Ga Moran 2016

Does Racial Isolation in School Lead
to Long-Term Disadvantages? Labor

Market Consequences of High School
Racial Composition1
Adam Gamoran
William T. Grant Foundation
Sarah Barfels
Kirkwood Community College
Ana Cristina Collares

Universidade de Brasilia
School racial composition has modest effects on test score gaps, but
evidence of a longer-term impact is scarce. Perpetuation theory suggests
that blacks who attend schools with higher proportions of white classmates may have better job outcomes. Multilevel analyses of two national longitudinal surveys reveal no effects of high school racial composition on occupational status, employment, or annual earnings for
blacks or whites. For other minority groups, attending schools with
more whites impedes occupational advancement. For all groups, however, school racial composition predicts workplace racial composition:
Whites who attend high schools with higher proportions of white students have higher proportions of white coworkers, while nonwhites
who attend schools with higher proportions of whites have fewer samerace coworkers. The ndings are modest in size but robust to alternative specications, and sensitivity analyses support a causal interpretation for same-race coworkers. These results support perpetuation
theory for workplace composition but not for stratication outcomes.
Particularly in the case of African-Americans, the history of race relations
in the United States is largely a history of exclusion. Despite the major mile1
This research was supported by a grant from the William T. Grant Foundation while
the rst author was on the faculty of the University of WisconsinMadison. The au-
2016 by The University of Chicago. All rights reserved.

0002-9602/2016/12104-0004$10.00
1116
AJS Volume 121 Number 4 (January 2016): 111667
This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
Racial Isolation in School

stones of the election and reelection of the nations rst African-American
president, blacks remain underrepresented in positions of status and power
and substantially separated from whites in everyday life Mouw and Entwisle 2006; Alba 2009; National Urban League 2010; DiPrete et al. 2011.
Although race was less predictive of social standing at the end of the 20th
century than it was at the beginning, it remains an essential dividing line in
U.S. society Wilson 1978, 1987; Massey and Denton 1993; Fischer et al.
2004. As the 21st century unfolds, unemployment rates for blacks are
approximately twice those of whites BLS 2013, college graduation rates
are about 50% greater for whites than for blacks National Urban League
2010, and despite some decline in residential segregation, segregation rates
remain high for African-Americans compared to members of other groups in
suburban as well as urban communities across the United States Charles
2003; Alba 2009.
Since the middle of the last century, public schools have served as a major battleground for efforts to combat racial exclusion. The U.S. Supreme
Courts landmark decision in Brown v. Board of Education 347 U.S. 483
1954 declared that separate educational facilities are inherently unequal and required school districts to integrate their student populations.
At rst reinforced through authorization of cross-town busing to desegregate schools in Swann V. Charlotte-Mecklenberg Board of Education 402
U.S. 1 1971, the power of Brown was subsequently eroded when in
Millikin v. Bradley 418 U.S. 717 1974 the Supreme Court limited the
reach of desegregation to the boundaries of a school district. By the 1990s,
the courts had begun to relinquish supervision over mandatory desegregation with declarations that city school districts are unitary and thus no
longer required to maintain schools that are racially mixed Board of Education v. Dowell, 48 U.S. 237 1991. In 2007, the Supreme Court made
it difcult for school districts to maintain even voluntary systems of desegregation by disallowing race as a criterion for school enrollments Parents Involved in Community Schools v. Seattle School District No. 1, 551 U.S.
701 2007. Desegregation is now largely a policy of the past, and the school
segregation of African-Americans has increased in districts that have been
declared unitary Clotfelter, Vigdor, and Ladd 2006; Gamoran and Long
2007; An and Gamoran 2009; Reardon et al. 2012; Billings, Deming, and
Rockoff 2014; Gamoran and An, in press.
What are the implications of these policy shifts for racial inequality? By
some accounts, the consequences will be minimal. Over the years, school
thors are grateful for helpful comments on earlier versions of the article from James
Ainsworth, Brian An, Sarah Bruch, Thomas DiPrete, Ellen Goldring, Robert Granger,
Maureen Hallinan, and the AJS referees. Direct correspondence to Adam Gamoran,
William T. Grant Foundation, 570 Lexington Avenue, 18th oor, New York, New York
10022. Email: agamoran@wtgrantfdn.org
1117

