high risk youth. Journal of Quantitative
Criminology 14: 5-33.
Collins, P. Hill. (1998). Fighting words: Black women and the search for justice. Minneapolis,
MN: University of Minnesota Press.
Collins, P. Hill. (1990). Black feminist thought: Knowledge, consciousness and the politics of
empowerment. New York: Routledge.
Cottle, C. C., Lee, R. L., & Heilbrun, K. (2001). The prediction of criminal recidivism in
juveniles. Criminal Justice and Behavior, 28, 367–394.
Cullen, F. T., Fisher, B. S., & Applegate, B. K. (2000). Public opinion about punishment and
corrections. In M. Tonry (Ed.), Crime and justice: A review of research (Vol. 27, pp. 179). Chicago: University of Chicago Press.
Daly, K. (1992). A woman’s pathway to felony court. Review of Law and Women’s Studies, 2,
11-52.
Davidson, J. T. & Chesney-Lind, M. (2009). Discounting women: Context matters in risk and
need assessment. Critical Criminology, 17, 221-245.
Dishion, T.J., McCord, J., & Poulin, F. (1999). When interventions harm: Peer groups and
problem behavior. American Psychologist, 54(9), 755-764.
Emeka, T. Q. & Sorensen, J. R. (2009). Female juvenile risk: Is there a need for gendered
assessment instruments? Youth Violence and Juvenile Justice, 7(4), 313-330.
Erez, E., Adelman, M., and Gregory, C. (2009). Intersections of immigration and domestic
violence: Voices of battered immigrant women. Feminist Criminology 4(1): 32-56.
Feld, B. (2009). Girls in the juvenile justice system. In The Delinquent Girl, edited by Margaret
A. Zahn. Philadelphia, PA: Temple University Press, pp. 225-264.

82

Flores, A. W., Travis, L. F., & Latessa, E. J. (2004). Case classification for juvenile corrections:
An assessment of the Youth Level of Service/Case Management Inventory (YLS/CMI),
final report. Washington, DC: National Institute of Justice.
Freeman, R. B. (1991). Crime and the Employment Disadvantage of Youth. Cambridge, MA:
National Bureau of Economic Research.
Funk, S. J. (1999). Risk assessment for juveniles on probation: A focus on gender. Criminal
Justice and Behavior, 26(1), 44-68.
Gaarder, E., & Belknap, J. (2002). Tenuous borders: Girls transferred to adult court.
Criminology, 40, 481-587.
Gavazzi, S. M., Yarcheck, C. M., & Chesney-Lind, M. (2006). Global risk indicators and the role
of gender in a juvenile detention sample. Criminal Justice and Behavior, 33(5), 597-612.
Gavazzi, S. M., Yarcheck., C. M., & Lim, J. Y. (2005). Ethnicity, gender, and global risk
indicators in the lives of status offenders coming to the attention of the juvenile court.
International Journal of Offender Therapy and Comparative Criminology, 49, 696-710.
Goodkind, S., Wallace, J. M., Shook, J. J., Bachman, J., & O’Malley, P. (2009). Are girls really
becoming more delinquent? Testing the gender convergence hypothesis by race and
ethnicity, 1976-2005. Children and Youth Services Review, 31, 885-895.
Grisso, T. (2008). Adolescent offenders with mental disorders. The Future of Children, 18(2),
143-164.
Grove, W. M., & Meehl, P. E. (1996). Comparative efficiency of informal (subjective,
impressionistic) and formal (mechanical algorithmic) prediction procedures: The clinicalstatistical controversy. Psychology, Public Policy, & Law, 2(2), 293-323.
Hanley, J. A., & McNeil, B. J. (1983). A method of comparing the areas under receiver operating
characteristic curves derived from the same cases. Radiology, 148(3), 839-843.
Hannah-Moffat, K. (2009). Gridlock or mutability: Reconsidering “gender” and risk assessment.
Criminology & Public Policy, 8(1), 209-219.
Hannah-Moffat, K. (1999). Moral agent or actuarial subject: Risk and Canadian women’s
imprisonment. Theoretical Criminology, 3, 71–94.
Hannah-Moffat, K. (2001). Punishment in disguise: Penal governance and federal imprisonment
of women in Canada. Toronto: University of Toronto Press.\
Hannah-Moffat, K. (2004). Losing ground: Gendered knowledge, parole risk, and responsibility.
Social Politics, 11, 363–385.

