Children and Youth Services Review

26 (2004) 63–91

Childhood risk and protective factors

and late adolescent adjustment in
inner city minority youth
Paul R. Smokowskia,*, Emily A. Mannb, Arthur J. Reynoldsc,
Mark W. Frasera
a
UNC-CH School of Social Work, University of North Carolina at Chapel Hill, CB 噛3550,
301 Pittsboro St., Chapel Hill, NC 27599-3550, USA
b
Graduate School of Education, Harvard University, USA
c
Waisman Center and School of Social Work, University of Wisconsin-Madison, USA

Abstract

This investigation examined longitudinal relationships among childhood risk and protective
factors and academic, social, and mental health outcomes in late adolescence. Data were
drawn from the Chicago Longitudinal Study, a research project that has tracked a cohort of
1539 impoverished inner-city youth from birth to young adulthood. An ecological model
containing information on child characteristics, family processes, early childhood intervention
program participation, and middle childhood indicators of competence and problems was
used to predict depression, juvenile delinquency, and high school or GED completion during
late adolescence or young adulthood. Multivariate negative binomial and logistic regression
analyses were used to show that cumulative family risk from birth to age 12 significantly
predicted increases in juvenile court petitions and decreases in high school or GED
completion. Early childhood intervention in preschool had the widest ranging protective
effects on all three adolescent outcomes. The probability of high school or GED completion
was significantly increased by preschool intervention, by parent(s) participating in the child’s
early elementary school experiences, by satisfactory elementary school grades, and by the
child’s ability to be task oriented. Being female, participating in preschool intervention,
displaying shy or anxious behavior, and having higher grades in middle school predicted
lower rates of juvenile court involvement while acting out behavior in middle school
increased court involvement rates. Preschool intervention, peer social skills, early classroom
adjustment, and shy or anxious behavior in middle school were protective factors against

*Corresponding author.
E-mail address: smokowsk@email.unc.edu (P.R. Smokowski).

0190-7409/04/$ - see front matter 䊚 2003 Elsevier B.V. All rights reserved.
doi: 1 0 . 1 0 1 6 / j . c h i l d y o u t h . 2 0 0 3 . 1 1 . 0 0 3
64 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91

adolescent depression while being female and having higher grades in early elementary
school were associated with higher rates of adolescent depression. Implications for social
work practice and future research were discussed.
䊚 2003 Elsevier B.V. All rights reserved.

Keywords: Early childhood intervention; Protective factors; Child development; Childhood risk; Middle
school; Poverty; Preschool intervention; Longitudinal research

1. Introduction

Over the past three decades, research on childhood risk and resilience has
burgeoned. The foundation for resilience research was formed from seminal studies
that examined enigmatic groups of individuals who were able to ‘overcome the
odds’ by maintaining adaptive functioning in the wake of various risk situations
(Anthony, 1974; Farber & Egeland, 1987; Cowen, Wyman, Work & Parker, 1990;
Masten, Best & Garmezy, 1990; Werner & Smith, 1992; Luthar, Doernberger &
Zigler, 1993 among others). Risk and resilience research has also benefited from
integrative literature reviews, which have critiqued the conceptual and methodolog-
ical techniques used in this area of inquiry (Rutter, 1987; Luthar & Zigler, 1991;
Masten & Coatsworth, 1995, 1998; Luthar, Cicchetti & Becker, 2000; Fraser, Kirby
& Smokowski, In Press). Taken together, these major studies and reviews of the
field have helped promote important advances, such as moving from concepts of
‘invulnerability’ to thresholds of resilience, distinguishing resilience which is
influenced by dynamic environmental transactions from ‘ego-resiliency’ which is a
personality trait, considering different categories of chronic vs. transitory risk factors,
examining resilience outcomes as domain-specific, and calling for specific attention
to be paid to mapping protective mechanisms.
Notwithstanding the extensive body of knowledge that has developed on resil-
ience, important questions remain unanswered. First, there is a serious lack of
studies investigating adaptive or resilient development in minority youth (Winfield,
1995; Luthar, 1997). Developmental models of minority and disadvantaged groups
commonly focus on negative outcomes, shedding little light on adaptive outcomes
and protective processes (McLoyd, 1990a,b; Nettles & Pleck, 1993). It is unclear,
for example, how risk and protective factors, which have been mapped for Anglo
children, function for African American youth growing up in impoverished inner-
city neighborhoods. Further, many of the studies that have been done on minority
youth rely on race comparative research designs. These designs have been criticized
for propagating a deficit perspective and for emphasizing a cultural equivalence
model that implicitly accepts the notion that some cultural styles are better than
others (Luthar, 1993; Cauce, 1995; Cauce, Ryan & Grove, 1998). Thus, comparative
studies of development that examine one racial or ethnic group relative to others
may compromise our knowledge of adaptive development which occurs within each
specific racial or minority group.
Second, few studies of resilience in adolescence have been conducted (Masten,
1994). Even more rare are ‘prospective studies linking multiple aspects of adaptation,
 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91 65

cumulative adversity exposure, and multiple resourceyprotective factors, particularly

over long time intervals’ (Masten et al., 1999, p. 144). Researchers are just
beginning to include multiple indicators of resilience in their investigations. This is
important because there is controversy in the resilience research literature concerning
how domains of resilient functioning interrelate. Only a few studies have integrated
multiple global and domain-specific indicators of resilience (see e.g. Reynolds,
1998; Masten et al., 1999). Thus, our knowledge base is limited concerning how
academic, social conduct, and psychological domains of resilience are associated
and if these disparate domains have unique risk and protective processes. These
issues are examined in more detail below.

1.1. Cumulative family risk

Risk factors are any variables that increase the probability of onset, exacerbation,
or maintenance of a problem condition (Fraser et al., In Press). Risk factors can
arise from any ecological level; however, it is commonly thought that distal risk,
such as neighborhood poverty, is largely mediated through family processes that are
closer to the individual (McLoyd, 1990b, 1998; Duncan & Raudenbush, 2001).
Risk factors may exert strong effects in some settings and at some times and weak
effects in other settings and times (Booth & Crouter, 2001). Consequently, there is
an important and complex set of interactions among risk factors, context, timing of
onset during development, and the length or duration of the risk (see e.g. Elder,
1974y1999).
As the number of risk factors increases, the accumulation appears to exert an
increasingly strong influence on children (Seifer, Sameroff, Baldwin & Baldwin,
1992; Garmezy, 1993; Dishion, Capaldi & Yoerger, 1999; Greenberg, Speltz,
DeKlyen & Jones, 2001; Rutter, 2001). Rutter (1979) found that the presence of a
single family stressor had a negligible effect on the rate of psychiatric disorder
among children. The presence of two or more risk factors, however, multiplicatively
increased the rate of disorder among children. In a longitudinal study, Fergusson
and Lynskey (1996) developed an index of family stress and then, using this index,
they compared youths at age 15 on measures of delinquency, substance abuse, and
other social problems. The relationship between the number of family stressors and
the presence of social problems was positive but non-linear. One or two family
stressors seemed to make little difference, but several created high odds for serious
behavioral problems. Similarly, in a multi-state study of 78 710 children in 6th
through 12th grades, Pollard, Hawkins and Arthur (1999), p. 151) found substance
abuse, school problems, and delinquency to be strongly related to risk exposure,
with ‘steep increases in prevalence associated with the highest levels of risk.’ Thus,
the effect of exposure to several risk factors may not simply be additive. Although
the effect of a single stressor may be negligible, the effect of three or four stressors
may be far greater than a three-fold or four-fold increase in vulnerability. Children
exposed to the least cumulative risk appear to fare the best. Those children with the
highest levels of exposure are the least likely to produce adaptive behavior (Pollard
et al., 1999; Rutter, 2000). Two questions concerning the functioning of risk factors
66 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91

require further clarification. First, are specific single risk factors more potent than
others, thus contributing relatively more to the deleterious effect of accumulated
risk? Few investigators have examined the effects of single risk factors and
cumulated risk indices in the same analysis. Even fewer have done so across
multiple adjustment domains. Second, does accumulated risk have a linear or non-
linear impact on adjustment outcomes and does this relationship vary across
adjustment domains and in different development stages? The second question
provides a transition to examine the developmental dynamics of different adjustment
domains.

