You are on page 1of 33

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/236025515

Staying Back and Dropping Out: The Relationship Between Grade Retention and
School Dropout

Article · July 2007


DOI: 10.1177/003804070708000302

CITATIONS READS

53 327

4 authors, including:

Elizabeth Stearns Stephanie Moller


University of North Carolina at Charlotte University of North Carolina at Charlotte
49 PUBLICATIONS   609 CITATIONS    63 PUBLICATIONS   1,618 CITATIONS   

SEE PROFILE SEE PROFILE

Stephanie Potochnick
University of Missouri
17 PUBLICATIONS   363 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Elizabeth Stearns on 16 April 2016.

The user has requested enhancement of the downloaded file.


Staying Back and Dropping Out:
The Relationship Between
Grade Retention and School Dropout
Elizabeth Stearns
University of North Carolina at Charlotte
Stephanie Moller
University of North Carolina at Charlotte
Judith Blau
University of North Carolina at Chapel Hill
Stephanie Potochnick
University of North Carolina at Charlotte

Students who repeat a grade prior to high school have a higher risk of dropping out of high
school than do students who are continuously promoted. This study tested whether standard
theories of dropout—including the participation-identification model and the social capital
model—explain this link. Although the presence of variables, including academic achievement
and disciplinary problems, reduces the higher probability of retained students dropping out,
existing models of dropping out do not adequately explain the markedly higher probability of
dropping out for retained students. Regression decomposition reveals differences between
promoted and retained students in the importance of resources and illustrates that various
resources hold different levels of importance for white, black, and Latino students.

W
ith a greater number of school dis- filled these gaps by using data from the
tricts mandating grade retention for National Education Longitudinal Study of
students who do not perform well 1988 (NELS:88) to explore why retained stu-
on standardized tests, an old problem in edu- dents are more likely than are continuously
cation has gained increasing attention. promoted students to drop out of school. We
Students who repeat a grade are significantly investigated the efficacy of various theories of
more likely than are continuously promoted dropout in explaining the link between reten-
students to drop out. Numerous studies have tion and high school dropout, both early and
established this link (for a thorough review, late, for white, black, and Latino students.
see Jimerson, Anderson, and Whipple 2002). The link between retention and dropout is
Yet, few researchers have explicitly examined well established (Alexander, Entwisle, and
the reasons behind the relationship between Horsey 1997; Janosz et al. 1997; Roderick
retention and dropping out (Gottfredson, 1994; Rumberger 1995; Teachman, Paasch,
Fink, and Graham 1994; Roderick 1994) or and Carver 1996).1 In fact, in a review of 17
whether the probability of dropping out dif- studies, Jimerson et al. (2002) reported that
fers for Latino, black, and white retained stu- every study found a significant link between
dents throughout high school. Our study prior grade retention and dropping out. By

Sociology of Education 2007, Vol. 80 (July): 210–240 210


Staying Back and Dropping Out 211

some estimates, 15 percent of students aged most have not been systematically tested to
6–17 are retained in grade (Hauser 1999). explain why retained students drop out.
Furthermore, black students are almost twice Instead, these theories explain, more generally,
as likely as white students to be retained why students leave school. Second, it is unclear
(Moller et al. 2006). These are children who how useful these theories may be in explaining
have an elevated risk of dropping out. differential associations between retention and
Less well established, however, are the dropout among black, white, and Latino stu-
explanations of why retention may substantial- dents, especially considering research that has
ly influence dropping out, if indeed a causal shown how the dropout process differs by stu-
relationship exists (Gottfredson et al. 1994; dent’s race and ethnicity (Crowder and South
Kaplan, Peck, and Kaplan 1997; Roderick 1994; 2003; Rumberger and Rodriguez 2002; Stearns
Shepard and Smith 1989). Some scholars have and Glennie 2006). Third, researchers have not
posited that retention serves as an early warn- assessed the possibility that the utility of these
ing of impending academic failure (Alexander, theories may differ, depending on the timing of
Entwisle, and Dauber 2003). Others have spec- high school dropout. This is an important over-
ulated that the association reflects the difficulty sight, given evidence that various predictors of
that overage retainees may face in integrating dropping out vary in strength and significance
themselves into peer and school cultures at different stages of high school (Carbonaro
(Kaplan et al. 1997; Roderick 1994). 1998; Rumberger and Larson 1998; Rumberger
Nevertheless, few have systematically tested and Thomas 2000; Stearns and Glennie 2006;
their speculations (Gottfredson et al. 1994). Swanson and Schneider 1999).
What remains incontrovertible is that there is a In this article, we fill these gaps in the
substantial gap in the probability of dropout research by examining retained students’
between retained and continuously promoted greater probability of dropping out of school.
students. Our research questions include the following:
Meanwhile, several theories have been How do the three aforementioned models,
advanced to explain why all students who along with retention as an early warning,
drop out of school do so, including discus- explain retained students’ higher probability of
sions of how school context interacts with dropping out? Do these models explain drop-
students’ experiences to produce dropout ping out among black and Latino retained stu-
behavior. These models include the frustra- dents as well as they explain dropping out
tion–self-esteem model, which posits that among white retained students? Do they
school failure lowers students’ self-esteem explain both early and late dropout? We further
(Finn 1989); the participation-identification explore the link between retention and drop-
model, which contends that students who are ping out for both early and late dropouts
more engaged with school are less likely to through regression decomposition, a statistical
drop out (Mahoney and Cairns 1997; McNeal technique that has been used extensively in
1995; Newman 1991; Randolph et al. 2004; examinations of wage differentials between
Tinto 1993); and a social capital model, black and white workers (Blinder 1976; Fairlie
which examines the resources that students 2003; Oaxaca 1974). This method allows us to
draw from their relationships with teachers, determine which portion of the link between
parents, and peers and argues that teenagers retention and dropping out is accounted for by
with fewer academic-related relationship deficiencies in resources, defined by the afore-
resources are more likely to drop out of mentioned theories.
school (Gottfredson et al. 1994).
Further research is necessary to clarify
whether these theories help explain the associ-
ation between retention and dropping out. LITERATURE REVIEW AND
First, although some of these theories have HYPOTHESES
been applied in an ad hoc manner to explain
why retained students are more likely than are Several theories offer the potential for
continuously promoted students to drop out, explaining the link between grade retention
212 Stearns, Moller, Blau, and Potochnick

and dropping out. In this section, we proceed existing academic or disciplinary problems,
by explaining the process of retention and however, the experience of grade retention in
discussing why it is necessary to clarify why the educational system has the potential to
retained students are more likely to drop out introduce or exacerbate a number of prob-
than are continuously promoted students. lems for students. As Jimerson et al.
Then, we review three psychological and soci- (2002:442) stated: “The experience of being
ological theories that explain why all students retained may influence numerous factors
who drop out do so: the frustration–self- determined to be associated with dropping
esteem theory, the participation-identification out of high school (e.g., student’s self-
theory, and social capital theory. We discuss esteem, socioemotional adjustment, peer
how these theories are applied to retained relations, and school engagement).” First,
students. We argue that retained students are retained students have been shown to have
particularly at risk of dropping out, since lower self-esteem than their classmates even
these theories highlight risk factors that are before the retention decision (Alexander et al.
especially acute for them. 2003). Then, there is potential damage to the
children’s self-esteem resulting from the
Retention as a Social Process retention decision, which Byrnes (1989) high-
lighted. In an interview with “George,” a
Retention in grade is designed to identify stu- child who had been retained during elemen-
dents who lack the academic qualifications to tary school, Byrnes asked how he felt about
move to the next grade level (Alexander et al. the retention. George responded that he felt
2003; Dauber, Alexander, and Entwisle 1993; “sad” about the retention “because every-
Rodney et al. 1999). In reality, many factors, body laughs at [him].” (p. 123). He also
including race, socioeconomic status (SES), reported that the worst thing associated with
and family structure, play a role in deciding being retained was that “kids say ‘flunker’”
which students are retained (Dauber et al. (p. 123). Although Alexander et al. (2003)
1993; Guo, Brooks-Gunn, and Harris 1996; did not find damage to retained students’
Hallinan 2000; Randolph et al. 2004; Rodney self-esteem, this finding may have been due
et al. 1999). Nevertheless, the important dis- to the fact that they primarily studied ele-
tinction is that, on average, retained students mentary school students in an urban school
have lower achievement levels and/or more system where retention is normative. In addi-
disciplinary problems than do students who tion, academic self-concept may be particu-
are promoted continuously throughout larly harmed by retention. Differences in self-
school (Alexander et al. 2003; Dennenbaum esteem and academic self-concept, whether
and Kulberg 1994; Holmes and Matthews they exist prior to the retention decision, after
1984; Jimerson 1999; McCoy and Reynolds it, or both, also mark retained students as dif-
1999; Owings and Magliaro 1998; Phelps et ferent from continuously promoted students.
al. 1992; Walker and Madhere 1987). One Second, retention may lead retained stu-
would then reasonably expect that students dents to view the school system as one in
who are retained are more likely to drop out which they have experienced failure. As a
of school; research has borne out this expec- result, retained students may pursue success
tation (Jimerson 1999; Roderick 1994; in other, non-school-related, avenues. This
Rumberger 1995). If the main difference difficulty with engagement may only worsen
between retained and continuously promot- as the students age (Roderick 1994). Because
ed students lies in their various levels of retained students are older than their peers,
achievement or retained students’ early they may also be more susceptible to societal
inability to follow disciplinary norms, then pressures that pull them out of school
controlling for achievement or for disciplinary (Stearns and Glennie 2006). In other words,
problems should substantially reduce the sig- they may take on adult responsibilities in the
nificant relationship between retention and labor market and family realms in earlier
dropping out. grades because they are older than their class-
In addition to identifying those with pre- mates. Furthermore, these adult responsibili-
Staying Back and Dropping Out 213

ties may serve to pull them out of school as tion–self-esteem model, the participation-
the identity associated with that of “worker” identification model, and the social capital
or “parent” or “caregiver” becomes more model. These models predict that students
attractive than the identity associated with with deficiencies in self-esteem, attachment
“retained student.” Their age and the possi- to and engagement with school, and social
ble effects that retention has on retained stu- capital, respectively, are more likely to drop
dents’ success in engaging with school also out. Earlier, we discussed evidence that
mark retained students as different from con- retained students have particular difficulties in
tinuously promoted students. each of these three areas, so it is informative
Third, retention may rupture social bonds to examine the extent to which these theories
with peers and hurt students’ ability to bond can explain the link between retention and
with teachers later in their educational careers. dropping out.
Retention separates students from their same-
aged peers and may end friendships. Retained Frustration–Self-esteem Model As Finn
students must develop new peer groups (1989) described, the frustration–self-esteem
among their new classmates and overcome model explains why students who have expe-
their label of “flunker” while doing so. In fact, rienced academic difficulties drop out of
retained students may have fewer friends than school. In this model, unsuccessful school
continuously promoted students (“Repeating a outcomes, such as grade retention, lead to a
Grade” 1986). Since retained students reach reduction in students’ self-esteem. In an effort
puberty in earlier than modal grades, the dif- to boost their self-esteem, students turn away
ference between them and continuously pro- from academic effort and toward other
moted students may become more obvious to behaviors and venues that will boost their
their classmates, as well as to their teachers. self-esteem. This attempt to find activities in
Teachers may therefore be subconsciously influ- which they will be successful often results in
enced by the retention status, in that they may delinquency of various types and problem
expect and demand less academic achieve- behaviors, including dropping out of school.
ment from retained students and put less effort Thus, as Finn stated, “the youngster’s self-
into reaching them. view is a central mediator of problem behav-
In sum, there is good reason to suspect that ior” (p. 120). Nevertheless, self-esteem has
the experience of retention, in and of itself, is a not been shown to predict dropout status
risk factor for dropping out. There is a strong consistently among all students. Since it is dis-
evidentiary basis for considering retained stu- proportionately low among retained students
dents as more than simply overage under- (Alexander et al. 2003), we hypothesized that
achievers. Unfortunately, in the absence of pre- it may explain part of their higher probability
retention measurements, we cannot be certain of dropping out.
that the differences between retained students
and continuously promoted students are the Participation-Identification Model Whereas
result of retention. Hence, we do not make a the frustration–self-esteem model focuses
causal argument about the “effects” of reten- mostly on internal psychological processes, the
tion. What we can say, however, is that there participation-identification model moves
are differences, whether they are present prior beyond psychological processes to consider the
to retention or whether the process of retention student in the context of his or her relationships
introduces them, in these two populations. We in the school. Under ideal circumstances, stu-
ask how the differences in these two popula- dents can engage with school through the aca-
tions account for the gap in their probability of demic and the social arenas (Newmann 1991;
dropping out of high school. Tinto 1993). Academic engagement includes
arriving at school on time and going to class
Three Models of Dropout Behavior prepared with books, homework, and school
supplies (Smerdon 1999), while social engage-
Next, we discuss three models that attempt ment includes participation in a range of
to explain dropout behavior—the frustra- extracurricular activities. Following Tinto’s
214 Stearns, Moller, Blau, and Potochnick