83

Harlow, C. W. (2003). Education and Correctional Populations. Washington, D.C.: Bureau of
Justice Statistics, Table 1. www.ojp.usdoj.gov/bjs/pub/pdf/ecp.pdf
Haynie, D. L. (2001). Delinquent peers revisited: Does network structure matter? American
Journal of Sociology, 106(4), 1013-1057.
Hewitt, L. E., & Jenkins, R. L. (1946). Fundamental patterns of maladjustment: The dynamics of
their origins. Springfield, IL: D. H. Green.
Hoge, R. D., & Andrews, D. A. (1996). Assessing the youthful offender: Issues and techniques.
New York: Plenum.
Hoge, R. D., & Andrews, D. A. (2002). Youth Level of Service/Case Management Inventory
(YLS/CMI). Toronto, Ontario, Canada: Multi-Health Systems.
Holman, B., & Ziedenberg, J. (2006). The Dangers of Detention: The Impact of Incarcerating
Youth in Detention and Other Secure Facilities. Washington, D.C.: Justice Policy
Institute.
Holsinger, K. (2000). Feminist perspectives on female offending: Examining real girls' lives.
Women & Criminal Justice, 12(1), 23-51
Holsinger, A. M., Lowenkamp, C. T., & Latessa, E. J. (2003). Ethnicity, gender, and the Level of
Service Inventory-Revised. Journal of Criminal Justice, 31, 309-320.
Holtfreter, K., & Cupp, R. (2007). Gender and risk assessment: The empirical status of the LSI-R
for women. Journal of Contemporary Criminal Justice, 23, 363-382.
Hoyt, S., & Scherer, D. G. (1998). Female juvenile delinquency: Misunderstood by the juvenile
justice system, neglected by social science. Law and Human Behavior, 22, 81-107.
Hubbard, D. J. & Pratt, T. C. (2002). A meta-analysis of the predictors of delinquency among
females. Journal of Offender Rehabilitation, 34, 1-13.
Ilacqua, G. E., Coulson, G. E., Lombardo, D., & Nutbrown, V. (1999) Predictive validity of the
Young Offender Level of Service Inventory for criminal recidivism of male and female
young offenders, Psychological Reports, 84(3/2), 1214-1218.
Javdani, S., Sadeh, N., & Verona, E. (2011). Gendered social forces: A review of the impact of
institutionalized factors on women and girls’ criminal justice trajectories, Psychology,
Public Policy, and Law, 17(2), 161-211.
Jones, N. (2010). Between good and ghetto: African American girls and inner-city violence. New
Brunswick, NJ: Rutgers Univeristy Press.