American Journal of Sociology

desegregation has yielded modest and inconsistent effects on the achievement of African-Americans e.g., Entwisle and Alexander 1992, 1994, in
contrast to the clear benets that were initially anticipated e.g., Coleman
et al. 1966. Indeed, many African-American political and educational leaders have advocated an end to desegregation policies, arguing that it is the
quality of schooling, and not the racial composition of schools, that matters
for student achievement Bates 1990; Schmidt 1994. From the early 1970s
to the late 1980s, test scores for white and African-American students became more equal, but only a small portion of this convergence can be attributed to desegregation Grissmer et al. 1994; Hedges and Nowell 1999.
Supporters of desegregation argue that the focus on student achievement
is too narrow Wells et al. 2004. Wells and Crain 1994, for example,
suggested that to assess the impact of school desegregation policy on the
status attainment of African American adults, researchers and policymakers need to look beyond the short-term effects, especially standardized
test scores, and focus more on long-term social and economic outcomes
pp. 53233. According to this view, the point of mixing students of different racial and ethnic backgrounds in schools is not just to reduce achievement inequality but to prepare young persons to break down barriers to
advancement in adult society Coleman 1976 1999; Granovetter 1986;
Wells and Crain 1994; Wells et al. 2004; Braddock 2009. Compared to research on achievement, however, studies of the relation between school composition and long-term outcomes are rare.
This article advances research on school racial composition in several
ways. First, in contrast to a literature that focuses mainly on short-term
outcomes with limited samples, this article explores whether high school
racial composition has long-term effects on workplace outcomes as evidenced by nationally representative survey samples. Second, following
recent contributions to the achievement literature e.g., Goldsmith 2004;
Lee 2007, this study examines school composition effects for other minority
groups as well as for African-Americans. Third, rather than assuming that
school composition effects are invariant across schools, as is typically the
case in survey research, this study assesses whether the effects depend on
key elements of the school context. Fourth, by examining two national surveys conducted a decade apart the high school classes of 1982 and 1992,
this study allows for the possibility that school composition effects changed
as the politics of desegregation shifted.
HOW MIGHT LONG-TERM EFFECTS OCCUR?
Drawing on Allports 1954 contact theory and Pettigrews 1965 analysis of barriers to desegregation, Braddock 1980 developed perpetuation
theory to explain how young persons experiences of racial isolation may
1118


lead them to live and work in largely segregated environments as adults.
Initially, Braddock offered a social-psychological account, asserting that
blacks who grow up in segregated environments tend to remain in segregated circumstances because they feel unprepared to navigate the largely
white realms of high-status educational and occupational tiers. Attending
an integrated school, Braddock reasoned, helps blacks prepare for social
advancement by enhancing their condence that they can succeed in predominantly white environments and by building up their experience with
handling the tensions of interracial contact Braddock and McPartland
1989. Support for this perspective comes from evidence that interracial
contact during childhood reduces racial stereotyping and helps build social
skills for interracial environments Ellison and Powers 1994; Wood and
Sonleiter 1996; see Pettigrew and Tropp 2006 for a review. With this socialpsychological mechanism in mind, Braddock intended perpetuation theory
to address both socioeconomic advancement and workplace integration, as
both outcomes are constrained by the experience of segregated schooling.
Moreover, these outcomes are linked because more equality in status attainment is likely to result in more mixed-race workplaces, and vice versa.
Extensions of Perpetuation Theory
Other writers such as Granovetter 1986 and Wells and Crain 1994 extended perpetuation theory to incorporate a structural mechanism: social
networks that offer contacts and information that may help young persons
nd places in postsecondary education and employment see also Braddock
2009. Granovetter 1986 suggested that even if interracial friendships in
integrated schools were fragile, they might nonetheless offer the weak ties
that enhance the chances of nding a job. Further fuel for this perspective
came from subsequent research showing that interracial friendships were
nearly as stable as same-race friendships and that school characteristics inuenced friendship formation Hallinan and Williams 1987, 1989. Building on these ndings, Wells and Crain 1994 explained that perpetuation
theory posited two mechanisms: an individual mechanism that operates
through attitudes and dispositions and a collective mechanism that works
through network ties and the ow of information that such ties facilitate see
also Wells et al. 2004. Like the social-psychological mechanism, the network mechanism may operate similarly for both workplace integration and
status attainment outcomes.
Mixed Evidence of Long-Term Effects
Case studies have marshaled evidence to support perpetuation theory.
Wells and Crain 1994 cited several small-scale, often retrospective, ac1119


counts that tied school segregation to limited social and economic opportunities and outcomes e.g., Crain 1970, 1971. More recent work has suggested, correspondingly, that racially mixed schools advance black students
opportunities to make their way in a white-dominated society Wells et al.
2004. Like the earlier work, this research has relied on historical and retrospective analyses that, though provocative, may be limited in their generalizability because recollections of past barriers and supports may be colored
by later successes and failures.
Experiments and longitudinal surveys have yielded mixed results for the
long-term effects of school racial composition. An important landmark was
Kaufman and Rosenbaums 1992 analysis of the impact of Hills v. Gautreaux 425 U.S. 284 1976, a housing discrimination case. As a result of
this case, some Chicago families had the chance to relocate from the inner
city to largely white suburban neighborhoods while others moved to largely
black city neighborhoods; and because families had no control over the relocation site they were offered, one can treat this as a case of random assignment to largely white or largely black schools. Kaufman and Rosenbaum
showed that black youths who moved to the white suburbs were more likely
to complete high school and obtained higher wages than those who remained
in the cities. These ndings are consistent with those from earlier desegregation experiments, which also showed positive effects Armor 1972; Crain
and Strauss 1985. However, a more recent housing experiment called
Moving to Opportunity MTO, in which families were offered vouchers to
move from high-poverty to low-poverty neighborhoods, yielded few lasting
benets overall Vigdor and Ludwig 2008; Ludwig et al. 2013. Weak effects
for MTO may reect a restricted range on changes in neighborhood racial
composition: although participants moved to neighborhoods with signicantly less poverty, changes in neighborhood racial composition were modest,
as were changes in school conditions Clampet-Lundquist and Massey 2008;
DeLuca and Rosenblatt 2010. Yet a reanalysis of MTO data suggested that
although effects for those who moved as teenagers were null or even negative,
those who moved as younger children experienced long-term benets to
income as adults Chetty, Hendren, and Katz 2015. Hence, social experiments have left questions about racial segregation and long-term outcomes
unresolved Sampson 2008.
Studies that have focused on the timing of school desegregation, as opposed to variation in school racial composition, have uncovered long-term
benets for blacks. For example, examining high school completion rates at
the district level, Guryan 2004 found that the implementation of desegregation plans elevated the chances that black students would graduate
from high school. Using individual-level, retrospective data, Ashenfelter,
Collins, and Yoon 2006 reported that southern-born blacks who would
have nished their schooling just prior to desegregation in the South were
1120


