The author(s) shown below used Federal funds provided by the U.S.
Department of Justice and prepared the following final report:

Document Title:

Crime during the Transition to Adulthood: How
Youth Fare as They Leave Out-of-Home Care

Author:

Gretchen Ruth Cusick, Mark E. Courtney, Judy
Havlicek, Nathan Hess

Document No.:

229666

Date Received:

February 2010

Award Number:

2005-IJ-CX-4031

This report has not been published by the U.S. Department of Justice.
To provide better customer service, NCJRS has made this Federallyfunded grant final report available electronically in addition to
traditional paper copies.

Opinions or points of view expressed are those
of the author(s) and do not necessarily reflect
the official position or policies of the U.S.
Department of Justice.

Crime during the Transition to
Adulthood
How Youth Fare as They Leave Out-of-Home Care

Gretchen Ruth Cusick
Mark E. Courtney
Judy Havlicek
Nathan Hess
2010

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Acknowledgments
The authors wish to thank our three state partners, the Illinois Department of Children and Family
Services, the Wisconsin Department of Health and Family Services, and the Iowa Department of
Human Services. The study would not have been possible without their cooperation and financial
support. We also want to recognize our other funders of the Midwest Study, the William T. Grant
Foundation, the National Institute of Justice, and the National Institute of Mental Health. We also
thank the University of Wisconsin Survey Center in Madison, Wisconsin, for their invaluable
contribution to this study. We thank the Illinois Criminal Justice Information Authority and the
Iowa Division of Criminal and Juvenile Justice Planning and Statistical Analysis Center for
providing criminal history data.
We also thank our colleagues Noel Bost, for his help with the literature review and conceptual
design, and Wayne Osgood, for his insightful comments on earlier drafts of this work.
This project was supported by Grant No. 2005-IJ-CX-0031/2005-IJ-CX-4031 awarded by the
National Institute of Justice, Office of Justice Programs, and US Department of Justice. Points of
view in this document are those of the author and do not necessarily represent the official position
or policies of the US Department of Justice.

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Table of Contents
Introduction ...........................................................................1
Review of Relevant Literature ..........................................2
The Impact of Maltreatment and Child Welfare
Experiences on Offending.............................................2
The Importance of Social Bonds at the Transition to
Adulthood .....................................................................7
Contributions of the Current Study.............................10
Research Questions .........................................................12
Methods ...............................................................................15
Data and Sample .............................................................15
Survey Instruments .........................................................17
Analytic Methods ............................................................18
Variables .........................................................................21
Dependent Variables ...................................................21
Independent Variables ................................................24
Missing Data ...............................................................30
Findings ...............................................................................31
Offending Patterns ..........................................................31
Sample Comparisons ..................................................33
Variation in Offending by Gender ..............................37
Offending Over Time among Foster Youth ................41
Predictors of Criminal Behavior .....................................48
Descriptive Statistics...................................................48
Results of the Poisson/Negative Binomial Models.....51
Predictors of Criminal Justice Involvement: The Risk for
Arrest...............................................................................59
Descriptive Statistics...................................................59
Results of the Hazard Model.......................................60
Limitations of the Study..................................................68
Policy and Practice Implications .........................................71
References ...........................................................................77

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

List of Figures
Figure 1. Self Reported Criminal Behavior: Midwest Study
vs. Add Health (17-18 year olds).........................................34
Figure 2. Self Reported Criminal Behavior: Midwest Study
vs. Add Health (19 year olds)..............................................35
Figure 3. Self Reported Criminal Behavior: Midwest Study
vs. Add Health (21 year olds)..............................................36
Figure 4. Cumulative Probability of First Arrest.................60
Figure 5. Conditional Probability of First Arrest.................61

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

List of Tables
Table 1. Self Reported Arrests: Midwest Study vs. Add
Health Study ........................................................................37
Table 2. Percentage Engaging in Selected Offenses (17-18
year olds) .............................................................................38
Table 3. Percentage Engaging in Selected Offenses (19 year
olds) .....................................................................................39
Table 4. Percentage Engaging in Selected Offenses (21 year
olds) .....................................................................................40
Table 5. Percentage of Foster Youth Engaging in Selected
Offenses Over Time (n=438)...............................................42
Table 6. Profiles of Latent Classes (n=438) ........................44
Table 7. Characteristics of Latent Classes (n=438) .............47
Table 8. Descriptive Statistics: Dependent Variables..........48
Table 9. Descriptive Statistics: ............................................50
Table 10. Estimates from Poisson Regression of Violent
Criminal Behavior on Covariates (n=504) ..........................52
Table 11. Estimates from Poisson Regression of Violent
Crime on Covariates with ....................................................54
Table 12. Estimates from Negative Binomial Regression of
Non-Violent Criminal Behavior ..........................................56
Table 13. Estimates from Poisson Regression of NonViolent Crime on Covariates with .......................................58
Table 14. Descriptive Statistics: Covariates in Hazard
Models Predicting Arrest (n=728)......................................59
Table 15. Estimates from Cox Proportional Hazard
Regression of Time to First Arrest on .................................62
Table 16. Estimates from Cox Proportional Hazard
Regression of Time to First Arrest on Covariates By Gender
.............................................................................................67

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Introduction
While the transition to adulthood has arguably become a more ambiguous and
complex period for all youth (Furstenberg, Rumbaut, and Settersten, 2005), it may be
particularly difficult for youth aging out of the child welfare system. Many youth placed
under the care of the child welfare system have grown up in high risk families and
neighborhoods. They enter adulthood with low educational status, little employment
experience, mental health problems, high rates of homelessness, and limited social and
financial resources (Collins, 2001). On top of these disadvantages, the transition to
adulthood may be more abrupt for foster youth than for youth more generally. Whereas
many young people move gradually toward independent adulthood, while continuing to
rely on the support and guidance of families, foster youth “aging out” 1 of care lose the
support of the child welfare system that has assumed the role of parent or guardian when
they reach a particular age of majority. Although formal independent living services
exist to help youth formerly in out-of-home 2 care move into adulthood, these youth are
often on their own when making the transition (Courtney and Hughes-Heuring, 2005).
In making the transition, youth in out-of-home care may be particularly at risk for
engagement in crime or involvement with the criminal justice system (see Barth, 1990;

1

Aging out refers to the process of being discharged from the child welfare system due to age ineligibility
rather than being discharged due to adoption or reunification with one’s family of origin.
2

We use the term “out-of-home care” to describe all youth placed under the supervision of the child
welfare system for reasons of abuse or neglect rather than delinquent behavior. This includes youth placed
in relative (kinship care) and non-relative foster homes, group homes/residential treatment facilities,
adoptive placements, and independent living arrangements.
1
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Courtney et al., 2001; Cusick and Courtney, 2007; Jones and Moses, 1984; Zimmerman,
1982). Yet not all former foster youth end up in trouble. Many follow pathways into
adulthood marked by more success than failure (Hines, Merdinger, and Wyatt, 2005).
However, very little is known about the factors associated with adaptive functioning, and
avoidance of criminal behavior or criminal justice contact in particular, of former foster
youth.
In this study, we examine criminal behavior and criminal justice system
involvement among youth making the transition from out-of-home care to independent
adulthood. We consider the importance of earlier experiences with maltreatment and
within the child welfare system on criminal behavior during the transition to adulthood.
In addition, we examine whether social bonds predict criminal behavior and the risk for
criminal justice involvement among former foster youth.
The results of this research are intended to provide a theoretically and empirically
driven basis for future efforts at reducing negative outcomes for youth formerly in out-ofhome care as they become adults and for preventing crime among a particularly high risk
population that has received limited attention by criminal justice researchers.

Review of Relevant Literature

The Impact of Maltreatment and Child Welfare Experiences on Offending
The link between childhood experiences of maltreatment and later delinquency
represents an area of growing interest among researchers and child welfare practitioners.
Although heavily debated due to methodological issues, studies generally show a positive

2
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

association between maltreatment, including physical abuse and neglect, and crime
(McCord, 1983; Widom, 1989; Smith and Thornberry, 1995; Maxfield and Widom,
1996). While such research examines the link between the childhood experiences of
maltreatment, which if detected may lead to involvement with the child welfare system,
and later crime, few studies look at experiences within the system that may influence
offending. Youth who are placed in out-of-home care are exposed to a system of services
that impact behavior. There is, however, considerable variation across individuals in the
duration, number, and type of placements they experience (Courtney and Wong, 1996;
Wulczyn, Kogan, and Harden, 2003). For example, the length of stay in out-of-home
care across all children who exited care in 2001 varied from less than 1 month (19
percent) to greater than 5 years (9 percent), with a median duration of 12 months (U.S.
Department of Health and Human Services, 2003). Of the 542,000 children in out-ofhome care in 2001, 48 percent were in non-relative foster family homes, 24 percent were
in relative family homes, 18 percent were in group homes or institutions, 4 percent were
in pre-adoptive homes, and 6 percent were in other placement types (U.S. Department of
Health and Human Services, 2003). 3
It is in relation to varying placement experiences such as these that a small
number of child welfare researchers have examined the role of youths’ placement
experiences on offending behaviors. An early study by Runyan and Gould (1985) found
that number of foster home placements was positively correlated with criminal behavior

3

These figures are based on all children in foster care in 2001, and therefore, these figures may differ from our sample
based on differences in age distribution. The median age at entry and exit of all foster youth in 2001 was 8.7 and 10.2,
respectively. We expect that placement experiences of our sample, containing only youth between the ages of 17-21,
will differ from the total population of foster children. For example, older foster children are much less likely to live in
a foster family setting than younger children (see Wertheimer, 2002).

3
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

among youth aged 11 to 18 at a bivariate level. The authors attribute this relationship,
however, to earlier behavior problems; children were removed from homes, and therefore
experienced additional placements due to their behavior problems. Widom (1991)
suggests a similar pattern. More specifically, she found that children with a lower
number of placements and for whom the first placement was longer than ten years had
the lowest rates of delinquency and adult criminality. Similarly, Jonson-Reid and Barth
(2000) found that children with multiple placements had a higher risk of incarceration
during adolescence.
By viewing offending among former foster youth through a life course
perspective (Elder, 1985), we recognize that the path one takes during the transition
adulthood is largely dependent on earlier life events. For youth leaving out-of-home
care, behavior during the transition may be partially explained by earlier experiences of
abuse or neglect as well as unstable or negative experiences within the child welfare
system. It may be that these negative experiences decrease child well being and result in
greater behavior problems among youth. It may also be that instability or placement in a
group home setting is an indicator of a difficult case; a youth already showing significant
problem behavior, a parent, caregiver, or household environment with problems, or both.
Independent Living Services
In addition to experiences with out-of-home care placements, specific services
provided by the child welfare system to assist in the transition to adulthood may be
important for later behavior. In response to the needs of youth once they leave care, the
federal government has, since 1986, provided funding for independent living programs
intended to assist in the transition to adulthood. In 1985, the Independent Living

4
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Initiative (Public Law 99-272), under title IV-E of the Social Security Act, provided
states federal funds to help youth develop the skills needed to live independently after
leaving care. In 1993, funding for the Independent Living Program (ILP) was
reauthorized (Public Law 103-66), which allowed for longer-term planning of state
programs. Services included outreach programs to attract eligible youth, training in daily
living skills, education and employment assistance, counseling, case management, and a
written transitional independent living plan. As an amendment to Title IV-E, The Foster
Care Independence Act (FCIA) of 1999 (Public Law 106-169) doubled federal funding to
$140 million per year and provides states greater flexibility in the use of these funds,
extends age restrictions, and provides funding for education and training vouchers.
Despite these legislative efforts, the number of youth who actually benefit from
independent living services as they transition to adulthood is unknown. A report by the
United States General Accounting Office (GAO) found that only about 60 percent of all
eligible youth received some type of independent living service in 1998 (GAO, 1999).
Moreover, the efficacy of independent living services is, as of yet, uncertain. Although
Collins (2001) suggests that there is empirical evidence that youth who receive
independent living services have greater success transitioning to adulthood, she cautions
that the sparse and methodologically flawed state of research precludes a definitive
statement of service usefulness. No study to date has examined how variation in receipt
of independent living services affects criminal behavior during early adulthood.
Race, Placement, and Crime

A neglected aspect in research noted above on the link between out-of-home care
experiences and delinquency or crime is racial disparity. Studies of delinquency show

5
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

that African American youth have higher rates of both official and self-reported
offending than Caucasian youth (Elliot and Ageton, 1980; Hawkins et al., 2000),
particularly for serious violent behaviors (McNulty and Bellair, 2003). In addition to
higher rates of criminal behavior, African American children are disproportionately
represented within the child welfare system (Shyne and Schroeder, 1978; Stehno, 1982;
Lu et al., 2004). A recent volume of works on the overrepresentation of African
American children in the child welfare system reports that although African American
children comprised 15 percent of the U.S. child population in 1999, they constituted 45
percent of the children in out-of-home care. Conversely, white children, who comprised
60 percent of the child population, accounted for only 36 percent of children in out-ofhome care (Derezotes, Poertner and Testa, 2004).
Of those children requiring out-of-home care, experiences within the system also
differ by race. In an extensive review of the literature, Courtney et al (1996) found that
African American children received fewer services from the child welfare system and
experienced poorer case outcomes, including longer stays in care, greater reentry rates,
less placement stability, and lower adoption rates. Moreover youth who are placed in
kinship care, an arrangement that is predominantly composed of African American youth
(Testa, 1997), typically remain separated from their parents, and linger in the child
welfare system longer than those who are not placed with kin.
Although African American youth are both more likely to engage in criminal
behaviors and to have more negative experiences within the child welfare system, few
studies to date have examined how the potential variation in specific placement
experiences by race is, or is not may, related to crime during the transition to adulthood.

6
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Given the disparities between the experiences of African Americans and other groups in
the child welfare system, it is important to consider the possibility that placement
experiences may differentially influence later criminal behavior and justice system
involvement of these groups.

The Importance of Social Bonds at the Transition to Adulthood
Drawing on the literature noted above, we recognize that experiences, both from
being abused or neglected and from involvement with the child welfare system, are
unique to youth placed in out-of-home care and variation in these experiences may
predict later offending. Yet, maltreatment histories and foster care experiences alone
likely do not explain why some former foster youth engage in criminal behavior into
adulthood while others do not. To more fully understand offending within this
population, we turn to what is known to set youth toward a successful transition to
adulthood and to avoid engagement in crime more specifically.
Although there are many pathways into adulthood (Maow, 2005; Osgood, Ruth
Eccles, Jacobs, and Barber, 2005; Shanahan, 2000) successfully navigating the transition
is largely a function of the support, guidance, and resources offered by families. More so
than in a previous generation, transitioning to adulthood now requires parental assistance
(Furstenberg, Rumbaut, and Settersten, 2005). Young people rely on their families for a
range of emotional and tangible support (Schoeni and Ross, 2005). Young people may
also receive a “patchwork of institutional support as they leave adolescence”
(Furstenberg, Rumbaut, and Settersten, 2005, p. 20). Education and employment
institutions are two common institutions. Completing high school and pursuing a college
education increase employment opportunities in adulthood. Similarly, employment
7
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

opportunities nurture the development of work related skills and set the stage for future
responsibilities and careers as adults.
Social bond theory (Hirschi, 1969) suggests that attachments to adults and
commitment to and involvement with conventional social institutions reduce crime and
delinquency by acting as agents of informal social control (Sampson and Laub, 1990).
Bonds to parents or school discourage delinquency during adolescence (Cernkovich and
Giordano, 1992; Felson and Staff, 2006; Hirshi, 1969; Sampson and Laub, 1993) while
bonds to spouses and jobs discourage offending during adulthood (Horney et al.,1995;
Laub et al., 1998; Sampson and Laub, 1993). Attachment and commitment to and
involvement with families and educational or employment institutions are in many ways
the building blocks to a successful transition to adulthood because they facilitate the
exploration and gradual movement toward independence that has increasingly come to
define the transition to adulthood. However, they also may play a role in discouraging
negative behavior more specifically.
Here we review detail the many social bonds that may be related to crime, with a
focus on the notion that establishing these bonds at the transition to adulthood may be
particularly difficult for foster youth.
Parents and Caregivers

By being placed in out-of-home care, foster youth all have had disrupted family
relationships, which could lead to weaker social bonds to families during the transition to
adulthood. In spite of this, many former foster youth report have ongoing contact with
their families of origin, including parents (Barth, 1990; Cook et al., 1991; Courtney et al.,
2001; Courtney and Dworsky, 2006; Festinger, 1983; Frost and Jurich, 1983; Harai,
1980; Jones and Moses, 1984; Zimmerman, 1992). However, compared to a general
8
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

population, foster youth are less likely to report being close to parents (Perry, 2006). A
recent study of foster youth found that only 52 percent reported being somewhat or very
close to mothers and only 28 percent reported being somewhat or very close to fathers
during the transition to adulthood (Courtney et al., 2001).
For youth placed in out-of-home care, a parent can include individuals beyond
biological parents. Foster parents, a group home worker, or even the larger child welfare
system that has legally assumed guardianship may play an important role for foster youth.
Yet, as with biological parents, not all youth feel attached to these parental figures.
Former foster youth generally report feeling satisfied with their experiences in foster care
(Chapman, Wall, and Barth, 2004; Courtney et al., 2001; Courtney and Dworsky, 2006)
and some report close relationships with foster parents or other caregivers. In this study,
we examine attachment not only to biological parents, but also this range of individuals
that play a parental role to foster youth.
Education and Employment

Education and work are also central to the lives of young people during the
transition to adulthood. Although changes in federal legislation increased the funding
provided for education and training opportunities for former foster youth making the
transition to adulthood 4 , studies suggest that many former foster youth are disconnected

4

The Education and Training Voucher (ETV) Program resulted from the passage of child welfare
legislation which provided states with greater funding and flexibility to assist youth during the transition to
adulthood. The Foster Care Independence Act of 1999 (Public Law 106-169), which established the John
H. Chafee Foster Care Independence Program, doubled the funding provided to states for concrete living
needs, such as housing, and extended that age youth can remain in foster care from 18 to 21 years (Kessler,
2004; Massinga and Pecora, 2004). The Promoting Safe and Stable Families Amendment of 2001 provided
the federal dollars for Education and Training Vouchers (ETV).