1.2. Relationships among adjustment domains

Researchers define competence as successfully achieving developmental tasks

that are considered normative for a particular age and environmental context
(Garmezy & Masten, 1991; Masten & Coatsworth, 1995, 1998). Adequate coping
is not necessarily accompanied by high levels of adaptation within all spheres of
functioning (Luthar & Zigler, 1991; Garmezy, 1993). Being resilient or competent
does not imply that children survive without pain or anxiety. Significant psycholog-
ical distress has been observed in children who were behaviorally competent (Farber
& Egeland, 1987; Luthar, 1991; Luthar et al., 2000). Similarly, definitions of
resilience based on competence in a single functional domain – for example,
academic success – may give a false impression, because competence in one area
represents only a small part of what may be considered successful coping (Luthar
et al., 1993).
Research findings mapping the linkages among domains of adjustment have been
inconclusive. For example in the Kauai study, Werner and Smith (1992) found that
many ‘invulnerable’ children had painful memories, nightmares, and other adjust-
ment problems. For high risk urban adolescents, Luthar (1991) reported high levels
of internal distress in individuals who were externally highly competent in academic
accomplishments and social behavior. Mental health symptom levels reported by
these resilient youth were not significantly different from symptom levels in high-
stress low competence youth. Likewise, in a study of 150 ninth-grade youth from
inner city schools, Luthar et al. (1993) found that high performing, resilient children
had depressive symptoms comparable to low-performing, highly stressed youths.
Luthar (1997) once again found highly-stressed but behaviorally competent youth
displaying vulnerability to emotional distress. Over 70% of youth who showed
competence in one domain also displayed significant problems in another sphere of
functioning. Specifically, she wrote, ‘bright inner-city teens appeared to do very
well in the presence of positive psychosocial forces. However, in the presence of
negative forces, they seemed to be hurt (e.g. emotionally distressed) more than did
their less intelligent peers’. (Luthar, 1997, p. 465, italics added). Thus, academically
competent children and adolescents may be internalizing symptoms, causing them
to be emotionally troubled.
Academic competence may have complex interrelations with social functioning
for disadvantaged minority children. For example, Cauce, Felner and Primavera
 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91 67

(1982) found that high levels of informal social support were negatively associated
with academic performance in disadvantaged inner-city males. Luthar (1995) also
found inner-city peer group integration was associated with declines in academic
functioning, lending support to the negative association between academics and
social functioning in disadvantaged environments. However, after studying 205
children from an urban area, Masten et al. (1999) reported that high functioning
children who had high levels of exposure to adversity differed little from high
functioning children who had low levels of exposure to adversity. Clearly, more
research is needed to determine how adaptation domains are interrelated.

1.3. Protective factors

Protective factors are internal and external resources that modify or buffer the
impact of risk factors (Rutter, 1987; Luthar et al., 2000; Fraser et al., in press).
Like risk factors, protective factors may be domain specific. That is, they may
operate principally within specific domains of development and have limited
spillover to other domains. While some protective factors, such as self-efficacy,
appear to have a widespread effect on functioning, other protective factors are more
specific. For instance, academic achievement, one of the most prominent develop-
mental outcomes through childhood and adolescence, appears to be promoted by
having more individual resources and social capital (Coleman, 1988). Competence
in the academic domain appears to be influenced by IQ, motivation to succeed,
beliefs in one’s abilities, and positive attitudes about school (Masten & Coatsworth,
1998). Cognitive patterns attributing success to hard work and effort (and failure as
the lack of hard work) also facilitate academic achievement by fostering an internal
sense of control over personal achievement (Stevenson, Chen & Lee, 1993). In
contrast, socioemotional functioning, another important developmental outcome has
been associated with higher IQ and positive academic achievement (Hartup, 1983;
Masten & Coatsworth, 1995). However, factors that promote adaptive development
in this domain appear to be more rooted in environmental interactions with parents,
teachers and peers (Masten & Coatsworth, 1998).
Three broad categories of protective variables have been found to promote
resilience in childhood (Garmezy, 1985). The first refers to individual dispositional
attributes, including temperamental factors, social orientation and responsiveness to
change, cognitive abilities, and coping skills. The second general category of
protective factors is the family milieu. A positive relationship with at least one
parent or a parental figure serves an important protective function. Other important
family variables include cohesion, warmth, harmony, supervision, and absence of
neglect. The third category of protective influences in childhood encompasses
attributes of the extrafamilial social environment. These include the availability of
external resources and extended social supports as well as the individual’s use of
those resources. The two most prominent predictors of resilience throughout
childhood and adolescence are having a strong prosocial relationship with at least
one caring adult and having good intellectual capabilities (Werner & Smith, 1982;
Rutter, 1990; Masten & Coatsworth, 1998).
68 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91

Protective factors have been examined in two ways – as additive models and in
interactive models. Additive models, in which protective factors are said to exhibit
main or compensatory effects, suggest that the presence of a risk factor directly
increases the likelihood of a negative outcome and the presence of a protective
factor directly increases the likelihood of a positive outcome (Masten, 1987;
Pellegrini, 1990; Luthar, 1991). Risk and protection are often seen as polar opposites
with protective factors promoting normative developmental outcomes (for an
exception, see Ladd & Burgess, 2001). In interactive models, protective factors
have effect only in combination with risk factors. In other words, protective factors
are thought to exert little effect when stress is low. Their effect emerges when stress
is high (Masten, 1987). This protective factor modeling approach utilizes statistical
interaction terms. Because statistical interaction terms can be unstable and difficult
to replicate (Rutter, 1987; Luthar & Zigler, 1991; Luthar, 1993), additive or main
effects models are more common in resilience research.
Reynolds (1998) provided an useful example of a main effect examination of
risk and protective factors and their differential impact across multiple outcome
domains. He examined middle childhood resilience in the Chicago Longitudinal
Study using a sample of 1170 economically disadvantaged, inner city African
American children. He investigated academic, social, and total (academic plus
social) resilience during middle childhood as dichotomous outcomes coded from
academic tests and teacher ratings. In order to account for variability in risk
exposure, Reynolds used a risk index as a covariate in his analyses. Measuring sixth
grade competence outcomes, he found Child Parent Center intervention was
significantly associated with scholastic and total, but not social, resilience. Gender,
parental expectations, early classroom adjustment, and early academic achievement
were found to be important predictors of sixth grade resilience. Path modeling
revealed that early childhood intervention participation and parent expectations
promoted middle childhood academic and social resilience outcomes and mediated
the effects of cumulative risk. The current study builds upon Reynolds (1998)
investigation by examining longitudinal risk and resilience processes from childhood
into late adolescence in the same sample of high risk, minority, inner city children.
To summarize, our understanding of resilience has grown substantially over the
past three decades. At the same time, more longitudinal research is needed on
minority children in diverse ecological contexts. The deleterious effect of cumulative
risk needs to be further examined. Longitudinal dynamics among adjustment domains
need to be assessed. Protective factors need to be more closely mapped; detailing
which of these factors cut across and which are unique to specific adjustment
domains.
The current study addressed these concerns by investigating academic, social, and
psychological adjustment from childhood to late adolescence in a sample of inner
city minority youth. The relationships among single risk factors, cumulative risk,
and an array of protective factors were examined to see which factors inhibited or
promoted functioning across different domains of functioning.
 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91 69

2. Research design and methods

2.1. Sample

Data for this investigation was drawn from the Chicago Longitudinal Study (CLS:
Reynolds, 1991, 1998; Smokowski, Reynolds & Bezrucko, 1999; Reynolds, Temple,
Robertson & Mann, 2001). The CLS follows a cohort of 1539 disadvantaged,
minority children (93% African American, 7% Latino or Other) who were born in
1980 and attended kindergarten programs within the Chicago Public School System
in 1985–1986. Out of the full sample of 1539, a subset of 989 children (64% of
the sample) received preschool services from one of Chicago’s 20 Child-Parent
Center (CPC) programs. An additional set of 550 children (36% of the sample) did
not attend CPC preschool and serve as a non-CPC comparison group. The full
sample represented 17 neighborhood areas and 25 schools, many of which were in
the most impoverished areas in Chicago.
The sample was evenly split between girls and boys. Approximately, 76% of the
children lived in neighborhoods where 60% or more of the residents were considered
to have low incomes. Table 1 presents information on family risk factors from birth
through adolescence.
The sample was clearly at high risk based on indicators of socioeconomic
disadvantage. For example, at birth 75% of the children were in single parent or
non-married families. Thirty-five percent of the children had teenage mothers and
40% of these parents had not graduated from high school. From age 8 to 17, family
risk for the entire sample was remarkably stable. Slightly more than four out of ten
children had parents who had not graduated from high school. Approximately, 80–
87% of the children were eligible for a free or reduced lunch at school, a commonly
used indicator of family socioeconomic status. From birth through adolescence,
more than 60% of the sample children lived in single or non-married parent homes
at any of the measured intervals. Finally, from age 8 to 12, at least 50% of parents
in the sample were unemployed. This percentage only decreased below 50% (to
43%) when children were age 17. These family risk indicators show that this sample
was indeed made up of high risk families. Adolescent outcomes also reflected this
disadvantage. At age 16, 45% of the sample with data showed signs of depression,
19% had juvenile court petitions filed against them, and only 61% of the sample
children had graduated from high school or had gotten a GED by age 22.

