Journal of School Psychology, Vol. 38, No. 6, pp.

525549, 2000
Copyright 2000 Society for the Study of School Psychology
Pergamon Printed in the USA
0022-4405/00 $see front matter

PII S0022-4405(00)00051-0

A Prospective Longitudinal Study of

High School Dropouts
Examining Multiple Predictors Across Development
Shane Jimerson
University of California

Byron Egeland, L. Alan Sroufe, and Betty Carlson

University of Minnesota

Prior studies report a variety of demographic, school, individual, and family charac-
teristics that are related to high school drop out. This study utilizes data from a 19-
year prospective longitudinal study of at-risk children to explore multiple pre-
dictors of high school dropouts across development. The proposed model of drop-
ping out emphasizes the importance of the early home environment and the quality
of early caregiving influencing subsequent development. The results of this study
demonstrate the association of the early home environment, the quality of early
caregiving, socioeconomic status, IQ, behavior problems, academic achievement,
peer relations, and parent involvement with dropping out of high school at age 19.
These results are consistent with the view of dropping out as a dynamic develop-
mental process that begins before children enter elementary school. Psychosocial
variables prior to school entry predicted dropping out with power equal to later IQ
and school achievement test scores. In our efforts to better understand processes
influencing dropping out prior to high school graduation, early developmental fea-
tures warrant further emphasis. 2000 Society for the Study of School Psychology.
Published by Elsevier Science Ltd

Keywords: Dropouts, Development, Longitudinal study, Early caregiving, Home en-

vironment, Parent involvement, Behavior problems, IQ, SES, Academic achieve-
ment, Peer relations.

The seriousness of the drop out problem among American youth is well
documented including both personal and societal implications (Cairns &
Cairns, 1994; Rumberger, Ghatak, Poulos, Ritter, & Dornbusch, 1990). The
estimated 3.4 million nongraduating youth (National Center for Education
Statistics, 1994) are ill-equipped for the modern work force, thus ultimately
paying less tax, adding costs to welfare programs, and being disproportion-
ately represented in crime and incarceration statistics (Kirsch, Jungeblut,

Received August 23, 1999; accepted February 1, 2000.

Address correspondence and reprint requests to Shane R. Jimerson, University of Califor-
nia, Graduate School of Education, 2208 Phelps Hall, Santa Barbara, CA 93106-9490. E-mail:
Jimerson@education.ucsb.edu

525
526 Journal of School Psychology

Jenkins, & Kolstad, 1993; Rumberger, 1987). Nearly a decade ago, the an-
nual financial cost of the drop out problem was estimated at $240 billion
(Dryfoos, 1990).
In some ways, dropping out is no longer mysterious. Five decades of re-
search have uncovered numerous correlates of withdrawal from high
school. Prior research highlights various demographic status variables, indi-
vidual characteristics, psychological and behavioral measures, and family
factors associated with high school drop out (Rumberger, 1987, 1995;
Rumberger et al., 1990). They are now well-known but not always useful.
Demographic factors such as low socioeconomic status (SES), neighbor-
hood-level variables, gender, ethnic minority status, and low parental edu-
cation, for example, are consistently found to be related to school with-
drawal (Cairns, Cairns, & Neckerman, 1989; Ensminger, Lamkin, &
Jacobson, 1996; Fine, 1989; Oakland, 1992; Weis, Farrar, & Petrie, 1989).
However, such broad status variables leave considerable variance unex-
plained and are not very informative with regard to the processes of drop-
ping out. Of course, achievement problems and failing grades also are
strong correlates (Borus & Carpenter, 1984; Ekstrom, Goertz, Pollack, &
Rock, 1986; Ensminger & Slusarick, 1992; Garnier, Stein, & Jacobs, 1997;
Lloyd, 1974, 1978), but these may be viewed as early indicators of dropping
out itself rather than as root causes.
In contrast to the above status variables, other studies have identified
more direct behavioral influences associated with drop out status such as
measures of behavior problems, poor peer relationships, and certain family
variables (Cairns & Cairns, 1994; Cairns, Cairns, & Neckerman, 1989; Ens-
minger & Slusarick, 1992; Feldhusen, Thurston, & Benning, 1973; Garnier,
Stein, & Jacobs, 1997; Parker & Asher, 1987). These measures (most ob-
tained in middle school or late elementary school) have predicted later
dropping out quite well, often with some specificity. For example, Cairns
and Cairns (1994) found that association with others on the pathway to
dropping out increased the likelihood that a student would drop out. Stud-
ies that include family factors have isolated variables such as parental
school involvement, monitoring of the child, quality of parentchild inter-
actions, and family lifestyles and values (Alpert & Durham, 1986; Brooks-
Gunn, Guo, & Furstenburg, 1993; Dornbusch, Ritter, Leiderman, Rob-
erts, & Fraleigh, 1987; Garnier, Stein, & Jacobs, 1997; Morris, Ehren, &
Lenz, 1991; Rumberger, 1995). However, most of these studies relied on
questionnaires or interview data, and with the exception of the Garnier et
al. (1997) study, none of them began in the early years.
It is also the case that many of the factors predicting dropping out are
interrelated. Peer problems, behavior problems, and achievement prob-
lems are strongly correlated with each other. Therefore, sorting out their
causal role in later behavior is challenging. Moreover, the predictors of
dropping out have known antecedents. For example, observed quality of
 Jimerson et al. 527

care in infancy and early childhood has well-demonstrated links to peer

relationships, behavior problems, and achievement in elementary school
and high school (Carlson et al., 1998; Elicker, Englund, & Sroufe, 1992;
Renken, Egeland, Marvinney, Mangelsdorf, & Sroufe, 1989; Sroufe, Ege-
land, & Carlson, 1999; Teo, Sroufe, & Carlson, 1996). Thus, what are cur-
rently considered to be predictors may be midpoint markers of a lengthy
developmental pathway to dropping out.
Rumberger (1987, 1995) provided a review of the literature regarding
dropouts and concludes that although it is useful, the literature, in general,
has several shortcomings. First, few studies have examined multiple influ-
ences of dropping out in a comprehensive fashion (e.g., individual, family,
and peer effects). Second, the longitudinal component often is very brief
such as from an early high school period to a later high school period. Fi-
nally, many of the factors (especially those related to family background)
focus on structural characteristics rather than process (i.e., one-parent fam-
ilies versus parentchild interactions). Rumberger (1987, 1995) argued
that further research efforts should focus on developing multivariate, longi-
tudinal, and comprehensive models of the causes of dropping out.
The present study was designed to further explore the developmental
processes and precursors that lead to dropping out of high school. The pur-
pose of this study was to examine developmental adaptation and the factors
that influence adaptation that may lead to a pathway toward dropping out.
Lower achievement, poorer peer relations, behavior problems, and less
parent involvement during elementary school have all been associated with
students who later drop out (relative to students who graduate). Prior re-
search illustrates the influence of the family on each of these elementary
school factors. However, most current literature examining early predictors
of dropping out of high school has been limited to information from ele-
mentary school and high school. In an effort to better understand trajector-
ies toward dropping out, developmental features prior to elementary
school were examined. This study provides a conceptual model of drop-
ping out as a dynamic developmental process that begins before children
enter elementary school. From a developmental perspective, peer rela-
tions, behavior problems, and academic achievement do not simply appear
in elementary school; instead, development in each of these domains is re-
lated to prior development. Dropping out is hypothesized to occur as a re-
sult of current circumstances and prior development (i.e., experiences and
adaptations).
We have been guided by a developmental transactional model (Samer-
off, 1992; Sroufe, Egeland, & Carlson, 1999). In this perspective, early de-
velopmental history is given some priority, not because it ineluctably causes
later outcomes, but because what the child takes forward from these experi-
ences, in part, frames subsequent transactions with the environment. Be-
havior always is a product not only of current circumstances but also of de-
velopmental history. There is a homeorhetic tendency in development
528 Journal of School Psychology

Figure 1. Variables examined in this study of dropping out as a developmental process.