9
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

from educational and employment pursuits. Studies show that individuals formerly
placed in out-of-home care have fewer years of education and many foster youth do not
graduate from high school (Barth, 1990; Cook et al., 1991; Courtney et al., 2001;
Courtney and Dworsky, 2007; Festinger, 1983; Frost and Jurich, 1983; Jones and Moses,
1984; Zimmerman, 1992). A study by Courtney et al. (2001) found that 37 percent of
former foster youth had not completed high school or obtained a GED, and only 9 percent
had taken any college courses within 12 to 18 months of discharge from care.
Unemployment rates are higher among former foster youth than the general population of
young people (Cook et al., 1991; Goerge et al., 2001; Jones and Moses, 1984;
Zimmerman, 1982) and having serious financial trouble during the transition to adulthood
is common among this population (Dworsky, 2005).

Contributions of the Current Study
Of course, not all former foster youth have weak bonds to parents or educational
and employment institutions. Some have ongoing, close relationships with biological or
foster families, attend college, and obtain full-time, stable employment as they leave the
child welfare system and enter adulthood. The focus of this study is to examine which
factors, above and beyond maltreatment and out-of-home care placement experiences are
predictive of criminal behavior and criminal justice involvement among these young
people.
This study contributes to the knowledge base on crime among foster youth
transitioning to adulthood in a number of ways. The first concerns our study design and
data sources. We use data from largest longitudinal study of young people aging out of
foster care and transitioning to adulthood since the passage of the John Chafee Foster

10
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Care Independence Act in 1999. The use of both official and self-reported data allows us
to examine behaviors that may not come to the attention of authorities as well as those
which elicit legal action. The use of both data sources also allows us to examine
offending for the full sample, including those leaving the study over time. Although
relevant recent work (see Ryan, Testa, and Zhai, 2008) examines crime among male
foster youth only, our study includes both males and females. We feel this is another
important contribution of our study as all too often crime among females and, in
particular among female foster youth, is given less attention because it is less common,
although no less troubling, than among males.
Second, out study expands on prior research that has examined the link between
experiences with the child welfare system and outcomes. For example, previous studies
have examined the relationship between placement instability and offending (see Runyan
and Gould, 1985; Jonson-Reid and Barth, 2000; Ryan, Testa, and Zhai, 2008). Yet these
studies focused on outcomes during childhood and adolescence rather than during the
transition to adulthood, as is the focus of our study. In addition, our study is the first to
examine the link between receipt of independent living services from the child welfare
system and criminal behavior.
Third, our study contributes substantively to the growing body of research on both
social bonds and crime during the transition to adulthood more generally and social
bonds and crime among foster youth more specifically. Although studies have examined
the effect of social bonds on crime during the transition to adulthood (Piquero, Brame,
Mazerolle, and Haapanen, 2002), the focus has tended to be on job stability and marriage
in the mid- to-late 20’s and early 30’s rather than factors that may be more salient in the

11
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

very early transition years. Less is known about the influence of bonds to families,
particularly parents or caregivers, education, and early employment, on crime as one
prepares for the transition to adulthood. In addition, although recent work has examined
the relationship between social bonds and crime among foster youth (see Ryan, Testa,
and Zhai, 2008), such work is limited by focusing on adolescence rather than the
transition to adulthood and considers only bonds to foster caregivers rather than to
biological parents.

Research Questions
1) Are offending patterns during the early transition to adulthood (ages 17-22)
among youth formerly in out-of-home care different from those of the general
population?
Before examining what predicts variations in criminal behavior or criminal justice
involvement, we first identify whether offending patterns during the early transition to
adulthood among foster youth are the same or differ from those seen in the general
population. The early transition to adulthood marks an important stage in the life course
to examine offending patterns as it is during this period from the end of high school
through the early twenties that offending, on average, declines. Data show that offending
increases during adolescence, reaches a peak around age 16 (for property offenses) and
18 (for violent offenses) and then declines thereafter (Hirschi and Gottfredson, 1983;
Farrington, 1986). We examine whether foster youth display this same general pattern.

2) Do experiences in out-of-home care, including number of placements, placement
type, age at entry, and receipt of independent living services predict later criminal
12
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

behavior or criminal justice involvement during the transition to adulthood among
youth aging out of the child welfare system? Is the relationship between out-ofhome care experiences and crime moderated by race?
The limited body of prior literature on the relationship between placement
experiences and criminal behavior suggests that youth who experience more “negative”
placements are those who are more likely to offend. Similarly, it has been hypothesized,
but never examined, that youth who receive aid from the child welfare system in making
the transition to adulthood will have fewer negative outcomes. It could be that the
relationship between out-of-home care experiences and subsequent offending is spurious,
and that both are influenced by already existing problem behaviors. Those youth who
exhibit problem behaviors may be moved into and out of placements, placed in group
homes, have strained relationships with caregivers, and not receive the support of the
child welfare system. Controlling for prior delinquency as well as background individual
and family demographic factors and maltreatment history, we examine how foster care
experiences are related to criminal behavior during the transition to adulthood.
In addition, we examine whether the relationship between placement experiences
and criminal behavior varies by race. As noted earlier, African American youth have
different experiences in out-of-home care than youth of other races. Negative effects of
placement experiences may be greater among African American youth. Therefore, we
test for statistical interactions between placement experiences and race.

3) Above and beyond prior experiences with maltreatment and within the child
welfare system, do the bonds to parents or caregivers, education, or employment

13
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

that foster youth have as they approach the transition to adulthood predict later
criminal behavior or criminal justice involvement?
With this final research question, we examine whether social bonds that youth
have as they approach the transition to adulthood predict criminal behavior or criminal
justice involvement during the transition. Because our focus is on the early part of the
transition, we examine social bonds that we hypothesize may be most important at this
point in the life stage. In particular, we examine bonds to parents or caregivers, school,
work, and the child welfare system, while controlling for earlier experiences with
maltreatment and out-of-home care, background risk factors, and demographic
characteristics.

14
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Methods
Data and Sample
Data for this project come from The Midwest Study of the Adult Functioning of
Former Foster Youth, which is a longitudinal panel study that is part of a collaborative
effort of the state public child welfare agencies in Illinois, Iowa, and Wisconsin, Chapin
Hall at the University of Chicago, and the University of Washington. The larger purpose
of the Midwest Study is to gather information about services provided to selected youth
served in participating states and the adult self-sufficiency outcomes achieved by the
youth. The participating states fund and/or operate the full range of services supported
by the Chafee Foster Care Independence Program (e.g., life skills training, mentoring,
tutoring, employment services, transitional housing), but these services vary considerably
in availability between and within the states. In addition, the policy regimes differ across
the three states in ways that affect the supports available to youth who age out of care.
For example, at the time of this study, youth in Illinois were legally permitted to remain
in care until age 21, while youth in Iowa and Wisconsin remained in care until
approximately age 18.
Youth from the three states who met the following basic criteria were eligible for
inclusion in the original sample: (1) they reached the age of 17 years old while in out-ofhome care, (2) they had been in care for at least one year prior to their 17th birthday, and
3) they were placed in out-of-home care for reasons of abuse or neglect rather than
delinquency. In the states of Iowa and Wisconsin, the sample reflects the universe of

15
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

youth in out-of-home care who fit these criteria. In Illinois, a random sample of youth
were selected, who represent two-thirds of those youth in out-of-home care that fit the
sampling criteria. A group of 767 youths who met the sample selection criteria was
identified.
Survey data were collected directly from the youth in the sample over three
waves. Respondents participated in in-person interviews that were approximately ninety
minutes long. The interview schedule was made up almost entirely of fixed response
questions. During the first wave of interviews, conducted between May 2002 and March
2003, 732 interviews were completed, for a response rate of approximately 95 percent.
Youth were between the ages of 17 and 18 when they were first interviewed. The second
interview took place in March through December 2004, between respondents’19th and
20th birthdays. At Wave 2, 603 youth were located, for a follow-up response rate of 83
percent. The third wave of interviews, conducted between March 2006 and January
2007, took place as soon as possible after their 21st birthday. At Wave 3, 590 of the
original sample took part, for a response rate of 81 percent.
To supplement the survey data, we accessed official arrest 5 data from each state
for this study. Youth were matched to arrest records based on identifying information
such as name, date of birth, social security number, gender, and race. In each of the three
state databases, we included all criminal arrests that occurred between respondents’ Wave
1 interview and August 31, 2007, by which time all of the study participants were at least

5

We note that only data on arrests are used in this study; not all arrests result in a conviction.

16
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

21 years old. 6 Official arrest information was obtained for 728 of the original 732
sample. 7

Survey Instruments
A main source of information for this research comes from survey data collected
over three waves. As will be discussed further, independent variables examined in this
study come largely from the Wave 1 interview. The survey instrument at Wave 1
targeted a number of areas that prior research suggests might be related to later offending
for this population, including: family background, experiences prior to and during out-ofhome care, maltreatment history, health status, mental health status, delinquency,
substance abuse and other high risk behaviors, teen parenthood, receipt of independent
living services, and social support systems. In developing the survey, emphasis was
placed on using standardized measures that have been employed in other large-scale
studies of youth (e.g., the National Longitudinal Study of Adolescent Health). Questions
about respondents’ experiences in out-of-home care including ongoing relations with
foster care providers and the youth’s family of origin were taken from Dr. Courtney’s
earlier study in Wisconsin. Questions about independent living services were based on
the range of potential services identified by the Service Standards and Outcome Measures
Workgroup appointed by the U.S. Children’s Bureau to help create implementation
regulations for the John Chafee Foster Care Independence Program. Hence, these
questions reflect current practice in the field.

6

We chose August 31, 2007 as the end of the study period in order to maximize the amount of time post-wave 1
interview in which to measure arrests that were available in the data.

7

Four respondents did not provide permission to access administrative records.

17
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Analytic Methods
To answer our research questions, we used a combination of statistical methods,
including bivariate and multivariate approaches. To address the first research question,
we first compared the prevalence of offending across a range of criminal behaviors from
our sample at ages 17-18, when offending should be at it’s peak, through ages 19 and 21,
when offending should be declining, to those from the Add Health study 8 . We also
compared self reported arrest rates of our sample to the national sample. We also
examined changes in offending within our sample of foster youth that participated in all
three interviews and responded to questions about criminal behavior (n=438). T-tests for
differences in proportions are used to compare foster youth to the Add Health sample and
within the Midwest Study sample over time. We next identified groups of foster youth
who appear on different offending or desisting paths as they enter adulthood. Previous
research has identified categories of offenders (Moffitt, 1993; Chung et al., 2002) and
developed methods to predict different offending trajectories (see Nagin and Land, 1993;
Nagin and Trenblay, 1999; Bushway et al., 2001). Based on the notion that there are
different categories of offenders, we conducted a latent class analysis to determine
whether our sample of foster youth consist of distinct classes of offenders during the
early transition to adulthood. These descriptive analyses are helpful in providing a
picture of the overall pattern of offending during the early transition to adulthood for

8

This research uses data from Add Health, a program project designed by J. Richard Udry, Peter S. Bearman, and
Kathleen Mullan Harris, and funded by a grant P01-HD31921 from the Eunice Kennedy Shriver National Institute of
Child Health and Human Development, with cooperative funding from 17 other agencies. Special acknowledgment is
due Ronald R. Rindfuss and Barbara Entwisle for assistance in the original design. Persons interested in obtaining data
files from Add Health should contact Add Health, Carolina Population Center, 123 W. Franklin Street, Chapel Hill, NC
27516-2524 (addhealth@unc.edu). No direct support was received from grant P01-HD31921 for this analysis.

18
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

former foster youth, as well as identifying the proportion of the sample that appear to be
continuing or desisting from crime.
To address the second and third research questions, we used two multivariate
approaches, First, we estimated a series of Poisson and negative binomial regression
models to examine Wave 1 predictors of self-reported criminal behavior (violent and
non-violent) at age 21 (Wave 3). We conducted Poisson and negative binomial
regression analyses because the dependent variables for criminal behavior, described
below, had a large number of zero (no criminal behavior) values and a positive skew.
However, rather than use a logistic regression which would treat all levels of offending
the same (no criminal behavior vs. any criminal behavior), the Poisson and negative
binomial analyses allow us to model variation among “high” and “low” offenders. When
the deviance statistic for a Poisson model indicated over dispersion (when the true
variance is bigger than the mean), the negative binomial model was favored. We
controlled for background risk factors and demographic characteristics and tested for
statistical interactions between race and placement experiences. Our sample included
504 youth who responded to questions about criminal behavior during the Wave 3
interview. 9

9

Of those 590 youth interviewed at Wave 3, 86 youth refused or did not respond to the questions on
criminal behavior. Because some of the questions in the interview dealt with sensitive topics that study
participants might not have felt comfortable talking with the interviewer about, a portion was administered
using Audio Computer Aided Self Interviewing (ACASI). Study participants listened to a recording of
these questions through headphones and entered their responses into a computer. Questions about criminal
behavior were part of the ACASI. Therefore, not taking the ACASI explains the majority (n=53) of those
86 who did not respond to the criminal behavior questions. This included 35 youth who were interviewed
by telephone and 18 who refused to complete the ACASI or were unable to do so (e.g. they were unable to
read). Although being incarcerated did not necessarily preclude youth from taking the ACASI (7
incarcerated youth did not complete the ACASI), 29 of the remaining 33 youth who did complete the
ACASI but did not respond to criminal behavior questions were incarcerated during the Wave 3 interview.
Twelve of these youth had been incarcerated for more than 12 months.
19
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Second, we estimated Cox proportional hazard models, or “survival models” to
examine whether out-of-home care experiences and social bonds predict the risk for
arrest during the early transition to adulthood. Compared to the survey data analyses, in
which self reported criminal behavior is the outcome, these hazard models offer three
benefits. First, all survey respondents can be included in the models, rather than only
those who participated in the Wave 3 interview. Second, we are able to measure official
arrests for a longer period than was available with the survey data. Third, these models
allow us to examine the timing of arrest, rather than simply the occurrence of an arrest.
Such models are useful for examining time to event data, such as the time to a first arrest,
because they allow for censored observations of the experiences of individual s that do
not experience the event (Allison, 1995). The model estimated is a semi-parametric
survival model that does not require specification of a baseline hazard function.10 The
hazard can be thought of as the instantaneous risk of experiencing an arrest at time t,
given that an arrest has not already occurred prior to time In our analyses, t = 0 represents
the time at which the Wave 1 interview was conducted. The general form of the Cox
proportional hazard regression equation is

h(t i ) = h0 (t ) exp( β1 χ 1i + β 2 χ 2i + ... + β k χ ki ) ,

10

Prior research on time to arrest as a measure of re-arrest or recidivism suggest the use of a parametric estimates
which allows for the risk for arrest to be highest immediately following release and then decrease, or to be highest
immediately following release, to then decrease, and subsequently increase again. Although we are examining arrest as
an outcome, we are not examining arrest as an indicator of recidivism, as was the focus of these earlier studies.
Therefore, we have no theoretical reason to hypothesize that the hazard function follows a particular parametric
distribution. The age crime curve would suggest that the hazard function of arrests from late adolescence into early
adulthood might take a functional form that is monotonically decreasing, thus following a Weibull distribution. We
found that our data more closely follows a Weibull distribution than the log-normal distribution that has been used in
recidivism studies. Regression analyses specifying a Weibull distribution and using a Cox model, however, produced
the same results. Because our interest is in the theoretical relationship between covariates and the risk for arrest rather
than the specific form of the duration dependency, we chose to present results of the Cox model which is less sensitive
to the functional form of the model (Box-Steffensmeier and Jones, 2004).