2.2. Measures of independent variables

Family risk – A variety of family risk indicators were collected from birth to age
17. Specifically, five dichotomous indicators of family risk were collected when the
child was age 8, 10, 12 and 17. These variables are shown in Table 1. They are;
(a) parent did not graduate from high school; (b) child was eligible for a free lunch
subsidy (during school-age data collection); (c) there were four or more children
in the family; (d) parent was not working full or part time; and (e) child lived in
single parent or non-married family. Data from birth records were also collected.
 70
Table 1
Descriptive statistics and correlations for family risk from birth to late adolescence

Spearman’s non-parametric correlation

P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91
Juvenile High school
court or GED
Depression petitions completion
Risk factor N Reporter1 Ageyyear Min Max Mean S.D. (Age 16) (-Age 18) (Age 22)
Birth to age 3
Cumulative risk factors – 0–3 1523 1–5 0–3y1980–1983 0.00 5.0 1.67 1.10 y0.038 0.074** y0.132***
N of missing risk factors 1539 1–5 0–3y1980–1983 0.00 6.0 0.426 1.11 y0.005 y0.123*** 0.080**
Parent is not a HS graduateq 1456 4 0y1980 0.00 1.0 0.404 0.491 0.043 0.089** y0.114***
4 or more children in homeq 1346 1, 2, 3, 4 0y1980 0.00 1.0 0.102 0.303 y0.056 y0.001 y0.039
Any indicated child neglect or 1408 3, 5 0–3y1980–1983 0.00 1.0 0.021 0.142 0.024 0.024 y0.043
abuse (DCFS or court)q
Child lives in single parent or 1456 3, 4 0y1980 0.00 1.0 0.751 0.432 0.053 0.087** y0.078**
non-married familyq
Low birthweight (-2500 g)q 1456 4 0y1980 0.00 1.0 0.124 0.330 y0.019 y0.017 y0.042
Mother is 19 years of age or 1456 4 0y1980 0.00 1.0 0.346 0.476 0.063 0.007 y0.045
less at child’s birthq
Child age 8
Cumulative risk factors age 8 1539 1, 2, 3, 4 8y1988 0.00 5.0 2.54 1.29 0.068* 0.152*** y0.148***
N of missing risk factors 1539 1, 2, 3, 4 8y1988 0.00 5.0 0.445 0.915 y0.008 y0.089*** y0.043
Parent is not a HS gradq 1510 1, 4 8y1988 0.00 1.0 0.434 0.496 0.043 0.084** y0.203***
4 or more children at homeq 1453 1, 2, 3, 4 8y1988 0.00 1.0 0.363 0.481 0.037 0.068** y0.095**
Any free lunch for childq 1265 1, 6 8y1988 0.00 1.0 0.875 0.331 0.043 0.068* y0.150***
Child lives in single parent or 1500 1, 2, 3 8y1988 0.00 1.0 0.602 0.490 y0.011 0.069** y0.050
non-married familyq
Parent reported not being 1282 1, 3 8y1988 0.00 1.0 0.559 0.497 0.054 0.051 y0.011
employed full or part time
Child age 10
Cumulative risk factors age 10 1539 1, 2, 3, 4 10y1990 0.00 5.0 2.58 1.33 0.069 0.167*** y0.185***
N of missing risk factors 1539 1, 2, 3, 4 10y1990 0.00 5.0 0.406 0.898 0.007 y0.081** y0.055*
Table 1 (Continued)
Spearman’s non-parametric correlation
Juvenile High school
court or GED

P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91
Depression petitions completion
Risk factor N Reporter1 Ageyyear Min Max Mean S.D. (Age 16) (-Age 18) (Age 22)
Parent reported not being 1299 1, 2, 3 10y1990 0.00 1.0 0.503 0.500 0.028 0.068* y0.059*
employed full or part time
Any free lunch for childq 1299 1, 6 10y1990 0.00 1.0 0.841 0.366 0.016 y0.051 y0.099***
Child lives in single parent or 1506 1, 2, 3 10y1990 0.00 1.0 0.653 0.476 0.036 0.098*** y0.071*
non-married familyq
Parent is not a HS graduateq 1513 1, 4 10y1990 0.00 1.0 0.454 0.498 0.088* 0.087** y0.199***
4 or more children in homeq 1453 1, 2, 3, 4 10y1990 0.00 1.00 0.377 0.485 y0.034 0.071** y0.105***
Child age 12
Cumulative risk factors 1539 1, 2, 3, 4 12y1992 0.00 5.0 2.71 1.30 0.058 0.160*** y0.183***
N of missing risk factors 1539 1, 2, 3, 4 12y1992 0.00 5.0 0.210 0.593 y0.020 y0.068* y0.084**
Parent is not a HS graduateq 1513 1, 4 12y1992 0.00 1.0 0.439 0.496 0.086* 0.066* y0.192***
Parent reported not beingq 1301 1, 2, 3 12y1992 0.00 1.0 0.514 0.500 0.031 0.076** y0.080*
employed full or part timeq
Any free lunchq 1323 1, 2, 6 12y1992 0.00 1.0 0.8360 0.370 0.000 0.064* y0.122***
Child lives in single parent or 1509 1, 4 12y1992 0.00 1.0 0.6779 0.467 0.024 0.106*** y0.067*
non-married familyq
4 or more children in homeq 1461 1, 2, 3, 4 12y1992 0.00 1.0 0.3470 0.476 0.008 0.084** y0.127***
Child age 17
Cumulative risk factors age 17 1539 1, 2, 3, 4 17y1997 0.00 5.0 2.44 1.33 – 0.119*** y0.150***
N of missing risk factors 1539 1, 2, 3, 4 17y1997 0.00 5.0 0.365 0.879 – y0.074** y0.061*
Parent is not a HS graduateq 1519 1, 4 17y1997 0.00 1.0 0.415 0.493 – 0.057* y0.211***
Parent reported not being 1307 1, 2, 3 17y1997 0.00 1.0 0.434 0.496 – 0.046 y0.081**
employed full or part timeq
Any free lunch for childq 1323 1, 2, 6 17y1997 0.00 1.0 0.792 0.406 – 0.070* y0.153***
Child lives in single parent or 1510 1, 4 17y1997 0.00 1.0 0.694 0.461 – 0.043 y0.031
non-married familyq
4 or more children in homeq 1474 1, 2, 3, 4 17y1997 0.00 1.0 0.314 0.464 – 0.099** y0.139***

71
 72
Table 1 (Continued)
Spearman’s non-parametric correlation
Juvenile High school
court or GED

P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91
Depression petitions completion
Risk factor N Reporter1 Ageyyear Min Max Mean S.D. (Age 16) (-Age 18) (Age 22)
Any child neglect or abuse 1539 3, 5 0–17y1980–1997 0.00 1.0 0.09 0.29 0.023 0.095*** y0.073**
(birth to 16)
Cumulative risk factors 1539 1, 2, 3, 4 0–12y1980–1992 0.00 20 9.34 4.34 0.063 0.166*** y0.195***
Birth to age 12
N of missing risk factors 1539 1, 2, 3, 4 0–12y1980–1992 0.00 20 1.66 3.42 y0.009 y0.097** y0.005
Birth to age 12
Adolescent Outcomes
Depression 801 2 16y1996 0.00 1.0 0.45 0.50 – 0.080* y0.113**
Juvenile arrest 1404 5 -18y1980–1998 0.00 1.0 0.21 0.41 – y0.329***
High school completion 1334 1, 2, 6 By 22y2002 0.00 1.0 0.62 0.49 –
qNon-parametric bivariate correlations: Spearman’s Rho – *P-0.05; **P-0.01; ***P-0.001.qincluded in risk index.
1
Note: 1sParent, 2sChild, 3sIDHHS: Illinois Department of Health and Human Service, 4sIDPH: Illinois Department of Public Health, 5sCCJJ:
Cook County Juvenile Justice, 6sChicago Public School system.
 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91 73

From birth to age 3, risk factors were slightly different. Parent education, family
size and single parent family status were identical. However, indicators of low birth
weight (below 2500 g), whether or not the child’s mother was 19 years old or less
at the child’s birth, and any report of child neglect or abuse to Child Protective
Services or the court system were also included.
Because the accumulation of risk factors may pose a stronger threat to functioning
than any particular risk factor on its own (Garmezy, 1993; Seifer et al., 1992;
Rutter, 2001), cumulative adversity present in the child’s life was measured using a
Family Risk Index. The Family Risk Index provided an additive scale score that
was the sum of the five dichotomously defined risk indicators collected when the
child was age 8, 10, 12 and 17. From birth to age 3, the scale had six indicators.
The five main indicators in this index have been similarly used in prior studies
(Rutter, 1987; Bendersky & Lewis, 1994; Reynolds, 1998). Scale indicators are
associated with lower developmental adjustment and decreased academic achieve-
ment (Natriello, McDill & Pallas, 1990). Shown in Table 1, the sample had an
average of approximately two and one half family risk factors at each of the school
age data collection intervals.
A comprehensive childhood family risk index was also calculated by adding the
number of family risk factors at each data collection interval from birth to age 12.
This resulted in a scale with a range from 0 to 20, a mean of 9.34 and a standard
deviation of 4.3. Childhood family risk was normally distributed in the sample.
Child attributes and characteristics that were used as independent variables in
this study are listed in Table 2. Gender was a dichotomous variable coded 0 for
males and 1 for females. The child’s early academic achievement was measured by
averaging reading and mathematics scores on the Iowa Test of basic skills. Scores
from grades 1, 3 and 6 were used as three separate continuous variables. Scores
were measured as ‘grade equivalents’. Scoring 3.8 or above, for e.g. demonstrated
performance equal to the 8th month of third grade. ITBS tests were typically
administered in the 8th month of the school year.
Early perceived school competence was measured by a 10 item child self
perception scale administered during grades 5 and 6. Internal consistency reliability
for the scale was 0.74.
Early classroom adjustment was measured by teacher ratings of the child’s
socioemotional maturity. The scale is the sum of six items rated from poor (1) to
excellent (5). Items were: concentrates on work, follows directions, is self confident,
gets along well with others, participates in group discussions and takes responsibility
for actions. Internal consistency reliability for the scale was 0.94. The scale was
administered yearly from grade 1 to grade 6. For a longitudinal indicator of
childhood adjustment, a continuous variable was created that assessed the number
of years teachers rated the child’s classroom adjustment average or above from
grades 1 to 6.
Middle childhood competencies and problems were assessed using the Teacher-
Child Rating Scale (TCRS; Hightower, Spinnell & Lotyczewski, 1989). The T-CRS
includes 20 competence items (rated from 1, ‘not at all’ to 5, ‘very well’) that
create four subscales, measuring socio-emotional adjustment in the areas of frustra-
 74
Table 2
Protective factors and child characteristics from birth to late adolescence