(1993) argument, engagement with either the focused on a variety of behaviors that indicate
academic or the social realm may be enough to engagement with schooling, as well as mea-
keep students in school. Retention is thus a par- sures of the bonds that students form with a
ticular risk factor with respect to the participa- variety of people, including peers. Thus, we
tion-identification model, since retained stu- hypothesized that students who are seen as
dents may feel more alienated from the school popular are less likely to drop out of school
and be less likely to participate in both the for- than are others. Given the differences in
mal curriculum and the extracurriculum. Thus, retained and promoted students’ peer net-
we hypothesized that students, both retained works (“Repeating a Grade” 1986), this lack
and continuously promoted, who regularly of popular esteem may particularly harm
attend class prepared are less likely to drop out retained students.3
than are other students. Social capital can also be located in rela-
Furthermore, students who are engaged tionships with teachers. Positive bonds with
with school through participation in extracur- teachers reduce the probability of students
ricular activities are less likely to drop out of dropping out between the 10th and 12th
school than are students who do not partici- grades, particularly for students from socioe-
pate in extracurricular activities (Mahoney conomically disadvantaged backgrounds who
and Cairns 1997; McNeal 1995; Randolph et have struggled academically in the past
al. 2004). This relationship is particularly (Croninger and Lee 2001). At the same time,
strong for children whose teachers consider however, there is some reason to expect that
them to be at risk of dropping out early retained students may be less likely to seek
(Mahoney and Cairns 1997). Thus, we help from their teachers. The experience of
hypothesized that students who are socially retention may shape retained students’ per-
engaged with schooling are less likely to drop ceptions of their past schooling experiences
out of school than are those who are not. and alter the ways in which retained students
Given that retention may be a risk factor for deal with schooling in the future (Jimerson et
engagement, this lack of engagement may al. 2002). Thus, retained students may view
explain part of retained students’ higher the school and agents of the school as partic-
probability of dropping out. ularly intimidating and avoid seeking contact
with those agents later in their educational
Weakening Social Capital In contrast to careers. We hypothesized that a lack of these
the first two models, which have been tan- pro-social bonds with teachers increases the
gentially applied to the case of retained stu- probability of dropping out. If retention does,
dents, social capital theory has been explicit- indeed, break the bonds between students
ly applied to retained students. Social capital and schools, we would expect to find that a
can be located in a variety of different rela- lack of positive contact with teachers particu-
tionships, including those between students larly disadvantages retained students.
and teachers and those between students Aside from ties with peers and teachers,
and their parents. A social capital model social capital can also be located in familial
posits that retention may weaken students’ ties. Parents can increase their children’s lev-
pro-social bonds, thus leading to a higher els of social capital by interacting positively
probability of dropping out. Drawing on with their children; by implementing closure
Hirschi’s (1969) control theory, Gottfredson in their children’s networks; or by interacting
et al. (1994) argued that although retention closely with schools, other institutions, and
may increase the bond between students and other adults in their children’s lives
academic subject matter, it may also inhibit (Carbonaro 1998; Teachman et al. 1996). In
students’ abilities to form later bonds with pursuing these strategies, parents add to the
teachers and other students because of the social capital of their children and help to pre-
stigma that comes with retention.2 Although vent disengagement from schooling and tru-
their findings did not strongly support this ancy.
theory, their study concentrated on attach- In fact, there have been mixed findings
ment instead of engagement. In contrast, we about the importance of parental involvement
Staying Back and Dropping Out 215

in students’ educational careers. Parent-based growing body of research on the influence of


social capital influences educational and social educational structures on children’s achieve-
outcomes, but only for children who are tradi- ment, attainment, and aspirations has shown
tionally advantaged (McNeal 1999). Teachman that educational structures, such as retention,
et al. (1996) found that the nature of children’s affect black, Latino, and white children in dif-
and parents’ interaction about academic mat- ferent ways (Blau 2003; Moller et al. 2006).
ters predicts early dropout, but the amount of Blau (2003) argued that aspects of the formal
contact between parents and schools does not, educational system, such as retention and
net of other factors. In an analysis of 12th-grade tracking decisions, do not have as negative an
dropout status, Carbonaro (1998) discovered impact on the future aspirations and attain-
that communication between parents and ment of black students as they do on white
schools is positively associated with the proba- students because black communities are able
bility that students will stay in school, but that to provide support for children who are not
parental participation with the school is not sig- identified as “high achievers” in school.
nificantly associated with the probability of stu- Furthermore, while retention is often synony-
dents dropping out. We hypothesized that the mous with individual failure in white culture,
experience of retention may harm already-frag- it may symbolize institutional discrimination
ile bonds between parents and children or in other cultures (Blau 2003). Students may
between parents and schools and thus that react to this discrimination by persevering
these factors may explain part of the increased through greater participation in school or by
probability of dropout among retained stu- acting out and engaging in behavior that
dents. requires disciplinary action (Blau 2003; Moller
Students who experience a great deal of et al. 2006; Ogbu 1978). Thus, the experi-
mobility and frequent changes of schools can ence of retention may differ by race.
experience a dramatic loss in social capital as Furthermore, racial differences in the rea-
well. As Ream (2003:239) pointed out, these sons why students drop out and the influ-
moves can “disrupt social root systems, chal- ences on students’ decisions suggest that the
lenging the development and maintenance process of dropping out differs by racial
of social capital . . . by inhibiting students’ groups (Crowder and South 2003; Stearns
efforts to make new friends, adjust socially to and Glennie 2006; Velez 1989). These racial
a new school situation, and develop recipro- differences in processes that are associated
cal relations with school personnel.” These with high school dropout may be attributable
transfers are thus associated with dropping to the frustration–self-esteem, participation-
out of school at both early and late stages identification, and social capital models of
(Lee and Burkam 1992; Rumberger and dropping out because each of these theories
Larson 1998; Swanson and Schneider 1999). highlights large racial differences in resources.
School transience may be particularly harmful Indeed, black children have been consistently
for students who have characteristics, such as shown to have higher levels of self-esteem
prior grade retention, that suggest their lower than have white children (Bankston and Zhou
probability of establishing strong ties with 2002; Pallas et al. 1990). In contrast, whites
school personnel. Thus, we hypothesized that are generally more likely to participate in
changes of schools and dropping out are extracurricular activities than are children of
especially related among retained students. other racial/ethnic groups and to benefit aca-
demically from participation in such activities
(Brown and Evans 2002; Davalos, Chavez,
Race, Retention, and Dropping and Guardiola 1999; Eitle 2005; Eitle and Eitle
Out 2002). In addition, blacks and Latinos are less
likely than are students of other racial/ethnic
In addition to considering the link between groups to attend class prepared with home-
retention and dropping out, we also consid- work, books, and papers (Blau 2003).
ered the possibility that the mechanisms Finally, researchers have identified persis-
behind the link differ across racial groups. A tent differences with respect to social capital,
216 Stearns, Moller, Blau, and Potochnick

with various results for the dropout behavior, dents more likely to leave school for discipli-
academic achievement, and aspirations of nary reasons and older students more likely to
teenagers from different racial groups do so because they are struggling with acad-
(Crosnoe, Johnson, and Elder 2004; Faircloth emics (Stearns and Glennie 2006). Further-
and Hamm 2005; McNeal 1999; Ream 2003; more, other research has shown a divergence
Valadez 2002; Yan and Lin 2005). For in the size and significance of various predic-
instance, Latino students may not benefit tors at various stages of dropping out
from having teacher- and parent-based social (Carbonaro 1998; Rumberger and Larson
capital as much as students of other 1998; Swanson and Schneider 1999). Taken
racial/ethnic groups (Ream 2003; Valadez together, these studies have suggested that
2002). Therefore, we hypothesized that the there are qualitative differences between the
link between retention and dropping out, processes of dropping out early and dropping
including the factors that influence dropout out late that would be obscured by our con-
decisions, differ for members of different sidering all dropouts together.
racial groups, a possibility that we explore For the analyses of early dropout, all inde-
with regression decomposition. pendent variables were measured in the 8th
grade. For the analyses of late dropout, most
independent variables were measured in the
DATA AND METHODS 10th grade. When 10th-grade data were
unavailable, we used the 8th-grade measure.
We used data from NELS:88, a nationally rep- The key variables of interest were race and
resentative sample of eighth graders in 1988 retention. Black was coded 1 if the student
who were resurveyed in 1990, 1992, 1994, self-identified as black, as opposed to white,
and 2000. The National Center for Education Latino, or Asian; Latino was coded 1 if the stu-
Statistics (NCES) conducted the survey. We dent self-identified as Latino; and white was
included black, white, and Latino students coded 1 if the student self-identified as white.
who participated in the 1988, 1990, and Retention was coded 1 if the student was
1992 waves. We did not include Asian stu- retained at least once prior to the 8th grade.
dents in our analyses because of their low In the NELS, students and parents were asked
probabilities of being retained and of drop- whether the students had been retained dur-
ping out. This constraint prevented a reliable ing elementary school. We used students’
decomposition of dropout status by retention self-reports of retention and filled in missing
for Asians. cases with the reports from their parents.4
The analyses focused on two dependent Since we were interested in determining why
variables: early dropout and late dropout. retention is associated with dropout for black,
Early dropout was calculated from the 1990 Latino, and white students, we combined
observations on the status, whether dropout these two variables into black retained, Latino
or currently enrolled, of students who were in retained, white retained, black promoted, and
the 8th grade in 1988. Late dropout was cal- Latino promoted. White promoted was the
culated from the 1992 student-status vari- excluded category.
able. Late dropouts included students who We also controlled for two sociodemo-
were in school in the 8th and 10th grades but graphic variables that are correlated with
dropped out by the 12th grade. Students selection into retention: SES and family struc-
who were dropouts in the 10th grade were ture. We measured SES with the NCES-pro-
excluded from the analyses of late dropout. vided variable that captures father’s and
We analyzed early dropouts and late mother’s education and occupation and fam-
dropouts separately for two reasons. First, as ily income. The final demographic variable
students move through high school and was two-parent household, which is a dichoto-
through adolescence, expectations about mous variable indicating whether the student
their proper roles may change. Thus, they lived in a household with two parents, at least
may drop out for different reasons at different one of whom had to be a biological parent.
points during high school, with younger stu- The analyses controlled for characteristics
Staying Back and Dropping Out 217