less likely to nish high school and earned less income as adults than those
who would have nished their schooling a few years after desegregation
occurred. Taking the timing of court battles over desegregation as an instrument for the implementation of desegregation plans, Johnson 2011 reported that school desegregation led to more years of schooling and greater
earnings for blacks who attended school after desegregation compared to
those who attended school earlier. Importantly, however, Johnson noted
that the actual racial composition of schools did not serve as a mechanism for
the desegregation effects he uncovered. Instead, improvements in school
quality e.g., increased per pupil expenditures appeared more likely to account for the benets of desegregation. Billings et al. 2014 offered a similar
explanation for why the end of court-ordered desegregation in CharlotteMecklenburg, North Carolina, did not reduce school outcomes for black
students despite the resulting increase in racial isolation.
In contrast to comparisons of students before and after desegregation,
studies with direct evidence on the racial composition of schools have
yielded inconsistent conclusions. Early on, Crain and Mahard 1978 found
that attending schools with a higher percentage of white students contributed positively to the college attendance of black students in the North, but
not in the South. Similarly, Braddock and McPartland 1989 reported that
black students in the North who attended more integrated high schools
tended to nd jobs in more integrated work environments, a key prediction
of perpetuation theory, but the nding did not hold in the South. Using a
different national survey, however, with retrospective data on school composition and employer reports of workplace composition, Braddock, Dawkins, and Trent 1994 found that blacks, whites, and Mexican Americans
who had attended integrated schools were more likely to be employed in
mixed-race workplaces. On the basis of the same retrospective data, Boozer,
Krueger, and Wolkon 1992 reported that blacks who attended high school
with more black schoolmates had more black coworkers as adults, completed fewer years of education, and earned lower wages. However, the
coworker effects disappeared and the wage effects diminished in models
that used instrumental variables to adjust for selection bias. Meanwhile,
Dawkins 1994 also identied a range of associations between school racial
composition and long-term outcomes, but most of them dissipated when
background conditions were controlled; Wilson 1979 concluded that school
integration had no long-term effects once the social background characteristics of students were taken into account.
Research on more recent surveys has added to the inconsistent ndings
on long-term effects. After taking account of selectivity into racially isolated versus racially mixed high schools, Grogger 1996 reported that the
racial composition was unrelated to the wages of black high school graduates. Similarly, Rivkin 2000 found no effect of school racial composition
1121


on educational attainment or income of blacks 10 years after their scheduled high school graduation. Yet Goldsmith 2009 found that minority
students who attended high school with higher proportions of other minority students attained less education, and Stearns 2010 reported that minority students in high-minority high schools tended to have more coworkers of their own race/ethnic groups after high school.
Owing to limitations of the study designs, these results cannot be taken
as denitive. On the one hand, early analyses commonly used retrospective
rather than prospective data, and many of the longitudinal studies either
provided few modeling details or did little to control for selection into schools
with different compositions, so effects of school racial composition may
have been overestimated. On the other hand, if attending a racially mixed
school increases the likelihood of high school completion for AfricanAmericans, studies that began with high school graduates e.g., Grogger
1996 may have underestimated the benets of reducing racial isolation.
Because Rivkin 2000 constructed a measure of school quality based on
the academic performance of nonblacks, his analysis omitted all schools
that were fully segregated. Thus, he compared schools with larger and
smaller proportions of minority students, but he did not compare racially
mixed schools with those in which blacks were completely isolated. Yet
this may be the crucial comparison. Consequently, further work is warranted on the effects of high school racial composition on labor market
outcomes.
LIMITATIONS ON LONG-TERM EFFECTS
Despite continued interest in perpetuation theory, recent studies offer at

least four reasons why school racial composition may not have long-term
effects, which may account for the empirical inconsistencies. First, attending school with same-race classmates may have benets that counteract
losses due to isolation. Second, variation in the contexts of racially mixed
schools may lead to heterogeneous effects of attending school with classmates of other races. Third, racial isolation in school may perpetuate racial
isolation in the workplace, but the latter may not inhibit socioeconomic
advancement. Finally, inconsistent ndings may reect different effects of
racial composition during time periods that featured different policy responses to concerns over racial isolation.
Possible Benets of Same-Race Classmates
Although most writers emphasize the disadvantages of school racial isolation, a few studies have noted advantages of attending school with others of the same racial or ethnic background. Carter 2010, for example,
1122