Spearman’s non-parametric correlation

P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91
Juvenile High school
Depression arrest or GED completion
N Reporter Ageyyear Min Max Mean S.D. (Age 16) (-Age 18) (Age 22)
Protective factors
Gender (1sFemale) 1539 Parent 0y1980 0 1 0.50 0.50 0.096** y0.316*** 0.165***
Child parent center 1539 CPS 3–9y1983–1985 0 1 0.64 0.48 y0.066 y0.094** 0.112***
preschool participation
Average of child’s math 1327 CPS 6y1986 y0.90 4.05 1.56 0.673 y0.025 y0.162*** 0.262***
and reading scores in Grade 1
Average of child’s math and 1299 CPS 8y1988 y0.80 6.15 3.05 1.00 0.022 y0.172*** 0.306***
reading scores in Grade 3
Average of child’s math and 1208 CPS 11y1991 1.30 9.45 5.20 1.33 y0.005 y0.231*** 0.358***
reading scores in Grade 6
Number of years teachers rated 1539 CPS-teachers 6–11y1986–1991 0 6 1.84 1.52 y0.038 y0.125*** 0.265***
parent participation in school
average or above (Grades 1–6)
Number of years teachers rated 1539 Teachers 6–11y1986–1991 0 6 2.24 1.73 y0.053 y0.148*** 0.305***
child’s classroom adjustment
Average or above (Grades 1–6)
Child’s self perception of competence 1054 Child 8–10y1988–1990 15 48 35.87 5.12 y0.074 y0.135*** 0.140***
in school (Grades 5–6)
Grade 6–7 child characteristics
Peer social skillsa 1058 Teachers 12–13y1991–1992 y2.7 1.9 0 1.0 y0.137*** y0.197*** 0.259***
Shy or anxious behaviora 1058 Teachers 12–13y1991–1992 y1.0 4.5 0 1.0 y0.016 0.076* y0.165***
Assertiveness skillsa 1058 Teachers 12–13y1991–1992 y2.4 2.2 0 1.0 y0.023 y0.109*** 0.202***
Table 2 (Continued)
Spearman’s non-parametric correlation
Juvenile High school
Depression arrest or GED completion

P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91
N Reporter Ageyyear Min Max Mean S.D. (Age 16) (-Age 18) (Age 22)
a
Task orientation 1058 Teachers 12–13y1991–1992 y1.8 2.0 0 1.0 y0.051 y0.259*** 0.309***
Acting out behaviora 1058 Teachers 12–13y1991–1992 y1.0 2.8 0 1.0 0.114** 0.320*** y0.303***
Learning Problemsa 1058 Teachers 12–13y1991–1992 y1.3 2.3 0 1.0 0.039 0.283*** y0.321***
Frustration tolerancea 1057 Teachers 12–13y1991–1992 y2.0 2.3 0 1.0 y0.117** y0.252*** 0.277***
Child’s total competence 1058 Teachers 12–13y1991–1992 y2.6 2.5 60.75 15.75 y0.087* y0.243*** 0.308***
Child’s total problems 1058 Teachers 12–13y1991–1992 y1.3 3.3 37.03 14.33 0.064 0.292*** y0.327***
a
sStandardized Scores; *P-0.05; **P-0.01; ***P-0.001; CPSsChicago Public School system.

75
76 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91

tion tolerance, task orientation, assertiveness skills and peer social skills. The T-
CRS includes 18 problems items (scored from 1 ‘not a problem’ to 5 ‘very serious
problem’) with subscales assessing acting-out behavior, shy-anxious behavior, and
learning problems. Internal consistency reliabilities average over 0.90. Teachers
filled out the T-CRS in grades 6 and 7. To minimize missing data, TCRS subscale
scores from grades 6 and 7 were standardized and averaged.
Parental involvement in elementary school was measured by the number of years
teachers rated parents’ participation in school activities as average or above. Data
were collected from teacher surveys administered during grades one to six.
Early childhood intervention was measured by participation in one of Chicago’s
Child Parent Center programs. Program participation was scored as a dichotomous
variable – whether or not the child attended a CPC preschool. The sample sizes for
preschool participation vary by outcome domain. For adolescent depression, 542
children (68%) attended CPC preschool and 259 (32%) did not. For juvenile court
involvement, 911 (65%) children attended CPC preschool and 493 (35%) did not.
Finally, for high school or GED completion, 869 (65%) children attended CPC
preschool and 465 (35%) did not. For each domain, the percentage of participants
vs. non-participants closely approximates the percentages in the full sample (65 vs.
35%).

2.3. Measures of dependent variables

Three dependent variables were selected to assess key indicators of educational

attainment, social functioning, and mental health in adolescence and early adulthood.
Graduation from high school or receiving a General Education Diploma (GED) by
age 22 data collection was a dichotomous outcome that was considered a key
indicator of educational attainment. The number of juvenile court petitions filed
against the adolescent was a highly skewed continuous variable (e.g. most adoles-
cents had none) and was considered to be a key indicator of poor social functioning
or low rule abiding behavior. Finally, adolescent depression was assessed during
grade 10, approximately age 16, using three ‘yes’ or ‘no’ items from a student
survey. The three items were: ‘In the past year, I have felt really down about life in
general’, ‘In the past year, I have felt seriously depressed’, and ‘In the past year, I
have felt hopeless about the future’. Internal consistency reliability for this scale
was 0.61. This was also a highly skewed continuous variable. Most students
answered ‘no’ to all three of these questions. Answering ‘yes’ to any of these
questions was considered to be an indication of possible depression. No clinical
diagnosis of major depression was implied, but rather this indicator was used as a
rough estimate of mental health functioning. Descriptive statistics for the three
dependent variables are provided at the end of Table 1.

2.4. Data analyses

Missing data analyses were conducted to examine whether sample participants

with available data were similar to sample participants with missing data. 1406 out
 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91 77

of the 1539 sample participants (91%) had juvenile arrest data. For high school or
GED completion, 1334 of the 1539 sample participants had data and 205 (13%)
were lost to attrition. Adolescent depression had the most serious problem with
missing data. This dependent variable was derived from a survey sent to adolescents
in the sample during 10th grade. Of the 1539 surveys sent out, 801 (52%) were
returned. In order to determine if the sample participants missing on adolescent
depression, juvenile arrest, and high school completion were different from the rest
of the sample, dichotomous missing data indicators were computed for adolescent
depression, juvenile arrest, and high school or GED completion. Chi-square and T-
tests were performed using the missing vs. not missing indicator as the grouping
variables and childhood risk and protective factors as the test variables.
Bivariate correlational analyses were used to begin to examine relationships
between childhood risk and protective factors and adolescent adjustment outcomes.
Family structure risk factors from birth to age 17 were examined as individual
factors, as aggregated risk indices from each time period (birth to age 3, 8, 10, 12
and 17), and as a single childhood risk index that was the summation of the
individual risk factors from birth to age 12. Correlational analyses specifically
evaluated if cumulated family risk displayed a stronger negative association with
adjustment outcomes than individual family risk factors did on their own.
Cumulative family risk was further examined in the second stage of the study’s
analyses. This second stage of data analyses sought to determine if cumulative risk
had a linear or non-linear impact on childhood and adolescent adjustment outcomes.
The cumulative childhood risk index was broken into thirds to identify children
with low, moderate, and high levels of family structure risk. As shown in Table 1,
the childhood risk index had a range of 0–20 with a mean of 9.3 and a standard
deviation of 4.3. Children with fewer than 8 family structure risk factors were
considered low risk. Those with 8–11 were considered to have experienced moderate
risk. Children with 12 or more family structure risk factors were considered to have
had high or severe levels of family risk. These high risk children spent nearly all
of their childhoods in large families with single or unmarried parents, who did not
finish high school, and who were chronically (rather than episodically) unemployed.
These risk factors are consistently associated with low socioeconomic status and
appear to influence parenting practices and child development by raising parent
stress (McLoyd, 1990b, 1998). A series of univariate ANOVAs were run to see if
there were differences in middle childhood and adolescent outcomes based on these
levels of risk. High school or GED completion rate was tested with a Chi-square
test because it was a dichotomous outcome cross-classified with a categorical
variable.