of the student’s school. Public indicates activities. The variables that make up these
whether the child was attending a public (1) indices are described in detail in Appendix B.
or private (0) school in the 8th and 10th The principal- components analyses of the
grades, and urban shows whether the child 8th-grade variables produced two factors that
was living in an urban area in the 8th and explain 66 percent of the variation in these
10th grades. We also used control variables variables. The first factor, students’ participa-
that indicate in which area of the country the tion, represents students’ participation in
child was living in the 8th and 10th grades, extracurricular activities. The second factor,
since regional differences in dropout have school preparedness, reflects students’ atten-
been shown to be persistent (McMillen and dance and preparedness for class. The princi-
Kaufman 1996). ple-components analyses of the 10th-grade
Additional educational background vari- variables produced two similar factors. These
ables that are correlated with retention were two factors account for 58 percent of the vari-
also included in the analyses—scores on ation in the 10th-grade variables
achievement tests, educational aspirations, For social capital, we used measures of
and students’ misbehavior in school. bonds with parents, teachers, and peers. We
Achievement test scores are a composite of replicated Teachman et al.’s (1996) measures
math and reading scores from the 8th and of parent-school and parent-child connectivi-
10th grades that were used to predict early ty. Briefly, Teachman et al. constructed five
and late dropout, respectively. Educational index scores of parent-school and parent-
pessimism is a dichotomous variable measur- child involvement using the questions and
ing whether the student thought that he or response categories shown in Appendix A.
she would graduate from high school (coded The index scores were as follows: (1) the
0) or drop out of high school (coded 1), mea- child’s reports of parent-child interactions, (2)
sured in the 8th and 10th grades. The ques- the parent’s reports of parent-child interac-
tion was slightly different in the two waves.5 tions, (3) the child’s reports of parent-school
Because conduct is an important aspect interactions, (4) the parent’s reports of par-
of students’ educational experience and has ent-school interactions that involved contact-
been shown to predict significantly the prob- ing the school, and (5) the parent’s reports of
ability of students dropping out of school, we parent-school interactions that involved
also measured how well behaved in school doing things at the school.
students were in the 8th and 10th grades. Teachman et al. (1996) then conducted a
Factor loadings and variables6 that were used principal-components factor analysis on these
in creating the factors are presented in five scale scores and found that the first two
Appendix C.7 index scores formed a factor of parent-child
Other independent variables included interaction and the last three index scores
those that were generated from the frustra- formed a factor of parent-school interaction.
tion–self-esteem model, the participation- We replicated their steps with the parent-
identification model, and the social capital child and parent-school connectivity mea-
model of dropping out. Self-concept is an sures. We used the same measures of this vari-
NCES-created composite of variables, includ- able for the early-dropout and late-dropout
ing “I feel good about myself” and “I’m a per- analyses because data on parents were not
son of worth, equal to others.”8 We also cre- collected in the 10th grade, and these data
ated factor scores of students’ engagement account for a substantial part of the variable.
with academics and extracurricular activities Students can also derive social capital
at their schools. We created the student-par- through relationships with teachers. Thus, fol-
ticipation and student-preparedness factors lowing Croninger and Lee (2001), we created
from a principal-components analysis of the a factor score of the following variables: “the
index scores for (1) school attendance, (2) teaching is good,” “teachers praise my
class preparedness, (3) participation in gener- efforts,” “teachers are interested in students,”
al extracurricular activities, and (4) participa- “teachers put down students” (reverse
tion in academically oriented extracurricular coded), and “most of my teachers listen to
218 Stearns, Moller, Blau, and Potochnick

me.” High scores on this factor indicate that extent to which gaps in outcomes, such as
the students were not well bonded with the income, are due to gaps in resources between
teachers. two populations. Oaxaca (1974) and Blinder
The final variables measure aspects of peer (1976) developed this technique to examine
relations. Peer popularity is a dichotomous the extent to which the gender gap in earn-
variable coded 1 if the student reported that ings, among others, was due to the different
peers saw him or her as “very” or “some- resources that men and women bring to the
what” popular and 0 if he or she was “not at labor market.
all” popular. School change was measured as Fairlie (1999, 2003) revised the Oaxaca-
the number of times the student changed Blinder technique to predict nonlinear out-
schools prior to the eighth grade. Means and comes. To make these models more consis-
standard deviations of all variables for the tent with the earnings analogy, we coded the
early- and late-dropout analyses are present- dependent variable so dropouts received a
ed in Table 1. value of 0 and students who had an uninter-
rupted high school education (without drop-
Analytic Strategy ping out) received a value of 1. In notational
format, the gap in the average value of the
We used logistic regression to analyze the dependent variable between promoted and
dependent variables. Throughout the analy- retained students can be written as
ses, we used nested models to examine the  Np Nr

–p
moderating effects of our control variables on
Σ Σ
r ^ ^
Y –Y = F(XPi β R)– F(Xri β R) +
the relationship between grade retention and p R
dropping out. We also used a Huber-White  Np
i=1 N i=1 N
N
p 
Σ Σ
^ ^
estimation of variance that corrects standard F(XPi β P) – F(XPi β R) ,
errors for correlation within clusters (STATA i=1 i=1
NR NR
Corporation 2003). In the early-dropout
logistic regression analyses, clusters are where NP is the sample size for promoted stu-
defined as the student’s 8th-grade school, dents; NR is the sample size for retained stu-
and in the late-dropout analyses, clusters are dents; and µR are the coefficient estimates for
defined as the student’s 10th-grade school. retained students, associated with the inde-
This correction produced robust standard pendent variables Xi. The first term in the
errors that account for the clustering of stu- solution represents the part of the gap
dents within schools.9 between retained and promoted students
We also used regression decomposition to that results from group differences in the dis-
examine the relationship between retention tributions of X. The second term reflects dif-
and dropping out. Regression decomposition ferences in processes determining levels of Y
allowed us to clarify sources of the gap in (i.e., the unexplained portion of the model).
uninterrupted high school education betwe- We focus on the first term to explain how dif-
en retained and continuously promoted stu- ferences in resources account for the gap in
dents. Here, an analogy can be drawn uninterrupted high school education
between the issues of high school dropout or between previously retained and continuous-
success in school and earnings or success in ly promoted students. The first term repre-
the labor market.10 In the labor market, peo- sents how much of the gap in the uninter-
ple convert their resources into valued goods, rupted high school education between
much as students convert their resources into retained and promoted students is due to dif-
valued goods in school. For example, people ferences in resources when retained students
who have higher levels of education typically have the same distribution of resources (X) as
earn higher wages in the labor market, while continuously promoted students. We
people with higher achievement scores in obtained the logistic regression decomposi-
school typically are more likely to earn high tion results from STATA, using the FAIRLIE
school diplomas. Economists have developed command. These results are sensitive to the
decomposition techniques to examine the ordering of independent variables (Fairlie
Staying Back and Dropping Out 219

Table 1. Weighted Means and Standard Deviations of the Independent Variables, by


Period: NELS: 88–92
Variables Early Dropout Analyses Late Dropout Analyses

Dropouts .05 .09


(.21) (.28)
Black retained .04 .03
(.19) (.18)
White retained .13 .12
(.34) (.32)
Latino retained .03 .02
(.16) (.14)
Black continuously promoted .08 .08
(.27) (.27)
Latino continuously promoted .07 .07
(.26) (.25)
Male .49 .50
(.50) (.50)
SES -.07 -.03
(.76) (.74)
Two-parent family .80 .81
(.40) (.39)
Public .88 .90
(.33) (.30)
Urban .24 .25
(.43) (.44)
Northeast .19 .19
(.39) (.39)
North Central .28 .29
(.45) (.45)
West .17 .16
(.38) (.37)
Educational pessimism .01 .06
(.10) (.24)
Achievement scores 51.17 51.29
(9.91) (9.79)
Misbehavior factor -.01 .01
(.99) (1.00)
Self-concept .01 .00
(.72) (.76)
Student participation factor -.01 .02
(.91) (.99)
Student preparedness factor .01 -.03
(.97) (.94)
Parent-school connectivity -.04 -.03
(.96) (.96)
Parent-student connectivity .00 .04
(.99) (.94)
Lack of bonds with teachers factor .07 .02
(.97) (1.00)
Popularity with peers .83 .84
(.38) (.37)
School changes 1.20 1.10
(1.51) (1.43)
N 13,356 12,343
220 Stearns, Moller, Blau, and Potochnick

2003). Thus, we examine alternative order- uously promoted students to drop out of
ing. Most results are robust, and we discuss school early. Indeed, white retained students
the inconsistencies in the notes. are 25 times more likely, black retained stu-
dents are 15 times more likely, and Latino
retained students are 24 times more likely to
RESULTS drop out early than are white continuously
promoted students (see Table 5).
First, we examine differences in the resources Model 2 of Table 3 adds demographic and
that retained and continuously promoted stu- contextual variables, including SES, sex, and
dents of various racial groups have. Table 2 region, to the baseline model. Once these
shows that retained students are more likely demographic and contextual variables are
than are continuously promoted students to controlled for, black continuously promoted
drop out both early and late and that white students are not significantly different from
and Latino retained students are more likely white continuously promoted students in
than are black retained students to drop out their probability of dropping out. At the same
early. Retained students also come from poor- time, however, black, Latino, and white
er households than do continuously promot- retained students are significantly more likely
ed students, with fewer hailing from house- to drop out of school early than are white
holds with two parents. It is not surprising continuously promoted students, net of all
that retained students tend to have lower other variables.
achievement scores, to be more pessimistic Model 3 of Table 3 adds educational back-
about their educational futures, and to have ground variables to the previous model. This
more disciplinary problems. In addition, set of variables taps into the notion that
white and Latino retained students have a retention is an early warning for students
lower self-concept than do their same-race whose academic achievement or behavior
continuously promoted counterparts, but does not meet school standards. Model 3
there is no difference in self-concept between reveals differences in the expected direction
black retained and continuously promoted for students’ educational aspirations, achieve-
students. Although black retained students ment scores, and conduct. Yet differences in
are less likely to participate in extracurricular educational background do not fully explain
activities than are black continuously promot- the relationship between retention and drop-
ed students, they are more likely to partici- ping out, since black, white, and Latino
pate than are white and Latino continuously retained students are all significantly more
promoted students. Finally, Table 2 shows dif- likely than are white continuously promoted
ferences in the social capital resources that students to drop out, although the magni-
are available to retained and continuously tude of the coefficients is reduced.
promoted students, since continuously pro- Another model, the frustration–self-esteem
moted students of all racial groups have more model, predicts that deficiencies in self-
parent- and teacher-based social capital than esteem explain why students drop out. Thus,
do their same-race counterparts and have Model 4 of Table 3 adds self-esteem to the
experienced fewer changes in schools. baseline and demographic-contextual mod-
Tables 3 and 4 present the results from els. Model 4 does not show the expected rela-
weighted multivariate logistic models predict- tionship between self-esteem and early-
ing dropping out of school early (see Table 3) dropout behavior. It also shows that black,
and late (see Table 4), with correction for Latino, and white retained students are still
clustering within schools. Table 5 converts the significantly more likely than are white con-
coefficients to odds ratios for selected inde- tinuously promoted students to drop out of
pendent variables. In Table 3, Model 1 gives a school early net of these demographic-con-
baseline for the relationship between the textual and self-esteem variables. In other
probability of early dropout and retention sta- words, the frustration–self-esteem model
tus. This model shows that all our comparison does not explain why retained students are
groups are more likely than are white contin- more likely to drop out.
Staying Back and Dropping Out 221