(Sameroff & Fiese, 1989) such that when a pathway is enjoined, numerous
factors conspire toward its continuation. Individuals engage, select, and in-
terpret experiences within the previously established framework, so that
later experiences often support prior experiences. As other researchers of
drop out have proposed,
of special importance is the developmental nature of the patterns, so that
events occur which then set other events in motion. This snowballing effect is
one that will have to be mapped out in order to intervene at critical points in
time before a trend or development has moved into a less malleable phase.
(Evans & DiBenedetto, 1990, p. 68)
Dropouts present a variety of profiles; however, certain early characteristics
may increase the likelihood that a student will drop out. The hypothesis of
this investigation is that ones early developmental history (e.g., quality of
caregiving and home environment) will predict not only drop out anteced-
ents (i.e., academic achievement, problem behaviors, peer relations, and
parent involvement), but dropping out of high school itself (see Figure 1).
Although there is converging evidence illustrating the role of the family
upon decisions to drop out, no longitudinal studies have illustrated the im-
portance of the early quality of caregiving and home environment for sub-
sequent school withdrawal. By focusing on earlier characteristics than prior
studies afforded, the power of family factors will be examined.
 Jimerson et al. 529

METHODS
Participants
The sample consisted of 177 children and their families from an original
sample of 267 first-time pregnant women in their last trimester of preg-
nancy who were followed from the birth of their child through age 19 (Ege-
land & Abery, 1991; Egeland & Brunnquell, 1979). The sample was at risk
due to poverty and associated risk factors such as age, education, and single
parenthood. All had received prenatal care through public assistance at the
Maternal and Infant Care Clinic of the Minneapolis Health Department.
To examine the antecedents of school drop out, data collected at age 1
through age 19 was used. A primary concern with longitudinal research is
participant attrition. For the period of 12 months of age to kindergarten,
22 subjects left the study. Comparing the group of 22 to those who re-
mained in the study did not yield any differences on race, mothers age, ed-
ucation, or occupation. During the period first grade to age 19, 22 partici-
pants left the study. Comparing the group of participants who left the study
with the remaining sample of 177 revealed no differences on the basic de-
mographics including race, mothers age, education, or occupation at the
time of the childs birth.
The students were classified in terms of their high school graduation sta-
tus at age nineteen. The classification was determined by reviewing forms
completed by the high school counselors, high school teachers, annual in-
terviews with the students, and phone calls to verify academic enrollment
at age 19. In this study, students who were not enrolled in any form of an
educational program and were not making progress toward a high school
diploma or GED were classified as Dropouts (n 43). Students who were
currently enrolled in full-time attendance and making progress toward a
high school diploma within a traditional setting or had graduated were clas-
sified as Traditional students (n 100). Some students withdrew from a
traditional educational program and entered alternative programs to con-
tinue their education (n 34); because these students did not fit the crite-
ria for either of the above groups they were not included in this study.
Within this subsample (n 143), mothers age at the time of birth
ranged from 1624 years (M 19.90; SD 2.08) for the Dropouts, and
1534 years (M 21.01; SD 4.04) for the Traditional students. Educa-
tional attainment of the mothers in both groups ranged from junior high
school to college. Whereas 28% of the mothers of Dropouts had not com-
pleted 12 years of education at the time of the babys birth, 33% of the
mothers of Traditional students had not completed 12 years of education
at the time of the babys birth. For the Dropouts, 63% had Caucasian par-
ents, 14% had African American parents, and 14% were of mixed race. For
the Traditional students, 68% had Caucasian parents, 8% had African
530 Journal of School Psychology

American parents, and 21% were of mixed race. The only significant differ-
ence regarding the composition of these two groups was that only 28% of
the dropouts were females (i.e., 12 of the 43), whereas 52% of the Tradi-
tional students were females.
Of 97 Caucasian students in this subsample, 28% dropped out, and of 29
mixed-race children, 24% dropped out. In contrast, 46% of the African
American children dropped out. The disproportionate percentage of Afri-
can American students who dropped out relative to Caucasian students
found in this study is consistent with prior research (McMillen, Kaufmen,
Hausken, & Bradby, 1993).

Measures and Procedures: Family Factors

Early quality of caregiving composite. This composite included (a) a rat-
ing of overall maternal sensitivity (an observation of play and two observa-
tions of feeding) at age 6 months, (b) the quality of the infant and mother
attachment relationship at age 12 and 18 months, and (c) the structure and
limits set by the mother on a series of tasks and the quality of the instruction
provided by the mother on these same tasks at age 42 months. This compos-
ite is not only a measure of the caregiver, but also reflects the dyadic rela-
tionship established between the caregiver and infant. Z-score transforma-
tions were performed, and the three components were combined and
averaged. Each of the three initial variables is described below.
Overall maternal sensitivity. At age 6 months, mothers were observed in
two feeding situations and one play situation with their infants. Sensitivity
ratings were made for each session and an overall maternal sensitivity score
was produced by combining the scores from each of the two sessions. The
sensitivity ratings were based on the scale of Ainsworth, Blehar, Waters, and
Wall (1978) and included four essential components: the mothers aware-
ness of the infants signals, an accurate interpretation of them, an appro-
priate response to them, and a prompt response to them. Each session was
scored on a 9-point scale from highly sensitive to highly insensitive. In-
terrater agreement was calculated using the Lawlis-Lu index, with agree-
ment defined as a 2-point discrepancy among raters on the 9-point scale.
The T-value of agreement was .87. The overall maternal sensitivity was in-
cluded in the early quality of caregiving composite.
Quality of infantmother attachment relationship. At 12 and 18 months,
the quality of the motherchild attachment relationship was assessed in the
Strange Situation (Ainsworth & Wittig, 1969). The Strange Situation con-
sists of a standardized sequence of eight episodes involving brief separa-
tions from and reunions with the mother and contact with a stranger. Rat-
ings are made of the infants response to the stress of separation, response
to the reunion, and response to the stranger (with and without the mother
 Jimerson et al. 531

present). At each assessment, the infantmother relationship was catego-

rized as secure, insecureavoidant, or insecureresistant. A secure classifi-
cation indicates that the infant is able to rely on the caregivers supportive
presence as a means of reducing distress. Classification as insecureavoidant
indicates that instead of turning to the mother at times of distress, the in-
fant actively avoids interaction with the mother. Classification as insecure
resistant indicates that the infant seeks the mother when distressed, but
signs of anger or continued distress are manifested while seeking contact.
Attachment classifications have been extensively validated as early mea-
sures of important individual differences (Bretherton & Waters, 1985).
This study used a composite of the 12- and 18-month classifications in the
early quality of caregiving composite. Videotaped Strange Situations were
coded by two expert coders who independently classified each of the in-
fants (89% agreement). Infants classified as secure were given a score of 1;
infants classified as anxious were given a score of 0. Thus, considering the
scores of infants at both 12 and 18 months, the maximum score was 2 and
the minimum score was 0.
Quality of problem-solving support. A series of teaching task situations
were used to assess the quality of mothers teaching and problem-solving
support to infants at age 42 months (Erickson, Sroufe, & Egeland, 1985).
Each motherchild dyad participated in a series of four problem-solving sit-
uations that required them to work together. The four tasks were (a) build-
ing block towers of specific proportions, (b) naming things with wheels, (c)
matching colors and shapes on a form board, and (d) tracing a preset pat-
tern through an etch-a-sketch maze. The mother was instructed to give the
child instructions and whatever help she thought was appropriate for the
task. The procedure was videotaped and then the mothers behaviors were
rated on a series of 7-point scales including structure and limit setting and
quality of instruction. Two independent raters viewing 87 tapes produced
an average correlation across all rating scales of .78. Scores on structure
and limit setting and quality of instruction were included in the early qual-
ity of caregiving composite.