studied cultural exibility, an attitude of openness to interaction with
persons of other backgrounds that corresponds closely to the sort of psychological preparedness for social success that Braddock 1980 argued
was forestalled by the experience of segregated schools. Yet Carter found a
greater presence of cultural exibility among blacks in schools with predominantly minority populations than among blacks whose schools had
largely white populations. She attributed this pattern to differences in selfesteem, which she also observed to be higher in high-minority schools than
in largely white schools, possibly reecting a segregated tracking structure
in the latter. Similarly, Tyson, Darity, and Castellino 2005, investigating
the concern that social pressure causes black youths to refrain from excelling in school to avoid the appearance of acting white, found evidence
of such pressure only in a racially mixed school, not in schools populated
primarily by minority students. Tyson 2011, p. 67 elaborated: Unlike
their high-achieving black counterparts at racially diverse and predominantly white schools, students in these majority black schools did not
report any problems with students casting achievement as acting white or
other racialized ridicule and ostracism. Evidence of the generalizability of
these patterns comes from work by Fryer 2006, who observed in a national survey analysis that a tendency among blacks to gain popularity by
underperforming in school was twice as large in the more-integrated racially mixed schools as in the less-integrated ones pp. 5758.
With respect to other minority groups, Portes and his colleagues Portes
and McLeod 1996; Portes and Hao 2004 have reported that the
achievement disadvantages of Mexican American students are attenuated
in schools with larger proportions of co-ethnic students. Similarly, Goldsmith 2004 found that both black and Hispanic students had more positive attitudes and better relationships with teachers in schools with smaller
proportions of non-Hispanic whites and that these attributes tended to
reduce the achievement gaps between minority and white students.
Clearly, racial isolation limits the opportunities for minority students to
have interracial contacts. However, recent work on peer relations indicates
that friendship patterns do not follow a linear trend that corresponds to
school composition. Moody 2001 reported that increasing heterogeneity in
school composition was linked to higher levels of friendship segregation, a
pattern that was reversed only at the top levels of heterogeneity. Likewise,
Mouw and Entwisle 2006 observed a high degree of racial segregation in
friendships even among students who attend the same school. Thus, while
racially mixed schools offer the potential for increasing minority students
contacts with persons of other races and their ease in heterogeneous environments as perpetuation theory argues, racially mixed schools have
countervailing pressures that may also limit minority students academic
performance and social success.
1123


The Context of School Racial Composition
Even if perpetuation theory correctly predicts that racial isolation constrains long-term opportunities, the effects of school racial composition may
be obscured by varying contextual conditions. Few survey studies attend
to the underlying school policies or contexts through which school racial
compositions are formed, but understanding these conditions may be crucial
for identifying long-term effects. At least two conditions have been identied that may induce variation in the observed effects of school racial
composition: the degree of racial conict in the school environment and the
extent of resegregation within racially mixed schools.
Desegregation policies and racial conict.A mixed racial composition
that occurs in a climate of hostility and conict is less likely to promote
positive outcomes than racial mixing that comes about without conict
Schoeld 1995. Both contemporary and retrospective accounts indicate
that outcomes are tied to the local context in which desegregation occurred
Eaton 2001; Hochschild and Scovronick 2003; Wells et al. 2004. Conict
may occur in response to busing policies or court orders that desegregate
schools that would be racially isolated if attendance were based solely on
neighborhood zoning patterns, or it may occur in schools that are racially
mixed without such policy levers. To identify the effects of composition
accuratelywhether positive or negativeit is important to take account
of such contextual conditions.
Although our focus is on school racial composition, desegregation efforts
of the 1970s, 1980s, and 1990s provide the policy context for our investigation, even as perpetuation theory provides the sociological context. It is
worth noting that changing racial composition is not the only mechanism
through which desegregation effects may operate Schoeld 1995; Johnson
2011. By examining the effects of racial composition in the presence and
absence of desegregation policies as manifested in court orders and busing
intended to alter school racial composition, we are able to assess the impact of racial composition in varied policy contexts.
Resegregation within schools.Some scholars have argued that resegregation within schoolsthrough tracking, ability grouping, or segregated
extracurricular activitiesmay have forestalled any benets that might have
accrued to desegregation Eyler, Cook, and Ward 1983; Mickelson 2001;
Clotfelter 2002. Although this argument was originally offered to explain
the absence of short-term desegregation effects, Wells et al. 2004 proposed
that it may hold for long-term effects as well. Inattention to context and
policy may have led researchers to miss the long-term benets of attending
racially mixed schools that can occur in some cases, as indicated in the
retrospective accounts. When assessing the impact of school racial composition, it is important to account for differences in the racial compositions
1124