Table 2. Weighted Means and Standard Deviations of the Independent Variables, by Race and
Retention Status: NELS: 88–92

White Black Latino White Black Latino


Variables Retained Retained Retained Promoted Promoted Promoted

Early dropouts .19 .12 .19 .01 .05 .05


(.39) (.33) (.39) (.10) (.22) (.21)
Late dropoutsa .23 .22 .22 .06 .06 .10
(.42) (.41) (.42) (.23) (.23) (.29)
Male .63 .59 .49 .47 .39 .45
(.48) (.49) (.50) (.50) (.49) (.50)
SES -.28 -.59 -.84 .12 -.35 -.48
(.76) (.68) (.78) (.69) (.74) (.76)
Two-parent family .76 .47 .73 .86 .55 .80
(.42) (.50) (.44) (.35) (.50) (.40)
Public .94 .95 .92 .85 .90 .90
(.42) (.22) (.28) (.35) (.29) (.30)
Urban .17 .48 .39 .18 .48 .43
(.38) (.50) (.49) (.39) (.50) (.50)
Northeast .15 .16 .10 .21 .13 .13
(.36) (.36) (.30) (.41) (.34) (.33)
North Central .29 .20 .14 .33 .15 .12
(.45) (.40) (.35) (.47) (.36) (.32)
West .14 .02 .26 .16 .06 .46
(.35) (.15) (.44) (.37) (.25) (.50)
Educational pessimism .04 .01 .03 .00 .01 .02
(.19) (.10) (.16) (.06) (.10) (.12)
Achievement scores 45.21 41.21 43.28 54.20 46.89 47.73
(7.70) (5.75) (7.31) (9.50) (8.37) (8.34)
Misbehavior factor .48 .52 .27 -.16 .04 .01
(1.27) (1.22) (1.12) (.86) (.87) (1.01)
Self-concept -.08 .27 -.12 -.01 .26 .03
(.74) (.62) (.73) (.72) (.70) (.71)
Student participation factor -.21 .17 -.05 .00 .23 -.05
(.94) (1.09) (1.23) (.82) (1.07) (1.07)
Student preparedness factor -.30 -.27 -.35 .12 .06 -.15
(1.07) (1.04) (1.14) (.91) (.94) (1.05)
Parent-school connectivity -.19 -.10 -.21 .01 -.03 -.15
(.86) (.95) (.92) (.97) (1.03) (.97)
Parent-student connectivity -.25 -.35 -.47 .12 -.05 -.24
(1.12) (1.03) (1.38) (.88) (1.01) (1.23)
Lack of bonds with teachers factor .15 -.06 .00 .08 -.01 -.07
(1.03) (.95) (.96) (.96) (1.01) (.89)
Popularity with peers .82 .91 .73 .83 .85 .77
(.38) (.28) (.45) (.37) (.35) (.42)
School changes 1.93 1.63 1.57 .99 1.32 1.27
(1.79) (1.65) (1.52) (1.38) (1.46) (1.53)
N = 13,356 1,780 502 339 8,679 1,093 963

aThe late-dropout measure comes from the third NELS wave, reducing the N to 12,343. The N for
each respecitve category is as follows: 1,445, 425, 237, 8,371, 1,004, and 862.
222 Stearns, Moller, Blau, and Potochnick

Table 3. Logistic Regression Coefficients Predicting Early School Dropout: NELS: 88–92 (n = 13,356)

Variable Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7

Black retained 2.69*** 1.77*** 1.26** 1.74*** 1.74*** 1.68*** 1.12**


(.30) (.39) (.43) (.36) (.38) (.36) (.36)
White retained 3.22*** 2.88*** 2.46*** 2.89*** 2.76*** 2.65*** 2.20***
(.22) (.25) (.32) (.25) (.25) (.23) (.28)
Latino retained 3.19*** 2.08*** 1.78*** 2.09*** 1.98*** 2.09*** 1.73***
(.24) (.35) (.37) (.35) (.36) (.34) (.36)
Black continously promoted 1.73*** .93 .70 .91 .99 .97 .67
(.48) (.59) (.59) (.57) (.59) (.54) (.48)
Latino continously promoted 1.61*** .77* .60 .76* .71 .86** .55
(.27) (.36) (.36) (.35) (.36) (.33) (.34)
Demographic and Contextual Variables
Male -.32 -.63** -.34 -.43 -.31 -.61*
(.21) (.23) (.23) (.22) (.20) (.25)
SES -1.00*** -.88** -1.01*** -.97*** -.87*** -.76**
(.25) (.28) (.24) (.26) (.23) (.22)
Two-parent family -.34 -.26 -.34 -.30 -.26 -.18
(.20) (.21) (.20) (.20) (.19) (.19)
Public 1.25*** 1.03** 1.25*** 1.15** 1.01** .70
(.35) (.38) (.35) (.37) (.38) (.40)
Urban .30 .23 .30 .24 .19 .12
(.24) (.26) (.23) (.24) (.24) (.24)
Northeast -.33 -.33 -.33 -.33 -.15 -.16
(.36) (.36) (.36) (.35) (.36) (.36)
North Central -.21 -.21 -.21 -.19 -.17 -.12
(.28) (.30) (.28) (.28) (.27) (.28)
West .07 .02 .08 .04 -.21 -.21
(.35) (.39) (.35) (.36) (.27) (.32)
Educational Background
Educational pessimism .60* .72*
(.30) (.31)
Achievement scores -.04* -.05**
(.02) (.02)
Misbehavior factor .44*** .39***
(.06) (.07)
Self-esteem
Self-concept .08 .04
(.18) (.16)
Participation-Identification Model
Student participation -.11 -.09
(.10) (.09)
Student preparedness -.42*** -.21**
(.06) (.08)
Social Capital Model
Parent-school connectivity -.16 -.28*
(.11) (.11)
Parent-student connectivity -.24** -.16*
(.07) (.07)
Lack of bonds with teachers -.08 -.28**
(.12) (.11)
Popularity with peers .59** .45*
(.19) (.19)
School changes .29*** .28***
(.06) (.06)
Constant -4.66*** -5.48*** -3.24* -5.47*** -5.42*** -6.27*** -3.53**
(.14) (.49) (1.34) (.48) (.50) (.55) (1.18)
-2 log likelihood -2078.59 -1900.40 -1786.31 -1899.60 -1846.21 -1800.22 -1674.42

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


Staying Back and Dropping Out 223

Hypotheses from the participation-identifi- ables.11 The magnitude of the coefficient for
cation model predict that students who are white retained students has decreased by 31
more engaged with the social and/or acade- percent over its magnitude in the baseline
mic aspects of schooling are less likely to drop model, the magnitude for black retained stu-
out of school and that the lack of engage- dents has decreased by 58 percent, and the
ment accounts for retained students’ higher magnitude for Latino retained students has
probability of early dropout. The results for decreased by 46 percent, but the models that
this model are mixed (see Model 5 of Table are considered here do not entirely explain
3). The findings show that students who are the higher probability of early dropout for
academically engaged are significantly less retained students.
likely to drop out of school early, but the coef- It is possible, however, that the process of
ficient for participation in extracurricular dropping out of school prior to completion of
activities is not significant. Once these vari- the 10th grade is different from the process of
ables are included, the relationship between dropping out of school later. Thus, we con-
retention and dropping out remains for sider the relationship between late dropout,
white, black, and Latino retained students. or dropping out between the 10th and 12th
Therefore, the participation-identification grades, and explanatory variables from the
model does not fully explain the higher prob- 10th grade in Table 4. We also expect to find
ability of dropout for retained students. differences in the probabilities of dropping
Thus, we must turn to other independent out between the early- and late-dropout peri-
variables to explain the link between reten- ods, given that retained students may be eli-
tion and early dropout. Model 6 of Table 3 gible to drop out of school in earlier grades
includes variables that measure social capital. than may continuously promoted students.
As we hypothesized, children whose parents Our samples of retained students are tainted
speak with them more often about school are a bit by selection in that retained students
also less likely to drop out of school. In addi- have been exposed to the risk of dropping
tion, we found that students who have out legally longer than have continuously
changed schools more frequently are more promoted students, since they reached the
likely to drop out, as are students with more states’ minimum dropout age at an earlier
friends. This finding may be due to adoles- grade.
cent popularity hierarchies that reward fac- Model 1 in Table 4 shows that, as with
tors other than academic excellence early dropouts, there is a significant relation-
(Coleman 1961; Cusick 1973), but it should ship between retention and dropping out late
be investigated further. Somewhat more puz- for Latino, white, and black students. Given
zling, however, is the result that shows that that they are still enrolled in school in the
parent-school connectivity has no significant 10th grade, white retained and black retained
impact on students’ probability of dropping students are, respectively, 5.10 and 4.85
out early. Nonetheless, with the introduction times more likely to drop out of school late
of these social capital variables, Latino, black, than are white continuously promoted stu-
and white retained students are still signifi- dents, while Latino retained students are 4.95
cantly more likely to drop out of school than times more likely to drop out of school (see
are white continuously promoted students. Table 5).
Thus, differences in social capital do not Model 2 of Table 4 adds demographic and
explain away the link between retention and contextual variables to the baseline model.
dropping out. Again, black, Latino, and white retained stu-
Finally, Model 7 of Table 3 includes all the dents are significantly more likely to drop out
potential explanatory variables together in of school late than are white continuously
one model. Most notable for our purposes are promoted students, results that hold up with
the results for retained students, all of whom black continuously promoted and Latino con-
are still significantly more likely to drop out of tinuously promoted used as the reference cat-
school early than are white continuously pro- egories. As with the analyses that considered
moted students, even net of all the other vari- early dropouts, adding the academic back-
224 Stearns, Moller, Blau, and Potochnick

Table 4. Logistic Regression Coefficients Predicting Late School Dropout: NELS: 88–92 (n = 12,343