Early home environment. The Home Observations for Measurement of

the Environment (HOME) for infants and toddlers (Caldwell & Bradley,
1984) was completed by trained observers at 30 months. The HOME is a
semistructured interview and observation instrument designed to measure
the quality of the childs home environment. The 48-item inventory con-
sists of six subscales: Emotional and Verbal Responsivity of the Parent, Ac-
ceptance of Childs Behavior, Organization of Physical and Temporal Envi-
ronment, Provision of Appropriate Play Material, Parent Involvement with
Child, and Opportunities for Variety in Daily Stimulation. The HOME is a
532 Journal of School Psychology

reliable and valid indicator of dimensions of the home environment impor-

tant for cognitive development (Bradley & Caldwell, 1981, 1982, 1984;
Bradley, Caldwell, & Elardo, 1979; Bradley, Caldwell, & Rock, 1988). This
study used the total score based on a composite of the six subscale scores
in the early home environment composite.

SES. ES scores at third grade were used in this study. An overall household
index of SES was calculated as the mean of Z-scores from at least two of
three relevant sources. The SES index included the revised Duncan Socio-
economic Index household score (Duncan, 1961; Stevens & Featherman,
1981), mothers level of education, and household income. SES indices
based on Z-score means were transformed into T-scores to produce posi-
tively scaled distributions for these variables.

Parent involvement at school. In both first and sixth grade, teachers pro-
vided information regarding whether or not the students parents were in-
volved in conferences or contacted the school. Classroom teachers used a 6-
point rating scale that provided descriptions ranging from a highly involved
parent to an uninvolved parent. The scale ranged from actively involved (6)
to uninvolved (1).

Measures and Procedures: Child Factors

Wechsler Intelligence Scale for Children-Revised (WISC-R). Three sub-

tests from the WISC-R (Block Design, Vocabulary, and Similarities; Wech-
sler, 1974) were administered to children as part of the third-grade assess-
ment, and the subtests yield a derived IQ score. The scale scores from this
short form are highly correlated with Full Scale IQs, and this brief form has
been demonstrated to have adequate reliability and validity for research
purposes (Sattler, 1982).

Peabody Individual Achievement Test (PIAT). The PIAT (Dunn & Markw-
ardt, 1970) was administered as part of the assessments conducted follow-
ing the first and sixth grade. The PIAT provided an assessment of achievement
in five areas: Mathematics, Reading Recognition, Reading Comprehension,
Spelling, and General Information. All of the subtests were administered in
the first-grade assessment and all but the general information subtest were
administered in the sixth-grade assessment. Grade standard scores were
used for analyses to avoid biases due to early grade retention.

Problem behaviors. The Child Behavior ChecklistTeacher Form (CBCL-

T; Edelbrock & Achenbach, 1985) was used to assess problem behaviors dis-
played by students in the classroom setting. This measure consists of short
 Jimerson et al. 533

descriptions of 113 problem behaviors associated with psychopathology in

children ages 611 years. The CBCL-T scores have been reported to have
excellent reliability and validity (Achenbach & Edelbrock, 1980; Edel-
brock & Achenbach, 1985). The total problem behavior scores from Grades
1 and 6 and age 16 were used in this study.

Peer competence. For the peer competence measure, the classroom

teacher read a description of a socially competent and popular child. The
teacher was then asked to rank order the children in his/her classroom,
with the child most closely resembling the description to be ranked at the
top. The childs scores on these measures consisted of a ratio of the childs
rank divided by the number of students in the class. Because only a single
teacher did these rankings, reliability figures are not available. However, in
a separate study, multiple counselors independently performed these rank-
ings following four weeks of summer day camp, with interrater reliability
ranging from .64 to .81 on peer competence (Elicker et al., 1992). The
teacher rankings themselves show significant stability from year to year, sig-
nificant correlations with behavior problems, and significant correlation
with observed peer competence (Hiester, Carlson, & Sroufe, 1993). Peer
competence ratings from Grades 1 and 6 and age 16 were used in this study.

Woodcock-Johnson Achievement Test-Revised (WJ-R). Two sections (Vo-

cabulary and Math) of the WJ-R (Woodcock & Johnson, 1990) were admin-
istered to participants during the interview at age 16. Grade standard scores
were used in comparisons to avoid biases due to early grade retention. The
achievement score at age 16 was used as a contemporaneous measure of ac-
ademic achievement in this study.

Analyses
Initially, descriptive statistics and t-tests between groups were completed for
each variable in this study. Data analyses to explore the proposed develop-
mental model were completed in two phases. (Preliminary hierarchical
multiple regression models demonstrated a significant relationship be-
tween the early quality of caregiving and the early home environment with
academic achievement, problem behaviors, peer relations, and parent
involvement in sixth grade. The results of these initial analyses demon-
strated the association between the early developmental history variables
and those variables purported in the literature to be primary predictors of
dropping out, that is, academic achievement, problem behaviors, peer rela-
tions, and parent involvement). Phase one of the analyses utilized four hier-
archical logistic regression models (Hosmer & Lemeshow, 1989) to exam-
ine the association of the early and contemporaneous variables in specific
domains (i.e., academic achievement, problem behaviors, peer relations,
534 Journal of School Psychology

and parent involvement) with high school status (e.g., Dropout or Tradi-
tional). Phase two used two discriminant function analyses (Klecka, 1980)
to examine the relative influence among the early predictors and the in-
fluence of the early predictors relative to the contemporaneous variables.
Thus, the models in the second phase of analyses force the most significant
predictors to emerge when considering all variables across domains and
years.
The first phase used hierarchical logistic regression models because this
method allows for a dichotomous dependent variable (i.e., high school sta-
tus: Traditional or Dropout) and both continuous and dichotomous inde-
pendent variables (Ensminger & Slusarick, 1992; Fagan & Pabon, 1990).
Logistic regression is similar to multiple regression, but instead of pre-
dicting the amount of variance accounted for in the dependent variable, it
predicts category membership within the dependent variable and com-
pares the predictions to the observed values in the data. To evaluate the in-
dividual steps within the logistic regression model, the chi-square statistic
compares the predicted and observed values of the dependent variable. In
the second phase, the discriminant function analyses provide an estimate
of the relative predictive power of the variables included in each model by
selecting those variables that are most highly associated with dropping out.
The relative small number of female participants in the dropout group
(n 12) prohibited separate analyses by gender. Thus, the hierarchical re-
gression models included gender as the first variable in order to force the
other variables to account for variance above and beyond gender. More-
over, regression models placing the gender variable last in the model dem-
onstrated the powerful predictive value of the other variables in the model
without the overall correct classification yielded by the gender variable
(e.g., early home environment at Step 1 generated an overall correct classi-
fication of .70 when gender was the last variable, compared to .75 when
gender was included at Step 1).

RESULTS
Means, Standard Deviations, and t-Tests
Each of the dependent and independent variables from the regression
models are listed in Table 1. With the exception of parent involvement at
first grade, t-tests demonstrated a significant difference between the means
for the two groups on all variables (p .01; see Table 2). No outliers were
identified in considering the distribution of scores on the variables in this
study.