of the settings that students actually experienced within schools Tyson
2011.
School Racial Composition and Ethnic Labor Markets
Even if attending racially isolated schools results in racial isolation in the
labor market, such patterns of isolation may be less consequential for labor
market outcomes among immigrant minority groups as compared with
African-Americans. Urban sociologists have noted that immigrant groups
such as Hispanics and Asian Americans who work in segregated communities can take advantage of ethnic enclaves to nd opportunities for
work and occupational advancement Portes and Jensen 1989; Bohon
2001; Edin, Fredricksson, and Aslund 2003. As posited by segmented assimilation theory, strong social networks in a co-ethnic community provide resources that help buffer community members from the negative effects of
disadvantaged social backgrounds such as low levels of parental education
Portes and Zhou 1993; Portes and Fernandez-Kelly 2008. High levels of
social capital in ethnic communities may increase the availability of information for workers and employers in the sector, shaping their work relations
and producing advantages for some workers Wilson and Martin 1982;
Bailey and Waldinger 1991.
Extending this reasoning to school segregation, immigrant youths who
attend high schools with few whites may develop networks of mutual support that help them succeed in the job market, in occupational status if not
in earnings. Attending segregated schools may even stave off the acculturation that sometimes leads successive generations of immigrants to attain
less than the previous generation Portes and Rumbaut 2001. For these
reasons, attending schools with a high proportion of whites may actually
limit opportunities for occupational success among youths from immigrant
minority groups, in contrast to African-Americans for whom segregated
communities often mean diminished opportunities Wilson 1978; Anderson 1990.
Temporal Variation in Composition Effects
Recent survey studies of school composition effects follow students longitudinally; in the best cases, students are observed throughout their high
school careers and beyond Rivkin 2000; Goldsmith 2009; Stearns 2010.
These studies, however, do not examine multiple cohorts of students to
allow for the possibility that composition effects may change over time.
The inconsistent ndings from past research do not show a clear time
trend, but the wide variety of sampling frames and statistical models used
in past studies make it difcult to discern a temporal pattern. In this study,
1125


by contrast, we apply the same statistical models to nearly identical survey
data collected from two cohorts a decade apart. The class of 1982 completed high school when the racial isolation of blacks in schools was at a
low ebb Oreld 2001; Clotfelter 2004: mandatory desegregation plans
were widespread across the southern states, and voluntary programs were
in place in cities across the country. By 1992, the dismantling of desegregation had begun; school districts were beginning to be relieved of court
oversight, and a trend of increasing segregation had commenced Oreld
2001. If attending a school with a higher proportion of white students
helps blacks in the long run, these effects may have been more potent for
the class of 1982 than for the class of 1992.
Attending to Variation in the Present Study
Considerations of possible heterogeneous effects have shaped the research
design for this study. In addition to examining two different cohorts that
exited high schools at different times in the desegregation debate, we examine varied policy conditions within cohorts. We consider the presence of
busing and court orders, internal resegregation within schools, and school
racial climate, to allow for possible variability in the effects of composition.
Further, we consider the relation between school composition and longterm outcomes not only for blacks and whites but also for students of other
ethnic backgrounds.
DATA
Data for this study come from two national longitudinal studies conducted
by the U.S. Department of Education. High School and Beyond HSB
began in 1980 with a sample of more than 20,000 high school sophomores in
more than 1,000 schools. Students were followed up in 1982, 1984, 1986, and
1992. This study draws on the base year and the 1992 follow-up, focusing on
the association between high school characteristics and labor market outcomes 10 years after the scheduled high school graduation.2 In 1984, the
sample was reduced to a random subsample of about 14,000 respondents,
who constitute the sample for this study.
The National Education Longitudinal Survey NELS began in 1988 with
a sample of nearly 25,000 eighth graders in more than 1,000 schools. A representative subsample of about 15,000 students was selected for follow-up
to more than 1,000 high schools in 1990, and subsequent follow-up surveys
2
Students who changed schools or who dropped out of school were retained in the
analytic sample because dropout and mobility may be endogenous to school racial
composition. In both cases, students school characteristics were taken from the schools
they attended in tenth grade.
1126


were conducted in 1992, 1994, and 2000. By using sample weights and
freshening additions to the sample, the 1990 survey wave can be made to
represent the population of tenth graders in 1990 which is not identical to
the two-year follow-up of the 1988 eighth-grade population. We analyze
the tenth-grade NELS freshened sample for comparability with HSB,
which began with tenth graders.3 To take advantage of the fact that NELS
rst surveyed students at the outset of their high school careersan
important advantage over HSBwe also analyze the NELS eighth-grade
sample. Thus, we analyze results for three samples of students: HSB 1980
tenth graders, followed up in 1992; NELS 1990 tenth graders, followed up
in 2000; and NELS 1988 eighth graders, followed up in 2000. We have
the most condence in the latter results because student control variables
e.g., test scores can be taken into account at the beginning of high school,
rather than two years into high school when they may have been affected by
racial composition. Also, the eighth-grade sample captures the entire age
cohort because almost no students drop out before eighth grade, whereas by
tenth grade a small fraction have dropped out. The tenth-grade samples are
still useful, however, because they permit comparisons across cohorts.
Variables
With few exceptions, the same variables are used in analyses of both data
sets. Note, however, that outcomes in HSB were obtained 10 years after
expected high school completion, whereas NELS outcomes date to eight
years after high school completion. Means and standard deviations for all
variables in the three samples are found in table 1.
Dependent variables.Four labor market outcomes are used as dependent variables: occupational status, employment, earnings, and workplace
racial composition. These variables were reported by respondents on the
1992 HSB and 2000 NELS surveys. Occupational status is based on
responses to survey categories coded according to Hauser and Warrens
1997 values derived from a combination of occupational education,
occupational wages, and occupational prestige from the 1990 census.
Employment is a dummy variable indicating whether the respondent
worked more than 20 weeks NELS; or ve months in HSB.4 For earnings,
we used the log of annual earnings in the previous calendar year, or during
the most recent of the two preceding years if earnings were unreported for
3
It should be clear that sample freshening is not a response to attrition, but rather is a
means to represent a different population, in this case the population of tenth graders in
1990, as contrasted with the population of eighth graders in 1988. These populations are
not the same because of grade retention, dropout, and international migration. To
address attrition, sampling weights are used, as described below.
4
Other specications of weeks employed yielded similar results.
1127