Variable Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7

Black retained 1.58*** .99* .48 1.07* 1.06* 1.00* .59


(.38) (.50) (.51) (.50) (.49) (.46) (.44)
White retained 1.63*** 1.41*** 1.00*** 1.40*** 1.25*** 1.26*** .89***
(.14) (.14) (.15) (.14) (.14) (.14) (.15)
Latino retained 1.60*** .87*** .47 .85*** .78*** .87*** .50*
(.20) (.22) (.25) (.22) (.22) (.22) (.23)
Black continously promoted .06 -.50* -.71* -.42 -.45 -.39 -.56*
(.20) (.24) (.24) (.24) (.24) (.23) (.24)
Latino continously promoted .59*** -.02 -.19 -.02 -.10 .07 -.13
(.15) (.19) (.22) (.19) (.20) (.19) (.21)
Demographic and Contextual Variables
Male -.07 -.46*** -.02 -.22 -.11 -.45***
(.11) (.13) (.12) (.12) (.12) (.13)
SES -.75*** -.52*** -.74*** -.71*** -.69*** -.50***
(.09) (.09) (.09) (.09) (.09) (.10)
Two-parent family -.15 -.12 -.15 -.09 -.06 -.03
(.14) (.14) (.14) (.14) (.14) (.14)
Public .18 -.01 .18 .02 -.12 -.32
(.47) (.48) (.47) (.48) (.48) (.49)
Urban .12 .06 .14 .01 .06 -.07
(.20) (.20) (.20) (.19) (.19) (.19)
Northeast -.49* -.45* -.49* -.55* -.44* -.42
(.21) (.22) (.21) (.22) (.21) (.22)
North Central -.19 -.26 -.22 -.18 -.19 -.23
(.20) (.20) (.21) (.20) (.19) (.18)
West .11 .05 .10 -.03 .03 -.04
(.16) (.18) (.16) (.17) (.17) (.18)
Educational Background
Educational pessimism .71*** .64***
(.16) (.15)
Achievement scores -.04*** -.04***
(.01) (.01)
Misbehavior factor .56*** .44***
(.05) (.05)
Self-esteem
Self-concept -.25*** -.01
(.07) (.07)
Participation-Identification Model
Student participation -.52*** -.24***
(.05) (.06)
Student preparedness -.37*** -.27**
(.10) (.09)
Social Capital Model
Parent-school connectivity -.11 -.13
(.08) (.08)
Parent-student connectivity -.14* -.03
(.06) (.06)
Lack of bonds with teachers .36*** .10
(.06) (.06)
Popularity with peers .23 .19
(.14) (.15)
School changes .20*** .18***
(.04) (.04)
Constant -2.84*** -2.75*** -.44 -2.80*** -2.68*** -3.04*** -.94
(.09) (.51) (.58) (.50) (.50) (.51) (.55)
-2 log likelihood -3412.92 -3269.04 -2926.94 -3252.96 -3084.00 -3146.83 -2832.92

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


Staying Back and Dropping Out 225

ground variables in Model 3 produces results icant difference between Latino and white
that are consistent with theory. White retained students and white continuously
retained students, however, continue to promoted students remains, with white
demonstrate their higher probability of drop- retained students more than twice as likely to
ping out late, but there are no longer signifi- drop out as white promoted students (see
cant differences in the probability of late Table 5).12
dropout between black and Latino retained
students and white continuously promoted Decomposing Difference
students.
Model 4 of Table 4 adds self-esteem to the In Tables 6 and 7, we further clarify the utility
baseline and demographic and contextual of standard theories of dropout in explaining
models. Again, self-esteem marginally affects the association between retention and high
the relationship between retention and drop- school dropout among white, black, and
ping out late, in that retained students of all Latino students. These tables present esti-
racial groups are still more likely than are con- mates from logistic regression decomposi-
tinuously promoted white students to drop tions that predict uninterrupted high school
out late. At the same time, the results of the education (the inverse of high school
independent variables are consistent with the dropout). They indicate, for different races,
predictions from the hypotheses, whereas the extent to which the gap in uninterrupted
they were not consistent in the early-dropout high school education would be reduced if
model. Including variables that are designed retained students had the same distribution
to measure the participation-identification of resources (as identified by the independent
model in Model 5 of Table 4 does not sub- variables) as continuously promoted stu-
stantively change the relationship between dents. These tables clearly illustrate that
retention and dropping out for Latino, black, resources offer different benefits to students,
and white students: With these other vari- depending on the students’ race and timing
ables included, retained students are signifi- of dropout. The regression decompositions
cantly more likely than are white continuous- are based on Model 7 in Tables 3 and 4; how-
ly promoted students to drop out. The fact ever, logistic regressions are conducted sepa-
that black and Latino retained students show rately for each race. The coefficients present-
significant differences in these models, ed in Tables 6 and 7 reflect the contributions
whereas they did not in the models that that groups of variables make to gaps in unin-
included educational background and demo- terrupted high school education between
graphic and contextual differences, indicates retained and continuously promoted stu-
that these students may be disproportionate- dents.
ly advantaged in terms of their self-esteem, Table 613 shows that 99 percent of contin-
extracurricular participation, and class pre- uously promoted white students attain an
paredness. uninterrupted high school education, or do
The results in Model 6 of Table 4 from the not drop out, by the 10th grade, compared
introduction of variables from the social capi- to 86 percent of previously retained white
tal model are also largely consistent with students—a gap of 13 percentage points.
those from the analyses of early dropouts, Similarly, 98 percent of continuously promot-
with the exception that the lack of bonds ed black students attain an uninterrupted
with teachers significantly predicts the proba- high school education by the 10th grade,
bility of dropout for late dropouts, but peer compared to 87 percent of retained stu-
popularity does not. Model 7 includes all the dents—a gap of 11 percentage points. Finally,
explanatory variables in the same model, pre- 97 percent of continuously promoted Latino
dicting late dropout. In contrast to the early- students attain an uninterrupted high school
dropout model, once all variables are con- education by the 10th grade compared to 82
trolled, black retained students are not signif- percent of retained Latino students—a gap of
icantly more likely to drop out than are white 14.4 percentage points. All the included vari-
continuously promoted students, but a signif- ables in the model explain 72 percent of the
Table 5. Odds Ratios for Predicting Early and Late School Dropout, NELS: 88–92
226

Variable Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7

Early School Dropout


Black retained 14.73 5.87 3.53 5.70 5.70 5.37 3.06
White retained 25.03 17.81 11.70 17.99 15.80 14.15 9.03
Latino retained 24.29 8.00 5.93 8.08 7.24 8.08 5.64
Black continously promoted 5.64 NS NS NS NS NS NS
Latino continously promoted 5.00 2.16 NS 2.14 NS 2.36 NS
N = 13,356

Late School Dropout


Black retained 4.85 2.69 NS 2.92 2.89 2.72 NS
White retained 5.10 4.10 2.72 4.06 3.49 3.53 2.44
Latino retained 4.95 2.39 NS 2.34 2.18 2.39 1.65
Black continously promoted NS 0.61 0.49 NS NS NS 0.57
Latino continously promoted 1.80 NS NS NS NS NS NS
N = 12,343

Note: Model 1, baseline; Model 2, demographic and control variables; Model 3, educational background; Model 4, self-esteem; Model 5, participa-
tion-identification; Model 6, social capital; Model 7, all included variables.
Stearns, Moller, Blau, and Potochnick
Staying Back and Dropping Out 227

gap in uninterrupted high school education high school education by the 10th grade
among white students (.09/.13), 51 percent among white students, 30 percent of the gap
of the gap among black students (.06/.11), among black students, and 28 percent of the
and 50 percent of the gap among Latino stu- gap among Latino students. Thus, if white
dents (.07/.14). Thus, together, these theo- retained students had the same educational
ries are most effective at explaining early background as white continuously promoted
dropout among white students. students, the gap in uninterrupted high
The demographic and contextual control school education would be reduced by 37
variables account for 20 percent of the gap in percent. An interesting finding is that
uninterrupted high school education resources that are associated with the frustra-
between white continuously promoted and tion–self-esteem and the participation-identi-
retained students (.03/.13), 15 percent of the fication models do not significantly reduce
gap between black continuously promoted the gap in continuous high school promotion
and retained students, and 30 percent of the regardless of race. In addition, the social cap-
gap between Latino continuously promoted ital model has a significant impact on the gap
and retained students. In addition, group dif- in uninterrupted high school education for
ferences in educational background account white and Latino students, but the effects
for 37 percent of the gap in uninterrupted diverge. When white retained students have

Table 6. Logistic Regression Decompositions of the Difference Between Retained and


Continuously Promoted Students in Uninterrupted High School Education Between the 8th
and 10th Grades, by Race

Variable White Black Latino

Uninterrupted education rate for promoted students .991 .981 .966


Uninterrupted education rate for retained students .861 .872 .822
Promoted/retained gap .130 .109 .144

Contributions from Differences in


Demographic and contextual variables .026*** .016* .044**
(.004) (.008) (.013)
20.2% 15.1% 30.2%

Educational background .048*** .033* .040**


(.007) (.013) (.013)
37.2% 30.4% 27.8%

Self-esteem .000 .001 -.001


(.001) (.001) (.005)
0.0% 0.6% -0.7%

Participation-identification model .005 .005 .011


(.003) (.007) (.008)
3.9% 4.6% 7.4%

Social capital model .014** .000 -.021*


(.004) (.005) (.009)
10.9% -0.4% -14.7%

All included variables .094 .056 .072


72.1% 51.2% 50.0%

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


228 Stearns, Moller, Blau, and Potochnick

the same levels of social capital as do white students. Variables that are associated with
continuously promoted students, 11 percent the participation-identification model are sig-
of the gap in uninterrupted high school edu- nificant only for white students, indicating
cation is explained. In contrast, when Latino that if white retained students had the same
retained students have the same levels of levels of participation in school and prepared-
social capital as Latino continuously promot- ness for school as do white continuously pro-
ed students, the gap in uninterrupted high moted students, the gap in uninterrupted
school education is larger (since the coeffi- high school education would be 10 percent
cient is negative, suggesting that differences smaller. This finding supports Eitle’s (2005)
between the groups increase the gap). This contention that white students disproportion-
finding supports prior research that found ately benefit from participation in school, par-
that Latinos do not benefit from social capital ticularly in extracurricular activities. Self-
in the same way as whites do (Ream 2003; esteem and social capital are not significant in
Valadez 2002).14 any of these models.15
Table 7 presents similar results with unin-
terrupted high school education through the
12th grade as the dependent variable. It DISCUSSION AND CONCLUSIONS
shows that 96 percent of white continuously
promoted students who did not drop out by High school students face the cumulative
the 10th grade continue through high school impact of their prior experiences in elemen-
uninterrupted, compared to 83 percent of tary and middle school. In this article, we
white retained students. Furthermore, 93 per- attempted to answer several overlooked
cent of black continuously promoted students questions centering on the larger question of
continue through high school uninterrupted, why retained students are more likely than
compared to 82 percent of black retained stu- are continuously promoted students to drop
dents. Finally, 93 percent of Latino continu- out of high school. We asked how well-estab-
ously promoted students continue through lished models of dropping out explain
high school uninterrupted, compared to 78 retained students’ heightened probability of
percent of Latino retained students. Thus, the dropping out; whether these models explain
gaps between retained and continuously pro- dropping out equally well for Latino, white,
moted students in uninterrupted high school and black retained students; and whether
education are 13 percentage points for white these models explain both early and late
students, 11 percentage points for black stu- dropout among retained students. What is
dents, and 16 percentage points for Latino clear from our results is the following:
students. Retention predicts both early and late
All the variables in the models account for dropout for white, black, and Latino students,
69 percent of the gap among white students even net of differences in resources and con-
and 58 percent of the gap among black stu- trols for sociodemographic and educational
dents, but only 42 percent of the gap among background. But these differences in
Latino students. Demographic and contextu- resources affect the gap in the probability of
al background variables significantly explain dropping out between retained and continu-
the gap for white and Latino students, but ously promoted students differently for stu-
not for black students. In addition, educa- dents of different racial groups. On the
tional background is a significant predictor of whole, the models explain both early and late
the gap in uninterrupted education for white dropout equally well. Next, we briefly sum-
and, most notably, for black students. Indeed, marize our results and propose other variables
group differences in educational background that may cause the relationships we observe.
explain 61 percent of the gap in uninterrupt- We conclude with policy recommendations
ed high school education between continu- that follow from our work.
ously promoted and retained black students. To explain why retained students are more
Differences in educational background likely to drop out of school than are continu-
explain 35 percent of the gap among white ously promoted students, we turned to sever-
Staying Back and Dropping Out 229

Table 7. Logistic Regression Decompositions of the Difference Between Retained and