20
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

and the hazard rate for the ith individual is
hi (t ) = h0 (t ) exp( β ′x) ,
where h0 (t ) is the baseline hazard function. Once an individual experiences an arrest,
they are removed from the risk set. We present model results for the full sample and
separately by gender.

Variables
Dependent Variables

Our main outcomes are self-reported criminal behavior and official arrest. First,
self reported criminal behavior was measured during each wave of the survey, based on
questions from the National Longitudinal Study of Adolescent Health (Add Health). For
the bivariate analyses, we examined ten items, coded as ‘yes’ if the respondent reported
engaging in the behavior during the past 12 months and ‘no’ if they had not reported the
behavior. These include a range of criminal behaviors from stealing something worth
less than $50, to vandalism, to using or threatening to use a weapon to get something.
For multivariate (negative binomial) regression analyses, we examined two selfreported criminal behavior dependent variables: 1) any violent behavior, and 2) any nonviolent behavior, 11 following the example of LaGrange and Silverman (1999) and
Haynie and Giordano (2005) who each used a similar self-reported crime outcomes.
Respondents were asked a series of questions at Wave 3 about their criminal behavior.

11

We also estimated models for an outcome of any criminal behavior, which combined violent and non-violent
behavior, and determined that the model using the full criminal behavior scale was not as informative as the models
that looked at violent and non-violent crime separately. Therefore, separate violent and non-violent scales are
presented as outcomes in this report.

21
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Each item was measured on a scale of zero to three for incidence frequency, with zero
indicating they had never engaged in the activity and 3 indicating they had engaged in the
activity more than 5 times. 12 Questionnaire items were grouped into non-violent and
violent scales; items were summed to create outcome measures for any non-violent crime
(8 items, alpha = .703) and any violent crime (4 items, alpha = .736). Non-violent crime
included deliberately damaging property, stealing something worth less than $50, stealing
something worth more than $50, going into a building or house to steal something, selling
marijuana or other drugs, buying, selling, or holding stolen property, writing a bad check,
and using someone else’s credit card, bank card, or automatic teller card without their
permission or knowledge. The violent crime scale included using or threatening to use a
weapon to get something, taking part in a group physical fight, using a weapon in a fight,
and hurting someone badly enough in a physical fight that he/she needed medical
attention. Outliers were recoded to help reduce skew. As noted above, of the 590
respondents who participated in the Wave 3 survey, only 504 responded to items about
criminal behavior. Therefore, these regression analyses were conducted on these 504
respondents with valid outcome data.
To estimate the risk for arrest, we used official criminal arrest records to measure
the dependent variable. For these analyses, our outcome of interest is the first arrest 13 , 14

12

0 = never, 1 = 1 or 2 times, 2 = 3 or 4 times, 3 = more than 5 times.

13

Forty-two respondents were arrested for minor offenses, mainly traffic offenses such as speeding or driving without
a valid license, or public order offenses, such as disorderly conduct. We did not include such arrests in this study as
these minor offenses may be systematically recorded differently across the three states.
14

Only arrests that occurred in the state in which respondents were residing during the wave 1 interview are included.
Therefore, our data do not contain arrests that occurred outside of the residing state at Wave 1. An arrest in another
state is possible, particularly if an individual moved out of state. As noted by Schmidt and Witte (1988), covariates that
correlate positively with mobility may therefore have a spurious correlation with the outcome, yet data limitations for
such analyses do not permit an easy solution and the use of arrest data from only one state in survival analyses is
22
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

during the transition from adolescence to adulthood. We assume that all of the
respondents are at-risk for arrest at the time of their first interview and estimate a hazard
model that predicts when that first arrest occurs. Hazard models utilize two dependent
variables to capture both the occurrence and timing of an event. First, the survival time
reflects the duration of time between the beginning of the observation period (i.e., the
date of the Wave 1 interview 15 ) and either the date of first arrest or the end of the
observation period (i.e., August 31, 2007). A second variable indicates whether an arrest
occurred during the observation period, or in other words, if the respondent “failed”.
This variable is coded ‘1’if the respondent was arrested and ‘0’ if he/she was not. As all
respondents were 17 or 18 years old at the Wave 1 interview, our dependent variable
captures arrests that occurred during the early transition to adulthood, between age 17 at
the Wave 1 interview to nearly 24 years at the end of the observation period 16 . Unlike
survey data analyses, which only included respondents who were interviewed at Wave 3,
these analyses included all respondents for which official arrest records were available
(n=728).

common (see recidivism studies by Bottcher and Ezell, 2005; Schmidt and Witte, 1988; Visher, Lattimore, and Linster,
1991). However, out-of-state mobility is low among this population. We found that at least 30 individuals in our
sample moved out of state at some point after their first interview, although the date is unknown. Analyses excluding
these individuals resulted in no differences across survival models, suggesting that our substantive results are not
impacted by unobserved out-of-state arrests.
15

Wave 1 interviews were conducted throughout 2002 and early 2003.

16

At the end of the observation period (8/31/2007) the mean age was 22.7, with ages ranging from 21.8 to 23.6 years.

23
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Independent Variables

In both the Poisson/negative binomial regression and proportional hazard
regression analyses, independent variables were measured from the Wave 1 survey.
Measurement of all independent variables is described below.
Social Bonds

We measured bonds to parents as the degree of closeness respondents reported
feeling toward parents, based on a 4-point ordinal scale (‘1’= not at all close; ‘2’ = not
very close; ‘3’ = somewhat close; ‘4’ = very close). Separate measures were included for
closeness to mother (biological or step-mother) and closeness to father (biological or
step-father). 17 Because young people placed in out-of-home care also have parental or
family figures in the form of their out-of-home caregiver, we also created a measure of
closeness to caregiver (non-relative foster parents, relatives in kinship care placements, or
adults working in group homes in which the respondent was placed during the Wave 1
interview), also coded on the same 4-point ordinal scale. 18
We measured bonds to educational and employment institutions through three
variables. First, school enrollment was measured as a dichotomous variable coded ‘1’ if
the respondent was enrolled in some form of school (high school, vocational school, or
college) at the time of the Wave 1 interview and ‘0’ otherwise. We measured
commitment to educational pursuits through a dichotomous variable of educational
17

We also tested the effect of closeness to grandparents, which was not significant in any models, and therefore, not
included in the final models reported here.

18

Respondents without a living mother or father or who did not know if their parent was alive did not receive questions
about closeness to biological parent. For these cases systematically missing a value, we coded the closeness to mother
or closeness to father variables as ‘not at all close’ to represent the lack of this connection. In regression models, we
included an indicator variable coded ‘1’ if the respondent did not have the specified parental figure on the item and ‘0’
otherwise. Similarly, respondents in independent living settings were coded as ‘not at all close’ to caregiver because no
parental figure or guardian is present in this type of placement.

24
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

expectations, coded ‘1’ if the respondent reported that he/she would like to graduate from
college or beyond and ‘0’ otherwise. Having bonds to employment was measured
through employment status at Wave 1, coded ‘1’ if the respondent was employed and ‘0’
otherwise. 19
In addition to the measures of social bonds noted above, we used a global
measure of perceived social support 20 , which captures support across emotional, tangible,
positive social interaction, and affectionate domains. This measure is based on the MOS
Social Support Survey (Sherbourne and Stewart, 1991). Respondents were asked to
indicate on a 5-point scale how often each type of support was available to them (i.e., ‘1’
= none of the time; ‘2’ = a little of the time; ‘3’ = some of the time; ‘4’ = most of the
time; ‘5’ = all of the time). Each of 19 items were summed and converted to a score on a
scale of 0-100 21 , with higher scores indicating greater social support. 22 The alpha
reliability score for these 19 items is .96.
In addition to attachment to parents, school, and employment, we measured
attachment to the child welfare system. We asked respondents how likely they were to
turn to the child welfare system for assistance with personal, employment, family,
housing, health, or other problems or for financial help after being discharged.

19

In analyses not shown, we examined the effect working more than 20 hours per week, as prior research has found
that high intensity employment is related to offending among adolescents. We found no relationship between this
measure and the risk for arrest.
20

Although empirical support is limited, scholars have argued that social support, both expressive and instrumental,
may reduce an individual’s likelihood of offending (Colvin, Cullen, and Vander, 2002; Cullen, 1994).

21

Documentation on this social support scale can be found in Sherbourne and Stewart, 1991 and at
http://www.rand.org/health/surveys_tools/mos/mos_socialsupport_scoring.html.
22

In addition to the global measure, we also tested the effect of each social support sub-scale (emotional, tangible,
positive social interaction, and affectionate). We found similar effects of each sub-scale on the outcome, and therefore
opted to include only the full social support scale in the final models.

25
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Responses for each of the seven types of assistance were coded on a 4-point ordinal scale
(‘1’ = very unlikely; ‘2’ = unlikely; ‘3’ = likely; ‘4’ = very likely). We created an overall
scale of the likelihood of turning to the child welfare system for assistance after discharge
as the mean of these seven items. The alpha reliability score for these seven items is .92.
Maltreatment and Out-of-Home Care Experiences

Although all the young people in the sample entered out-of-home care due to
reasons of abuse or neglect, and therefore, represent a maltreated population, there was
variation in the extent of respondents’ abuse history. Therefore, we included a measure
of the number of different acts of physical abuse the respondent reported having
experienced prior to entering care, as well as a measure of the number of different acts of
neglect experienced. 23 Prior sexual abuse or assault was also included as a dichotomous
variable coded ‘1’ if the respondent ever experienced either type of victimization and ‘0’
otherwise.
The goal of out-of-home care is to protect children and youth and assist in
positive development, yet certain placement experiences may negatively affect
individuals and impede a successful transition to adulthood, and more specifically, may

23

Physical abuse includes being thrown or pushed, hit hard with a fist or kicked or slapped really hard,
being beaten up (hit or kicked repeatedly), being attacked with a weapon, being tied up, held down, or
blindfolded so that they could not protect themselves, attempted to be choked, strangled, or smothered, and
being locked in a room or closet. Neglect includes that the respondent had a serious illness or injury or
physical disability, but his/her caretakers ignored it or failed to obtain necessary medical or remedial
treatment for it; caretakers failed to help respondent with washing and grooming so that he/she was often
dirty, had uncombed hair, or wore dirty clothes; caretakers often failed to provide regular meals for
respondent so that he/she had to go hungry or ask other people for food; respondent had to go without
things that he/she needed because family's paycheck was spent on the adult's interests; respondent was
required to do chores that were too difficult or dangerous; respondent was actually abandoned by a
caretaker; respondent’s caretakers were physically or emotionally ill to the extent that he or she was unable
to care for respondent or pay attention to respondent because of the illness; respondent missed school
because he/she had to stay home to take care of a parent, grandparent, brother or sister, or to do chores;
respondents caretakers failed to protect you from being physically harmed by someone else.

26
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

be related to arrests. For example, arrest has been associated with multiple placement
moves (Johnson-Reid and Barth, 2003; Runyan and Gould, 1985; Widom, 1989) and
group home care. To control for these different experiences within the child welfare
system, we included measures of out-of-home care experiences that reflect the stability
and type of placements. First, we measured placement instability as the total number of
out-of-home care placements the respondent had experienced to date. Second, we
measured type of out-of-home care placement as one prepares for the transition to
adulthood as the current placement type during the Wave 1 interview. 24 Dummy
variables for kinship care, group care, and independent living setting or other placements
were compared to the reference placement of traditional foster care. Type of care that
young people are placed in just before they begin the transition to adulthood may also
indicate different access to support from caregivers. We also controlled for the age, in
years, at which respondents first entered care.
The John H. Chafee Foster Care Independence Program of 1999 provides states
federal funds to help prepare young people during the transition from out-of-home care to
independent living. These funds may be used for educational services, vocational
training or employment services, budgeting and financial management services, health
education services, housing services, or services to promote their positive youth
development (Collins, 2001). We asked respondents if they received any training or
services in preparation for independent living across domains of education (8 items),
employment (11 items), financial management (7 items), housing (9 items), health

24

Although we feel current placement is the best indicator of how different placements might impact behaviors
following discharge, we also examined the effect of ever being placed in group care and ever being placed in kinship
care. Neither of the lifetime placement variables was related to the risk for arrest.

27
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

education (9 items), and youth development (3 items). We measured independent living
services (ILS) support as a count of the number of services received across the six
domains. 25 The alpha reliability score for these 47 items is .96.
Finally, because our sample consists of young people across three states that
follow different child welfare practices that could impact behavior and in which three
different reporting systems were utilized to measure arrests, we included two dummy
variables to represent respondents who were placed in care in the states of Iowa and
Wisconsin, with respondents from Illinois serving as the reference category.
Demographic and Background Control Factors

We controlled for several demographic characteristics, background risk factors,
and out-of-home care placement experiences of respondents. Gender was included as a
dichotomous variable with males coded ‘1’ and females coded ‘0’. We measured race of
respondent through a series of dummy variables, with African American and other 26
racial groups being compared to the reference category of White respondents. Although
we attempted to interview all respondents near to their 17th birthday, the logistics of
locating respondents and scheduling interviews resulted in variation in the timing of
Wave 1 interviews. To account for age and potential developmental differences in
respondents based on the timing of their first interview, we included a variable measuring
respondents’ age (in years) at their Wave 1 interview. We also included a variable
25

A small amount of data on independent living services was missing. For missing items, we imputed an individual’s
mean score within the relevant domain. If all items in a domain were missing, we imputed the sample mean.
26

Our decision to collapse all other racial groups was based on the distribution of race in the sample, with
approximately 12% of respondents reporting a race other than White or African American (.5% “Asian/Pacific
Islander”, 1.4% “American Indian/Native Alaskan”, 9.7% mixed, .5 “other”). In addition, we controlled for Hispanic
ethnicity in earlier models (youth of Hispanic ethnicity comprised 8.6% of the sample). This measure had no
relationship with the outcome and was subsequently removed from the final model presented in our results.

28
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

indicating whether the respondent was a teen parent, coded ‘1’ if he/she had children and
‘0’ if he/she did not have children at the Wave 1 interview.
We also included background risk measures that are associated with crime. First,
in negative binomial regression models where we predict self-reported criminal behavior,
we controlled for prior self-reported delinquency, which we calculated as a summed scale
based on 11 items that range from minor theft to using a weapon, each coded from zero
(never) to three (5 or more times). In the hazard models where we predicted time to first
arrest, we controlled for prior self-reported arrest. Prior arrest was coded ‘1’ if
respondents reported that they had been arrested at any point prior to the Wave 1
interview and ‘0’ otherwise. 27
We also included two measures of mental and behavioral health disorders. These
measures were gathered using the Composite International Diagnostic Interview (CIDI;
World Health Organization, 1998), which is a highly structured interview designed for
use by non-clinicians that renders both lifetime and current psychiatric diagnoses
according to definitions and criteria of the Diagnostic and Statistical Manual of Mental
Disorders, Fourth Edition (DSM-IV). We distinguished between alcohol and drug related
diagnoses and other mental health diagnoses through two measures taken from the CIDI.
First, we measured whether respondents had any lifetime alcohol or substance
dependence diagnosis. Second, we measured whether respondents had any other lifetime
mental health disorder diagnosis, including major depression, panic disorder, social
phobia, generalized anxiety disorder, or post-traumatic stress disorder.

27

We chose to use the self reported measure of prior arrest because official arrest records for juveniles are not
consistently available.