2.5. Multivariate modeling of risk and protective factors

Finally, stepwise regression analyses were used to see which individual, family,
and program participation variables predicted adolescent adjustment outcomes in
multivariate models. Independent variables were individually added to the model in
blocks to examine mediation processes. High school or GED completion was a
78 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91

dichotomous outcome, making logistic regression the optimal analytic strategy. This
analysis was conducted using SPSS 11.0.
Adolescent depression and juvenile arrest records were count data. Both of these
outcomes were highly skewed, continuous variables. To examine delinquency
prediction using the number of delinquency petitions and depression using the
number of positive answers to depression survey items, a negative binomial
regression model was estimated. This method of data analysis has previously been
used in social science research to model the frequency of delinquency (Nagin &
Land, 1993). The use of a negative binomial model has been recommended for
highly skewed dependent variables where the variance of the dependent variable is
sufficiently larger than this variable’s mean. While the negative binomial regression
model assumes a poisson distribution and characterizes data as a poisson model
would, the negative binomial model allows for the probability of Y to be unequal
among the study sample (Land, McCall & Nagin, 1996). In the example of juvenile
delinquency, individuals who have received one or more petitions to the juvenile
court may be more likely to experience further arrests than individuals who have
never had a juvenile arrest. The use of a negative binomial model is more
conservative than the use of a conventional possion model, which may decrease the
standard error estimates and overstate statistical significance of certain factors (Land
et al., 1996). To further specify this assumption, a zero-inflated probability (ZIP)
version of the negative binomial model was estimated. The ZIP model allows for
the possibility that the individuals with zero arrests are fundamentally different from
those with one or more arrests. This analysis was conducted using the statistical
package STATA.

3. Results

3.1. Missing data analyses

Chi-square tests were used to cross-classify the dichotomous missing data indicator
and dichotomous family risk variables. T-tests were used with continuous cumulative
family risk or protective factor variables. These tests indicated that sample partici-
pants with missing data on adolescent depression had significantly more cumulative
family risk at ages 8, 10, 12 and 17. Compared to sample participants with
depression data, more of these children came from single parent homes with parents
who did not finish high school. Fewer of these children had attended CPC preschool.
They had lower test scores in grades 1, 3 and 6, less total competence and more
total problems in grades 6–7. In adolescence, sample participants with missing data
on depression had more arrests and more of them had dropped out of high school.
Clearly, adolescents who returned the 10th grade survey were more advantaged than
those who did not. Consequently, results for adolescent depression may be under-
estimated because the highest risk adolescents were less likely to be included in the
analyses, lowering the variation in the sample.
High school or GED completion had a different profile of missing vs. not missing
data. Most of the individual family risk factor comparisons revealed no significant
 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91 79

differences between missing and not missing groups. Sample participants who did
not have high school or GED completion data were more likely to have four or
more children in their homes and more indicated reports of child abuse or neglect
from birth to age 3. As a group, they also had significantly lower averages on
cumulative family risk at ages 8, 10, 12 and 17. While this indicated that they
experienced less cumulative family risk, they also had significantly less parent
participation in elementary school, lower ratings of classroom adjustment, and less
total competence in grades 6–7. Compared with those participants with data, fewer
of the sample participants with missing data on high school or GED completion
had attended CPC preschool. The implications of this pattern of differences are not
immediately clear. Participants with missing data on high school or GED completion
experienced less cumulative family risk and, at the same time, had fewer childhood
protective factors.
Compared to participants with data, missing cases on juvenile arrest experienced
significantly fewer family risk factors, especially single parent or non-married
household status, number of children in the home, and cumulative family risk. Once
again, this loss of lower risk cases may cut down the variance in the sample,
rendering results that are underestimated. Detailed tables for the missing data
analyses are available upon request from the first author.

3.2. Bivariate correlational analyses

Non-parametric correlation coefficients are shown in Table 1. Non-parametric

correlations revealed few significant associations between adolescent depression and
family risk factors. All of the correlations were below 0.10. The only relationship
that reached statistical significance was between the parent not graduating from high
school by the time the child was age 10 (rs0.088, P-0.05) or age 12 (rs0.086,
P-0.05) and subsequent adolescent depression at age 16. The only cumulative risk
index that showed a statistically significant association with adolescent depression
was family cumulative risk at age 8 (rs0.068, P-0.05). Cumulative family risk
from birth to age 12 did not have a significant correlation with adolescent depression.
Family risk factors displayed stronger relationships with juvenile court petitions.
For parsimony, only correlations above 0.10 will be discussed. However, all the
correlations are presented in Table 1. The strongest relationships that surfaced were
between juvenile court petitions and cumulative family risk when the child was age
8 (rs0.152, P-0.001), age 10 (rs0.167, P-0.001), age 12 (rs0.160, P-
0.001), and age 17 (rs0.119, P-0.001). Cumulative family risk from birth to age
12 displayed a statistically significant association with juvenile court petitions (rs
0.166, P-0.001).
The associations between high school completion and family risk factors revealed
a different pattern of relationships. Cumulative family risk indicators at the different
time intervals had statistically significant, negative correlations with high school or
GED completion. However, the strongest and highly consistent correlation across
time intervals was between the parent not having completed high school and
adolescent high school completion (e.g. age 8, rsy0.203, P-0.001 or age 17,
80 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91

rsy0.211, P-0.001). The association between cumulative family risk from birth
to age 12 and high school or GED completion was nearly as strong (rsy0.195,
P-0.001).
Correlations between the adolescent outcome variables are provided at the end of
Table 1. The strongest association was between high school or GED completion
and juvenile court petitions (rsy0.329, P-0.001). Adolescent depression and
high school or GED completion were negatively associated (rsy113, P-0.01)
and adolescent depression and juvenile court petitions were positively associated
(rs0.080, P-0.05).
Non-parametric correlations between childhood protective factors and adolescent
outcomes are provided in Table 2. Bivariate relationships between child character-
istics and adolescent depression were stronger than the associations between family
risk factors and adolescent depression. Adolescent depression was associated with
being female (rs0.096, P-0.01), having lower levels of peer social skills (rs
y0.137, P-0.001), frustration tolerance (rsy0.117, P-0.001) and displaying
higher levels of acting out behavior (rs0.114, P-0.001) in grades 6–7.
Juvenile court petitions were associated with all of the protective factors and
child characteristics listed in Table 2. The strongest relationships were between
juvenile court petitions and acting out behavior (rs0.320, P-0.001), learning
problems (rs0.283, P-0.001), and total problem behavior (rs0.292, P-0.001)
in grades 6–7.
High school or GED completion also had strong relationships with protective
factors and child characteristics. The strongest relationships were between high
school or GED completion and childhood standardized test scores from 3rd and 6th
grades (rs0.306, P-0.001 and rs0.358, P-0.001, respectively). Acting out
behavior, learning problems, and total problem behavior in grades 6–7 all had
statistically significant correlations above y0.3 with high school or GED completion.
Teacher ratings of the child’s adjustment in school from grades 1–6, task
orientation, and total competence in grades 6–7 all had positive associations above
0.3 (P-0.001) with high school or GED completion.

3.3. Cumulative family risk

Results examining cumulative family risk as a categorical variable are presented

in Table 3. To guard against the risk of making a Type I error from running multiple
tests, only results that were statistically significant at the P-0.01 level were
reported.
There were statistically significant differences between the mean scores for the
three cumulative risk levels on most of the child characteristics from grades 6 to 7
(P-0.01). Specifically, low risk children had the lowest scores and high risk
children had the highest scores on shy or anxious behavior, learning problems, and
T-CRS ratings of total problems. Conversely, the low risk children had the highest
scores and high risk children had the lowest scores on peer social skills, task
orientation, and T-CRS ratings of total competence. High risk children also had
significantly lower academic test scores in grades 6 and 8. Adolescent depression
 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91
Table 3
Late childhood and late adolescence outcomes by level of risk

Childhood risk (birth to age 12)

Low Medium High
(0–7 risk (8–11 risk (12–20 risk
factors) factors) factors)
Variable N Reporter Ageyyear M (S.D.) M (S.D.) M (S.D.) F Post hoc comparisons
Grade 6–7 characteristics
Peer social skillsa 1058 Teachers 12–13y1991 0.130 (1.0) 0.026 (1.0) y0.117 (0.97) 5.54** Low different from high
Shy or anxious behaviora 1058 Teachers 12–13y1991 y0.082 (0.97) y0.080 (0.91) 0.126 (1.1) 5.42** Low-Med different from high
Task orientationa 1058 Teachers 12–13y1991 0.111 (1.0) 0.041 (0.98) y0.115 (0.99) 4.91** Low different from high
Learning problemsa 1058 Teachers 12–13y1991 y0.131 (0.98) y0.034 (0.99) 0.125 (1.0) 6.10** Low different from high
Total competencea 1058 Teachers 12–13y1991 0.132 (1.0) 0.039 (0.99) y0.130 (0.97) 6.47** Low-Med different from high
Total problemsa 1058 Teachers 12–13y1991 y0.120 (1.0) y0.055 (0.97) 0.133 (1.0) 6.43** Low-Med different from high
Gr. 6 reading and math avg. 1208 CPS 12y1991 5.73 (1.3) 5.35 (1.3) 4.97 (1.3) 34.8*** All groups different
Gr. 8 reading and math avg. 1096 CPS 14y1993 7.57 (1.6) 7.10 (1.5) 6.74 (1.7) 24.2*** All groups different
Outcomes in Adolescence
Depression 801 Child 16y1996 0.66 (0.91) 0.70 (0.93) 0.80 (1.0) 1.5 No differences
Juvenile arrest 1406 CCJJ -18y-1998 0.28 (1.0) 0.51 (1.5) 0.93 (2.3) 18.7*** Low-Med different from high
High school or GED 1334 CPS, Parent, 22y2001 0.71 (0.46) 0.65 (0.48) 0.50 (0.50) x2(2) s44.1*** Low-Med different from high
completion rate Child
a
sStandardized scores: One way ANOVA; **P-0.01; ***P-0.001. Post hoc comparisons were Tukey’s HSD when group variances were equal and Dunnet’s C
when variances were unequal. CPSsChicago Public School system, CCJJ: Cook County Juvenile Justice system. High school or GED completion rate was tested
with a Chi-square test because it was a dichotomous outcome cross classified with a categorical variable. Follow-up comparisons were performed by selecting two
out of the three levels and running the Chi-square test again.