Continuously Promoted Students in Uninterrupted High School Education Between the 10th
and 12th Grades, by Race

Variable White Black Latino

Uninterrupted education rate for promoted students .959 .930 .932


Uninterrupted education rate for retained students .828 .816 .773
Promoted/retained gap .131 .114 .159

Contributions from Differences in


Demographic and contextual variables .023*** .014 .035*
(.005) (.010) (.013)
17.3% 12.1% 21.6%

Educational background .046*** .070** .037


(.008) (.021) (.022)
35.3% 61.3% 23.5%

Self-esteem .000 -.003 -.005


(.000) (.006) (.005)
0.2% -2.8% -2.9%

Participation-identification model .013*** .002 .001


(.003) (.005) (.005)
9.9% 2.1% 0.4%

Social capital model .009 -.016 -.002


(.005) (.010) (.010)
6.8% -14.3% -1.3%

All included variables .090 .067 .067


69.0% 58.4% 42.2%

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

al established models of dropping out. Then, retention and dropping out. We turned there-
we explained why these models may be par- fore to regression decomposition techniques
ticularly appropriate to the case of retained to determine how much of the greater prob-
students. These explanations include differ- ability of dropout among retained students is
ences in resources in academic achievement due to these students’ relatively lower levels
and disciplinary problems, self-esteem, of resources. The dropout models that we
engagement with schooling, and social capi- tested performed markedly better in explain-
tal. We found that these differences account ing white retained students’ dropout behav-
for a large portion of the probability of high- ior than that of black and Latino students. In
er dropout among all retained students. the analyses of early dropout, demographic
Specifically, the magnitude of the difference and contextual variables, as well as educa-
in the probability of dropout between tional background, stand out as important for
retained and continuously promoted students whites, blacks, and Latinos. There are several
declined with the introduction of variables divergences in the importance of social capi-
measuring students’ academic background, tal, however. For white students, it appears
engagement with school, and social capital. that retained students are harmed by their
Differences in resources do not, however, relative lack of social capital, but the results
fully explain racially specific links between suggest no such effect for black students. For
230 Stearns, Moller, Blau, and Potochnick

Latino students, the relationship appears to was due to the fact that they had education-
work in the opposite direction, with social al backgrounds that would have contributed
capital disproportionately advantaging Latino to their dropout behavior, regardless of reten-
retained students over their continuously pro- tion. The larger point is that the meaning of
moted counterparts. retention is different and thus that its poten-
In the analyses of late dropout, demo- tial impact is somewhat lessened for black
graphic and contextual variables partially students.
explain the gap between retained and con- In contrast, Blau (2003) argued that the
tinuously promoted students for whites and white community believes that the signals
Latinos. For white students and black stu- that retention decisions are sending are accu-
dents, achievement scores, educational pes- rate, and the children act accordingly. Thus,
simism, and disciplinary behavior significantly for white students, the labels of success and
contribute to the gap between retained and failure are viewed in stark terms that are
continuously promoted students. In addition, reflective of individual ability, motivation, and
the lack of engagement for white students effort, and the labels are considered valid, sta-
appears to contribute to the gap in the prob- ble indicators of individual worth. In the case
ability of dropout. For black students, only of retention, the label also disadvantages stu-
academic background significantly explains dents who have preexisting differences in
the gap between retained and continuously resources and may impede their ability to
promoted students. For Latino students, only form positive relationships with their families,
demographic and contextual variables do so. friends, and teachers. Thus, we found that
Why might we find these racial differ- white continuously promoted students are
ences? The discussion of the racialization of more engaged with school, have more posi-
school processes by Blau and her colleagues tive bonds with teachers during postelemen-
(see Blau 2003; Blau, Moller, and Jones 2004; tary education, and are more popular with
Moller et al. 2006) points to a possible expla- their peers in the postelementary years, and
nation. Blau (2003) argued that white stu- their parents more frequently interact with
dents’ experiences with schooling and with them and with their schools about academic
schools’ organizational decisions differ from matters. All these differences significantly
those of blacks. She attributed this racial dif- explain why white retained students are more
ference to subtle psychological and commu- likely to drop out than are white continuous-
nity responses to these decisions: The black ly promoted students, in both the early and
community places less emphasis on these late periods.
decisions because they are perceived to Although Blau’s (2003) theory does not
reflect a legacy of discrimination. More devot- address how the culture of Latino students
ed to inclusive norms, the black community may serve as a protective force in the way
emphasizes talent in a diverse array of dimen- that black culture may do for black students,
sions, rather than in a single dimension of our results suggest this possibility. In addition,
academic ability, often measured by retention the fact that the standard theories of dropout
decisions. Thus, as Tables 6 and 7 showed, explain the gap between retained and con-
differences in self-esteem, school engage- tinuously promoted students better for whites
ment, and social capital do not significantly than for blacks and Latinos indicates the need
explain the gap in the probability of dropout to refine these theories and to develop others
for black retained and continuously promoted that are more racially and ethnically specific.
students. One interpretation of this finding is Further research, perhaps of a qualitative
that the retention decision marks black stu- nature, is needed to explore the exact
dents as less different than it does retained processes through which grade retention
students of other racial/ethnic groups, partic- affects whites, blacks, and Latinos in the
ularly whites. Other findings in Table 7 sup- increasingly diverse school-aged population
port this interpretation. Although black and to determine the extent to which reten-
retained students dropped out at higher tion changes trajectories.
rates, the majority of this dropout behavior One other potential explanation for the
Staying Back and Dropping Out 231

racial differences that we uncovered is not differences between early and late dropout.
mutually exclusive with the first. Whites, For example, because of data limitations, the
Latinos, and blacks attend high schools of predictors of early dropout originated in the
widely varying quality, on average (Duncan eighth-grade survey, which was when the stu-
and Brooks-Gunn 1997; Mickelson 2003; dents were enrolled in middle or junior high
Orfield and Lee 2005). It is possible that schools. Indeed, many of the early dropouts
school-level variables, such as the percentage drop out from high school, a school context
of the student body in poverty or the per- in which the students’ bonds with middle
centage of the student body that has been school teachers and their parents’ interaction
retained, are moderating the effects of reten- with middle school is less important than
tion on dropping out. In fact, there is ample their bonds with high school teachers and
evidence that school-level variables have a their parents’ interaction with the high
significant impact on students’ dropout school.
behavior (see, e.g., Lee and Burkam 2003), In sum, our results suggest that differences
attachment to school (Ma 2003), and in SES, school sector, academic achievement,
engagement with school (Johnson, Crosnoe, aspirations, behavior, self-esteem, engage-
and Elder 2001). Additional research is neces- ment with schooling, and social capital partly
sary, however, to examine the relationship explain why retained students are more likely
between school climate and engagement, to drop out of school than are continuously
particularly for children who are retained. promoted students. Therefore, our results
Future investigations should expand Lee and point to the particular attention that schools
Smith’s (1995) work on school engagement that are interested in minimizing their
and discuss whether the reform strategies dropout rates should give to retained stu-
that Lee and Smith found increased students’ dents. Besides their lower achievement rates
engagement are equally effective for retained and more disciplinary problems, retained stu-
and promoted students. dents have lower self-esteem, are more pes-
There is also the possibility, however, that simistic about their future, are less engaged
there are other individual-level differences with school, and have fewer bonds with
between retained and continuously promot- teachers than do continuously promoted stu-
ed students that we did not measure, such as dents. Alternatives to retention, including
friendship with peers who are oriented to summer schools and academies that keep
school success. It is also possible that there children with their same-age peers while
are aspects of engagement with schooling ensuring that they gain mastery of the acad-
that our measures did not adequately cap- emic content and skills that had previously
ture. Furthermore, the possibility of gender caused them difficulty, can be attractive. In
differences in the link between retention and addition, schools should also consider attack-
dropping out remains to be investigated. ing the mediating factors that partially
Our third research question focused on the explain why retained students drop out. Both
potential differences in dropout behavior research and policy efforts should focus on
between the early and late periods. We sug- how schools facilitate students’ engagement
gested that examining both dropout periods with academics and with the social arena. It is
was necessary because considering all only with this knowledge that schools will be
dropouts together would obscure qualitative able to reach out to some of the most at-risk
differences in the early- and late-dropout students.
processes. Our results, however, do not
diverge widely between the two periods, sug-
gesting that the predictors of dropping out NOTES
late are more similar to the predictors of
dropping out early than they are dissimilar. It 1. It is important to note that a debate
is also possible, however, that a greater tem- continues over the extent to which retention
poral matching of our independent and is disadvantageous for students’ later achieve-
dependent variables would reveal additional ment (see, e.g., Alexander et al. 2003;
232 Stearns, Moller, Blau, and Potochnick

Jimerson 1999; Shepard, Smith, and Marion the factor scores for missing cases. We tested
1996, 1998). our results against variables with listwise dele-
2. It is also possible that the causality works tion of missing cases and found similar
in the opposite direction and that those who results.
have a low level of social capital are more like- 7. The original subsample of black, Latino,
ly to be retained than are others. and white cases included 22,509 students.
3. Peers may serve as a form of social cap- Many of these cases did not participate in the
ital, but these ties may not always be orient- second wave, which was needed for the
ed to the student’s future achievement. analysis of early dropout, or in the third wave,
Unfortunately, these different types of peer- which was needed for the analysis of late
based social capital are not those that we can dropout. We deleted cases that did not par-
parse out with the NELS data, since we do not ticipate in the first two waves for the early-
have information on the grade point averages dropout data set, leaving 16,043 cases. For
of the peers. the late-dropout data set, we deleted cases
4. Researchers have also found that the that did not participate in all three waves,
timing of retention is an important compo- leaving 15,201 cases. After selecting cases
nent of dropping out (Alexander et al. 2003). with data available for at least two variables in
Because of small subsamples when consider- each factor, we lost 856 cases for the early-
ing the timing of retention in conjunction dropout data set and 650 for the late-
with race, we were unable to examine the dropout data set. Finally, we performed a list-
timing of retention. For example, in the early wise deletion of variables, for final samples of
dropout sample, about 60 black students 13,356 for early dropout and 12,343 for late
were retained early, and about 10 of these dropout. To check for potential sample bias,
students dropped out. These subsamples we compared our final weighted samples to
were even smaller in the late dropout sample, the weighted samples of the full data set.
making it difficult for us to understand how Owing to missing cases on the weight vari-
each of our resources affects the probability able, the full data set included individuals
of students dropping out. who participated in the first two waves for
5. The question for the 8th-grade wave early dropout and all three waves for late
was, “How far in school do you think you will dropout. The final samples had a similar
get?” The options were won’t finish high dropout rate, in comparison to the original
school, will finish high school, vocational/ subsample of black, white, and Latino stu-
trade/business school, will attend college, will dents. We found some significant differences
finish college, and higher school after college. between the means of our independent vari-
For the 10th-grade wave, the question was ables in the unweighted original data set
“Think about how you see your future. What compared to our unweighted subsample.
are the chances you will graduate from high Our sample had a lower retention rate for
school?” The options were very low, low, Latinos (at 3 compared to 4), a lower promo-
about 50–50, high, and very high. tion rate of blacks (at 8 compared to 9), and
6. All factor variables in the analyses were a lower promotion rate for Latinos (at 7 com-
created through principal-components analy- pared to 10). Our sample was also biased
ses. We allowed factors to be correlated toward two-parent families (80 versus 78),
through promax rotation. To ensure that we public schools (88 versus 79), nonurban
adequately fit the data, we permitted the regions (24 versus 30), and North Central
principal-component analyses to create all states (28 versus 25). Students in our sample
factors with eigenvalues higher than 1. In had less educational pessimism (1.0 versus
each model, the number of factors that were 1.4) and higher standardized test scores (51.2
created corresponded with our expectations. versus 50.6). For our factor variables, our stu-
We included all cases with data that were dents scored lower on school trouble and par-
available for at least two variables included in ent-school connectivity but higher on stu-
each factor. For cases with some missing data, dents’ preparation. Our students also had
we used STATA’s impute command to impute fewer bonds with their teachers than did
Staying Back and Dropping Out 233