29
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Missing Data

Within each wave of data collection, respondents may have missing data for some
items. For regression analyses, respondents were excluded from analyses in which the
dependent variable was missing. We imputed the sample mean for missing values on
independent variables. 28 For scale items (e.g. independent living services or social
support measure), we first imputed the respondent’s individual mean on those missing
items before calculating the scale. For respondents missing values for all items in a scale,
we imputed the sample mean. Missing data for independent variables was minor. For
the full sample of 732 youth, no independent variable was missing on greater than 4
percent of cases.
For the analyses examining self reported offending within each of the three
waves, only those youth with valid responses to the questions about delinquent and
criminal behavior were included. Missing values were not imputed as these data were
not missing at random. Rather this occurred either because youth did not participate in a
particular interview or because youth did not participate in the ACASI portion of the
interview. Also, when examining offending over time, for example in the latent class
analysis, only those youth who participated in all three waves of data collection and had
valid data on the crime measures were included.

28

We note that the mean from the specific sample being used in each analysis was used for imputation.

30
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Findings
Offending Patterns
1) Are offending patterns during the early transition to adulthood (ages 17-22)
among youth formerly in out-of-home care different from those of the general
population?

In examining whether offending patterns during the early transition to adulthood
among youth formerly in out-of-home care differ from those in the general population,
we address three main issues: 1) how do youth in out-of-home care compare to youth
more generally in terms of self-reported offending during the early transition to
adulthood; 2) does offending decline overall between late adolescence and early
adulthood among former foster youth as has been shown in previous research on general
populations; and 3) is there variation in offending patterns during the early transition to
adulthood among former foster youth, again as has been shown in previous research on
general populations.
To address these issues, we first compared self reported offending among
respondents in the Midwest Study to same aged peers from in a national survey. We
present data from each of the three waves of the Midwest Study. During the first
interview when respondents were 17-18 years old, questions on self-reported offending
were answered by all but two youth in the sample (n=730). After attrition, the sample
during the second wave of interviews, at which point respondents were 19 years old,
consisted of 603 respondents (82 percent of participants in the first interview), of which

31
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

95 percent reported on criminal behavior (n=574). By the third wave at which point
respondents were about 21 years old, 590 respondents were interviewed, of which 85
percent reported on criminal behavior (n=504).
We compared self reported offending among foster youth to a sample of same
aged peers from the National Longitudinal Study of Youth (Add Health). Add Health is a
nationally representative study that examines the causes of health-related behaviors of
adolescents and their outcomes in young adulthood. 29 The data cited in this report were
collected from the Add Health study participants in the core sample who were 17-18
years old during the first wave of the study (n=1,938), and 19 (n=498) or 21 (n=740)
years old at the time of a third interview in order to compare to our 17-18 year olds at
Wave 1, 19 year olds at Wave 2, and 21 year olds at Wave 3, respectively. We note that
the although Add Health is a longitudinal study, in order to select youth approximately
the same age as those in our Midwest Study, the sample analyzed here does not include
the same youth over time.
The Midwest Study and the Add Health study had significantly different racial
distributions at each of the time points, with the Midwest Study being represented by a
higher proportion of African American and multiracial respondents. 30 The high
percentage of African American youth in the Midwest sample is not surprising, given that
African American children are historically and currently overrepresented in the child
welfare system (Shyne and Schroeder, 1978; Stehno, 1982; Lu et al., 2004). These
different racial distributions, however, make comparing the samples on criminal behavior
29

Several groups were over-sampled (e.g., African American youth from highly educated families or a
parent with a college degree), but only youth in the core sample were included in our analyses.
30
For example, at age 17-18, 57% of the Midwest Study was African American and 10% multiracial
compared to only 18% and 4% respectively in the Add Health Study.
32
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

problematic because African American youth tend to have higher rates of both official
and self-reported offending than Caucasian youth (Elliot and Ageton, 1980; Hawkins et
al., 2000), particularly for serious violent behaviors (McNulty and Bellair, 2003). Given
the differences in both the hypothesized crime rates and racial distributions of the two
samples, it would be more appropriate to compare youth aging out of the child welfare
system to a sample in the general population with a more similar racial distribution. In
order to account for differences in the racial distributions of the two samples, we
computed a case weight for youth in the Add Health sample by dividing the percentage of
Midwest Study youth who identified themselves as belonging to a particular race by the
percentage of Add Health youth who identified themselves as belonging to that same
race. 31 32

Sample Comparisons

Figure 1 shows the percentage of youth in each sample who reported engaging in
each of the delinquent or criminal behaviors during the last 12 months before the
interview when youth were approximately 17-18 years old. This represents offenses
committed when youth were about 16-17 years old. Behaviors ranged from minor
property offenses, such as taking something from a store worth less than $50 to serious
acts of violence, such as such using or threatening to use a weapon to get something from
someone and shooting or stabbing someone. Using independent samples t-tests to
31

For example, 57 percent of the Midwest Study youth were African American compared with
approximately 19 percent of Add Health youth. The weight assigned to each African American youth in
the Add Health sample was 57/19=3.00.

32

Although by weighting for race we attempt to make the two samples more directly comparable, we recognize that the
Add Health sample is not a formal comparison group. These two groups were sampled at different points in time, using
different methods. Therefore, we cautiously use the Add Health sample to estimate a general population of youth,
compared to a sample of foster youth.

33
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

compare the proportions, we found that a significantly higher proportion of youth in the
Midwest sample reported engaging in each offense. For most offenses, the proportion of
youth in out-of-home care engaging in each offense was at least twice that found in the
Add Health study.

Figure 1. Self Reported Criminal Behavior: Midwest Study vs. Add Health (17-18 year olds)
50%

% of sample reporting behavior in past 12 months

45%
40%
35%
30%
25%
20%
15%
10%
5%
0%
Damaged
property*

Stole
Stole
something < something >
$50*
$50*

Went into
house or
building to
steal
something*

Sold drugs*

Midwest Study

Hurt
someone
badly
enough to
need
medical
care*

Used or
threatened
to use a
weapon to
get
something*

Participated Pulled knife
in a group
or gun on
fight*
someone*

Shot or
stabbed
someone*

Add Health

We found fewer differences in offending between the youth in out-of-home care
and their Add Health peers as reported at age 19 than found two years earlier. As shown
in Figure 2, offending at about age 18 was lower in general for both foster youth and
youth in general than offending at ages 16-17. Yet, a significantly higher proportion of
youth in out-of-home care reported damaging property, stealing something worth more
34
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

than $50, taking part in a group fight, and pulling a knife or gun on someone. Although
the sample sizes are smaller during this second period, and thus, the power to detect
variation across samples is reduced, it does not appear that this explains the lack of many
statistically significant differences across the groups. Unlike the comparisons made
during late adolescence, few offenses differed by more than two to three percent across
the two groups. We note, however, that youth in out-of-home care are more likely to
engage in some violent offenses, with nearly a quarter participating in a group fight and
six percent having pulled a knife or gun on someone.

Figure 2. Self Reported Criminal Behavior: Midwest Study vs. Add Health (19 year olds)
50%

% of sample reporting behavior in past 12 months

45%
40%
35%
30%
25%
20%
15%
10%
5%
0%
Damaged
property*

Stole
Stole
something < something >
$50
$50*

Went into
house or
building to
steal
something

Sold drugs

Used or
Hurt
threatened
someone
to use a
badly
weapon to
enough to
get
need
medical care something*

Midwest Study

Participated Pulled knife
in a group
or gun on
fight*
someone*

Shot or
stabbed
someone

Add Health

35
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

By Wave 3, when respondents were 21-22 years old, we found even fewer
differences between foster youth and their peers more generally. The only significant
differences between the two samples were for damaging property, going into a house to
steal something, and pulling a knife or gun on someone. Respondents from the Midwest
Study were more likely than their peers in Add Health to report engaging in these three
criminal behaviors.
Figure 3. Self Reported Criminal Behavior: Midwest Study vs. Add Health (21 year olds)
50%

% of sample reporting behavior in past 12 months

45%
40%
35%
30%
25%
20%
15%
10%
5%
0%
Damaged
property*

Stole
Stole
something < something >
$50
$50

Went into
house or
building to
steal
something*

Sold drugs

Used or
Hurt
threatened
someone
to use a
badly
weapon to
enough to
get
need
medical care something

Midwest Study

Participated Pulled knife
in a group
or gun on
fight
someone*

Shot or
stabbed
someone

Add Health

Large differences across the two groups are seen when comparing self reported
arrests by age 19 and since age 18 for those interviewed at age 19. 33 As shown in Table

33

A comparable figure of self report arrest between the Midwest Study and Add Health Study was not available at age
21.

36
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

1, a significantly higher proportion of youth in out-of-home care reported having ever
been arrested by age 19 and having been arrested since age 18 when compared to youth
in general. This is true for both males and females. For arrests since age 18, we found
even larger differences between the Midwest and Add Health studies. Even more
striking, the percentage of female foster youth ever experiencing an arrest was not only
higher than females in the Add Health sample, but also higher than that for Add Health
males. The higher levels of self-reported arrest among foster youth could reflect the
tendency for youth in out-of-home care to engage in violent behaviors at higher rates than
youth in general, as shown above, though the differences between the two groups in selfreported arrest are much higher that the differences in self-reported criminal behavior.
Alternatively, higher arrest rates could be a result of greater scrutiny being applied to
foster youth by child welfare authorities and the police.
Table 1. Self Reported Arrests: Midwest Study vs. Add Health Study

Males
Midwest
Add Health
Study
Study
Ever arrested by age 19
Arrested since age 18

57.0%
35.9%

20.1%
1.9%

Females
Midwest
Add Health
Study
Study
34.4%
17.9%

2.8%
1.1%

Variation in Offending by Gender

We next compared offending in the Midwest and Add Health samples by gender
to determine if the differences noted above were true for both males and females
separately. As shown in Table 2, we found a significantly higher proportion of female
foster youth reported all offenses when compared to female youth in general at ages 1718. A higher proportion of male foster youth also reported most offenses. Exceptions
included pulling a knife or gun on someone and shooting or stabbing someone, for which
the two samples did not significantly differ.
37
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Table 2. Percentage Engaging in Selected Offenses (17-18 year olds)

Damaged property
Stole something < $50
Stole something > $50
Went into house or building
to steal something
Sold drugs
Hurt someone badly enough
to need medical care
Used or threatened to use a
weapon to get something
Participated in a group fight
Pulled knife or gun on
someone
Shot or stabbed someone

Midwest
Study
15.7
31.9
5.0

Females
Add
Health
8.0
10.9
3.0

*
*

Midwest
Study
26.6
42.7
14.9

Males
Add
Health
17.8
20.7
6.5

*
*
*

5.8
14.1

2.4
4.7

*
*

16.9
27.9

6.5
14.7

*
*

27.1

8.6

*
*

50.0

29.7

*

5.9
28.5

3.2
12.1

*

14.7
42.4

6.9
28.6

*
*

9.0
4.0

3.1
.1

*
*

16.7
7.1

13.9
5.6

*p<.05; one-tailed test

When comparing offending as reported at age 19 across the two samples by
gender (see Table 3), we found fewer differences among males, while female foster youth
continued to report more offending. A significantly higher proportion of female foster
youth reported all offenses except stealing something worth less than $50, selling drugs,
and shooting or stabbing someone. On the other hand, the only significant difference
between male foster youth and male youth in general was for participating in a group
fight.

38
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Table 3. Percentage Engaging in Selected Offenses (19 year olds)

Damaged property
Stole something < $50
Stole something > $50
Went into house or building
to steal something
Sold drugs
Hurt someone badly enough
to need medical care
Used or threatened to use a
weapon to get something
Participated in a group fight
Pulled knife or gun on
someone
Shot or stabbed someone

Midwest
Study
17.6
6.9
4.7

Females
Add
Health
4.2
4.6
1.0

*
*
*

Midwest
Study
25.8
15.2
10.2

Males
Add
Health
20.7
14.5
8.3

1.9
6.3

.3
6.2

*
*

6.3
21.1

9.5
16.3

8.5

2.9

*

23.1

22.6

3.8
14.8

.5
4.5

*
*

5.9
34.4

6.2
19.6

4.4
1.0

.2
.0

*

8.2
3.1

5.8
4.1

*

*p<.05; one-tailed test

The gender differences at age 21 reflect those seen in the full sample (see Table
4). By age 21, even the female differences in self-reported offending declined, with only
one statistically significant difference. A higher percentage of female foster youth
reported pulling a knife or gun on someone. Similarly, there was only one difference
between males, with a higher percentage of male foster youth reporting going into a
house or building to steal something. In both of these cases, however, percentages in
each sample were low.

39
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Table 4. Percentage Engaging in Selected Offenses (21 year olds)

Damaged property
Stole something < $50
Stole something > $50
Went into house or building
to steal something
Sold drugs
Hurt someone badly enough
to need medical care
Used or threatened to use a
weapon to get something
Participated in a group fight
Pulled knife or gun on
someone
Shot or stabbed someone

Midwest
Study
8.8
3.5
2.8

Females
Add
Health
6.4
3.8
2.8

Midwest
Study
17.2
11.3
9.0

Males
Add
Health
12.3
10.7
6.0

1.1
4.9

.9
4.0

3.6
14.5

1.9
13.5

3.5

3.8

15.8

15.4

.4
6.0

.2
5.2

3.2
20.8

1.6
22.0

4.2
1.1

1.4
.7

5.9
.9

2.8
.3

*

*

*p<.05; one-tailed test

In comparing the samples over time, we note that the Add Health sample consists
of different individuals at each time point. The Midwest Study sample, however, consists
of individuals from the same base sample. Thus, it is possible that the Midwest Study
loses individuals over the three waves that are most at risk for offending. If this is the
case, offending at ages 19 and 21 among the Midwest Study may be underestimated. In
turn, differences to Add Health at these ages could be masked here.
To better gauge whether attrition in the Midwest Study results in underestimates
of self-reporting offending, we compared offending at ages 17-18 (Wave 1) between
those retained in the study and those not subsequently interviewed. Those who left the
study after the first wave of interviews were not significantly more or less likely to report
engaging in delinquent or criminal behaviors at Wave 1 than those who were
subsequently interviewed at Waves 2 and/or 3. We also compared official arrests rates of
those who were interviewed at Wave 3 to those who had left the study. Of those

40
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

interviewed at Wave 3, 46 percent had an arrest recorded compared to 47 percent of those
not interviewed (the difference is not statistically significant). These findings suggest
that the study did not lose those youth most likely to offend. We also found that 43
percent of those youth interviewed at all three waves had an official arrest. Again, this
suggests study attrition did not result in the selective loss of those displaying the worst
behavior.

Offending Over Time among Foster Youth

The analyses presented above suggest that like their peers more generally,
offending among foster youth tends to decline during the early transition to adulthood, at
least among those youth for whom offending information was available. Offending is
likely to be high among foster youth during late adolescence because, in general, it is
during this period that offending increases (Hirschi and Gottfredson, 1983; Farrington,
1986). Yet, little is known about whether young people aging out of the child welfare
system desist from offending as they transition to adulthood. The Midwest Study allows
us to examine changes in offending behavior among the same sample of former foster.
To do so, we compared self-reporting offending among respondents who completed all
three interviews. First we compared offending of respondents at age 17-18 to offending
at age 19, and then between offending of respondents at age 19 to offending at age 21.
Paired samples t-tests revealed that offending generally decreased from one time point to
the next (see Table 5). Exceptions included damaging property and stealing something
worth more than $50, in which we found no significant declines between age 17-18 and
19. Damaging property, selling drugs, hurting someone badly enough to need medical

41
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

care, using or threatening to use a weapon, and participating in a group fight all
significantly declined from age 19 to age 21.