84

Jones, N. (2008). Working ‘the code’: On girls, gender, and inner-city violence. The Australian
and New Zealand Journal of Criminology, 41(1), 63-83.
Jung, S., & Rawana, E. P. (1999). Risk and need assessment of juvenile offenders. Criminal
Justice and Behavior, 26, 69-89.
Justice Policy Institute (2009). The Costs of confinement: Why Good Juvenile Justice Policies
Make Good Fiscal Sense.
Kaufman, L., & Rousseeuw, P. J. (1990). Finding groups in data: An introduction to cluster
analysis. New York: John Wiley and Sons.
Kellam, S., Ling, G., Merisca, R. Brown, C., & Ialongo, M. (1998). The effect of the level of
aggression on first grade classroom on the course of malleability of aggressive behavior
in middle school. Development and Psychopathology, 10, 165-185.
Kroneman, L., Loeber, R., & Hipwell, A. E. (2004). Is neighborhood context differently related
to externalizing problems and delinquency for girls compared to boys? Clinical Child and
Family Psychology Review, 7, 109-122.
Lipsey, M. (1992). Juvenile delinquency treatment: A meta-analytic inquiry into the variability
of effects. In T. Cook et al. (Eds.), Meta-analysis for explanation (pp. 83-127). New
York: Russell Sage.
Lipsey, M., & Wilson, D. (1998). Effective intervention for serious juvenile offenders: A
synthesis of research. In R. Loeber & D. Farrington (Eds.), Serious and violent juvenile
offenders: Risk factors and successful interventions (pp. 3130345). Thousand Oaks, CA:
Sage.
Loeber, R., & Dishion, T. (1983). Early predictors of male delinquency: A review. Psychological
Bulletin, 94(1), 68-99.
Lowenkamp, C. T., & Latessa, E. J. (2002). Evaluation of Ohio’s community-based correctional
facilities and halfway house programs. Cincinnati, OH: University of Cincinnati.
Lowenkamp, C. T., Latessa, E. J., & Holsinger, A. M. (2006). The risk principle in action: What
have we learned from 13,676 offenders and 97 correctional programs? Crime &
Delinquency, 52, 77-93.
Lowenkamp, C. T., Latessa, E. J., & Smith, P. (2006). Does correctional program quality matter?
The impact of adhering to the principles of effective intervention. Criminology & Public
Policy, 5, 575-594.
MacDonald, J. M., & Chesney-Lind, M. (2001). Gender bias and juvenile justice revisited: A
multiyear analysis. Crime & Delinquency, 47, 173-191.

85

Marczyk, G. R., Heilbrun, K., Lander, T., & Dematteo, D. (2005). Juvenile decertification:
Developing a model for classification and prediction. Criminal Justice and Behavior,
32(3), 278-301.
Mazerolle, P. (1998). Gender, general strain, and delinquency: An empirical examination. Justice
Quarterly, 15(1), 65-91.
Miller, T. R., Cohen, M. A., & Wiersma, B. (1996). Victim Costs and Consequences: A New
Look. National Institute of Justice Research Report. NCJ-155282. Washington, D.C.:
National Institute of Justice.
Moffitt, T. E. (1993). Adolescence-limited and life-course-persistent antisocial behavior: A
developmental taxonomy. Psychological Review, 100, 674-701.
Moffitt, T. E., Capsi, A., Rutter, M., & Silva, P. A. (2001). Sex Differences in Antisocial
Behavior: Conduct Disorder, Delinquency, and Violence in the Dunedin Longitudinal
Study. New York, NY: Cambridge University Press.
Morash, M. (2006). Advances in understanding gender, crime, and justice. Thousand Oaks, CA:
Sage.
Morash, M. (2009). A great debate over using the Level of Service Inventory-Revised (LSI-R)
with women offenders. Criminology & Public Policy, 8(1), 173-180.
Morash, M. (2010). A feminist critique of community programs and service. Northeastern
University Press: Boston, MA.
Morash, M. (1986). Gangs, groups, and delinquency. British Journal of Criminology, 23(4), 309335.
Mulder, E., Brand, E., Bullens, R., & van Marle, H. (2010). Toward a classification of juvenile
offenders: Subgroups of serious juvenile offenders and severity of recidivism.
International Journal of Offender Therapy and Comparative Criminology, doi:
10.1177/0306624X10387518.
National Council on Crime and Delinquency. (1998). Risk Assessment Standards.
Norusis, M. J. (2010). PASW statistics 18 statistical procedures companion. Upper Saddle River,
NJ: Prentice Hall.
Obeidallah, D., Brennan, R. T., Brooks-Gunn, J., & Earls, F. (2004). Links between pubertal
timing and neighborhood contexts: Implications for girls’ violent behavior. Journal of the
American Academy of Child and Adolescent Psychiatry, 43, 1460-1468.