81
82 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91

was not statistically significantly different across the levels of risk. The average
number of juvenile arrest petitions for the high risk group was significantly higher
than the other two levels of risk. Finally, the high school or GED completion rate
for the high risk group was 50% compared to 65% for moderate risk, and 71% for
the low risk group (x2(2)s44.1, P-0.001).
Interestingly, Tukey’s HSD post hoc tests showed two major patterns. First, there
was a linear pattern where level of cumulative risk gradually increased negative
outcomes, and decreased positive outcomes. The low risk group mean was signifi-
cantly different from the high risk group mean while the moderate group was not
different from the other two. This linear relationship was the case for peer social
skills, task orientation, and learning problems. Grade 6 and 8 academic test scores
also had a negative linear relationship with increasing level of risk.
The second pattern revealed was non-linear. In this pattern, low and moderate
risk groups were comparable while high risk groups were significantly different.
This threshold effect was present for shy or anxious behavior, T-CRS total
competence and total problems scores, adolescent juvenile court petitions, and high
school or GED completion.

3.4. Predicting adolescent outcomes

The results obtained from running negative binomial regression models for
adolescent depression and juvenile arrest petitions are presented in Table 4. The
coefficients represented in Table 4 illustrate interpretable marginal effects estimated
using STATA. Marginal effects are the percentage point differences between groups
after adjusting for other variables in the model. They were derived from the partial
derivatives evaluated at the mean of the explanatory variable. Positive coefficients
suggest that the rates of delinquency or depression are higher for each unit change
of the explanatory variable, while negative coefficients illustrate lower rates of
juvenile delinquency or depression for each unit change of the explanatory variable.

3.5. Depression

In the multivariate model, several childhood risk and protective factors surfaced
as significant predictors of adolescent depression. Risk factors that predicted
adolescent depression at the trend level (P-0.10) were: being female, displaying
lower levels of classroom adjustment in grades 1–6, and exhibiting shy or anxious
behavior in grades 6–7. Each unit increase in the average of reading and math
scores during third grade was associated with a 15% point increase in adolescent
depression (P-0.05). Child Parent Center preschool participation was associated
with a 22 percentage point reduction in adolescent depression (P-0.05). Finally,
every unit increase in peer social skills in grades 6–7 was associated with a 20
percentage point decrease in adolescent depression (P-0.001).
Juvenile arrest—There were several childhood factors that predicted juvenile
arrest rates at the trend level (P-0.10). Child Parent Center preschool participation
was associated with a 23% point reduction in juvenile court petitions. CPC preschool
 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91 83

Table 4
Multivariate regression models for adolescent outcomes

Marginal effects from Odds ratios

Negative binomial from logistic
regression regression
Juvenile High school
court or GED
Depression petitions completion
(Age 16) (-Age 18) (Age 22)
Childhood risk and protective factors
Gender (1sFemale) 0.15q y0.87*** 1.20
Cumulative family risk (birth to age 12) 0.01 0.03* 0.93**
N of missing risk factors (birth to age 12) y0.01 y0.03 0.97
Child parent center preschool participation y0.22** y0.23q 1.39q
Average of child’s math and reading scores in Gr. 1 0.05 0.13 1.01
Number of years teachers rated parent participation in y0.01 y0.05 1.17*
school average or above (Gr. 1–6)
Average of child’s math and reading scores in Gr. 3 0.16* 0.04 0.80
Number of years teachers rated child’s classroom y0.05q 0.04 1.12
adjustment average or above (Gr. 1–6)
Child’s self perception of competence in school 0.00 y0.01 0.99
(Gr. 5–6)
Grade 6–7 characteristics
Peer social skillsa y0.20** y0.07 0.94
Shy or anxious behaviora y0.11q y0.13q 1.01
Assertiveness skillsa 0.03 y0.05 0.92
Task orientationa 0.01 0.11 1.67*
Acting out behaviora 0.08 0.36** 0.69*
Learning problemsa y0.03 0.09 1.25
Frustration tolerancea y0.01 y0.00 1.00
Gr. 6 reading and math avg. y0.07 y0.14q 1.58***
Note: q P-0.10, *P-0.05, **P-0.01, ***P-0.001. Marginal effects are the percentage point
differences between groups after adjusting for other variables in the model. They were derived from the
partial derivatives evaluated at the mean of the explanatory variable. This model was run using a max-
imum-likelihood zero-inflated negative binomial regression model where the dependent variable is a non-
negative count variable. All covariates and independent variables were assessed simultaneously.

was statistically significant (P-0.05) in the stepwise models until grade 6 test
scores entered on the final step. A unit change increase in either shy or anxious
behavior during grades 6–7 or grade 6 reading and math scores was associated with
a 13 or 14% point reduction in juvenile court petitions. Being female was associated
with an 87% point reduction in juvenile court petitions (P-0.001). Every increase
in family risk from birth to age 12 was associated with a 3% point increase in
juvenile court petitions (P-0.05) and finally, every increase in acting out behavior
in grades 6–7 was associated with a 36% point increase in juvenile court petitions.
High school or GED completion was regressed on childhood risk and protective
factors using logistic regression. Odds ratios for childhood risk and protective
factors that were included in the models are presented in Table 4. Holding other
84 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91

factors constant, each family risk factor that was added to the child’s life from birth
to age 12 was associated with a 7% (P-0.01) decrease in the probability of
completing high school or receiving a GED. Acting out in grades 6–7 was also an
important risk factor. Every increase in acting out behavior was associated with a
31% (P-0.05) reduction in the probability of graduating from high school or
completing a GED. There were also several significant protective factors. The
likelihood of graduating from high school or getting a GED certificate was 39%
higher for children who participated in the CPC preschool program compared to
children who did not (P-0.10). Each increase in the number of years of parent
participation in school during grades 1–6 was associated with a 17% increase in
the probability of high school or GED completion (P-0.05). Finally, every increase
in task orientation during grades 6–7 and in 6th grade standardized test scores was
associated with a 67% and a 58% increase in the chances of graduating from high
school or completing a GED (P-0.05 and P-0.001, respectively). The full model
fit the data well (x2(18)s191.35, P-0.001), correctly classifying 81% of individuals
who graduated and 57% of those who did not.

4. Discussion

This study investigated risk and resilience from childhood through adolescence
in a large sample of inner city minority youth. Family risk factors were not
significantly associated with adolescent depression, but had a positive relationship
with juvenile arrest rates and a strong negative association with high school or GED
completion. Some of the main effects for protective factors were far stronger than
the effects for cumulative family risk.

4.1. Risk factors

The strongest relationship between a family risk factor and an outcome surfaced
between parent’s educational status (i.e. not having finished high school) and
adolescent high school or GED completion. This correlation between the parent’s
low educational status and the adolescent not finishing school was even stronger
than the influence that cumulative family risk had on adolescent educational
attainment. This illustrated how risk can be passed down from generation to
generation.
Cumulative family risk during childhood exerted an important influence across
adolescent outcomes. In multivariate models, cumulative family risk significantly
increased the chances of juvenile court involvement and decreased the probability
of completing high school or getting a GED by age 22. This indication of the
strength of cumulative risk both confirms work completed by other resilience
researchers (Rutter, 1979; Seifer et al., 1992; Garmezy, 1993; Coie et al., 1993;
Dishion et al., 1999; Greenberg et al., 2001; Rutter, 2001) and extends our
understanding of the topic. Cumulative family risk was not a significant predictor
of adolescent depression. Further, much of the relationship it had with high school
or GED completion appeared to derive from the contribution of a single risk factor
 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91 85

– parent educational status. This suggests that although it is certainly important,

cumulative risk needs to be examined along with individual risk factors across
multiple domains of adolescent functioning. Risk dynamics are complex and need
to be further examined using diverse methods and models.
Further evidence for risk complexity was shown in this study’s analysis of low,
moderate, and high risk groups. Cumulative risk had both linear and non-linear
effects depending upon the outcome of interest. No clear pattern of linear vs. non-
linear effects surfaced. Moving from moderate to high cumulative family risk
resulted in a significant increase in shy or anxious behavior, and total problem
behavior while decreasing total competent behavior. This threshold effect was in
line with other research (Fergusson and Lynskey, 1996; Pollard et al., 1999),
prompting us to agree that cumulative risk may have a non-linear relationship with
childhood and adolescent outcomes. Future research should closely examine risk
thresholds for different outcome domains. At the same time, cumulative family risk
had a negative linear relationship with peer social skills, task orientation, and
standardized test scores in grades 6 and 8. It also had a positive linear relationship
with learning problems. This conflicting evidence leads us to conclude that some
children may have thresholds of tolerance for risk beyond which they should not be
pushed while other children may experience gradual deterioration of competencies
and elevations in problems with increasing accumulation of family risk.
In addition to cumulative family risk, there were several child characteristics that
heightened the chances of problematic adolescent outcomes. For example, being
female, and having higher standardized test scores in 3rd grade were risk factors
associated with increases in adolescent depression. Also, children with higher levels
of teacher-reported acting out behaviors during grades 6–7 had significantly higher
rates of juvenile court petitions in adolescence and lower chances of completing
high school or a GED program. We highlight these findings to illustrate that some
risk factors were domain specific (e.g. were only associated with depression) while
others were related to multiple outcomes (e.g. childhood acting out predicted both
juvenile arrest and decreased educational attainment).