Table 5. Percentage of Foster Youth Engaging in Selected Offenses Over Time (n=438)

Damaged property
Stole something < $50
Stole something > $50
Went into house or building
to steal something
Sold drugs
Hurt someone badly enough
to need medical care
Used or threatened to use a
weapon to get something
Participated in a group fight
Pulled knife or gun on
someone
Shot or stabbed someone

17-18 Year Olds
19.9
35.4
8.9

19 Year Olds
18.6
9.6***
6.8

21 Year Olds
11.9***
7.1
5.5

8.7
18.0

3.6***
11.9**

3.2
7.5*

34.7

15.3***

8.5***

7.8
34.3

3.7**
21.5***

1.6*
12.1***

11.9
4.3

5.7***
2.1*

4.6
1.0

Note: *p<.05, **p<.01, ***p<.001; one-tailed test (19 year olds vs. 17-18 year olds; 21 year olds vs. 19
year olds)

These findings suggest that patterns of offending among youth aging out of care
are similar to those seen in the general population, with offending peaking during late
adolescence and appearing to decline into adulthood. They do not, however, highlight
any possible variation in offending trends within our sample of foster youth. In other
words, while overall offending among our sample declines during the early transition to
adulthood, some youth may continue offending while others may even begin offending
after adolescence.
In order to further identify variations in offending patterns over time among our
sample of foster youth, we conducted a latent class analyses for the Midwest Study
respondents participating in all three interviews and who reported on criminal behavior
(n=438). Latent class analysis, or LCA, is a method that presumes observed variables are
42
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

indicators of a latent categorical variable that represents a mixture of distinct
subpopulations within the data (McCutcheon 1987). The result is an empirical
classification of individuals who share a common profile. In this case, we classify
individuals who share common patterns of offending across the three waves of data. The
analyses were conducted using Latent GOLD software (Vermunt and Magidson 2005).
In this LCA, we included six indicators of self-reported offending that represent
any violent offending and any non-violent offending at each wave. The analysis revealed
a five class model best fit the data. For reporting the profiles of the classes on the
indicators used in the analysis and for validating class distinctions across additional
characteristics of interest, cases were assigned to membership in the single class for
which they have the highest, or modal, probability. Profiles of the latent classes are
presented in Table 6. The first row shows the relative proportion of the sample in each
class. The remaining rows present the probabilities on the indicators used in the latent
class analysis.

43
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Table 6. Profiles of Latent Classes (n=438)

Class Size
Indicators
Any violent offense - w1
No
Yes
Any non-violent offense – w1
No
Yes
Any violent offense – w2
No
Yes
Any non-violent offense – w2
No
Yes
Any violent offense – w3
No
Yes
Any non-violent offense – w3
No
Yes

Rare or
NonOffenders
34%

Adolescent
Offenders
28%

Desisting
Offenders
19%

Chronic
Offenders
11%

Chronic
NonViolent
Offenders
8%

0.7686
0.2314

0.3998
0.6002

0.1504
0.8496

0.0852
0.9148

0.7328
0.2672

0.8891
0.1109

0.2593
0.7407

0.4392
0.5608

0.1422
0.8578

0.3739
0.6261

0.9898
0.0102

0.9927
0.0073

0.0166
0.9834

0.1659
0.8341

0.9935
0.0065

0.9709
0.0291

0.7169
0.2831

0.6212
0.3788

0.2351
0.7649

0.0812
0.9188

0.9212
0.0788

0.9982
0.0018

0.6725
0.3275

0.2954
0.7046

0.7496
0.2504

0.9212
0.0788

0.8864
0.1136

0.9286
0.0714

0.222
0.778

0.3518
0.6482

“Rare or Non-Offenders” - The largest class, representing approximately 34
percent of the sample, reported consistently low offending experiences across the three
waves when compared to respondents in other classes. For example, respondents in this
first class had only a .23 probability of reporting a violent offense and a .11 probability of
reporting a non-violent offense at Wave 1. Over the next two waves, the probability of
reporting either violent or non violent offenses dropped to below .10. Therefore,
although some respondents in this class did report offending across each wave, the
general pattern of this class, particularly compared to other classes, was to be nonoffenders.

44
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

“Adolescent Offenders” – The second largest class, with about 28 percent of the
sample, is distinguished from the other classes by relatively high levels of delinquent or
criminal behaviors at age 17-18, but lower levels over time. By age 19 and 21, the
probability of a violent offense dropped to almost zero. Non-violent offenses also
become much less likely than seen during adolescence. Thus, this class of Adolescent
Offenders appears similar to what Moffit (1993) referred to as “adolescent limited”,
meaning that offending is limited to what may be thought of as normative adolescent
behavior that decreases as youth enter adulthood.
“Desisting Offenders” - The third largest class, with about 19 percent of the
sample, is noteworthy for reporting delinquent and criminal behaviors at ages 17-18 and
age 19, but reporting a decrease in these behaviors by age 21. Whereas the Adolescent
Offenders limited their offending to adolescence, this group of Desisting Offenders
continued offending into late adolescence/early adulthood at age 19, but appears to
change, or desist from, their behavior as they enter their 20’s.
“Chronic Offenders” – The fourth class includes only 11 percent of the sample.
This class is distinguished from other classes by having the highest reported probability
of violent and non-violent behaviors over time. This group, although small, was most
likely to report engaging in both types of behavior at all three waves of the study. For
example, although decreasing slightly over time, the probability of reporting any violent
or non-violent offense remained above .70 at each of the three waves. Therefore, this
class appears to describe chronic offenders.
“Chronic Non-Violent” - The smallest class includes only about 8 percent of the
sample. Like the Chronic Offenders, respondents in this last class had a high probability

45
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

of engaging in non-violent offenses at each of the three waves. However, unlike the
Chronic Offenders and more similar to Non-Offenders, respondents in this class had a
relatively low probability of engaging in violent offenses at any time point. Therefore, it
appears that this class consists of chronic non-violent offenders.
To further describe the latent classes found in the LCA, we examined
demographic characteristics and background factors, foster care experiences, and social
bond factors for each class. As shown in table 7, Non-Offenders and Adolescent
Offenders were more likely than other classes to include females than males. With
respect to race, higher proportions of African American youth are found among NonOffenders, Desisting Offenders, and Chronic Offenders than in the other classes.
Respondents who reported experiencing childhood neglect were disproportionately
represented among Chronic Offenders. Adolescent Offenders were more likely to
include respondents who entered foster care as adolescents while Non-Offenders were
more likely to include respondents who entered foster care at younger ages. Desisting
Offenders were more likely to include respondents whose last placement was a relative
home while Chronic Offenders were overrepresented by respondents who reported last
placement was in a group home. Chronic Offenders were also overrepresented by
respondents who reported experiencing high levels of placement instability in foster care.
Non-Offenders were more likely to report being enrolled in school at ages 19 and 21 than
respondents in the other classes. Finally, Chronic Offenders were overrepresented by
respondents who reported lower social support. We found no statistically significant
difference in the state (Illinois, Iowa, or Wisconsin) in which foster youth were placed
across the latent classes.

46
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Table 7. Characteristics of Latent Classes (n=438)

Demographics
Female
Race:
White
Black
Other
State:
Wisconsin
Illinois
Iowa
Physical abuse
Neglect
Age at entry: 12 and older
Foster Care Experiences
Last placement type:
Kinship care
Foster care
Group care
Independent living/Other
5 or more placements
At-risk behaviors
Prior arrest
Substance abuse diagnosis
Mental health diagnosis
Teen parenthood
Social Bonds
Social support (MOS)
Close to maternal caregiver
Close to paternal caregiver
Close to foster caregiver
Wave One:
In school
Employed
Wave Two:
In school
Employed
Wave Three:
In school
Employed

Class 1
Rare or
Non
Offenders
(n=155)

Class 2

Class 3

Class 4

Adolescent
Offenders
(n=124)

Desisting
Offenders
(n=89)

Chronic
Offenders
(n=41)

%

%

%

70.3

66.9

32.3
56.8
11.0

Χ2

%

Class 5
Chronic
NonViolent
Offenders
(n=29)
%

Total

(n=438)
%

44.9

26.8

31.0

57.5

44.4
41.9
13.7

16.9
62.9
20.2

31.7
53.7
14.6

48.3
48.3
3.4

33.6
53.0
13.5

28.4
64.5
7.1
31.0
53.5
51.0

33.1
54.0
12.9
39.5
62.1
66.1

31.5
64.0
4.5
27.0
49.4
48.3

29.3
65.9
4.9
48.8
73.2
53.7

24.1
58.6
12.2
44.8
65.5
31.0

30.1
61.2
8.7
35.2
57.8
57.3

39.4
41.9
5.2
13.5
28.4

23.4
37.9
18.5
20.2
54.0

42.7
33.7
14.6
9.0
38.2

17.1
26.8
34.1
22.0
58.5

27.6
34.5
27.6
10.3
51.7

32.6
37.2
15.1
15.1
42.0

25.4***

29.0
9.0
21.9
19.4

54.8
28.2
30.6
9.7

56.2
18.0
22.5
12.4

70.7
39.0
17.1
17.2

62.1
24.1
24.1
14.2

47.9
20.1
24.2
14.2

37.8***
26.6
4.5
11.4

58.7
59.4
43.9
78.7

41.1
62.9
45.2
69.4

48.3
68.5
44.9
82.0

31.7
63.4
36.6
68.3

48.3
58.6
27.6
79.3

48.4
62.6
42.7
75.8

13.8**
2.2
3.9
6.9

85.8
40.6

82.3
42.7

79.8
34.8

78.0
29.3

79.3
32.9

82.4
38.8

2.4
3.2

56.8
47.1

37.9
45.2

39.3
36.0

43.9
36.6

41.4
44.8

45.7
43.2

12.4 *
3.8

32.2
59.4

23.4
56.9

16.9
58.4

12.2
35.0

13.8
69.0

23.5
56.9

13.2**
10.0*

44.8***
24.7**

10.5

9.4*
9.3*
15.2**
44.2***

*p<.05, **p<.01, ***p<.001

47
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Predictors of Criminal Behavior
Descriptive Statistics

Descriptive statistics for the criminal behavior dependent variables are presented
in Tables 8. Most respondents did not report engaging in any of the non-violent or
violent behaviors at Wave 3. Again, 8 items, measured from 0 (never) to 3 (more than
five times), were used to construct the non-violent scale. This could result in a maximum
possible score of 24 for someone who reported engaging in every behavior more than five
times. Four items were used to construct the violent scale, for a maximum possible score
of 12. However, as shown in Table 8, the maximum score for each outcome was 5.
These statistics indicate that most respondents did not report engaging in the targeted
behaviors. 34
Table 8. Descriptive Statistics: Dependent Variables

Non-violent Criminal Behavior
Violent Criminal Behavior

N
504
504

Mean
0.66
0.37

Standard
Deviation
1.33
1.00

Max
5
5

Descriptive statistics for independent variables included in the models predicing
non-violent and violent criminal behavior are shown in Table 9. Respondents reported
closer relationships with out-of-home caregivers than with either mothers or fathers.
Most of the sample was enrolled in some type of educational setting and nearly threequarters reported plans to graduate from college. Only 38 percent were employed at
Wave 1. Perceived social support was relatively high among the sample, with a mean
social support score of 73.15 out a possible 100. Attachment to the child welfare system,

34

Although the scales skewed toward zero, our analytic models (Poisson and negative binomial) do account for
variation beyond “yes” vs. “no” as would be the case in a binomial distribution.

48
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

however, appeared limited. For example, the average score on a scale indicating
likelihood to ask the child welfare system for help after discharge was only 2.4, on a scale
from 1-4. On average, respondents received 13.68 independent living services.
Considering 47 independent living services were potentially available, this figure
suggests that young people aging out of care receive limited preparation for making the
transition from adolescence to independent adulthood.

49
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Table 9. Descriptive Statistics:
Independent Variables in Violent and Non-Violent Criminal Behavior Models (n=504)
Males
Females
Total
(n=221)
(n=283)
(N=504)
s.d.
s.d.
s.d.
Min
Mean
Mean
Mean
Controls and Risk Factors
African American
.52
.50
.55
.50
.54
.50
0
White
.37
.48
.30
.46
.33
.47
0
Other race
.12
.32
.14
.35
.13
.34
0
Teen parenthood
.06
.24
.20
.40
.14
.35
0
Prior delinquency scale
5.16
3.82
3.42
3.28
4.18
3.62
0
Physical abuse
.92
1.53
1.14
1.79
1.04
1.69
0
Neglect
1.56
2.00
1.80
2.14
1.69
2.08
0
Sexual abuse/assault
.16
.37
.39
.49
.29
.45
0
Substance abuse diagnosis
.23
.42
.17
.38
.20
.40
0
Mental health diagnosis
.13
.34
.31
.46
.23
.42
0
Out-of-Home Care Experiences
Age at entry (in years)
10.71 3.87 10.92 4.04 10.83 3.96
0
Number of placements
6.10
6.39
5.12
5.10
5.55
5.72
1
Type of placement at wave 1
Foster care
.38
.49
.35
.48
.36
.48
0
Kinship care
.32
.47
.34
.47
.33
.47
0
Group care
.19
.40
.13
.34
.16
.36
0
Independent living/Other
.12
.32
.19
.39
.16
.36
0
Number of independent
living services received
14.53 12.11 13.01 11.87 13.68 11.99
0
Social Bonds
Closeness to mother
2.69
1.23
2.66
1.21
2.67 1.22
1
Missing mother indicator
.16
.37
.10
.30
.13
.33
0
Closeness to father
2.11
1.23
2.12
1.22
2.11
1.23
1
Missing father indicator
.30
.46
.21
.41
.25
.44
0
Closeness to foster caregiver 3.15
1.06
2.92
1.23
3.02
1.16
1
Social support
71.47 21.73 74.46 22.26 73.15 22.05
0
In school
.85
.36
.80
.40
.82
.39
0
College plans
.68
.47
.75
.43
.72
.45
0
Employed
.36
.48
.39
.49
.38
.48
0
Likely to ask child welfare
system for help
2.39
.72
2.43
.80
2.41
.77
1
State
Illinois
.56
.50
.65
.48
.61
.49
0
Iowa
.09
.28
.09
.28
.09
.28
0
Wisconsin
.35
.48
.27
.44
.30
.46
0

Max

1
1
1
1
11
7
9
1
1
1
16
42
1
1
1
1
47
4
1
4
1
4
100
1
1
1
4
1
1
1

50
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Results of the Poisson/Negative Binomial Models
Violent Crime

A Poisson distribution was chosen to model violent crime (deviance = 1.08). The
results for the full model can be seen in Table 10. As expected, there is a large and
statistically significant difference in delinquency by gender. Exponentiation of the
parameter estimate for males, β = 0.98, reveals that boys have an average level of violent
crime over 2.5 times the average for females. Youths who reported delinquency in Wave
1 also have higher criminal behavior in Wave 3, about 18 percent higher for each 1 unit
increase in Wave 1 delinquency.
Out-of-home care experiences were also significantly related to violent crime.
High numbers of foster care placements contributed to an increase of about 3 percent in
violent offending for each placement, and youths in group care at Wave 1 are expected to
report about 80 percent more crime than the average for youths in traditional foster care.
Youths who access independent living services however, have decreases in violent
behavior of about 2 percent for each service used.
Finally, we found that having college plans at Wave 1 decreases violent behavior
by about one-third compared to those without college aspiration. Being a victim of
sexual abuse is also related to a decrease of more than 50 percent in violent crime
reported.

51
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Table 10. Estimates from Poisson Regression of Violent Criminal Behavior on Covariates
(n=504)
Variables
Demographic and Background Risk Factors
Male (vs. female)
African American (vs. white)
Other race (vs. white)
Prior delinquency
Neglect
Physical abuse
Sexual abuse/assault
Substance abuse diagnosis
Mental health diagnosis
Teen parenthood
Iowa (vs. Illinois)
Wisconsin (vs. Illinois)

Β

SE

Exp(β)

.980***
.206
-.188
.161***
.064
-.011
-.808**
-.231
-.068
.251
-.180
-.538**

.199
.205
.284
.024
.044
.056
.245
.198
.213
.236
.388
.198

2.66
1.23
0.83
1.18
1.07
0.99
0.45
0.79
0.93
1.29
0.84
0.58

Out-of-Home Care Experiences
Age at entry
Number of placements
Type of placement at wave 1
(vs. traditional foster care)
Kinship care
Group care
Independent living/other
Number of independent living services

.013
.025*

.022
.012

1.01
1.03

.125
.582**
.471
-.019**

.214
.214
.399
.007

1.13
1.79
1.60
0.98

Social Bonds
Closeness to mother
Missing mother indicator
Closeness to father
Missing father indicator
Closeness to caregiver
Social support
In school
College plans
Employed
Likely to ask child welfare system for help

-.018
.776**
.001
-.129
.166
.007
.053
-.403*
-.003
-.038

.087
.255
.081
.211
.117
.004
.209
.160
.166
.108

0.98
2.17
1.00
0.88
1.18
1.01
1.05
0.67
0.99
0.96

*p<.05, **p<.01, ***p<.001 (two-tailed tests)

52
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Violent Crime by Race

To test the hypothesis that the relationship between out-of-home care placement
experiences and crime differ by race, we included interaction terms between a dummy
variable for race (African American = 1, white = 0) and each out-of-home care placement
variable 35 . Results for the interactions are shown in Table 11.
Contrary to our hypothesis, we did not find much evidence that the relationship
between out-of-home care placement experiences and violent criminal behavior was
moderated by race, specifically when comparing white and African American youth.
Only two interactions were significant – being in group care versus traditional foster care
and being in independent living or other care settings versus traditional foster care.
Being in group care is associated with a higher rate of violent behavior for African
Americans than for whites. In addition, being in an independent living or other
arrangement versus traditional foster care was associated with an increase in violent
crime for white youth, but not for African Americans. However, we note this finding
with caution due to the small number of youth who were living independently (or “other”
placement) in Wave 1.