those in the full sample. Last, our students moted students, not from Latino continuous-
changed schools fewer times than did those ly promoted or black continuously promoted
in the overall sample (1.2 versus 1.3). students.
8. Other questions included, “I’m able to 12. These results largely hold up with black
do things as well as others”; “on the whole, continuously promoted and Latino continu-
I’m satisfied with myself”; “I certainly feel use- ously promoted used as the reference cate-
less at times” (reverse coded); “at times I gories. The exception is that black retained
think I am no good at all” (reverse coded); students are significantly more likely to drop
and “I feel I do not have much to be proud out than are black continuously promoted
of” (reverse coded). students, net of all other variables. For this
9. In the early-dropout sample, students reason, we proceeded with the regression
were clustered in 957 schools. There were 14 decomposition for blacks, as well as for
students, on average, per school. This cluster- Latinos and whites.
ing necessitated the calculation of robust 13. The mean differences between the
standard errors. retained and promoted students in Table 2
10. Many thanks to Karl Alexander for compared to those reported in Tables 6 and 7
pointing this analogy out to us. are due to weighting effects.
11. These results were largely consistent 14. This result is not robust to the ordering
regardless of the racial identification of the of variables. All the remaining results present-
“continuously promoted group” used as the ed in Table 6 are robust to the ordering of
reference category, with one notable excep- variables, except that the demographic and
tion. When black continuously promoted stu- contextual control variables for black students
dents were used as the reference category, become nonsignificant when the variables are
white and Latino retained students were sig- reordered.
nificantly more likely to drop out early, as they 15. All the significant results presented in
were when Latino continuously promoted Table 7 are robust to the ordering of vari-
students were used as the reference category. ables. When the variables are reordered,
But black retained students were significantly social capital becomes significant for white
different only from white continuously pro- students.
234 Stearns, Moller, Blau, and Potochnick

APPENDIX A

Variables Used in Parent-School and Parent-Child Connectivity


Measures: National Education Longitudinal Study 1988–92

Parent-Child Interaction
Child’s Reports
Since the beginning of the school year, how often have you discussed the following with
either or both of your parents and/or guardian?
1. Selecting courses or programs at school
2. School activities or events of particular interest to the student
3. Things the child studied in class
4. and 5. School program (separate for father and mother)
Response options: 0 = not at all, 1 = once or twice or three or more times

Parent’s Reports
1. How often do you or your spouse/partner talk with your eighth grader about his or
her experiences with school?
2. How often do you or your spouse/partner talk with your eighth grader about his or
her plans for high school?
3. How often do you or your spouse/partner talk with your eighth grader about his or
her plans for after high school?
Response options: 0 = not at all or rarely, 1 = occasionally or regularly

Parent-School Interaction
Parent's Reports of Contacting School
Since your eighth grader's school opened last fall, how many times have you or your
spouse/partner contacted the school about each of the following?
1. The child's academic performance
2. The child's academic program for the year
3. The child's behavior in school
4. The parent participating in school fund-raising activities
5. The parent providing information for school records
6. The parent volunteering at school
Response options: 0 = none; 1 = once or twice, 3–4 times, or > 4 times
Parent’s Reports of School Activities
Do you or your spouse do any of the following at your eighth grader's school?
1. Belong to a parent/teacher organization
2. Attend meetings of a parent/teacher organization
3. Take part in activities of a parent/teacher organization
4. Act as a volunteer at the school
Response options: 0 = No, 1 = Yes
Child's Reports of Parent-School Interaction
Since the beginning of the school year, has either of your parents or guardians
1. Attended a school meeting
2. Phoned to speak to a teacher or counselor
3. Visited classes
4. Attended a school event
Response options: 0 = No, 1 = Yes
Staying Back and Dropping Out 235

APPENDIX B

Variables Used in the Student Academic and Extracurricular


Engagement Measures: National Education Longitudinal Study
1988–92

Student’s Academic Engagement


School Attendance
How many days of school did you miss over the past four weeks?
How often do you cut or skip classes?
How many times were you late for school over the past four weeks?
Response options: 0 = at least sometimes; 1 = none, never, or rarely
Class Preparedness
How often do you come to class and find yourself WITHOUT these things?
1. Pencil and paper
2. Books
3. Homework
Response options: 0 = usually, often, or seldom; 1 = never

Student’s Extracurricular Engagement


Extracurricular Activity
Have you or will you participate in any of the following activities during the current
school year, either as a member or as an officer?
1. School varsity sports 7. Debate or speech club
2. Intramural sports 8. Drama club
3. Cheerleading 9. Student newspaper
4. Band or orchestra 10. Student yearbook
5. Chorus or choir 11. Student council
6. Dance club
Response options: 0 = No, 1 = Yes
Academically Oriented Extracurricular Activities
Have you or will you participate in any of the following activities during the current
school year, either as a member or as an officer?
1. Science fair
2. History club
3. Science club
4. Math club
5. Foreign language club
6. Other subject matter club
7. Academic Honors Society
8. Computer club
Response options: 0 = No, 1 = Yes
236 Stearns, Moller, Blau, and Potochnick

APPENDIX C

Factor Loading Summary


Factor Early Dropout Late Dropout

Misbehavior .55 .47


Sent to office for misbehaving .84 —
Parents warned about attendance .55 .57
Parents warned about behavior .83 .62
Fought with another student .71 .77
Number of times in trouble — .76

Parent-School Connectivity .33 .33


Student's report of parent-child interaction -.12 -.12
Parent's rreport of parent-child interaction .10 .10
Student's report of parent-school interaction .44 .44
Parent's report of contacting the school .82 .82
Parent's involvement in school organizations/activities .77 .77

Parent-Child Connectivity .20 .20


Student's report of parent-child interaction .86 .86
Parent's report of parent-child interaction .66 .66
Student's report of parent-school interaction .12 .12
Parent's report of contacting the school -.11 -.11
Parent's involvement in school organizations/activities .01 .01

Student’s Participation .37 .33


School attendance .01 .04
Class preparedness -.01 -.04
General extracurricular participation .85 .80
Academically oriented extracurricular participation .85 .76

Student’s Preparedness .30 .26


School attendance .78 .75
Class preparedness .78 .78
General extracurricular participation .01 -.06
Academically oriented extracurricular participation -.01 .06

Lack of Bond with Teacher .53 .52


Teachers put down students (reverse coded) .54 .61
Teaching is good .76 .72
Teachers are interested in students .81 .80
Teachers praise My efforts .73 .69
Most of my teachers listen to me .77 .75

N 15,187 14,551
Staying Back and Dropping Out 237

REFERENCES The Variable Significance of Community


Context.” Social Science Research 32:659–98.
Alexander, Karl L., Doris R. Entwisle, and Susan L. Cusick, Philip. 1973. Inside High School: The
Dauber. 2003. On the Success of Failure: A Student’s World. New York: Holt, Rinehart, &
Reassessment of the Effects of Retention in the Winston.
Primary Grades (2nd ed.). New York: Dauber, Susan L., Karl L. Alexander, and Doris R.
Cambridge University Press. Entwisle. 1993. “Characteristics of Retainees
Alexander, Karl L., Doris R. Entwisle, and Carrie and Early Precursors of Retention in Grade:
Horsey. 1997. “From First Grade Forward: Who Is Held Back?” Merrill-Palmer Quarterly
Early Foundations of High School Dropout.” 39:326–43.
Sociology of Education 20:87–107. Davalos, Deana B., Ernest L. Chavez, and Robert J.
Bankston, Carl L., and Min Zhou. 2002. “Being Guardiola. 1999. “The Effects of Extracurricular
Well vs. Doing Well: Self-esteem and School Activity, Ethnic Identification, and Perception of
Performance Among Immigrant and School on Student Dropout Rates.” Hispanic
Nonimmigrant Racial and Ethnic Groups.” Journal of Behavioral Sciences 21:61–77.
International Migration Review 36:389–415. Dennebaum, Joanne M., and Janet M. Kulberg.
Blau, Judith R. 2003. Race in the Schools: 1994. “Kindergarten Retention and Transition
Perpetuating White Dominance? Boulder, CO: Classrooms: Their Relationship to Achieve-
Lynne Rienner Press. ment.” Psychology in Schools 31:5–12.
Blau, Judith R., Stephanie Moller, and Lyle V. Jones. Duncan, Greg J., and Jeanne Brooks-Gunn. 1997.
2004. “Talent Loss and Enrollment Loss: Why The Consequences of Growing Up Poor. New
Test?” Social Science Research 33:409–34. York: Russell Sage Foundation.
Blinder, Alan S. 1976. “On Dogmatism in Human Eitle, Tamela M. 2005. “Do Gender and Race
Capital Theory.” Journal of Human Resources Matter? Explaining the Relationship Between
11:8–22. Sports Participation and Achievement.”
Bonilla-Silva, Eduardo. 2001. White Supremacy and Sociological Spectrum 25:177–95.
Racism. Boulder, CO: Lynne Rienner Press. Eitle, Tamela M., and David J. Eitle. 2002. “Race,
Brown, Randall, and William P. Evans. 2002. Cultural Capital, and the Educational Effects of
“Extracurricular Activity and Ethnicity: Participation in Sports.” Sociology of Education
Creating Greater School Connection Among 75:123–46.
Diverse Student Populations.” Urban Faircloth, Beverly S., and Jill V. Hamm. 2005.
Education 37:41–58. “Sense of Belonging Among High School
Byrnes, Deborah A. 1989. “Attitudes of Students, Students Representing 4 Ethnic Groups.”
Parents, and Educators Toward Repeating a Journal of Youth and Adolescence 34:293–309.
Grade.” Pp. 108–31 in Flunking Grades: Fairlie, Robert W. 1999. “The Absence of the
Research and Policies on Retention, edited by African-American Owned Business: An
Lorrie A. Shepard and Mary Lee Smith. Analysis of the Dynamics of Self-employ-
London: Falmer Press. ment.” Journal of Labor Economics 17:80-108.
Carbonaro, William J. 1998. “A Little Help from My —. 2003. “An Extension of the Blinder-Oaxaca
Friend’s Parents: Intergenerational Closure Decomposition Technique to Logit and Probit
and Educational Outcomes.” American Models” (Discussion Paper No. 873). New
Sociological Review 71:295–313. Haven, CT: Economic Growth Center, Yale
Coleman, James. 1961. The Adolescent Society: The University. Available online at http://ssrn.com/
Social Life of the Teenager and Its Impact on abstract=497302
Education. New York: Free Press of Glencoe. Finn, Jeremy D. 1989. “Withdrawing from School.”
Croninger, Robert G., and Valerie E. Lee. 2001. Review of Educational Research 59:117–42.
“Social Capital and Dropping Out of High Gottfredson, Denise C., Carolyn M. Fink, and
School: Benefits to At-Risk Students of Nanette Graham. 1994. “Grade Retention and
Teachers’ Support and Guidance.” Teachers Problem Behavior.” American Educational
College Record 103:548–81. Research Journal 31:761–84.
Crosnoe, Robert, Monica K. Johnson, and Glen H. Guo, Guang, Jeanne Brooks-Gunn, and Kathleen
Elder. 2004. “Intergenerational Bonding in M. Harris. 1996. “Parents’ Labor Force
School: The Behavioral and Contextual Attachment and Grade Retention among
Correlates of Student-Teacher Relationships.” Urban Black Children.” Sociology of Education
Sociology of Education 77:60–81. 69:217–36.
Crowder, Kyle, and Scott J. South. 2003. Hallinan, Maureen T. 2000. “On the Linkages
“Neighborhood Distress and School Dropout: Between Sociology of Race and Ethnicity and
238 Stearns, Moller, Blau, and Potochnick