4.2. Protective factors

Protective factors were stronger predictors of adolescent outcomes than risk

factors. This was true for every multivariate model that we examined. Participation
in Child Parent Center preschool was the protective factor with the widest ranging
effects across outcomes. CPC participants had significantly lower rates of adolescent
depression, fewer juvenile court petitions, and their probability of completing high
school or their GED was 36% greater than non-participants. This robust relationship
between early childhood prevention and adolescent outcomes was striking, especially
considering the extensive array of child attributes and risk factors that were included
in the multivariate models. It was beyond the purview of this investigation to
delineate precise mechanisms or pathways for CPC effects. This has been discussed
at length elsewhere (Reynolds, 1991, 1998; Reynolds et al., 2001). However, the
CPC effects in this study are particularly noteworthy because it is the first time that
86 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91

program effects on adolescent depression have been examined. Further, the CPC
effect for adolescent depression may be underestimated. Missing data analyses
showed that higher risk adolescents were less likely to be included in the analyses,
lowering the variation in the sample. If these higher risk adolescents were included,
the effect of CPC preschool participation on adolescent depression may have been
even stronger.
Two other protective factors cut across two but not all three of the adolescent
outcomes. Children who displayed shy or anxious behavior during grades 6–7 had
lower rates of depression and fewer juvenile court petitions. Similarly, children with
higher standardized test scores during grade 6 had fewer juvenile court petitions
and significantly higher chances of finishing high school or getting a GED.
Each outcome also had unique protective factors. Classroom adjustment in
elementary school and peer social skills during grades 6–7 were associated with
lower rates of adolescent depression. Females had significantly lower rates of
juvenile court petitions. Finally, parent participation in elementary school and child
task orientation significantly increased the odds of completing high school or a
GED program.
These associations begin to illuminate complex relationships between academic
functioning, social adjustment (in the form of rule abiding behavior), and mental
health. There were universal protective factors (i.e. early intervention), common
protective factors that were associated with more than one outcome, and unique
protective factors within each outcome domain. Further, early indicators of social
skill, academic functioning, and mental health sometimes predicted later functioning
in other areas (e.g. childhood peer social skills predicted adolescent depression,
childhood acting out predicted high school or GED completion). Future research
should continue to delineate protective factor profiles for different domains of
functioning, paying special attention to protective factors that are unique to one
outcome vs. ones that are generally applicable to multiple outcomes. Future studies
should also try to replicate the patterns of relationships found in the current
investigation.

4.3. Implications for social work practice

This investigation’s most straightforward implication for social work practice is

that the results underscored the need to support high-quality early childhood
intervention efforts. CPC program participation was the protective factor with the
widest-ranging effects – even up to 17 years after the intervention was provided.
This, along with previous work cited above, makes a compelling case for both
intervention effectiveness and the lasting effect comprehensive services provided
early in life can have on educational attainment, social conduct, and mental health.
Social work practitioners should also welcome the evidence showing that
protective factors, compared to risk factors, have a stronger longitudinal influence
on adolescent outcomes. This lends support to strengths-based approaches to social
work practice wherein building assets and opportunities is just as important as
ameliorating risks. Once again, intervention appears to be optimal when provided
 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91 87

as early as possible in the lifespan. Supporting early classroom adjustment, building

peer social skills, encouraging parent participation in elementary school, and
increasing child task orientation are all targets for social work practice that appear
to have long-term benefits.
Finally, social workers should keep in mind the complexity of both risk and
protective factors. Risk factors may gradually lead to problems or may have a
serious threshold effect beyond which problems rapidly manifest. Cumulative risk
should always be assessed. Practitioners need to be familiar with risk and protective
factor profiles that are associated with specific outcome domains. Understanding
this complexity will help to make practice interventions specific and well-targeted.
4.4. Limitations
Missing data attrition was the most important weakness in this longitudinal
analysis. Because of missing data, the investigation’s sample size fluctuated between
the three multivariate models. Further, missing data analyses indicated that high risk
study participants were less likely to return the 10th grade survey, making our
adolescent depression data biased towards lower risk and higher functioning
adolescents. This decreased variation in the adolescent depression data may have
decreased the strength of our findings for this outcome.
Participants with missing data on high school or GED completion and juvenile
arrest experienced less cumulative family risk and, at the same time, had fewer
childhood protective factors. It is important to note this potential bias, but at the
same time, it is harder to interpret how this pattern may have influenced our results.
Because they experienced less family risk, these sample participants may have had
the family resources to move out of inner-city Chicago. Children who did not
participate in CPC preschool were more likely to have missing data on high school
or GED completion, indicating that the CPC program effect for this outcome may
be biased downward as well.
This study’s aim was to illuminate how risk and protective factors functioned for
inner city minority youth. Sample participants were all minority and were primarily
(93%) African American. While this attempted to fill a gap in our current knowledge
of development in minority children, it also limited the generalizability of the
results. Consequently, caution should be used when applying the study results
beyond their application to minority youth who grew up in impoverished neighbor-
hoods in Chicago.
Finally, our scale for adolescent depression had marginally acceptable psycho-
metrics and was not as rigorously measured as the other two dependent variables.
The score was based on responses to three dichotomous questions that may have
been useful signs of depression. However, it was not indicative of clinical depression.
Future studies considering the interplay of multiple adjustment domains in adoles-
cence would benefit from including a more rigorous measure of mental health.
5. Conclusions
Family risk factors had a strong influence across adjustment indicators in
childhood and adolescence. Cumulative family risk was particularly deleterious and
88 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91

its effects were either linear or non-linear depending upon the outcome under
consideration. The deleterious effects of cumulative family risk need to be further
examined in future research. Relationships between protective factors and adolescent
outcomes were even stronger than risk factor associations. Early childhood interven-
tion had the broadest protective effect and, for the first time, was shown to be
related to lower rates of adolescent depression. Each adolescent outcome had a
distinct profile of significant risk and protective factors. Future research should try
to replicate and refine these profiles so that they can ultimately be used as targets
for prevention programming.

Acknowledgments

Preparation of this paper was supported by National Research Service Adminis-

tration Pre- and Post-doctoral fellowships (No. T32MH12020) awarded to the first
author and funded through the National Institute of Mental Health. Support for the
Chicago Longitudinal Study also comes from the National Institute of Child Health
and Human Development (No. R01HD34294), The Office of Educational Research
Improvement (No. R305T990477) and The University of Wisconsin-Madison
Graduate School.

References

Anthony, E. J. (1974). The syndrome of the psychological invulnerable child. In E. J. Anthony, (pp.
529 –544) The child in his family children at psychiatric risk, vol. 3. New York: John Wiley and
Sons.
Bendersky, M., & Lewis, M. (1994). Environmental risk, biological risk, and developmental outcome.
Developmental Psychology, 30, 484 –494.
In Booth, A. & Crouter, A. C. (2001). Does it take a village?. Lawrence Erlbaum, Mahwah, NJ.
Cauce, A. (1995). Behavior problems in 5- to 11-year-old children from low-income families: Of norms
and cutoffs: Response. Journal of the American Academy of Child and Adolescent Psychiatry, 34(5),
537 –538.
Cauce, A., Felner, R. D., & Primavera, J. (1982). Social support in high-risk adolescents: Structural
components and adaptive impact. American Journal of Community Psychology, 10(4), 417 –428.
Cauce, A., Ryan, K., & Grove, K. (1998). Children and adolescents of color, where are you?
Participation, selection, recruitment and retention in developmental research. In V. McLoyd, L.
Steinberg, Studying minority adolescents: Conceptual, methodological and theoretical issues (pp.
147 –166). Mahwah, NJ: Lawrence Erlbaum.
Coie, J. D., Watt, N. F., West, S. G., Hawkins, J. D., Asarnow, J. R., Markman, H. J., Ramey, S. L.,
Shure, M. B., & Long, B. (1993). The science of prevention: A conceptual framework and some
directions for a national research program. American Psychologist, 48, 1013 –1022.
Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal of Sociology,
94(Suppl), S95 –S120.
Cowen, E. L., Wyman, P. A., Work, W. C., & Parker, G. R. (1990). The Rochester child resilience
project: Overview and summary of first year findings. Development and Psychopathology, 2, 193 –
212.
Dishion, T. J., Capaldi, D. M., & Yoerger, K. (1999). Middle childhood antecedents to progressions in
male adolescent substance use: An ecological analysis of risk and protection. Journal of Adolescent
Research, 14(2), 175 –205.
 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91 89