35

Because the interaction of interest was white versus African American youth, we limited the sample to only white
and African American youth (n=438) for the model testing interactions between race and placement experiences.

53
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Table 11. Estimates from Poisson Regression of Violent Crime on Covariates with
Interaction between Race (African American vs. White) and Out-of-Home
Placement Experiences (n=438)
Variables
Demographic and Background Risk Factors
Male (vs. female)
African American (vs. white)
Prior delinquency
Neglect
Physical abuse
Sexual abuse/assault
Substance abuse diagnosis
Mental health diagnosis
Teen parenthood
Iowa (vs. Illinois)
Wisconsin (vs. Illinois)
Out-of-Home Care Experiences
Age at entry
Number of placements
Type of placement at wave 1
(vs. traditional foster care)
Kinship care
Group care
Independent living/other
Number of independent living services
Social Bonds
Closeness to mother
Missing mother indicator
Closeness to father
Missing father indicator
Closeness to caregiver
Social support
In school
College plans
Employed
Likely to ask child welfare system for help
Interactions (with African American vs. white)
Age at entry*race
Number of placements*race
Type of placement at wave 1
(vs. traditional foster care
Kinship care*race
Group care*race
Independent living/other*race
Number of independent living services*race

Β
1.134***
1.074
.156***
.021
-.016
-.656*
-.421
.239
.165
-.519
-.552*

SE

Exp(β)

.226
.739
.028
.049
.069
.274
.233
.233
.286
.417
.219

3.11
2.93
1.17
1.02
.98
.52
.66
1.27
1.18
.60
.58

.063
.010

.048
.020

1.07
1.01

.513
-.397
1.466**
-.002

.426
.494
.497
.014

1.67
.67
4.33
1.00

-.124
.560
.062
.826*
.273*
.003
.068
-.406*
.195
.088

.076
.418
.078
.334
.126
.004
.234
.175
.191
.116

.88
1.75
1.06
2.28
1.31
1.00
1.07
.67
1.22
1.09

-.051
.030

.055
.025

.95
1.03

-.404
1.163*
-1.776*
-.030

.495
.566
.731
.016

.67
3.20
.17
.97

*p<.05, **p<.01, ***p<.001 (two-tailed tests)

54
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Nonviolent Crime

We estimated non-violent criminal behavior through negative binomial
regression, the results of which can be seen in Table 12. As seen in the violent crime
model, gender and Wave 1 delinquency are both significant and positive in the nonviolent crime model. Males are estimated to engage in 68 percent more non-violent
criminal behavior than females, a much less dramatic difference than seen in the violent
crime model. The effect of Wave 1 delinquency was also smaller, though still
significant, contributing a 12 percent increase in expected nonviolent crime for each 1
unit increase in prior delinquency. Study particpants with multiple foster care placements
are predicted to have higher levels of crime, as each placement is associated with a 4
percent increase in average criminal behavior. No social bonds variables were found to be
significant in the nonviolent model.

55
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Table 12. Estimates from Negative Binomial Regression of Non-Violent Criminal Behavior
on Covariates (n=504)
Variables
Demographic and Background Risk Factors
Male (vs. female)
African American (vs. white)
Other race (vs. white)
Prior delinquency
Neglect
Physical abuse
Sexual abuse/assault
Substance abuse diagnosis
Mental health diagnosis
Teen parenthood
Iowa (vs. Illinois)
Wisconsin (vs. Illinois)
Out-of-Home Care Experiences
Age at entry
Number of placements
Type of placement at wave 1
(vs. traditional foster care)
Kinship care
Group care
Independent living/other
Number of independent living services
Social Bonds
Closeness to mother
Missing mother indicator
Closeness to father
Missing father indicator
Closeness to caregiver
Social support
In school
College plans
Employed
Likely to ask child welfare system for help

Β

SE

Exp(β)

.519*
-.305
-.196
.115***
.031
.056
-.417
.108
-.043
-.314
.019
.013

.227
.272
.330
.030
.056
.073
.256
.262
.271
.392
.392
.239

1.68
0.74
0.82
1.12
1.03
1.06
0.66
1.11
0.96
0.73
1.02
1.01

-.015
.037*

.030
.017

0.99
1.04

-.428
.554
.387
-.012

.267
.309
.494
.009

0.65
1.74
1.47
0.99

.149
.362
-.150
-.249
.147
-.003
-.235
.109
-.165
-.074

.107
.359
.101
.272
.151
.005
.269
.222
.211
.134

1.16
1.44
0.86
0.78
1.16
0.99
0.79
1.12
0.85
0.93

*p<.05, **p<.01, ***p<.001 (two-tailed tests)

56
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Non-violent Crime by Race

As with the violent criminal behavior model, we tested for statistical interactions
between placement experiences and race. The regression results for the interaction terms
are shown in Table 13. Again, we found little support for the hypothsis that the
relationship between out-of-home care experiences and non-violent crime differs between
African American and white youth. Only one interaction term, number of placements by
race, was significant, such that more out-of-care placements increased non-violent crime
for African Americans. Conversely, non-violent crime was similar for white youth with
different numbers of placements. This suggests that moving through many out-of-home
care placements may be more detrimental for African American youth.

57
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Table 13. Estimates from Poisson Regression of Non-Violent Crime on Covariates with
Interaction between Race (African American vs. White) and Out-of-Home
Placement Experiences (n=438)
Variables
Demographic and Background Risk Factors
Male (vs. female)
African American (vs. white)
Prior delinquency
Neglect
Physical abuse
Sexual abuse/assault
Substance abuse diagnosis
Mental health diagnosis
Teen parenthood
Iowa (vs. Illinois)
Wisconsin (vs. Illinois)
Out-of-Home Care Experiences
Age at entry
Number of placements
Type of placement at wave 1
(vs. traditional foster care)
Kinship care
Group care
Independent living/other
Number of independent living services
Social Bonds
Closeness to mother
Missing mother indicator
Closeness to father
Missing father indicator
Closeness to caregiver
Social support
In school
College plans
Employed
Likely to ask child welfare system for help
Interactions (with African American vs. white)
Age at entry*race
Number of placements*race
Type of placement at wave 1
(vs. traditional foster care
Kinship care*race
Group care*race
Independent living/other*race
Number of independent living services*race

β

SE

Exp(β)

.247
.815
.034
.061
.083
.283
.283
.302
.359
.418
.253

2.24
.65
1.11
1.09
.94
.84
1.21
1.16
.58
1.13
1.26

.019
.005

.050
.028

1.02
1.01

-.909
.362
.679
-.031*

.553
.458
.625
.015

.40
1.44
1.97
0.97

.077
.126
-.115
-.133
.140
-.006
.106
-.022
-.198
-.120

.097
.737
.099
.532
.164
.005
.305
.238
.234
.148

1.08
1.13
.89
.88
1.15
.99
1.11
.98
.82
.89

-.076
.075*

.062
.036

.93
1.08

.703
.496
.102
.030

.641
.614
.635
.019

2.02
1.64
1.11
1.03

.805**
-.428
.100**
.089
-.058
-.179
.188
.146
-.544
.124
.229

*p<.05, **p<.01, ***p<.001 (two-tailed tests)

58
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Predictors of Criminal Justice Involvement: The Risk for Arrest
Descriptive Statistics

Descriptive statistics for the covariates predicting time to first arrest are presented
in Table 14. Results are similar to those presented in Table 9 for the Wave 3 sample.
Table 14. Descriptive Statistics: Covariates in Hazard Models Predicting Arrest (n=728)
Males
Females
Total
(n=352)
(n=376)
(N=728)
s.d.
s.d.
s.d.
Min
Max
Mean
Mean
Mean
Controls and Risk Factors
African American
.56
.50
.58
.49
.57
.49
0
1
White
.32
.47
.30
.46
.31
.46
0
1
Other race
.12
.32
.12
.33
.12
.33
0
1
Teen parenthood
.07
.26
.20
.40
.14
.35
0
1
Age at wave 1
17.9
.34
17.9
.37
17.9
.36
17.1 18.8
Prior arrest
.62
.49
.41
.49
.51
.50
0
1
Physical abuse (sum)
.39
.49
.40
.49
1.06
1.71
0
7
Neglect (sum)
.60
.49
.65
.48
1.75
2.06
0
7
Sexual abuse/assault
.14
.34
.40
.49
.30
.46
0
1
Substance abuse diagnosis
.23
.42
.16
.37
.20
.40
0
1
Mental health diagnosis
.13
.34
.32
.47
.23
.42
0
1
Out-of-Home Care Experiences
Age at entry
10.5
4.0
10.9
4.07
17.9
.36
17.1 18.8
Number of placements
6.3
6.4
5.3
5.4
5.8
5.9
0
42
Type of placement (wave 1)
Foster care
.36
.48
.36
.48
.36
.48
0
1
Kinship care
.29
.45
.32
.47
.31
.46
0
1
Group care
.21
.41
.15
.36
.18
.38
0
1
Independent living/Other
.07
.25
.09
.28
.08
.27
0
1
Number of independent
14.29 12.13 13.0
11.9
13.6
12.0
0
47
living services received
Social Bonds
Closeness to mother
2.65
1.2
2.69
1.2
2.67 1.2
1
4
Closeness to father
2.12
1.2
2.11
1.2
2.12
1.2
1
4
Closeness to foster caregiver 3.10
1.1
2.95
1.2
3.03
1.2
1
4
Social support
72.2
22.4
74.2 23.14 73.2
22.8
0
100
In school
.83
.38
.82
.39
.82
.38
0
1
College plans
.69
.46
.75
.43
.72
.45
0
1
Employed
.32
.47
.38
.48
.35
.48
0
1
Likely to ask child welfare
2.36
.74
2.42
.74
2.39
.79
1
4
system for help
State
Illinois
.61
.49
.68
.47
.65
.48
0
1
Iowa
.08
.49
.09
.29
.09
.28
0
1
Wisconsin
.31
.46
.23
.42
.27
.44
0
1

59
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Results of the Hazard Model

We first examined the time that young people aging out of care survived, on
average, before experiencing an arrest. Forty-six percent of the sample was arrested
between the Wave 1 interview (from 2002-2003) and the end of the study period (August
31, 2007), which is about 5.3 years. This figure is not surprising, given the high risk for
offending for maltreated youth in general (Maxfield and Widom, 1996; McCord,
McCord, and Thurber, 1983; Smith and Thornberry, 1995; Widom, 1989) and
considering that over half of the sample reported a previous arrest before the Wave 1
interview. Over the period, the probability of arrest reaches almost .50, as shown in
Figure 4 (the cumulative probability of first arrest).
Figure 4. Cumulative Probability of First Arrest
0.5
0.45

Cumulative Probability

0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
1

2

3

4

5

6

7

8

9

10 11 12 13 14 15 16 17 18 19 20 21 22

Duration in Quarters

60
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Figure 5 illustrates the conditional hazard probabilities of first arrest following the
Wave 1 interview. This is expressed as the proportion of individuals at risk who are
arrested during each time point, shown in 3-month, or quarterly, intervals. As shown in
Figure 5 the conditional probability of experiencing an arrest generally decreases over the
study period.
Figure 5. Conditional Probability of First Arrest
0.07

0.06

Conditional Probability

0.05

0.04

0.03

0.02

0.01

0
1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

Duration in Quarters

Next, we examined the relationship of covariates to the timing of first arrest.
Results of the Cox proportional hazards model for the full sample are presented in Table
15. A significant coefficient indicates the covariate is predictive of the time to first
arrest. Positive coefficients are associated with earlier time to first arrest (earlier failure),
while negative coefficients are associated with later time to first arrest (longer survival).
Hazard ratios are also provided. These hazard, or risk, ratios indicate how the risk of
61
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

arrest varies with a change in the value of a covariate relative to a reference value (for
categorical covariates) or with increasing values of a continuous covariate. Values over
one indicate an increase in the hazard rate and values less than one indicate a decrease in
the hazard rate. If one is subtracted from the hazard ratio and the remainder is multiplied
by 100, the result is equal to the percentage change in the hazard of arrest.
Table 15. Estimates from Cox Proportional Hazard Regression of Time to First Arrest on
Covariates: Full Sample (n=728)
Hazard
β
Variables
SE
Ratio
Demographic and Background Risk Factors
Age at wave 1 in years
-.053
.200
.95
Male (vs. female)
.692***
.133
2.00
African American (vs. white)
.719***
.155
2.05
Other race (vs. white)
.322
.197
1.37
Prior arrest
.814***
.135
2.26
Neglect
-.005
.035
1.00
Physical abuse
.047
.043
1.05
Sexual abuse/assault
.025
.153
1.02
Substance abuse diagnosis
.425**
.139
1.53
Mental health diagnosis
-.334*
.159
.72
Children
.055
.176
1.06
IA
-.054
.279
.95
WI
-.107
.155
.90
Out-of-Home Care Experiences
Age at entry
.015
.016
1.01
Number of placements
.035***
.009
1.04
Type of placement at wave 1
(vs. traditional foster care)
Kinship care
.194
.154
1.21
Group care
.458**
.168
1.58
Independent living/other
.295
.265
1.34
Number of independent living services
.002
.005
1.00
Social Bonds
Closeness to mother
.098
.061
1.10
Missing mother indicator
.495*
.205
1.64
Closeness to father
-.017
.058
.98
Missing father indicator
-.042
.164
.96
Closeness to caregiver
.034
.079
1.03
Social support
.005
.003
1.00
In school
.036
.153
1.04
College plans
-.270*
.123
.76
Employed
-.378**
.126
.68
Likely to ask child welfare system for help
-.048
.078
.95
*p<.05, **p<.01, ***p<.001 (two-tailed tests)
62
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

We found that being male versus female and African American versus White
increased the hazard rate for arrest by about 100 percent. In other words, the time to first
arrest was significantly shorter for males and African American respondents. The time to
first arrest was also significantly shorter among those with a prior arrest and those with a
history of alcohol or drug abuse diagnosis. That young people with substance abuse
problems would have a greater risk of arrest is not surprising given the association
between drug use and arrests among juveniles (National Institute of Justice, 2003). These
findings were independent of out-of-home care experiences. Having a history of one or
more of the mental health disorders we assessed was associated with a reduced hazard of
arrest. Since the most common diagnoses among the disorders we assessed were
depression and PTSD, in other words, disorders not normally reflective of externalizing
behavior problems, the negative association is not necessarily surprising.
The risk for arrest was significantly associated with out-of-home care placement
experiences. 36 Being in group home care versus traditional foster care at the time of the
Wave 1 interview led to significantly higher hazard rates, with a group care placement at
the time of the Wave 1 interview resulting in a 58 percent increase in the hazard rate
when compared to a traditional foster care placement. The need for a highly restrictive
placement setting at the onset of the transition may represent an important marker of risk
for criminal justice system involvement. Young people who experienced multiple
placements also had a higher hazard rate for arrest. Results show that each additional
placement after the first resulted in a 4 percent increase in the hazard rate. This finding is

36

As we did in the criminal behavior models, we tested for statistical interactions between race and placement
experiences. None of the relationships were significant and therefore were not included in the final model presented
here.