Phase 1. Hierarchical Logistic Regression Models

Phase one of the analyses examined the association of the early home envi-
ronment and the quality of early caregiving with later school status (e.g.,
 Table 1
Correlations of All Variables Included in This Study (n 143)

Variables 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

1. Home environment 1.00 .47 .44 .37 .35 .43 .28 .25 .09 .22 .31 .19 .40 .31 .35 .34 .22 .02 .22
2. Quality of caregiving 1.00 .49 .45 .40 .44 .27 .22 .24 .35 .22 .27 .41 .37 .39 .23 .24 .04 .33
3. WPPSI (64 months) 1.00 .72 .36 .34 .22 .25 .14 .28 .25 .29 .63 .54 .58 .23 .18 .09 .29
4. WISCR (Grade 3) 1.00 .27 .37 .23 .28 .15 .22 .16 .27 .69 .60 .63 .18 .13 .01 .27
5. SES (42 Months) 1.00 .67 .05 .29 .20 .22 .27 .27 .21 .38 .40 .14 .28 .11 .22
6. SES (Grade 3) 1.00 .18 .30 .16 .24 .34 .35 .30 .46 .51 .14 .33 .02 .30
7. Problem behaviors
(Grade 1) 1.00 .48 .18 .53 .44 .17 .31 .28 .28 .23 .29 .11 .35
8. Problem behaviors
(Grade 6) 1.00 .33 .45 .58 .36 .22 .33 .32 .06 .54 .10 .40
9. Problem behaviors
(16 years) 1.00 .28 .22 .38 .21 .31 .28 .01 .23 .10 .36
10. Peer competence
(Grade 1) 1.00 .42 .33 .31 .33 .33 .22 .32 .09 .30
11. Peer competence
Jimerson et al.

(Grade 6) 1.00 .37 .26 .21 .25 .18 .51 .22 .32
12. Peer competence
(16 years) 1.00 .30 .30 .32 .07 .25 .16 .37
13. Academic achievement
(Grade 1) 1.00 .60 .64 .21 .14 .08 .20
14. Academic achievement
(Grade 6) 1.00 .87 .15 .26 .03 .30
15. Academic achievement
(16 years) 1.00 .16 .24 .05 .32
16. Parent involvement
(Grade 1) 1.00 .01 .06 .14
17. Parent involvement
(Grade 6) 1.00 .39 .40
18. Gender 1.00 .11
19. High school status
535

(19 years) 1.00

WPPSI Wechsler Preschool and Primary Scale of Intelligence, WISCR Wechsler Intelligence Scale for ChildrenRevised, SES socioeconomic status.
536 Journal of School Psychology

Table 2
Descriptive Statistics for All Variables (n 143)

Dropouts Traditional
Variable M (SD) M (SD) t

Home environment
(30 months) 30.7 (7.3) 34.1 (6.7) 2.58*
Quality of caregiving
(1242 months) .4 (1.6) .4 (1.5) 3.01*
WISCR IQ (Grade 3) 100.7 (12.6) 108.6 (12.9) 3.38*
SES (Grade 3) 46.0 (7.1) 53.1 (11.5) 4.26*
Problem behaviors
(Grade 1) 60.2 (10.5) 51.8 (10.1) 4.41*
Problem behaviors
(Grade 6) 60.3 (8.1) 52.4 (8.6) 5.26*
Problem behaviors
(16 years) 59.4 (8.7) 53.0 (7.3) 4.26*
Peer competence
(Grade 1) 36.6 (28.3) 55.3 (26.4) 3.71*
Peer competence
(Grade 6) 41.1 (26.6) 61.9 (28.8) 4.19*
Peer competence
(16 years) 37.8 (26.4) 59.3 (23.8) 4.61*
Academic achievement
(Grade 1) 107.8 (12.7) 112.9 (11.4) 2.27*
Academic achievement
(Grade 6) 98.9 (9.4) 105.3 (9.3) 3.77*
Academic achievement
(16 years) 97.3 (9.8) 106.0 (12.7) 4.43*
Parent involvement
(Grade 1) 4.1 (1.1) 4.5 (1.2) 1.70
Parent involvement
(Grade 6) 3.7 (1.1) 4.7 (1.0) 4.96*

WISCR Wechsler Intelligence Scale for ChildrenRevised, SES socioeconomic

status; *p .01.

Dropout or Traditional) using four hierarchical logistic regression models.

In addition, these analyses included contemporaneous (age 16) variables
consistent with the elementary school predictors (i.e., problem behaviors,
peer relations, academic achievement, and parent involvement) in the re-
gression models to examine their association with later school status (i.e.,
Dropout or Traditional). The childs gender, early home environment (age
30 months), quality of early caregiving (age 642 months), IQ (third
grade), and SES (third grade) were entered into each of the models fol-
lowed by the elementary school predictors (first and sixth grade) and con-
temporaneous (age 16) variables. Because it was not possible to complete
separate analyses for males and females, gender was entered into each of
the hierarchical regression models first that required subsequent variables
to predict above and beyond this variable. Parsing the variables into models
 Jimerson et al. 537

by category (i.e., problem behaviors, peer competence, academic achieve-

ment, and parent involvement) allows for chronological analyses within
each of these four categories. For example, (a) Do problem behaviors at
first grade predict high school status above and beyond ones gender, early
home environment, quality of caregiving, IQ, and SES? (b) Do problem be-
haviors at sixth grade predict high school status above and beyond ones
gender, early home environment, quality of caregiving, IQ, SES, and prob-
lem behaviors in first grade? and (c) Do problem behaviors at age 16 pre-
dict high school status above and beyond ones gender, early home envi-
ronment, quality of caregiving, IQ, SES, problem behaviors at first grade,
and problem behaviors at sixth grade? Although each of the variables in-
cluded were independently and significantly related to high school status
at age 19, these logistic regression models provide further information by
placing the variables in the model chronologically, which examines the pre-
dictive power of later variables after earlier predictors have been entered.
The first logistic regression model considered the prediction of high
school status from early variables and problem behaviors through age 16
(see Table 3, Model 1). Results of the first model illustrated that at age 16,
the childs gender, early home environment, quality of early caregiving,
SES, and problem behaviors in first grade were each significantly associated
with high school status at age 19. The childs problem behaviors in sixth
grade did not contribute significantly, 2(7) 1.08. The correct prediction
rate reached its highest point at Step 3 with the childs gender, early home
environment, and quality of early caregiving (77%). After entering IQ, SES,
and aggressive behaviors in first grade, sixth grade, and age 16, the correct
prediction rate was also 77% (see Table 3, Model 1). If the latter variables
were better predictors than the early variables, we would expect a better
overall correct prediction rate. Although IQ was significant when entered
into the model (p .03), the overall correct classification (74%) actually
appears slightly lower than the previous step of the model. This is possible
because a discriminant function model actually considers four cells,
whereas the overall correct classification percentage emerges from only the
two cells containing correctly classified students (i.e., Dropouts and Tradi-
tional). Thus, although IQ significantly changes the number of students in
each cell, the overall correct classification is not improved; it is important
to note the direction of change in the overall correct classification per-
centage.
The second logistic regression model considered the prediction of high
school status from early variables and peer competence through age 16
(see Table 3, Model 2). Results of these analyses illustrated that the childs
gender, early home environment, quality of early caregiving, IQ, SES, peer
competence in first grade, and peer competence at age 16 were each sig-
nificantly associated with high school status at age 19. The childs peer com-
petence in sixth grade did not contribute significantly, 2(7) 1.86. The
 538

Table 3
Developmental History Predictors of High School Dropout Status (n 143)

Model 1. Model 2. Model 3. Model 4.