86

Onifade, E., Davidson, W., & Campbell, C. (2009). Risk assessment: The predictive validity of
the Youth Level of Service/Case Management Inventory with African Americans and
girls. Journal of Ethnicity in Criminal Justice, 7, 205-221.
Onifade, E., Davidson, W., Campbell, C., Turke, G., Malinowski, J. et al. (2008). Predicting
recidivism in probationers with the Youth Level of Service Case Management Inventory
(YLS/CMI). Criminal Justice and Behavior, 35(4), 474-483.
Onifade, E., Davidson, W., Livsey, S., Turke, G., Horton, C. et al. (2008). Risk assessment:
Identifying patterns of risk in young offenders with the Youth Level of Service/Case
Management Inventory. Journal of Criminal Justice, 36, 165-173.
Onifade, E., Nyandoro, A, Davidson, W. S., & Campbell, C. (2009). Truancy and patterns of
criminogenic risk in a young offender population. Youth Violence and Juvenile Justice,
16(1), 1-19.
Onifade, E., Petersen, J., Bynum, T. & Davidson, W. (2011). Multilevel recidivism prediction:
Incorporating neighborhood socioeconomic ecology in juvenile justice risk assessment.
Criminal Justice and Behavior, 38(8), 840-853.
Pasko, L. (2010). Damaged daughters: The history of girls’ sexuality and the juvenile justice
system. Journal of Criminal Law & Criminology, 100(3), 1099-1130.
Pasko, L. (2008). The wayward girl revisited: Understanding the gendered nature of juvenile
justice and delinquency. Sociology Compass, 2(3), 821-836.
Potter, H. (2006). An argument for black feminist criminology: Understanding African American
women’s experiences with intimate partner abuse using an integrated approach, Feminist
Criminology, 1, 106-124.
Puzzanchera, C., Adams, B., Sickmund, M. (2011). Juvenile Court Statistics 2008. Pittsburgh,
PA: National Center for Juvenile Justice.
Puzzanchera, C. & Sickmond, M. (2008). Juvenile Court Statistics 2005. Pittsburgh, PA:
National Center for Juvenile Justice.
Quay, H. C. (1964). Dimensions of personality in delinquent males as inferred from the factor
analysis of case history data. Child Development, 35, 479-484.
Reisig, M. D., Holtreter, K., & Morash, M. (2006). Assessing recidivism risk across female
pathways to crime. Justice Quarterly, 23, 384-405.
Reisig, M. D., Holtfreter, K., & Morash, M. (2002). Social capital among women offenders:
Examining the distribution of social networks and resources, Journal of Contemporary
Criminal Justice, 18(2), 167-187.