63
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

particularly salient given that over 50 percent of the respondents experienced four or
more placements during foster care.
We found some support for the hypothesis that having bonds as one approaches
the transition to adulthood reduces the risk of arrest. With respect to our measures of
connections to family and caregivers, however, the results indicate that contrary to our
expectations there appears to be little relationship with being close to one’s mother, father
or foster caregiver and crime. However, not having a mother was related to an increase
in the hazard rate. There was a 64 percent increase in the risk for arrest among those
youth coded as not having a mother.
Bonds to education and employment did predict timing to arrest in general, but
again, results were mixed. Having plans to graduate from college and being employed at
Wave 1 reduced the hazard rate by 24 and 32 percent respectively. Thus, those with
plans to attend college and those employed before the onset of the transition to adulthood
survived longer without experiencing an arrest during the transition than those
unemployed or without college plans. Being enrolled in some form of school just prior to
the transition, however, was not significantly related to the timing of arrest. These
findings suggest that commitment to one’s education may be more important than simply
the status of being in school for the timing to first arrest. However, a high percentage of
respondents (83 percent) were enrolled in some form of education at Wave 1. It may be
that we found no relationship between education status and the timing of arrest due to the
limited variation in this covariate. The likelihood of turning to the child welfare system
for assistance was not related to the hazard of first arrest.

64
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Time to Arrest by Gender

We also examined the hazard models separately by gender. The mean time to
first arrest differed for males and females. For males, the mean time to first arrest, for
those arrested, was 1040 days, or approximately 2.9 years. Over the observed range of
data (1927 days, or about 5.3 years) the probability of being arrested is .61 for males. For
females, the mean time to first arrest was longer, at 1408 days, or approximately 3.9
years. Over the observed range of data, the probability of being arrested is .32 for
females.
Results of the Cox proportional hazards model by gender are presented in Table
16. To determine whether the regression coefficients significantly differed across the
models, we calculated a Z-score using the formula Z =

b1 − b2
2

SEb1 + SEb2

2

(see Clogg,

Petkova, and Haritou, 1995; Paternoster, Brame, Mazerolle, and Piquero, 1998), which is
included in Table 16. Based on this test, we found little evidence of differences in the
factors predicting time to arrest between males and females. However, we note that there
may not be enough power to detect such differences. Therefore, although we discuss
these gender specific models, we note that we cannot be sure that the factors examined
here differentially predict the risk for arrest for males and females.
For males and females, having a prior arrest and experiencing placement
instability were significantly associated with a higher hazard rate for arrest. The high rate
of subsequent criminal system involvement among those with previous legal system
involvement suggests the need to take a closer look at the ways in which juvenile
criminal system involvement intersects with foster placement experiences in ways that
may influence subsequent criminal activity.

65
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

For males, being employed at the onset of the transition was associated with a
lower hazard rate of arrest; being employed resulted in a 31 percent decrease in the
hazard rate of arrest for males. Having plans to enroll in college also resulted in a 31
percent decrease in the hazard rate of arrest suggesting the protective effects of planning
for the future and a commitment to education. Given the uncertainty associated with
being in long-term foster care, the ability to plan likely reflects an important protective
factor for young males in meeting future goals. Being enrolled in school, however, was
not significantly related to the time to first arrest among males or females.
With regard to bonds to parents and caregivers, we found for females that a close
relationship with a biological mother resulted in a 24 percent increase in the hazard rate
of arrest. For males, not having a mother resulted in a 58 percent increase in the hazard
rate of arrest, suggesting that males who are without a mother experienced a shorter time
to first arrest. Again, differences for these coefficients were not statistically significant
suggesting the relationship between closeness to mother and arrest may be similar for
males and females.

66
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Table 16. Estimates from Cox Proportional Hazard Regression of Time to First Arrest on
Covariates By Gender
Males
Females

Variables
Demographic and Background
Risk Factors
Age at wave 1
African American
Other race
Prior arrest
Neglect
Physical abuse
Sexual abuse/assault
Substance abuse diagnosis
Mental health diagnosis
Children
IA
WI
Out-of-Home Care Experiences
Age at entry
Number of placements
Type of placement at wave 1
Kinship care
Group care
Independent living/other
Number of independent
living services
Social Bonds
Closeness to mother
Missing mother indicator
Closeness to father
Missing father indicator
Closeness to caregiver
Social support
In school
College plans
Employed
Likely to ask child
welfare system for help

SE

Hazard
Ratio

-.199
.931***
.213
.610**
.043
.044
-.094
.460**
-.674**
.470
-.186
-.028

.264
.192
.265
.182
.043
.055
.237
.173
.241
.265
.386
.184

.82
2.54
1.24
1.84
1.04
1.04
0.91
1.58
.51
1.60
.83
.97

.027
.029*

.021
.012

.166
.443 *
.106

Z-score
for diff

SE

Hazard
Ratio

.123
.338
.434
1.07***
-.116
.055
.028
.382
-.120
-.086
-.000
-.308

.327
.292
.321
.217
.064
.072
.224
.268
.234
.246
.431
.324

1.13
1.40
1.54
2.91
.89
1.06
1.03
1.47
.89
.92
1.00
.74

-.76
1.69
-.53
-1.62
2.06*
-.12
-.37
.24
-1.64
1.53
-.32
.75

1.03
1.03

-.025
.060***

.025
.016

.98
1.06

1.59
-1.55

.196
.223
.352

1.18
1.56
1.11

.215
.441
.167

.262
.290
.435

1.24
1.56
1.18

-.15
.01
-.10

.009

.006

1.01

-.008

.008

.99

1.70

.036
.681**
.018
-.067
-.076
.004
.188
-.373*
-.399*

.079
.255
.078
.200
.109
.003
.206
.156
.163

1.04
1.98
1.02
.93
.93
1.00
1.21
.69
.67

.213*
.060
-.022
-.031
-.024
.004
.032
.015
-.390

.102
.398
.091
.295
.129
.005
.258
.221
.220

1.24
1.06
.98
1.03
.98
1.00
1.03
1.02
.68

-1.37
1.31
.33
-.10
-.30
.00
.47
-1.43
-.03

-.078

.106

.92

.060

.133

1.06

0.81

b

B

*p<.05, **p<.01, ***p<.001 (two-tailed tests)

67
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Limitations of the Study
In presenting our findings, we note the following limitations to the study. First,
although this is the largest study of foster youth aging out of care since the Foster Care
Independence Act to date, youth from only three Midwestern states are represented.
While findings may not be generalizable to foster youth in other states, it is worth
comparing the Midwest Study sample to the characteristics on the foster care population
nationally. There are no readily available national data on older youth in foster care,
however, data provided by states to the Federal government in 2001 indicate that the
Midwest Study sample was comparable to the national foster care population in terms of
gender, but not race and ethnicity (U.S. Department of Health and Human Services,
2003). In 2001, just over half (52%) of children and youth in foster care nationally were
male, compared to 49 percent of the Midwest Study sample. African Americans made up
a larger percentage of the Midwest Study population (57%) than they did of the foster
care population nationally (37%), whereas Hispanics made up a smaller percentage of the
Midwest Study sample (9%) than they did of the overall foster care population (17%).
Because the sampling frame of the Midwest Study excluded youth under the care of the
child welfare system who were in psychiatric hospitals, incarcerated, or on runaway
status over the course of the study field period, the study might understate to some extent
crime outcomes for the overall population of older youth in care.
Second, although our study benefits from a longitudinal design with high
retention, some analyses are limited to only those youth interviewed at certain or all
waves. For example, only those youth with information on criminal behavior at Wave 3
were included in analyses predicting crime at age 21. Similarly we limited our latent
class analysis to only those youth present in all three waves and who answered questions
68
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

about their criminal behavior. There is the possibility that youth not interviewed and,
therefore, not included in these analyses engage in more criminal behavior than those
interviewed. Statistical checks on the data (e.g. comparing data at Wave 1 and official
arrest records over time) as described earlier do not suggest attrition of those most likely
to offend. However, we note this as a possible limitation.
Third, the purpose of our study was to examine whether social bonds one has as
he/she approaches the transition to adulthood predicts offending during the early
transition period. Therefore, we used social bond measures at Wave 1 to predict
subsequent behavior. It may be that the strength and type of social bonds change over
time. Future research would benefit from examining changes in social bonds and those
that are contemporaneous to the outcomes.
Fourth, our social bonds measures, while noteworthy for addressing attachment to
a variety of caregivers and for focusing on those bonds we hypothesize to be most critical
during the early transition to adulthood, have limitations. In particular, our measures of
attachment to parents and caregivers use only one question regarding closeness to the
specified individual. In addition, we have no measure of parental monitoring, which is a
focal measure in Hirschi’s theory. Again, although we hypothesized that certain social
bonds, notably attachment to parents and caregivers and commitment to education and
employment, would be most critical as one approaches the transition to adulthood, we
recognize that other bonds may important as well. For example, although marriage was
very rare among our sample, attachment to a romantic partner is a possible bond that may
predict criminal behavior. Future research, particularly that addresses youth as they

69
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

move further into the transition period, may benefit from considering the importance of
romantic partners.
Finally, future research would benefit from the use of multi-methods, possibly
including qualitative interviews, to gain a deeper understanding of the life course
experiences of youth aging out of foster care and their associated involvement in crime.

70
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Policy and Practice Implications
Since 1985, federal policy has long grappled with how to parent children in aging
out of foster care and help them to grow up to be productive adults. There have been
successive pieces of legislation which have attempted to address the needs of older
children in the child welfare system, including the Independent Living Program in 1985,
the Adoption and Safe Families Act of 1997, and the John Chaffee Foster Care
Independence Act in 1999. More recently, the Fostering Connections to Success and
Increasing Adoptions Act of 2008 (to be implemented in 2011) amended the Social
Security Act to provide states who extend the age at which youth may remain in foster
care past the age of 18 with federal funding.
Our findings have several implications for policy, practice, and research. First,
comparison of the Midwest Study and Add Health data on self-reported crime and arrest
provide reason for both optimism and concern. On the one hand, youth self reports
suggest that foster youth, like their peers, engage in less crime over time as they move
into adulthood. In fact, while foster youth report more criminal behavior than their peers
as they approach the age of majority, at age 19 and 21 there are few differences between
the groups in self-reported crime. Findings from our latent class analysis indicate that
only about one-fifth of the foster youth (19 percent; latent classes 4 and 5 combined) are
chronic offenders, with somewhat less than half of these being nonviolent chronic
offenders (8 percent of the sample). In other words, at least according to their own
reports, as a group foster youth in transition pose little threat to community safety. On

71
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

the other hand, in spite of these relatively small differences in self-reported criminal
behavior, foster youth remain much more likely than their peers to be arrested as they
make the transition to adulthood. These conflicting findings call for research designed to
better understand the circumstances that lead to foster youth in transition to being
arrested. Do their self reports mask behavior that understandably leads to arrest, or might
they be under more scrutiny than their peers or receive less sympathy from law
enforcement authorities?
Second, the finding that race is a significant predictor of arrest, but not selfreported crime 37 , is disturbing and calls for further investigation and explanation. That
race did not predict self reported non-violent nor violent crime in our sample when
controlling for other factors suggests that the racial difference in official arrest versus
self-reported outcomes may not simply be an artifact of a self reported crime measure
that reflects trivial and non-serious offenses (see Elliot and Ageton, 1980). In other
words, our findings are inconsistent with past work that finds significant race differences
when examining more serious self reported crime measures (Elliot and Ageton, 1980). If
our self-reported crime measures are not differentially valid by race (e.g. .if African
Americans are not less likely to report engaging in certain offenses 38 ), then our finding
that African American youth experience greater arrests but do not reporting engaging in
more crime suggests a possible bias in police response. Foster youth already face
considerable challenges making the transition to adulthood and African American foster

37

We note that our bivariate analyses showed a higher proportion of African American youth to be classified using
self-report data as chronic offenders as well as non-offenders. However, multivariate analyses did not reveal racial
differences in self reported offending.

38

See Hindelang, Hirschi, and Weis (1981) for a discussion on this methodological issue.

72
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

youth appear to face the added burden of heightened risk of arrest that may not reflect a
heightened engagement in crime.
Third, our findings suggest that greater attention is needed to ensure that older
foster youth are provided with opportunities, both before and after reaching the age of
majority, to become better connected to larger structures that traditionally support the
transition to adulthood, including employment and education. Presently, there are no
guidelines or practice standards that ensure foster youth are provided with opportunities
to gain skills to use in the work force. Yet, the limited research that has been done on
young adults who “aged out” of foster care has found that their labor market outcomes
are generally quite poor, which suggests that the employment-related needs of foster
youth are not currently being met (Courtney, Dworsky, Ruth Cusick, Keller, Havlicek,
and Perez, 2007). Similarly, educational outcomes for foster youth are generally poor
(Courtney and Hughes-Heuring, 2005). While the associations identified in our study
should not be regarded as causal, the fact that prior employment and plans to attend
college are associated with reductions in later crime argues for greater attention to
improving the education and employment experiences of foster youth. Given the lack of
evidence regarding the effectiveness of independent living programs directed towards
improving education and employment outcomes, resources should be devoted to rigorous
evaluation of interventions targeting these outcomes for foster youth (Montgomery et al,
2006).
Fourth, our findings on the relationships between the placement experiences of
foster youth and later crime have implications for child welfare practice. Our findings
suggest that it is not simply the case that youth with more difficult cases experience

73
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

multiple placements or are placed in group settings. Placement in group care near the
point of transition from care and a history of placement instability are strong predictors of
criminal behavior and arrest, even after controlling for prior problem behavior,
suggesting that placement instability and group care placement are not merely indicators
of existing problem behavior. 39 Although our analyses cannot confirm a causal
relationship between placement instability or group care and crime, our findings do point
to a possible opportunity for the child welfare system to act on these specific cases. The
child welfare system may consider providing extra supports in an attempt to mitigate the
possible detrimental impact of these particular placement experiences or those other
external factors that may be associated with placement instability or placement in group
care that were not examined here. Our findings suggest that efforts to prevent crime
among foster youth making the transition to adulthood are well targeted at those in group
care. That placement instability is associated with later crime may not be surprising, but
it calls for renewed efforts to minimize the instability that all too often characterizes the
lives of children and youth in care. Developing research on the effectiveness of treatment
foster care and training of foster parents to better cope with externalizing behavior should
help guide efforts to develop alternatives to group care and help reduce placement
mobility of foster youth (Chamberlain et al, 2008; Leve and Chamberlain, 2005).

39

We also note that we included a measure of caregiver problems and reentry to the child welfare system after a failed
attempt at reunification in previous models not shown in this report to control for the possibility that negative
placement experiences were indicators of antisocial behaviors of parents and continued difficulties with home
environments. Neither was significant predictors of the outcomes and the inclusion did not change our findings for
number of placements or group home placement.

74
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Fifth, although we found limited support for our hypothesis that greater receipt of
independent living services is related to less crime, our findings do suggest that these
services are meaningful. In particular, greater receipt of independent living services was
related to a reduction in violent behavior. This finding highlights the need for further
research on the nature of these services, including the quality and types of programs
being offered, availability of services, and program uptake (Naccarato and DeLorenzo,
2008; Montgomery et al, 2006).
Finally, our findings suggest that more attention ought to be given to
understanding the ways in which child welfare agencies assist older foster children to
navigate relationships with their biological parents and extended families. Contrary to
findings for vulnerable populations which have suggested that having a close and caring
relationship with at least one adult is an important protective factor for youth at risk, this
study suggests that greater understanding of the multiple bonds former foster youth have
with individuals and their communities is needed. While we found limited support for
the hypothesis that closeness of foster youth to their parents was related to crime, we
found stronger evidence that the absence of a parent increased the risk of self-reported
crime and arrest. Research consistently suggests that a large percentage of these young
people return to their families at discharge from the child welfare system (Wulcyzn and
Hislop, 2001). It may then follow that these young people will be better served by an
approach that attempts to minimize the risks presented by their caregivers, even when
youth are not reunified with their caregivers. A related issue has to do with the relatives
caring for these young people. Although the Fostering Connections to Success and
Increasing Adoptions Act will facilitate more permanent placements with relatives, there

75
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

is some evidence to suggest that helping relative caregivers to navigate relationships with
biological parents may go a long way in stabilizing placements (Beeman and Boisen,
1999) and potentially decrease familial obstacles facing foster youth during the transition
to adulthood.