Problem Peer Academic Parent
Behaviors Competence Achievement Involvement

Step Independent Variables 2 CC% 2 CC% 2 CC% 2 CC%

1. Gender 6.16** 69 6.16** 69 6.16** 69 6.16** 69

Home Environment
2. (30 months) 10.13*** 75 10.13*** 75 10.13*** 75 10.13*** 75
Quality of caregiving
3. (1240 months) 6.13** 77 6.13** 77 6.13** 77 6.13** 77
4. WISCR/IQ (Grade 3) 4.61* 74 4.61* 74 4.61* 74 4.61* 74
5. SES (Grade 3) 3.84* 75 3.84* 75 3.84* 75 3.84* 75
6. Grade 1 6.94** 77 3.61* 76 .42 75 .28 74
Journal of School Psychology

7. Grade 6 1.08 76 1.86 75 3.63* 77 8.00** 76

8. 16 years 5.90** 76 3.54* 76 1.08 76

Note. Dependent variables is high school status; CC% Correct classification percentage, WISCR Wechsler Intelligence Scale for ChildrenRevised, SES socioeconomic
status; *p .05, **p .01, ***p .001.
 Jimerson et al. 539

correct prediction rate at Step 3 with the childs gender, early home envi-
ronment, and quality of early caregiving (77%) was the highest in this
model (see Table 3, Model 2). Adding the peer competence variables in
place of the behavior problems variables in the fifth, sixth, and seventh
steps of this model yield similar results to the prior model.
The third logistic regression model considered the prediction of high
school status from early variables and academic achievement through age
16 (see Table 3, Model 3). Results of the fourth logistic regression model
illustrated that the childs gender, early home environment composite,
quality of early caregiving composite, IQ, SES, and academic achievement
at age 16 were each significantly associated with high school status at age
19. The childs academic achievement in first grade and academic achieve-
ment in sixth grade did not contribute significantly, 2(6) 0.42; 2(8)
1.08. At Step 3 with the childs gender, early home environment, and qual-
ity of early caregiving in the model, the overall correct classification rate
was 77%, which is the same correct prediction rate at Step 7 after entering
IQ, SES, academic achievement in first grade, and academic achievement
in sixth grade (see Table 3, Model 3). Prior studies consistently report low
achievement as a strong correlate of dropping out. Our results indicate that
the quality of the home environment and early caregiving prior to school
entry are powerful predictors of dropping out. Following the above vari-
ables, only achievement in sixth grade significantly contributed to the pre-
diction of dropping out when the early family and home variables were en-
tered.
The fourth logistic regression model considered the prediction of high
school status from early variables and parent involvement through sixth
grade (see Table 3, Model 4). Results of the fourth logistic regression
model illustrated that the childs gender, early home environment compos-
ite, quality of early caregiving, IQ, SES, and parent involvement in sixth
grade were each significantly associated with high school status at age 19.
The parent involvement in first grade did not contribute significantly, 2(8)
1.08. At Step 3 with the childs gender, early home environment, and qual-
ity of early caregiving, the correct classification rate was 77%, the highest
in the model (see Table 3, Model 4).
The results of the first phase of analyses demonstrate the importance of
the early home environment and the early quality of caregiving in pre-
dicting which students will later drop out. In addition, as found in prior re-
search, these models illustrate the association of IQ and SES with dropping
out and the relative influence (within category) of the elementary school
predictors from first and sixth grade and the age 16 variables.

Phase 2. Discriminant Function Analyses

In phase two of the analyses, two discriminant function analyses were calcu-
lated in order to identify the combination and order of variables maximiz-
ing the discrimination between Dropouts and Traditional students. The
540 Journal of School Psychology

discriminant function models differ from the prior hierarchical logistic re-
gression analyses in several ways. First, each discriminant function model
contains variables from across categories (i.e., problem behaviors, peer
competence, academic achievement, and parent involvement) rather than
examining one category at a time, as in the above logistic regression mod-
els. Including variables across categories in one model allows for between-
category comparisons (e.g., problem behaviors vs. academic achievement).
Second, the variables are considered simultaneously rather than being
forced into a prespecified hierarchical model. Thus, the results of these dis-
criminant function analyses yield the best overall predictors considering
the relative predictive power of all variables included in each model. For
example, although problem behaviors in sixth grade did not demonstrate
a significant contribution above and beyond the other variables considered
in the hierarchical regression model (above), problem behaviors emerged
as one of the best overall predictors in the discriminant function model
(below) where all variables (through sixth grade) were considered simulta-
neously. The first model included all of the variables through Grade 6: the
childs gender, early home environment (age 30 months), quality of early
caregiving (age 642 months), IQ (third grade), SES (third grade), peer
competence (first and sixth grade), problem behaviors (first and sixth
grade), academic achievement (first and sixth grade), and parent involve-
ment (first and sixth grade). In addition, a discriminant function model
was designed to examine the relative predictive power of the early variables
and the contemporaneous variables with later school status. The second
model included all of the above variables in addition to the age 16 variables
(i.e., peer competence, academic achievement, and problem behaviors).
The first discriminant function model examined the prediction of high
school status from early development and elementary school (see Table 4).
The results demonstrate that problem behaviors (first and sixth grade),
quality of caregiving (age 642 months), parent involvement (sixth grade),
the childs gender, and SES (third grade; Wilkss lambda .83, .77, .74, .72,
.71, and .70, respectively, p .001 for all variables selected) were the best
predictors of later school status when considering the variables from early
development and elementary school. The overall correct classification rate
was 75%. The correct category membership beat the base rates: for the
Dropouts, the above variables resulted in correct classification of 67% (base
rate 30%) of the students, whereas 79% (base rate 69%) of the Tradi-
tional students were correctly classified. Achievement did not emerge
among the most valuable predictors through Grade 6.
The second discriminant function model examined the prediction of
high school status considering early development, elementary school, and
the contemporaneous variables (see Table 5). The variables with the high-
est discriminating power were parent involvement (sixth grade), and prob-
lem behaviors (first grade), followed by peer competence (age 16), prob-
lem behaviors (age 16), the childs gender, quality of caregiving (age 642
 Jimerson et al. 541

Table 4
Discriminant Function Analysis Model 1: Includes Early Home
Environment, Early Quality of Caregiving, Gender, IQ, SES, Elementary
School Predictors at First and Sixth Grade Classifying High School Status
at Age 19 Years

Predicted Group Membership

Actual Group n Dropouts Traditional

Dropouts 43 29 (67%) 14 (33%)

Traditional 100 21 (21%) 79 (79%)
Overall % of cases grouped correctly: 75%
Highest Discriminating Variables Wilkss Lambda F
Problem behaviors (Grade 6) .83 29.17*
Quality of caregiving .77 20.61*
(12 & 40 months)
Parent involvement (Grade 6) .74 16.02*
Problem behaviors (Grade 1) .72 13.43*
Gender .71 11.25*
SES (Grade 3) .70 9.69*
Home environment (30 months) n.s. n.s.
Peer competence (Grade 1) n.s. n.s.
Peer competence (Grade 6) n.s. n.s.
WISCR IQ (Grade 3) n.s. n.s.
Academic achievement (Grade 1) n.s. n.s.
Academic achievement (Grade 6) n.s. n.s.

Note. n.s. (not significant) variables are listed in descending order of relative discrima-
tion; SES socioeconomic status; WISCR Wechsler Intelligence Scale for Children
Revised; *p .001.

months), and academic achievement (first grade and age 16; Wilkss
lambda .83, .75, .69, .67, .65, .64, .63, and .61, respectively, p .001 for
all variables selected). The overall correct classification rate was 78% (82%
of Dropouts, base rate 30%; 77% of Traditional students, base rate
69%; see Table 5).
The discriminant function analyses further illustrate the importance of
the early development, elementary school, and contemporaneous variables
in predicting which students will drop out at age 19. In the model consider-
ing variables through sixth grade, neither IQ nor achievement was selected
among the highest discriminating variables. Although the discriminant
function analyses and hierarchical logistic regression analyses answer differ-
ent questions, the results of both provide converging evidence suggesting
the importance of the early variables predicting high school status at age 19.