87

Rice, M. E., & Harris, G. T. (1995). Violent recidivism: Assessing predictive validity. Journal of
Consulting and Clinical Psychology, 63, 737-748.
Salisbury, E. J., Van Voorhis, P., Spiropoulos, G. V. (2009). The predictive validity of a genderresponsive needs assessment: An exploratory study. Crime & Delinquency, 55(4), 550585.
Schmidt, F., Hoge, R. D., & Gomez, L. (2005). Reliability and validity analyses of the Youth
Level of Service/Case Management Inventory. Criminal Justice and Behavior, 32, 329344.
Schumacher, M., & Kurz-Gwen, A. (2000). The 8% solution: Preventing serious, repeat juvenile
crime. Thousand Oaks, CA: Sage.
Schwalbe, C. S. (2008). A meta-analysis of juvenile justice risk assessment instruments:
predictive validity by gender. Criminal Justice and Behavior, 35(11), 1367-1381.
Schwalbe, C. S. (2006). Risk assessment for juvenile justice: A meta-analysis. Law and Human
Behavior, 31, 449-462.
Sickmund, M., Sladky, T. J., & Kang, W. (2008). Census of Juveniles in Residential Placement
Databook. http://ojjdp.ncjrs.gov/ojstatbb/cjrp/asp/State_Adj.asp
Simourd, L., & Andrews, D. A. (1994). Correlates of delinquency: A look at gender differences.
Forum on Corretions Research, 6(1), 26-31.
Simourd, D. J., Hoge, R. D., Andrews, D. A., & Leschied, A. W. (1994). An empirically-based
typology of male young offenders. Canadian Journal of Criminology, 36, 447-461.
Smith, P., Cullen, F. T., & Latessa, E. J. (2009). Can 14,373 women be wrong? A meta-analysis
of the LSI-R and recidivism for female offenders. Criminology & Public Policy, 8, 183208.
Snyder, H. (2008). Juvenile Arrests 2006. Bulletin. Washington, DC: U.S. Department of Justice,
Office of Justice Programs, Office of Juvenile Justice and Delinquency Prevention.
Snyder, H. N., & Sickmund, M. (2006). Juvenile offenders and victims: 2006 national report.
Washington, DC: U.S. Department of Justice, Office of Justice Programs, Office of
Juvenile Justice and Delinquency Prevention.
Somers, C. L., & Gizzi, T. L. (2001). Predicting adolescents’ risky behaviors: The influence of
future orientation, school involvement, and school attachment. Adolescent and Family
Health, 2, 3-11.

88

Steffensmeier, D., Schwarz, J., Zhong, H., & Ackerman, J. (2005). An assessment of recent
trends in girls’ violence using diverse longitudinal sources: Is the gender gap closing?
Criminology, 43, 355-406.
Stevens, T., Morash, M., & Chesney-Lind, M. (2011). Are girls getting tougher, or are we
getting tougher on girls? Probability of arrest and juvenile court oversight in 1980 and
2000. Justice Quarterly, 28(5), 718-744.
Strom, K., J., Warner, T. D., Tichavsky, L., Zahn, M. A. (2010). Policing juveniles: Domestic
violence arrest policies, gender, and police response to child–parent violence. Crime and
Delinquency, doi: 10.1177/0011128710376293
Swartz, J., Martinovich, Z., & Goldstein, P. (2003). An analysis of the criminogenic effects of
terminating the Supplemental Security Income, impairment category for drug addiction
and alcoholism. Contemporary Drug Problems, 30, 391-424.
Taylor, K. N. and Blanchette, K. (2009). The women are not wrong: It is the approach that is
debatable. Criminology & Public Policy, 8(1), 221-229.
Teplin, L. A., Abram, K. M., McClelland, G. M., Dulcan, M. K., & Mericle, A. A. (2002).
Psychiatric disorders in youth in juvenile detention. Archives of General Psychiatry, 59,
417-426.
Ullman, S. (2004). Sexual assault victimization and suicidal behavior in women: A review of the
literature. Aggression and Violent Behavior, 9, 331-351.
Van Voorhis, P., & Presser, L. (2001). Classification of women offenders: A national assessment
of current practices. Washington, DC: U.S. Department of Justice, National Institute of
Corrections.
Wong, T. M. L., Slotboom, A. M., & Bijleveld, C. J. H. (2010). Risk factors for delinquency in
adolescent and young adult females: A European review. European Journal of
Criminology, 7(4), 266-284.
Zahn, M., Hawkins, S., Chiancone, J., & Whitworth, A. (2008). The Girls Study Group—
Charting the Way to Delinquency Prevention for Girls. Bulletin. Washington, DC: U.S.
Department of Justice, Office of Justice Programs, Office of Juvenile Justice and
Delinquency Prevention.
Zahn et al. (2010). Causes and Correlates of Girls’ Delinquency. Bulletin. Washington, DC:
U.S. Department of Justice, Office of Justice Programs, Office of Juvenile Justice and
Delinquency Prevention.
Zhang, T., Ramakrishnon, R., & Livny, M. (1996). BIRCH: Method for very large databases.
Proceedings of the ACM. Management of Data. Montreal, Canada.

89