1128

Student level:
Proportion of coworkers same
Student test scores . . . . . . . .
Missing test scores . . . . . . . .
Student SES . . . . . . . . . . . .
Missing SES . . . . . . . . . . . .
Male . . . . . . . . . . . . . . . . . .
Black . . . . . . . . . . . . . . . . .
Other . . . . . . . . . . . . . . . . .
Native American . . . . . . . . .
Hispanic . . . . . . . . . . . . . . .
Asian . . . . . . . . . . . . . . . . .
Race relations . . . . . . . . . . .
Missing race relations . . . . . .
Number of students . . . . . . .
School level:a
School proportion white . . . .
School SES . . . . . . . . . . . . .
Private . . . . . . . . . . . . . . . .
North . . . . . . . . . . . . . . . . .
Midwest . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
race
....
....
....
....
....
....
....
....
....
....
....
....
....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.72
2.08
.11
.22
.28
9,300
.72
50.41
.07
2.04
.03
.48
.10
.14
Mean
HSB
.31
.46
.31
.41
.45
.30
8.62
.25
.71
.17
.50
.30
.35
SD
TABLE 1
Descriptive Statistics
NELS 10
.69
.01
.17
.21
.25
10,120
.65
51.39
.06
.02
.05
.49
.11
.14
Mean
.32
.57
.38
.41
.43
.31
9.59
.23
.75
.21
.50
.31
.35
SD
NELS 8
.32
.58
.38
.41
.43
.50
.32
.34
.11
.30
.19
.67
.22
.49
.11
.14
.01
.10
.04
3.09
.05
10,060
.69
.00
.17
.21
.25
.31
9.89
.19
.74
SD
.64
51.53
.04
2.05
Mean
1129

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
35.27
N 5 9,300
9.95
N 5 8,620
.90
N 5 9,020
870
.30
.20
.24
.28
.48
Mean
.30
.85
12.43
.46
.40
.45
.45
.50
SD
39.53
N 5 10,080
9.96
N 5 9,200
.96
N 5 9,360
1,070
.34
.19
.36
.26
.38
Mean
NELS 10
.20
.78
13.72
.48
.39
.48
.44
.49
SD
39.13
N 5 10,020
9.96
N 5 9,140
.96
N 5 9,300
.34
.19
.36
.27
.38
3.54
2.48
.07
.08
.04
.17
.08
.05
.35
1,070
Mean
NELS 8
.20
.79
13.70
.47
.40
.48
.44
.48
.59
1.28
.25
.27
.16
.38
.27
.17
.48
SD
NOTE.Student-level statistics are weighted as follows: for HSB, FU4WT; for NELS tenth-grade sample, F4QWT; and for
NELS eighth-grade sample, F4BYPNWT. In accordance with NCES security procedures for restricted data, unweighted
sample sizes have have been rounded to the nearest 10.
a
For cases with data on proportion of coworkers same race.
Employed > 20 weeks . . . . . . . . . . . . . . .
Log of income . . . . . . . . . . . . . . . . . . . . .
South . . . . . . . . . . . . . . . . . . . .
West . . . . . . . . . . . . . . . . . . . . .
Urban . . . . . . . . . . . . . . . . . . . .
Rural . . . . . . . . . . . . . . . . . . . .
Suburban . . . . . . . . . . . . . . . . .
Racial conict . . . . . . . . . . . . . .
Proporton white racial conict
Missing for school racial conict .
Busing . . . . . . . . . . . . . . . . . . . .
Proportion white busing . . . . .
Missing for busing . . . . . . . . . . .
Court order . . . . . . . . . . . . . . . .
Proportion white court order . .
Missing for court order . . . . . . . .
Number of schools . . . . . . . . . . .
Other dependent variables:
Occupational status . . . . . . . . . .
HSB
TABLE 1 (Continued )