DISCUSSION
This study examined dropping out of high school using longitudinal data
collected from age 6 months through age 19 years. Descriptive statistics, re-
gression models, and discriminant function analyses were used to examine
542 Journal of School Psychology

Table 5
Discriminant Function Analysis Model 2: Includes Early Home
Environment, Early Quality of Caregiving, Gender, IQ, SES, Elementary
School Predictors at First and Sixth Grade and Age 16 Years Classifying
High School Status at Age 19 Years

Predicted Group Membership

Actual Group n Dropouts Traditional

Dropouts 43 35 (82%) 08 (18%)

Traditional 100 23 (23%) 77 (77%)
Overall % of cases grouped correctly: 75%
Highest Discriminating Variables Wilkss Lambda F
Parent involvement (Grade 6) .83 26.66*
Problem behaviors (Grade 1) .75 21.64*
Peer competence (16 years) .69 18.95*
Problem behaviors (16 years) .67 15.79*
Gender .65 13.28*
Quality of caregiving .64 11.56*
(12 & 40 months)
Academic achievement (Grade 1) .63 10.18*
Academic achievement (16 years) .61 9.50*
Home environment (30 months) n.s. n.s.
Peer competence (Grade 1) n.s. n.s.
Peer competence (Grade 6) n.s. n.s.
WISCR IQ (Grade 3) n.s. n.s.
SES (Grade 3) n.s. n.s.
Problem behaviors (Grade 6) n.s. n.s.
Academic achievement (Grade 6) n.s. n.s

Note. n.s. (not significant) variables are listed in descending order of relative discrima-
tion; SES socioeconomic status; WISCR Wechsler Intelligence Scale for Children
Revised; *p .001.

the influence of individual, family, and school-related factors on dropping

out. Results are consistent with a conceptual model that suggests dropping
out is a process that begins early in development and continues through
the time a student formally withdraws from school. As educators and re-
searchers seek to better understand the processes of development, this
study provides important information toward this goal. School psycholo-
gists and other educational professionals are encouraged to provide early
intervention and prevention programs to influence subsequent develop-
ment and achievement trajectories.
The process of dropping out begins prior to the child entering school.
The first set of analyses illustrated the association of the early home envi-
ronment and the quality of early caregiving in predicting high school status
at age 19. The childs early home environment and quality of early care-
giving emerged as powerful predictors of whether a student remained in a
traditional program or dropped out of high school. Measures of SES, IQ,
problem behaviors, academic achievement, peer competence, and parent
 Jimerson et al. 543

involvement from elementary school through adolescence were also sig-

nificant predictors of high school status at age 19. However, it is notable
that these variables did not add to the overall predictive power of measures
from the first 3 years.
Further analyses explored the relative predictive power of variables in-
cluded in this study. The first discriminant function analysis including vari-
ables measured through sixth grade demonstrated that the most important
variables forecasting future dropout status were behavior problems, quality
of caregiving, parent involvement, peer competence, gender, and SES.
Note that problem behaviors in sixth grade was selected as the best pre-
dictor of dropping out. In addition, the relative importance of the early
quality of caregiving was illustrated, as this variable was selected second. Al-
though academic achievement and IQ are recognized as early predictors of
dropping out, it is notable that in contrast to other measures, academic
achievement and IQ were not selected from the variables in this model. In
the second discriminant function model, parent involvement at sixth grade
and problem behaviors at first grade were selected before any of the age 16
variables. In addition, the early quality of caregiving and achievement in
first grade were included within the selected variables. These models illus-
trate the importance of the early quality of caregiving, parent involvement,
and behavior problems and, to a lesser extent, academic achievement and
peer competence in predicting dropping out at age 19. It should be noted
that peer competence is highly correlated with behavior problems.
The data from this study provide evidence supporting the proposal that
the childs early home environment and the quality of early caregiving are
important for predicting elementary school predictors of students who
drop out of high school. Furthermore, the childs gender, early home envi-
ronment, and the quality of early caregiving are able to discriminate be-
tween students who remain in a traditional educational program and those
who drop out. Thus, development prior to the elementary school years war-
rants further attention in our efforts to design and implement drop out in-
terventions and prevention programs.

Dropping out as a Developmental Process

Consistent with the snowballing effect described by Evans and DiBened-
etto (1990), results from this study support the view of dropping out as a
developmental process wherein events that occur early have an influence
on subsequent events. For instance, early experiences may affect self-
esteem and sense of agency that may directly influence school performance
and decisions to stay in school, and may also lay foundations for behavioral
control and relationships with teachers and peers that further propel the
individuals along a pathway towards dropping out. Success in school calls
upon numerous capacities for self-regulation that begin to be formed in
544 Journal of School Psychology

the early years (Teo, Carlson, & Sroufe, 1996). Moreover, quality of early
care likely gains its predictive power from its affect upon the ability to make
use of later opportunities and supports in the environment (Sroufe & Ege-
land, 1991). The context from which the child emerges when entering ele-
mentary school provides a critical foundation for subsequent academic
success.
Previous drop out research is seen in a new light from this developmental
process perspective. Many established predictors may be better conceptu-
alized as markers of presence on the pathway. Thus, truancy, disciplinary
problems, and failing grades in high school mark an advanced stage in the
drop out process that, in many cases, began years before. Likewise, peer re-
jection, behavior problems, and even low school achievement are better
viewed as midcourse markers of a process rooted in even earlier development.
This is an important distinction because high school markers such as par-
ticipation, attendance, problem behaviors, peer competence, and academic
achievement may accurately identify students who will drop out, but yet,
may not provide an opportunity to effectively intervene. For example, iden-
tifying an academic achievement problem in 10th grade (e.g., a student has
flunked all five courses) does not provide us with as great of an opportunity
to assist the student as does identifying an academic achievement problem
during elementary school (e.g., student is falling behind in reading and
math in second grade). Early elementary predictors of later dropouts may
provide more meaningful information for designing intervention programs.
Recognizing the importance of prior development influencing subsequent
adaptation in elementary school provides additional considerations in our
efforts to design drop out prevention programs. Thus, research efforts to
identify predictors of dropping out and various pathways must explore
early development and the elementary school years.
In the previous discussion of the transactional view of development, the
importance of prior experiences influencing subsequent experiences in an
ongoing way was emphasized (Sameroff, 1975, 1992; Sameroff & Chandler,
1975). From a transactional perspective, it is the confluence of early factors
that propel individuals toward alternative pathways. The proposed model
suggests that pathways toward dropping out begin early in a childs life;
thus, efforts to combat this problem should begin as early as possible.

Implications for Intervention

Despite the impressive predictive power of the early variables used in this
study, we do not view dropping out as determined in the first few years
of life. Along with others (Cairns & Cairns, 1994; Evans & DiBenedetto,
1990), we view early predictors as probabilistic, and pathways to dropping
out as heterogeneous. By no means are all cases of drop out forecast by fam-
ily experiences in the preschool period. In subsequent research, it will be
 Jimerson et al. 545