Sociology of Education.” Pp. 65–84 in Ma, Xin. 2003. “Sense of Belonging to School: Can
Handbook of the Sociology of Education, edited Schools Make a Difference?” Journal of
by Maureen Hallinan. New York: Kluwer Educational Research 96:340–49.
Academic/Plenum. Mahoney, Joseph L., and Robert B. Cairns. 1997.
Hauser, Robert M. 1999. “Should We End Social “Do Extracurricular Activities Protect Against
Promotion? Truth and Consequences” Early School Dropout?” Developmental
(Working Paper No. 99-06). Madison: Center Psychology 33:241–53.
for Demography and Ecology, University of McCoy, Ann R., and Arthur J. Reynolds. 1999.
Wisconsin. Available online at http://www.ssc. “Grade Retention and School Performance:
wisc.edu/cde/cdewp/99-06.pdf An Extended Investigation.” Journal of School
Hirschi, Travis. 1969. Causes of Delinquency. Psychology 37:273–98.
Berkeley: University of California Press. McMillen, Marilyn M., and Phillip Kaufman. 1996.
Holmes, C. Thomas, and Kenneth M. Matthews. Dropout Rates in the United States: 1994 (NCES
1984. “The Effects of Non-promotion on 96-863). Washington, DC: U.S. Government
Elementary and Junior High School Pupils: A Printing Office.
Meta-Analysis.” Review of Educational Research McNeal, Ralph B., Jr. 1995. “Extracurricular
54:225–36. Activities and High School Dropouts.”
Janosz, Michel, Marc LeBlanc, Bernard Boulerice, Sociology of Education 68:62–77.
and Richard E. Tremblay. 1997. “Disentan- —. 1999. “Parental Involvement as Social Capital:
gling the Weight of School Dropout Differential Effectiveness on Science
Predictors: A Test on Two Longitudinal Achievement, Truancy, and Dropping Out.”
Samples.” Journal of Youth and Adolescence Social Forces 78:117–44.
26:733–62. Mickelson, Roslyn. 2003. “When Are Racial
Jimerson, Shane. 1999. “On the Failure of Failure: Disparities in Education the Result of Racial
Examining the Relationship Between Early Discrimination? A Social Science Perspective.”
Grade Retention and Education and Teachers’ College Record 105:1080–14.
Employment Outcomes During Late Moller, Stephanie, Elizabeth Stearns, Judith R. Blau,
Adolescence.” Journal of School Psychology and Kenneth C. Land. 2006. “Smooth and
37:243–72. Rough Roads to Academic Achievement:
Jimerson, Shane, Gabrielle E. Anderson, and Angela Retention and Race/Class Disparities in High
D. Whipple. 2002. “Winning the Battle and School.” Social Science Research 35:157–80.
Losing the War: Examining the Relation Newmann, Fred M. 1991. “Student Engagement
Between Grade Retention and Dropping Out in Academic Work: Expanding the Perspective
of High School.” Psychology in the Schools on Secondary School Effectiveness.” Pp.
39:441–57. 58–75 in Rethinking Effective Schools: Research
Johnson, Monica Kirkpatrick, Robert Crosnoe, and and Practice, edited by James R. Bliss, William
Glen H. Elder, Jr. 2001. “Students’ Attachment A. Firestone, and Craig E. Richards.
and Academic Engagement: The Role of Race Englewood Cliffs, NJ: Prentice-Hall.
and Ethnicity.” Sociology of Education Oaxaca, Ronald. 1974. “Sex Discrimination in
74:318–40. Wages.” Pp. 125–51 in Discrimination in Labor
Kaplan, Diane S., B. Mitchell Peck, and Howard B. Markets, edited by Orley Ashenfelter and
Kaplan. 1997. “Decomposing the Academic Albert Rees. Princeton, NJ: Princeton
Failure-Dropout Relationship: A Longitudinal University Press.
Analysis.” Journal of Educational Research Ogbu, John U. 1978. Minority Education and Caste:
90:331–43. The American System in Cross-Cultural
Lee, Valerie E., and David T. Burkam. 1992. Perspective. New York: Academic Press.
“Transferring High Schools: An Alternative to Orfield, Gary, and Chungmei Lee. 2005. “Why
Dropping Out?” American Journal of Education Segregation Matters: Poverty and Educational
100:420–53. Inequality.” Cambridge, MA: The Civil Rights
—. 2003. “Dropping Out of High School: The Role Project, Harvard University. Available online at
of School Organization and Structure.” http://www.civilrightsproject.harvard.edu/res
American Educational Research Journal earch/deseg/Why_Segreg_Matters.pdf
40:353–93. Owings, William A., and Susan Magliaro. 1998.
Lee, Valerie E., and Julia B. Smith. 1995. “Effects of “Grade Retention: A History of Failure.”
High School Restructuring and Size on Early Educational Leadership 56:86–88.
Gains in Achievement and Engagement.” Pallas, Aaron M., Doris R. Entwisle, Karl L.
Sociology of Education 68:241–70. Alexander, and Peter Weinstein. 1990. “Social
Staying Back and Dropping Out 239

Structure and the Development of Self-esteem Flunking Grades: Research and Policies on
in Young Children.” Social Psychology Retention. New York: Falmer Press.
Quarterly 53:302–15. Shepard, Lorrie A., Mary Lee Smith, and Scott F.
Phelps, LeAdelle, Nancy Dowdell, Frank G. Rizzo, Marion. 1996. “Failed Evidence on Grade
Peni Ehrlich, and Felicia Wilezenski. 1992. Retention.” Psychology in the Schools
“Five to Ten Years After Placement: The Long- 33:251–61.
Term Efficacy of Retention and Pre-grade —. 1998. ??On the Success of Failure’: A Rejoinder
Transition.” Journal of Psychoeducational to Alexander.” Psychology in the Schools
Assessment 10:116–23. 35:404?06.
Randolph, Karen, Roderick Rose, Mark Fraser, and Smerdon, Becky. 1999. “Engagement and
Dennis Orthner. 2004. “Promoting School
Achievement: Differences Between African-
Success Among At-Risk Youth.” Journal of
American and White High School Students.”
Poverty 8:1–22.
Research in Sociology of Education and
Ream, Robert K. 2003. “Counterfeit Social Capital
Socialization 12:103–34.
and Mexican-American Underachievement.”
Educational Evaluation and Policy Analysis STATA Corporation. 2003. STATA User’s Guide
25:237–62. Release 8. College Station, TX: Stata Press.
“Repeating a Grade: Does It Help?” 1986. Harvard Stearns, Elizabeth, and Elizabeth J. Glennie. 2006.
Education Letter 2:1–4. “When and Why Dropouts Leave School.”
Roderick, Melissa. 1994. “Grade Retention and Youth and Society 38:29–57.
School Dropout: Investigating the Association.” Swanson, Christopher B., and Barbara Schneider.
American Educational Research Journal 1999. “Students on the Move: Residential and
31:729–59. Educational Mobility in America’s Schools.”
Rodney, Laxley W., Betty Crafter, H. Elaine Rodney, Sociology of Education 72:54–67.
and Robert Mupier. 1999. “Variables Teachman, Jay D., Kathleen Paasch, and Karen
Contributing to Grade Retention among Carver. 1996. “Social Capital and Dropping
African American Adolescent Males.” Journal Out of School Early.” Journal of Marriage and
of Educational Research 92:185–90. the Family 58:773–83.
Rumberger, Russell. 1995. “Dropping Out of Tinto, Vincent. 1993. Leaving College: Rethinking
Middle School: A Multilevel Analysis of the Causes and Cures of Student Attrition.
Students and Schools.” American Educational Chicago: University of Chicago Press.
Research Journal 32:583–625. Valadez, James R. 2002. “The Influence of Social
Rumberger, Russell W., and Katherine A. Larson. Capital on Mathematics Course Selection by
1998. “Student Mobility and the Increased
Latino High School Students.” Hispanic Journal
Risk of High School Dropout.” American
of Behavioral Sciences 24:319–39.
Journal of Education 107:1–35.
Velez, William. 1989. “High School Attrition
Rumberger, Russell W., and Gloria M. Rodriguez.
Among Hispanic and Non-Hispanic White
2002. “Chicano Dropouts: An Update of
Research and Policy Issues.” Pp 114–46 in Youths.” Sociology of Education 62:119–33.
Chicano School Failure and Success: Research and Walker, Elaine M., and Serge Madhere 1987.
Policy Agendas for the New Millennium, edited by “Multiple Retentions: Some Consequences for
Richard R. Valencia. New York: Teachers College the Cognitive and Affective Maturation of
Press. Minority Elementary Students.” Urban
Rumberger, Russell W., and Scott L. Thomas. 2000. Education 22:85–102.
“The Distribution of Dropout and Turnover Yan, Wenfan, and Qiuyun Lin. 2005. “Parent
Rates Among Urban and Suburban High Involvement and Mathematics Achievement:
Schools.” Sociology of Education 73:39–67. Contrast Across Racial and Ethnic Groups.”
Shepard, Lorrie A., and Mary Lee Smith. 1989. Journal of Educational Research 99:116–27.
240 Stearns, Moller, Blau, and Potochnick

Elizabeth Stearns, Ph.D., is Assistant Professor, Department of Sociology, University of North


Carolina at Charlotte. Her main fields of interest are racial and gender differences in educational
outcomes, interracial friendships, and the longitudinal effects of high school experiences. Her cur-
rent work focuses on the long-term effects of high school experiences, particularly how track place-
ment and high school structural variables influence individuals’ success in postsecondary education
and the labor market. She is also investigating how the availability of extracurricular activities influ-
ences students’ decisions about whether to participate in activities.

Stephanie Moller, Ph.D., is Assistant Professor, Department of Sociology, University of North


Carolina at Charlotte. Her main fields of interest are education and stratification. She is currently
studying curriculum structure, students’ achievement, and postsecondary educational attainment.

Judith Blau, Ph.D., is Professor, Department of Sociology, University of North Carolina at Chapel
Hill, and President, Sociologists Without Borders. Her main field of interest is human rights. She is
currently undertaking theoretical work on human rights and collective goods.

Stephanie Potochnick, MA, is a doctoral student, Department of Public Policy, University of North
Carolina at Charlotte. Her main fields of interest are education and immigration. She is currently
examining the race, class, and gender predictors of college prestige and how predictors of students’
achievement differ across generations of Mexican Americans.

Many thanks to Linda Renzulli for her helpful comments; Benn Jann, Joseph Whitmeyer, and Charles
Brody for their statistical guidance; and Sara Kraft for her help in constructing the table. Address all
correspondence to Elizabeth Stearns, Department of Sociology, University of North Carolina at
Charlotte, 9201 University City Boulevard, Charlotte, NC 28223; e-mail: mestearn@ uncc.edu.
View publication stats

You might also like