76
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

References
Allison, P. (1995). Survival analysis using SAS: A practical guide. SAS Institute.
Barth, R. 1990. On their own: The experiences of youth after foster care. Child and Adolescent
Social Work 7: 419-440.
Bearman, P.S., Jones, J., and Udry, J.R. (1997). The National Longitudinal Study of Adolescent
Health: Research Design. Chapel Hill, NC: Carolina Population Center. Available at:
http://www.cpc.unc.edu/projects/addhealth/design
Beeman, Sandra, and Laura Boisen. (1999). Child Welfare Professionals' Attitudes toward
Kinship Foster Care. Child Welfare 78(3): 315-37.
Bottcher, J.B., and Ezel, M.E. (2005). The effectiveness of boot camps: A randomized
experiment with long-term follow-up. Research in Crime and Delinquency, 42, 309-332.
Box-Steffensmeier, J.M., and Jones, B.S. (2004). Event-history modeling: A guide for social
scientists. Cambridge: Cambridge University Press.
Bushway, S.D., Piquero, A.R., Broidy, L.M., Cauffman, E., and Mazerolle, P. (2001). An
empirical framework for studying desistance as a process. Criminology 39: 491-515.
Cernkovich, , S.A., and Giordano, P.C. (1992). School bonding, race, and delinquency.
Criminology, 30(2), 261-291.
Chamberlain, P., Price, J., Leve, L. D., Laurent, H., Landsverk, J. A., & Reid, J. B. (2008).
Prevention of behavior problems for children in foster care: Outcomes and mediation effects.
Prevention Science 9: 17-27.
Chapman, M.V., Wall, A., Barth, R., and the National Survey of Child and Adolescent WellBeing Research Group (2004). Children’s voices: The perceptions of children in foster care.
American Journal of Orthopsychiatry, 74(3), 293-304.
Chung, I., Hawkins, J.D., Gilchrist, L.D., Hill, K.G., and Nagin, D.S. (2002). Identifying and
predicting offending trajectories among poor children. The Social Service Review 76: 663713.
Clogg, C.C., Petkova, E., and Haritou, A. (1995). Statistical methods for comparing regression
coefficients between models. American Journal of Sociology 100: 1261-1293.

77
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Collins, M.E. (2001). Transition to adulthood for vulnerable youths: A review of research and
implications for policy. Social Services Review, 75(2), 271-291.
Colvin, M., Cullen, F., and Vander Ven, T. (2002). Coercion, social support and crime: An
emerging theoretical consensus. Criminology, 40, 19-59.
Cook, R., Fleishman, E., and Grimes, V. (1991). A National Evaluation of Title IV-E
Foster
Care Independent Living Programs for Youth: Phase 2, Final report, Volume 1. Rockville,
MD: Westat.
Courtney, M., Piliavin, I., Grogan-Taylor, A., and Nesmith, A. (2001). Foster youth transitions to
adulthood: A longitudinal view of youth leaving care. Child Welfare, 80(6), 685-717.
Courtney, M.E., Barth, R.P., Berrick, J. D., Brooks, D., Needell, B., & Park, L. (1996). Race and
child welfare services: Past research and future directions. Child Welfare 75: 99-136.
Courtney, M., and Dworsky, A. (2006). Early outcomes for young adults transitioning from outof-home care in the USA. Child and Family Social Work, 11, 209-219.
Courtney, M., Dworsky, A., Cusick, G.R., Havlicek, J., Perez, A., and Keller, T. (2007). Midwest
Evaluation of the Adult Functioning of Former Foster Youth: Outcomes at Age 21. Chicago:
Chapin Hall Center for Children at the University of Chicago.
Courtney, M.E. and Hughes-Heuring, D. (2005). The transition to adulthood for youth “aging
out” of the foster care system. In D. Wayne Osgood, E. Michael Foster, Constance Flanagan,
and Gretchen R. Cusick (Eds.) On Your Own Without a Net: The Transition to Adulthood for
Vulnerable Populations (pp. 27-67) Chicago: University of Chicago Press.
Courtney, M.E. and Wong, Y.I. (1996). Comparing the timing of exits from substitute care.
Children and Youth Services Review 18: 307-334.
Cullen, F. (1994). Social support as an organizing conception for criminology: Presidential
address to the Academy of Criminal Justice Sciences. Justice Quarterly, 11, 527-553.
Cusick, G., and Courtney, M. (2007). Offending During Late Adolescence: How do Youth Aging
Out of Care Compare with Their Peers? Chicago: Chapin Hall Center for Children at the
University of Chicago.
Dworsky, A. (2005). The economic self-sufficiency of Wisconsin’s former foster youth.
Children and Youth Services Review, 27(10), 1085-1118.
Elder, Glen H., Jr., ed. (1985). Life Course Dynamics: Trajectories and Transitions, 1968-1980.
NY: Cornell University Press.
Elliot, D.S. and S.S. Ageton. (1980). Reconciling race and class differences in self-reported and
official estimates of delinquency. American Sociological Review 45: 95-110.

78
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Farrington, D.P. (1986). Age and crime. In M. Tonry and N. Morris (Eds.) Crime and Justice: An
Annual Review of Research Vol. 7 (pp. 189-250). Chicago: University of Chicago Press.
Felson, R. and Staff, J. (2006). Explaining the academic delinquency relationship. Criminology,
44: 229-320.
Festinger, T. (1983). No one ever asked us: A postscript for foster care. New York: Columbia
University Press.
Frost, S., and Jurich, A., (1983). Follow up study of children living in the Villages. (Unpublished
report). The Villages, Topeka, KS.
Furstenberg, F., Rumbaut, R. and Settersten, R. (2005). On the frontier of adulthood: Emerging
themes and new directions. In F. Furstenberg, R. Rumbaut, R. Setterstein (Eds.), On the
frontier of adulthood (pp. 3-25). Chicago, IL: The University of Chicago Press.
Goerge, B., Bilaver, L., and Joo Lee, B., Needell, B., Brookhart, A., and Jackman, W. (2002).
Employment outcomes for youth aging out of foster care. Chicago: Chapin Hall Center for
Children at the University of Chicago.
Harari, T. (1980). Teenagers exiting from family foster care: A retrospective look (Doctoral
dissertation, University of California, Berkeley).
Hawkins, D. J., Laub, J.L., Lauritsen, J.L., and Cothern, L. (2000). Race, Ethnicity, and Serious
and Violent Juvenile Offending. Washington, D.C.: U.S. Department of Justice, Office of
Justice Programs, Office of Juvenile Justice and Delinquency Prevention.
Haynie, D., Giordano, P., Manning, W., and Longmore, M. (2005). Adolescent romantic
relationships and delinquency involvement. Criminology 43(1): 177 – 210.

Hindelang, M. J., Hirschi, T, & Weiss, J. G. (1981). Measuring delinquency. Beverly
Hills, CA: Sage.
Hines, A., Merdinger, J., and Wyatt, P. (2005). Former foster youth attending college: Resilience
and the transition to young adulthood. American Journal of Orthopsychiatry, 75(3), 381-394.
Hirschi, T. (1969). Causes of delinquency. Berkeley: University of California Press.
Hirschi, T. and M. Gottfredson, M. (1983). Age and the explanation of crime. American Journal
of Sociology 89: 552-584.
Horney, J., Osgood, D.W., and Marshall, I.H. (1995). Criminal careers in the short term: Intraindividual variability in crime and its relation to local life circumstances. American
Sociological Review, 60(5), 655-673.

79
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Hyde, J. and Kammerer N. (2009). Adolescents’ perspectives on placement moves and
congregate settings: Complex and cumulative instabilities in out-of-home care. Children and
Youth Services Review, 31(2):265-273.
Jones, M., and Moses, B. (1984). West Virginia’s former foster children: Their experiences in
care and their lives as young adults. New York: Child Welfare League of America.
Jonson-Reid, M. and Barth, R. (2000). From placement to prison: The path to adolescent
incarceration from child welfare supervised foster or group care. Children and Youth Service
Review, 22(7), 493-516.
Jonson-Reid, M. and Barth, R. (2003). Probation foster care as an outcome for children exiting
foster care. Social Work in Education, 48(3), 348-361.
Kessler, M., (2004). Educating youth in care: The first year of education and training vouchers.
National Resource Center for Youth Services at the University of Oklahoma.
Lagrange, T., and Silverman, R. (2006). Low self-control and opportunity: Testing the general
theory of crime as an explanation for gender differences in delinquency. Criminology, 37(1),
41-72.
Laub, J., Nagin, D., and Samson, R. (1998). Trajectories of Change in Criminal Offending: Good
Marriages and the Desistance Process, American Sociological Review, 63(2), 225-238.
Leve, L. D., & Chamberlain, P. (2005). Association with delinquent peers: Intervention effects
for youth in out-of-home care. Journal of Abnormal Child Psychology 33(3):339-347.
Lu, Y.E. Landsverk, J., Ellis-Macleod, E., Newton, R. Ganger, W., and Johnson, I. (2004). Race,
ethnicity, and case outcomes in child protective services. Children and Youth Services
Review 26: 447-461.
Maow, T. (2005). Sequences of early adult transitions: A look at variability and Consequences
(pp. ). In D.W. Osgood, E.M. Foster, C., Flanagan, and G.R. Cusick. On Your Own Without
a Net. Chicago, IL: University of Chicago Press.
Massinga, R., and Pecora, P. (2004). Providing better opportunities for older children in the child
welfare system. Future of Children, 14(1), 151-173.
Maxfield, M., and Widom, C. (1996). The cycle of violence revisited 6-years later. Archives of
Pediatrics and Adolescent Medicine, 150(4), 390-395.
McCord, J. (1983). A forty year perspective on effects of child abuse and neglect. Child Abuse
and Neglect, 7, 265-270.
McCord, J., McCord, W. and Thurber, E. (1960). The effects of foster home placement in the
prevention of adult antisocial behavior. Social Service Review, 34, 415-419.

80
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

McCutcheon, Alan. (1987) Latent Class Analysis. Sage University Paper series on Quantitative
Applications in the Social Sciences. Beverly Hills and London: Sage Publications.
McNulty, T.L. and Bellair, P.E. (2003). Explaining racial and ethnic differences in serious
adolescent violent behavior. Criminology 41: 709-748.
Moffit, T.E. (1993). Adolescence-limited and life-course-persistent antisocial behavior: A
developmental taxonomy. Psychological Review 100: 674-701.
Montgomery, P., Dunkoh, C., and Underhill, K. (2006). Independent living programs for young
people leaving the care system: The state of the evidence. Children and Youth Services
Review 28:1435-1448.
Naccarato T. and DeLorenzo E. (2008). Transitional youth services: practice implications from a
systematic review. Child & Adolescent Social Work Journal 25(4): 287-308.
Nagin, D.S. and Land, K. (1993). Age, criminal careers, and population heterogeneity:
Specification and estimation of a nonparametric, mixed Poisson model. Criminology 31: 327362.
Nagin, D.S. and Smith, D.A. (1990). Participation in and frequency of delinquent behavior: A test
for structural differences. Journal of Quantitative Criminology 6:335-356.

Nagin, D.S. and Tremblay, K.C. (1999). Trajectories of boys’ physical aggression, opposition,
and hyperactivity on the path to physically violent and nonviolent juvenile delinquency.
Child Development. 70: 1181-1196.
National Institute of Justice (2003). 2000 Arrestee Drug Abuse Monitoring: Annual Report. U.S.
Department of Justice, Office of Justice Programs.
Osgood, D.W. Foster, E.M., Flanagan, C., and Ruth, G. (2005). Introduction: Why focus on the
transition to adulthood for vulnerable populations? In D. Wayne Osgood, E. Michael Foster,
Constance Flanagan, and Gretchen R. Cusick (Eds.) On Your Own Without a Net: The
Transition to Adulthood for Vulnerable Populations (pp. 1-26) Chicago: University of
Chicago Press.
Osgood, D.W., Ruth, G.R., Eccles, J.S., Jacobs, J.E., and Barber, B.L. (2005). Six paths to
adulthood: Fast starters, parents without careers, educated partners, educated singles, working
singles, and slow starters. In F. Furstenberg, R. Rumbaut, R. Setterstein (Eds.), On the
frontier of adulthood (pp. 320-355). Chicago, IL: The University of Chicago Press.
Paternoster, R., Brame, R., Mazerolle, P., and Piquero, A. (1998). Using the correct statistical
test for the equality of regression coefficients. Criminology, 36 (4): 859-866.

81
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Perry, B. (2006). Understanding social network disruption: The case of youth in foster care.
Social Problems, 53(3), 371-391.
Piquero, A., Brame, R., Mazerolle, P., and Haapanen, R. (2002). Crime in emerging adulthood.
Criminology, 40(1), 137-170.
Derezotes, D., Poertner, J., & Testa, M. (2004). Race Matters in Child Welfare: The
Overrepresentation of African-American Children in the System. Washington DC: Child
Welfare League of America.
Runyan, D.K., and Gould, C.L. (1985). Foster care for child maltreatment: Impact on
delinquency. Pediatrics, 75(3), 562-566.
Ryan, J.P., Testa, M.F., and Zhai, F. (2008). African American males in foster care and the risk
of delinquency: The value of social bonds and permanence. Child Welfare 87(1): 115-140.
Sameroff, A.J., Barto, W.T., Baldwin, A., Baldwin, C., and Seifer, R. (1998). Family and child
influences on the development of competence. In M. Lewis and C. Ferring (Eds.), Families,
risk and competence (pp. 161-186). Mahwua, NJ: Lawrence Erlbaum.
Sampson, R.J. and Laub, J.H. (1990). Crime and deviance over the life course: The salience of
adult social bonds. American Sociological Review, 55, 609-627.
Sampson, R.J. and Laub, J.H. (1993). Crime in the Making: Pathways and Turning Points
through Life. Cambridge, MA: Harvard University Press.
Schmidt, P., and Witte, A.D. (1988). Predicting recidivism using survival models. New York:
Springer-Verlag.
Schoeni, R., and Ross, K. (2005). Marital assistance from families during the transition to
adulthood. In D. Wayne Osgood, E.M. Foster, C., Flanagan, and G.R. Cusick. On Your Own
Without a Net (pp. 396-416). Chicago, IL: University of Chicago Press.
Shanahan, M. (2000). Pathways to adulthood in changing societies: Variability and mechanisms
in life course perspective. Annual Review of Sociology, 26, 667-92.
Sherbourne, C.D., and Stewart, A.L. (1991). The MOS Social Support Survey. Social Science
Medicine, 32 (6), 705-714.
Shyne, A.W. and Schroeder, A.G. (1978). National Study of Social Services to Children and their
Families. Washington D.C.: National Center for Child Advocacy, U.S. Department of Health,
Welfare, and Education.
Smith, C. and Thornberry, T. (1995). The relationship between childhood maltreatment and
adolescent involvement in delinquency. Criminology, 33, 451-481.

82
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

Stehno, S.M. (1982). Differential treatment of minority children in service systems. Social Work
27: 39-45.
Testa, Mark. 1997. Kinship foster care in Illinois. In Jill Duerr Berrick, Richard P. Barth, and
Neil Gilbert (Eds). Child Welfare Research Review Vol. II (101–29). New York: Columbia
University Press.
U.S. Department of Health and Human Services. (2003). The AFCARS Report: Preliminary
FY2001 Estimates as of March 2003.
United States General Accounting Office (GAO) (1999). Foster care effectiveness of
independent living services unknown. Report No. GAO/HEHS-00-13. Washington, D.C.:
U.S. General Accounting Office.
Vermunt, J. K. and Magidson, J. (2005), Technical Guide for Latent GOLD 4.0: Basic and
Advanced, Belmont Massachussetts: Statistical Innovations Inc.
Visher, C. A., Lattimore, P.K., and Linster, R.L. (1991). Predicting the recidivism of serious
youthful offenders using survival models. Criminology, 35(4), 561-576.
Wertheimer, R. (2002). Aging Out of Foster Care: Young Adults with Special Needs.
Washington D.C.: Child Trends.
Widom, C.S. (1989). The cycle of violence. Science, 244,160-166.
__________. (1991). The role of placement experiences in mediating the criminal consequences
of early childhood victimization. American Journal of Orthopsychiatry. 61: 195-209.
World Health Organization (1998). The Composite International Diagnostic Interview (CIDI).
Geneva, Switzerland.
Wulczyn, Fred and Kristen Brunner Hislop. (2001). Children in Substitute Care at Age
16: Selected findings from the Multistate Foster Care Data Archive. (Unpublished
manuscript) Chapin Hall Center for Children. University of Chicago, April 2, 2001.
Wulczyn, F., Kogan, J. and Harden, B.J. (2003). Placement stability and movement trajectories.
Social Service Review 77: 212-236.
Zimmerman, R. (1982). Foster care in retrospect. New Orleans: School of Social Work, Tulane
University.

83
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

84
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.