important to examine factors that move individuals onto a drop out path-
way only in later school years or deflect individuals from patterns of adapta-
tion typically associated with dropping out. This latter goal also should be
central in intervention and prevention research.
In terms of intervention and prevention, this study should encourage
school psychologists, professional educators, and policy makers to design
and implement appropriate assessments during early development, prefer-
ably during the first 5 years. The basic idea would be to cast a wide net
or to over-identify children who appear at risk of later academic failure.
As prior longitudinal studies emphasize, the trajectories of childrens adap-
tations are relatively fixed by third grade (Alexander & Entwisle, 1988);
hence, the importance of early identification and intervention.
Traditionally, our educational system attempts to assess basic academic
skills during the elementary years and, by third grade, these assessments ad-
equately differentiate students who are at risk of academic failure. The first
6 years at home, Preschool, Kindergarten, and first grade may be conceptu-
alized as a sensitive period (Bornstein, 1989) for developing the skills,
knowledge, and attitudes critical for school success (Coleman, 1987). Con-
ceptualized as a sensitive period, these early years take on unique impor-
tance in providing the developmental infrastructure on which later experi-
ence will build (Pianta & Walsh, 1996).
The transactional model of development also emphasizes early interven-
tion and prevention (Sameroff, 1992; Sameroff & Fiese, 1989). Because the
emphasis is on development as a holistic and dynamic process that begins
at conception, the later the intervention is implemented, the more there is
to overcome. Whether one considers later behaviors a product of evolving
internal working models (Sroufe & Fleeson, 1986) or established behavior
patterns (Patterson, 1986), each has resulted due to prior experiences.
Thus, the more years that transpire, the greater the commitment to estab-
lished behaviors. Applying this rationale to decreasing the drop out rate
suggests that factors such as low grades, poor peer relations, and more be-
havior problems are most likely results of the ongoing processes that may
indeed lead to school withdrawal, which may lead to difficulty seeking em-
ployment, and which may affect ones subsequent lifetime earnings. Thus,
early intervention that attempts to put all students on a trajectory toward
graduation must be the goal.

Directions for Further Research

In this study, it was not possible to examine the systemic effects of schools
because only a few subjects were ever in the same school. In addition, the
frequency of data collection limited the number of participants. Thus, in
this study, it was not possible to examine males and females separately or
African Americans and Caucasians separately because the sample was not
546 Journal of School Psychology

large enough. It is believed that further study of pathways for males and fe-
males may yield valuable insights to our understanding of children who
drop out. In addition, processes leading to high school drop out may vary
for students from different SES and racial backgrounds. Although it would
be difficult, what is needed is a longitudinal sample that includes early de-
velopmental measures and more subjects, so that additional analyses would
be possible (e.g., gender, race, additional interactions, model testing).
Current statistical procedures limit the number of variables one may ex-
plore relative to the size of the sample. This is problematic, given that the
dynamic process of development likely includes a very large number of in-
fluences and that our efforts to examine a large number of variables limits
us to a relatively small sample. Studies with more children may be better
able to examine the interactions of many of the variables considered in this
study. In addition, it is clear from prior research that school-related factors
such as the curriculum and other institutional-level variables affect stu-
dents decisions to withdraw from school or seek alternative educational
placements (Bryk & Thum, 1989; Fine, 1991; Kronick & Hargis, 1990; Rod-
erick, 1993; Rumberger, 1995; Witte & Walsh, 1990). An important goal for
future research will be to distinguish which different factors play dominant
roles among various pathways to dropping out. Early developmental fea-
tures warrant further emphasis in our efforts to better understand proc-
esses affecting dropping out prior to high school graduation.

ACKNOWLEDGMENTS
Preparation of this work and the research described herein were supported
by grants from the Office of Education (USD/E R117E30249), the Spencer
Foundation (FDN/B1545), the National Institute of Mental Health
(MHO864), and the Irving Harris Foundation.

REFERENCES
Achenbach, T., & Edelbrock, C. (1980). Child Behavior ChecklistTeacher Report Form. Burl-
ington, VT: University of Vermont Child Psychiatry Associates.
Ainsworth, M., Blehar, M., Waters, E., & Wall, S. (1978). Patterns of attachment. Hillsdale,
NJ: Erlbaum.
Ainsworth, M., & Wittig, D. (1969). Attachment and exploratory behavior of one year
olds in a strange situation. In B. Foss (Ed.), Determinants of infant behavior (Vol. 4, pp.
111136). London: Methuen.
Alexander, K., & Entwisle, D. (1988). Achievement in the first two years of school: Pat-
terns and process. Monographs of the Society for Research in Child Development, 53(20, Se-
rial No. 21).
Alpert, G., & Durham, R. (1986). Keeping academically marginal youth in school. Youth
in Society, 17(4), 346361.
Bornstein, M. (1989). Sensitive periods in development: Structural characteristics and
causal interpretations. Psychological Bulletin, 105, 179197.
 Jimerson et al. 547

Borus, M., & Carpenter, S. (1984). Choices in education. In M.E. Borus (Ed.), Youth and
the labor market (pp. 81110). Kalamazoo, MI: W.E. Upjohn Institute for Employment
Research.
Bradley, R., & Caldwell, B. (1981). The HOME inventory: A validation of the preschool
scale for black children. Child Development, 52, 708710.
Bradley, R., & Caldwell, B. (1982). The consistency of the home environment and its re-
lation to child development. International Journal of Behavioral Development, 5, 445465.
Bradley, R., & Caldwell, B. (1984). The relation of infants home environments to
achievement test performance in first grade: A follow-up study. Child Development, 55,
803809.
Bradley, R., Caldwell, B., & Elardo, R. (1979). Home environment and cognitive devel-
opment in the first two years of life: A cross-lagged panel analysis. Developmental Psy-
chology, 15, 246250.
Bradley, R., Caldwell, B., & Rock, S. (1988). Home environment and school perfor-
mance: A ten-year follow-up and examination of three models of environmental ac-
tion. Child Development, 59, 852867.
Bretherton, I., & Waters, E. (1985). Growing points of attachment theory and research.
Monographs of the Society for Research in Child Development, 50, 147165.
Brooks-Gunn, J., Guo, G., & Furstenburg, F. (1993). Who drops out of and who contin-
ues beyond high school? A 20 year follow-up of urban black youth. Journal of Research
on Adolescence, 3(3), 271294.
Bryk, A., & Thum, M. (1989). The effects of high school organization on dropping out:
An exploratory investigation. American Educational Research Journal, 26(3), 353383.
Cairns, R., & Cairns, B. (1994). Lifelines and risks: Pathways of youth in our time. Cambridge,
England: Cambridge University Press.
Cairns, R., Cairns, B., & Neckerman, H. (1989). Early school drop out: Configurations
and determinants. Child Development, 60(6), 14371452.
Caldwell, B., & Bradley, R. (1984). Observation of the home environment. Little Rock, AK:
University of Arkansas.
Carlson, E. A., Sroufe, L. A., Collins, W. A., Jimerson, S., Weinfield, N., Hennighausen,
K., Egeland, B., Hyson, D. M., Anderson, F., & Meyers, S. E. (1998). Early environ-
mental support and elementary school adjustment as predictors of school adjustment
in middle adolescence. Journal of Adolescent Research, 14(1), 7294.
Coleman, J. (1987). Families and schools. Educational Researcher, 16, 3238.
Dornbusch, S., Ritter, P., Leiderman, P., Roberts, D., & Fraleigh, M. (1987). The relation
of parenting style to adolescent school performance. Child Development, 58, 1244
1257.
Dryfoos, J. (1990). Adolescents at risk: Prevalence and prevention. New York: Oxford Univer-
sity Press.
Duncan, O. (1961). A socioeconomic index for all occupations. In A. J. Reiss, Jr. (Ed.),
Occupations and social status (pp. 109138). New York: Free Press.
Dunn, L., & Markwardt, F. (1970). The Peabody Individual Achievement Test. Circle Pines,
MN: American Guidance Service.
Edelbrock, C., & Achenbach, T. (1985). Manual for the teachers report form and teacher ver-
sion of the child behavior profile. Burlington, VT: University of Vermont.
Egeland, B., & Abery, B. (1991). A longitudinal study of high-risk children: Educational
outcomes. International Journal of Disability, Development, and Education, 38(3), 271
287.
Egeland, B., & Brunnquell, D. (1979). An at-risk approach to the study of child abuse:
Some preliminary findings. Journal of the American Academy of Child Psychiatry, 18, 219
235.
548 Journal of School Psychology