Duncan, G. J., & Raudenbush, S. W. (2001). Neighborhoods and adolescent development: How can we
determine the links?. In A. Booth, A. C. Crouter, Does it take a village? (pp. 105 –136). Mahwah,
NJ: Lawrence Erlbaum.
Elder, G. H. (1974y1999). Children of the great depression: Social change in life experiences. Chicago:
University of Chicago Press.
Farber, A., & Egeland, B. (1987). Invulnerability among abused and neglected children. In E. Anthony,
B. Cohler, The invulnerable child (pp. 253 –288). New York: Guilford Press.
Fergusson, D. M., & Lynskey, M. T. (1996). Adolescent resiliency to family adversity. Journal of Child
Psychology and Psychiatry, 37, 281 –292.
Fraser, M. W., Kirby, L. D., Smokowski, P. R. (In Press). Risk and resilience in childhood. In: M. W.
Fraser (Ed.) Risk and Resilience in Childhood: An Ecological Perspective (2nd Edn). Washington
D.C.: National Association of Social Workers.
Garmezy, N. (1985). Stress-resistant children: The search for protective factors. In J. E. Stevenson,
Recent research in developmental psychopathology (pp. 213 –233). Tarrytown, NY: Pergamon Press.
Garmezy, N. (1993). Children in poverty: Resilience despite risk. Psychiatry, 56, 127 –136.
Garmezy, N., & Masten, A. (1991). ’The protective role of competence indicators in children at risk.
In E. M. Cummings, Lifespan developmental psychology: Perspectives on stress and coping (pp.
151 –176). Hillsdale, NJ: Lawrence Erlbaum.
Greenberg, M. T., Speltz, M. L., DeKlyen, M., & Jones, K. (2001). Correlates of clinic referral for
early conduct problems: Variable- and person-oriented approaches. Development and Psychopathology,
13, 255 –276.
Hartup, W. W. (1983). Peer relations. In: P. J. Mussen (Series Ed.) & E. M. Hetherington (Vol. Ed.).
Handbook of child psychology: socialization, personality, and social development, Vol. 4 New York:
Wiley pp. 103–196.
Hightower, A, Spinnell, A., Lotyczewski, B. (1989). Teacher-child rating scale (T-CRS) guidelines.
Rochester, NY: Primary Mental Health Project.
Ladd, G. W., & Burgess, K. B. (2001). Do relational risks and protective factors moderate the linkages
between childhood aggression and early psychological and school adjustment?. Child Development,
72(5), 1579 –1601.
Land, L. C., McCall, P. L., & Nagin, D. S. (1996). A comparison of poisson, negative binomial, and
semiparametric mixed poisson regression models with empirical applications to criminal careers
research. Sociological Methods and Research, 24, 387 –442.
Luthar, S. (1991). Vulnerability and resilience: A study of high-risk adolescents. Child Development,
62, 600 –616.
Luthar, S. (1993). Annotation: Methodological and conceptual issues in research on childhood resilience.
Journal of Child Psychology and Psychiatry, 34(4), 441 –453.
Luthar, S. (1995). Social competence in the school setting: Prospective cross-domain associations
among inner-city teens. Child Development, 66(2), 416 –429.
Luthar, S. (1997). Sociodemographic disadvantage and psychosocial adjustment: Perspectives from
developmental psychopathology. In S. Luthar, J. Burack, D. Cicihetti, J. Weisz, Developmental
psychopathology: Perspectives on adjustment, risk, and disorder (pp. 459 –485). New York, NY:
Cambridge University Press.
Luthar, S. S., Cicchetti, D., & Becker, B. (2000). The construct of resilience: A critical evaluation and
guidelines for future work. Child Development, 71(3), 543 –562.
Luthar, S. S., Doernberger, C. H., & Zigler, E. (1993). Resilience is not a unidimensional construct:
Insights from a prospective study of inner-city adolescents. Development and Psychopathology, 5,
703 –717.
Luthar, S., & Zigler, E. (1991). Vulnerability and competence: A review of research on resilience in
childhood. American Journal of Orthopsychiatry, 61(1), 6 –22.
Masten, A. (1987). Resilience in development: Implications of the study of successful adaptation for
developmental psychopathology. In D. Cicchetti, The emergence of a discipline: Rochester symposium
on developmental psychopathology (pp. 261 –294). Hillsdale, NJ: Lawrence Erlbaum.
90 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91

Masten, A. (1994). Resilience in individual development: Successful adaptation despite risk and
adversity. In M. C. Wang, E. W. Gordon, Educational resilience in inner-city America: Challenges
and prospects. Hilldale, NJ: Lawrence Erlbaum.
Masten, A., Best, K., & Garmezy, N. (1990). Resilience and development: Contributions from the study
of children who overcome adversity. Development-and-Psychopathology, 2(4), 425 –444.
Masten, A., & Coatsworth, J. D. (1995). Competence, resilience and psychopathology. In D. Cicchetti,
D. J. Cohen, (pp. 715 –752) Developmental psychopathology: Risk, disorder and adaptation, vol. 2.
New York: John Wiley and Sons, Inc.
Masten, A., & Coatsworth, J. D. (1998). The development of competence in favorable and unfavorable
environments: Lessons from research on successful children. American Psychologist, 53(2), 205 –
220.
Masten, A. S., Hubbard, J. J., Gest, S. D., Tellegen, A., Garmezy, N., & Ramirez, M. (1999).
Competence in the contest of adversity: Pathways to resilience and maladaptation from childhood to
late adolescence. Development and Psychopathology, 11, 143 –169.
McLoyd, V. C. (1990a). Minority children: Introduction to the special issue. Child Development, 61,
263 –266.
McLoyd, V. C. (1990b). The impact of economic hardship on black families and children: Psychological
distress, parenting and socioemotional development. Child Development, 61, 311 –346.
McLoyd, V. C. (1998). Socioeconomic disadvantage and child development. American Psychologist,
53(2), 185 –204.
Nagin, D. S., & Land, K. C. (1993). Age, criminal careers, and population heterogeneity: Specification
and estimation of a non-parametric, mixed poisson model. Criminology, 31, 327 –362.
Natriello, G., McDill, E. I., & Pallas, A. M. (1990). Schooling and disadvantaged children; racing
against catastrophe. New York: Teachers College Press, Columbia University.
Nettles, S. M., & Pleck, J. H. (1993). Risk, resilience, and development: The multiple ecologies of
black adolescents. US: Center for Research on Effective Schooling for Disadvantaged Students.
Pellegrini, D. S. (1990). Psychosocial risk and protective factors in childhood. Journal of Developmental
and Behavioral Pediatrics, 11(4), 201 –209.
Pollard, J. A., Hawkins, J. D., & Arthur, M. W. (1999). Risk and protection: Are both necessary to
understand diverse behavioral outcomes in adolescence?. Social Work Research, 23(3), 145 –158.
Reynolds, A. J. (1991). Early schooling of children at risk. American Educational Research Journal,
28, 392 –422.
Reynolds, A. J. (1998). Resilience among black urban youth: Prevalence, intervention effects and
mechanisms of influence. American Journal of Orthopsychiatry, 68(1), 84 –100.
Reynolds, A., Temple, J., Robertson, D., & Mann, E. (2001). Long term effects of an early childhood
intervention on educational achievement and juvenile arrest: A fifteen year follow-up of low income
children in public schools. Journal of the American Medical Association, 285(18), 2339 –2346.
Rutter, M. (1979). Protective factors in children’s responses to stress and disadvantage. In J. S. Bruner,
A. Garten, (pp. 49 –74) Primary prevention of psychopathology, vol. 3. Hanover, NH: University
Press of New England.
Rutter, M. (1987). Psychosocial resilience and protective mechanisms. American Journal of Orthopsy-
chiatry, 57(3), 316 –331.
Rutter, M. (1990). Psychosocial resilience and protective mechanisms. In J. Rolf, A. S. Masten, D.
Cicchetti, K. H. Nuechterlein, S. Weintraub, Risk and protective factors in the development of
psychopathology (pp. 181 –214). New York: Cambridge University Press.
Rutter, M. (2000). Resilience reconsidered: Conceptual considerations, empirical findings, and policy
implications. In J. P. Shonkoff, S. J. Meisels, Handbook of early childhood intervention (2nd ed)
(pp. 651 –682). New York, NY: Cambridge University Press.
Rutter, M. (2001). Psychosocial adversity: Risk, resilience, and recovery. In J. M. Richman, M. W.
Fraser, The context of youth violence: resilience, risk, and protection (pp. 13 –41). Westport, CT:
Praeger Publishers.
 P.R. Smokowski et al. / Children and Youth Services Review 26 (2004) 63–91 91

Seifer, R., Sameroff, A. J., Baldwin, C. P., & Baldwin, A. (1992). Child and family factors that
ameliorate risk between 4 and 13 years of age. Journal of the American Academy of Child and
Adolescent Psychiatry, 31, 893 –903.
Smokowski, P. R., Reynolds, A. J., & Bezrucko, N. (1999). Resilience and protective factors in
adolescence: An autobiographical perspective from disadvantaged youth. Journal of School Psychol-
ogy, 37(4), 425 –448.
Stevenson, H. W., Chen, C., & Lee, S. Y. (1993). Mathematics achievement of Chinese, Japanese, and
American children: Ten years later. Science, 259, 53 –58.
Werner, E., & Smith, R. (1982). Vulnerable but invincible: A longitudinal study of resilient children
and youth. New York: Adams, Bannister, and Cox.
Werner, E., & Smith, R. (1992). Overcoming the odds: High-risk children from birth to adulthood.
Ithaca, NY: Cornell University Press.
Winfield, L. F. (1995). The knowledge base on resilience in African–American adolescents. In L. J.
Crockett, A. C. Crouter, Pathways through adolescence: Individual development in relation to social
contexts. The Penn state series on child and adolescent development (pp. 87 –118). Hillsdale, NJ,
England: Lawrence Erlbaum.