the previous year. To facilitate comparison across cohorts, we report results
for both NELS and HSB in 1999 dollars.
A measure of workplace racial composition is derived from respondents
answer to the question, What percentage of people in your present most
recent workplace are were the same race and ethnicity as you? For white
respondents, a higher value on this variable means that their work environment was more white, whereas for blacks, a higher value signies a work
environment populated by a greater concentration of blacks, and similarly
for other groups. Since this variable has different meanings for different
subgroups, it is important to assess interactions, which is a central focus of
our analysis in any case i.e., we assess the effects of racial composition
separately for whites, blacks, and others for all dependent variables.
The percentage of same-race coworkers has been used to gauge the
perpetuation of racial segregation in a wide range of studies across the
decades e.g., Crain and Strauss 1985; Braddock and McPartland 1989;
Boozer et al. 1992; Trent 1997; Stearns 2010. Most studies have used a
self-reported indicator like the one available in these data, and some have
argued that a subjective measure offers a better test for perpetuation
theory than more objective indicators because the workers perception of
the workplace has bearing on social cohesion and feelings of isolation
Stearns 2010. In any case, prior studies that have compared self-reported,
employer-reported, and observed indicators of workplace racial composition have obtained similar results across indicators Braddock et al. 1994;
Trent 1997; Pettigrew and Tropp 2006.
School-level independent variables.The key predictor for our analysis
is the proportion of students in the high school who were white in 1980
HSB or 1990 NELS. This variable was reported by school principals
on a school survey. For three NELS schools that lacked principal reports,
we used the proportion white among students in the student survey sample for that school. We used two indicators of school policy related to
desegregation: whether the school used busing to achieve racial balance
and whether the school was under a court order to desegregate.5 Both were
reported by principals and are coded as dummy variables 1 5 yes, 0 5
no, with additional dummies to indicate nonresponse.
The percentage of white students in the school is not the only possible
indicator of racial composition, but we relied on it for three reasons. First is
the historic importance of % white in both the sociological traditions e.g.,
Coleman et al. 1966 and U.S. education policy e.g., Brown, Millikin,
Dowell, and Seattle. Second, our focus was on students experiences of
5
Both HSB and NELS asked whether students were bused for racial balance. Whereas
HSB asked whether schools were under a court order to desegregate, the analogous item
in NELS was Pupils are assigned from particular areas to achieve a desired racial or
ethnic composition in the school.
1130


their school environments, whereas indicators of segregation such as the
dissimilarity index that measure the evenness which with racial groups are
spread across schools are measures of school districts. We did, however,
compute dissimilarity indices for within-school segregation as described
below. Third, as Oreld 2001 and others have argued, the extent to
which students of minority backgrounds attend schools with white students is an essential marker of equality in education because white students
tend to be privileged in their educational opportunities. Whereas dissimilarity indices of segregation have been stable over the last two decades,
indicators of racial isolation, that is, the extent to which minority students attend school with few white students, have shown dramatic increases since
the early 1990s Oreld 2001; Gamoran and Long 2007, so the impact of
school % white takes on special salience.
To assess the context of desegregation, we sought indicators of racial
conict or hostility. HSB lacked a direct indicator, but NELS asked principals whether racial conict was a serious, moderate, small, or no problem
in their schools. We created a dummy variable to indicate whether racial
conict was a serious or moderate problem vs. small or not at all in the
NELS schools.
Although HSB did not contain direct indicators of racial conict, it
yielded information on the internal resegregation of schools, another important aspect of context. We constructed these variables only for HSB
because of the larger within-school sample size compared to NELS; sample
size constraints also limited these variables to blacks and whites only. Of
course, the role of resegregation can be examined only in schools that include both blacks and whites. Descriptive statistics for this subsample are
found in appendix table A1. We prepared three indicators of internal
resegregation: a the absolute value of the difference in the proportions of
blacks and whites assigned to the academic track mean 5 .30 such that
higher values reect more internal segregation; b the absolute value of
the difference in the proportions of blacks and whites who participate in
any extracurricular activity mean 5 .20; and c the index of dissimilarity
for the participation of blacks and whites across extracurricular activities mean 5 .06. The index of dissimilarity represents the proportion of
blacks or whites who would have to change type of extracurricular activity in order to balance out the representation of blacks and whites in
extracurricular activities in the school Farley and Taeuber 1974.6 We
6
The formula for the index of dissimilarity D is D 5 1=2ojwi =WT 2 bi =BT j, where
WT is the total number of whites in the school who participated in extracurricular
activities, BT is the total number of blacks in the school who participated, wi is the
number of whites in extracurricular activity i, and bi is the number of blacks in
extracurricular activity i.
1131