Ekstrom, R., Goertz, M., Pollack, J., & Rock, D. (1986). Who drops out of high school
and why? Findings from a national longitudinal study. Teachers College Record, 87(3),
356373.
Elicker, J., Englund, M., & Sroufe, L. A. (1992). Predicting peer competence and peer
relationships in childhood from early parentchild relationships. In R. Parke and G.
Ladd (Eds.), Family-peer relationships: Modes of linkage (pp. 77106). Hillsdale, NJ: Erl-
baum.
Ensminger, M., Lamkin, R., & Jacobson, N. (1996). School leaving: A longitudinal per-
spective including neighborhood effects. Child Development, 67, 24002416.
Ensminger, M., & Slusarick, A. (1992). Paths to high school graduation or drop out: A
longitudinal study of a first-grade cohort. Sociology of Education, 65(2), 95113.
Erickson, M., Sroufe, L. A., & Egeland, B. (1985). The relationship between quality of
attachment and behavior problems in preschool in a high risk sample. Child Develop-
ment Monographs, 50(12), 147166.
Evans, I., & DiBenedetto, A. (1990). Pathways to school drop out: A conceptual model
for early prevention. Special Services in School, 6, 6380.
Fagan, J., & Pabon, E. (1990). Contributions of delinquency and substance use to school
drop out among inner-city youth. Youth and Society, 21(3), 306354.
Feldhusen, J., Thurston, J., & Benning, J. (1973). A longitudinal study of delinquency
and other aspects of childrens behavior. International Journal of Criminology and Penol-
ogy, 1, 341351.
Fine, M. (1989). Why urban adolescents drop out of public high school. Teachers College
Record, 87(3), 393409.
Fine, M. (1991). Framing drop outs: Notes on the politics of an urban public high school. Albany,
NY: State University of New York Press.
Garnier, H., Stein, J., & Jacobs, J. (1997). The process of dropping out of high school:
A 19-year perspective. American Educational Research Journal, 34(2), 395419.
Hiester, M., Carlson, E., & Sroufe, L. A. (1993). The evolution of friendships in preschool, mid-
dle childhood, and adolescence: Origins in attachment history. Poster session presented at
the meeting of the Society for Research in Child Development, New Orleans, LA,
March.
Hosmer, D., & Lemeshow, S. (1989). Applied logistic regression. New York: Wiley.
Kirsch, I., Jungeblut, A., Jenkins, L., & Kolstad, A. (1993). Adult literacy in America: A first
look at the results of the national adult literacy survey. Report prepared by Educational
Testing Service with the National Center for Education Statistics, U.S. Department of
Education. Washington, DC: U.S. Government Printing Office.
Klecka, W. R. (1980). Discriminant analysis, Sage University Paper Series on Quantitative
Applications in the Social Sciences, No. 19. Beverly Hills, CA: Sage.
Kronick, R., & Hargis, C. (1990). Drop outs: Who drops out and whyand the recommended
action. Springfield, IL: Charles C. Thomas.
Lloyd, D. (1974). Analyses of sixth grade characteristics predicting high school drop out
or graduation. JSAS Catalog of Selected Documents in Psychology, 4, 90.
Lloyd, D. (1978). Prediction of school failure from third-grade data. Educational and Psy-
chological Measurement, 38, 11931200.
McMillen, M., Kaufmen, P., Hausken, E., & Bradby, D. (1993). Drop out rates in the United
States: 1992 (NCES No. 93-464). Washington, DC: Government Printing Office.
Morris, J., Ehren, B., & Lenz, K. (1991). Building a model to predict which fourth
through eight graders will drop out in high school. Journal of Experimental Education,
59(3), 286293.
National Center for Education Statistics. (1994). Drop out rates in the United States. Wash-
ington, DC: US Government Printing Office.
Oakland, T. (1992). School dropouts: Characteristics and prevention. Applied and Preven-
tative Psychology, 1(4), 201208.
 Jimerson et al. 549

Parker, J., & Asher, S. (1987). Peer relations and later personal adjustment: Are low-ac-
cepted children at risk? Psychological Bulletin, 102(2), 357389.
Patterson, G. (1986). Performance models for antisocial boys. American Psychologist, 41,
432444.
Pianta, R., & Walsh, D. (1996). High-risk children in schools: Constructing sustaining relation-
ships. New York: Routledge.
Renken, B., Egeland, B., Marvinney, D., Mangelsdorf, S., & Sroufe, A. (1989). Early child-
hood antecedents of aggression and passive-withdrawal in early elementary school.
Journal of Personality, 57, 257281.
Roderick, M. (1993). The path to dropping out: Evidence for intervention. Westport, CT: Au-
burn House.
Rumberger, R. (1987). High school dropouts: A review of issues and development. Re-
view of Educational Research, 57(2), 101121.
Rumberger, R. (1995). Dropping out of middle school: A multilevel analysis of students
and schools. American Educational Research Journal, 32(3), 583625.
Rumberger, R., Ghatak, R., Poulos, G., Ritter, L., & Dornbusch, S. (1990). Family influ-
ences on dropout behavior in one California high school. Sociology of Education, 63(4),
283299.
Sameroff, A. (1975). Transactional models in early social relations. Human Development,
18(12), 6579.
Sameroff, A. (1992). Systems, development, and early intervention. Monographs of the Soci-
ety for Research in Child Development, 57(6), 154163.
Sameroff, A., & Chandler, M. (1975). Reproductive risk and the continuum of care-tak-
ing casualty. In F. Horowitz, E. Hetherington, P. Scarr-Salapatek, & A. Siegal (Eds.),
Review of child development research (Vol. 4, pp. 187244). Chicago: University of Chi-
cago Press.
Sameroff, A., & Fiese, B. (1989). Transactional regulation and early intervention. In S.
J. Meisels & J. Shonkoff (Eds.), Early intervention: A handbook of theory, practice and analy-
sis. New York: Cambridge University Press.
Sattler, J. (1982). Assessment of children (2nd ed.). New York: Allyn & Bacon.
Sroufe, L. A., & Egeland, B. (1991). Illustrations of person and environment interaction
from a longitudinal study. In T. Wachs & R. Plomin (Eds.), Conceptualization and mea-
surement of organismenvironment interaction (pp. 6884). Washington, DC: American
Psychological Association.
Sroufe, L. A., Egeland, B., & Carlson, E. (1999). One social world: The integrated devel-
opment of parentchild and peer relationships. In A. Collins & B. Larson (Eds.), Min-
nesota symposium on child psychology. Hillsdale, NJ: Erlbaum.
Sroufe, L. A., & Fleeson, J. (1986). Attachment and the construction of relationships. In
W. Hartup & Z. Rubin (Eds.), Relationships and development (pp. 5172). Hillsdale, NJ:
Erlbaum.
Stevens, G., & Featherman, D. (1981). A revised socioeconomic index of occupational
status. Social Science Research, 10, 364395.
Teo, A., Sroufe, L. A., & Carlson, E. (1996). Socioemotional characteristics of academic
achievement. Journal of School Psychology, 36(2), 123138.
Wechsler, D. (1974). Wechsler Intelligence Scale for ChildrenRevised. New York: The Psycho-
logical Corporation.
Weis, L., Farrar, E., & Petrie, H. (1989). Dropouts from school: Issues dilemmas, and solutions.
New York: State University of New York Press.
Witte, J., & Walsh, D. (1990). A systematic test of the effective shoals models. Educational
Evaluation and Policy Analysis, 12(2), 188212.
Woodcock, R., & Johnson, M. (1990). Woodcock-Johnson Psycho-Educational BatteryRevised.
Allen, TX: DLM Teaching Resources.