examined black-white segregation across nine categories of extracurricular
activities: sports, cheerleading, debate/drama, band, chorus, hobby clubs,
subject clubs, vocational education clubs, and junior achievement.
Additional school-level controls included region of the country Midwest,
South, and West compared to the Northeast, private versus public, and
rural and suburban versus urban all reported by principals, and school
mean socioeconomic status SES; aggregated from student reports.
Student-level independent variables.Race/ethnicity was self-reported
by students in the base year of each survey. In HSB, we found too few Asian
Americans and Native Americans to break out these groups separately, so we
combine them with Hispanics in an other category. Subsamples are large
enough in NELS to break out the subgroups particularly Asian Americans,
who were oversampled, as well as Hispanics, so the NELS eighth-grade
analyses include all four racial/ethnic minority groups in comparison to nonHispanic whites. Like race/ethnicity, gender 1 5 male, 0 5 female was
reported and coded identically in the two surveys, but SES was not. Both
indicators of SES are unweighted standardized composites based on fathers
occupation, fathers and mothers education, family income, and household
possessions. In HSB, this information was reported by students as sophomores, while in NELS, it was obtained directly from parents when students
were in eighth grade, with student reports used only in cases of parent nonresponse 8.1% of cases. For NELS only, we also obtained a student report
of race relations: Students make friends with students of other racial and
ethnic groups. This variable, coded 14 on an agree/disagree scale with 4
reecting more positive relations, was drawn from the tenth-grade survey;
but if no response was available in grade 10, the response from grade 12
was used.
Test scores are the tenth-grade standardized composite reading and
mathematics scores or, in the case of the NELS eighth-grade sample, the
eighth-grade composite scores. The data also contain twelfth-grade test
scores, but these are more properly viewed as short-term outcomes of school
racial composition rather than as controls. The impact of racial composition
on high school test scores has been extensively studied in these data, and as
in the larger literature, the effects are modest. A consistent nding across
National Center for Education Statistics NCES surveys is that schools
with smaller proportions of minority students tend to exhibit higher test
scores in science Gamoran 1987; Konstantopoulos 2006, a pattern we
replicated in supplementary analyses. Overall, though, the absence of shortterm achievement effects across achievement domainsin these data sets
and othersis one of the main motivators for this study of long-term effects.
Design weights.Both NELS and HSB were stratied random samples, with schools selected rst and then students randomly selected within
schools, and oversampling occurred for some demographic categories e.g.,
1132


schools with high proportions of Hispanics in HSB, Asian Americans in
NELS. Consequently we use student weights to ensure that the student
data represent the national populations of students in each cohort. The
weights also adjust for survey nonresponse and attrition over time. For
analyses of tenth-grade cohorts, we used weights that pertain to the fourth
follow-up FU4WT in HSB and F4QWT in NELS, rather than panel
weights, to avoid dropping cases that would be excluded by panel weights
because they lacked data from survey waves we did not use, that is, the
rst, second, and third follow-ups. For analysis of the NELS eighth-grade
cohort, we used a panel weight designed specically for analysis of base year
and fourth follow-up data only F4BYPNWT. We applied the weights
to estimates of descriptive statistics as well as analytic results. While the
weights matter for descriptive statistics, the analytic results were largely
the same with or without the weights.
Missing value specications.Among cases with data present on the
dependent variables, no student-level independent variable had more than
5% missing data. We imputed missing values for SES and test scores using
each to impute values for the other, and we used school means to impute
values for student-reported racial conict. In each of these cases we included a dummy variable to indicate that data had been imputed. Less
than 1% of cases with missing data on race/ethnicity or on both test scores
and SES were dropped from the samples. In addition, because of our focus
on within-school as well as between-school inequalities, and for computational efciency, we eliminated schools that had only one student in the
sample. This occurred only in NELS and resulted in a loss of between
140 and 160 schools and students, depending on the subsample eighth or
tenth grade and dependent variables.
Overall, sample attrition reduced the analytic samples by 2,030 cases
14% in HSB and 3,850 cases 25% in NELS. The sample design weights
described above adjust the remaining cases to represent the national
population of students Zahs et al. 1995; Curtin et al. 2002. Further missing cases largely reect missing data on the dependent variables, as cases
with missing data on a dependent variable were dropped only for the
analysis pertaining to that dependent variable. Thus, analytic sample sizes
ranged from 8,620 for earnings in the HSB sample to 10,020 for occupational status in the NELS tenth-grade sample see table 1.7 As a check on
our missing data specications, we also estimated our main results for the
NELS eighth-grade sample using multiple model-based imputation, imputing the dependent variables for all missing cases as well as item-level
7
In accordance with NCES procedures for restricted data, unweighted sample sizes are
rounded to the nearest 10.
1133


missing data on control variables Allison 2001. These results, which rely
on a common case base for all analyses, yielded results virtually identical to
the single-imputation, outcome-specic results reported below see details
provided in n. 14.
METHODS
Studies such as this one face two primary methodological problems: nonindependence of students within schools and nonrandom selection of students to segregated and integrated schools. We address both of these challenges in our analyses.
Nonindependence and Multilevel Models
Ordinary least squares regression assumes that each case is independent of
all others, but when samples are drawn within schools, cases in the same
school are not really independent. This can lead to underestimated standard errors, as students in the same school may appear more alike than
would a totally random collection of students. The problem of nonindependence can be addressed with a multilevel model that distinguishes between effects that occur at the level of the individual student and effects
that occur at the level of the group the school, in our case. In the studentlevel model, a student posthigh school outcome is estimated as a function
of student-level background characteristics, including race. The intercept
from this model is an estimate of the average outcome for students in the
school, which in turn becomes an outcome at the school level. This schoollevel outcome is estimated as a function of school-level variables including
school composition e.g., % white, school policy e.g., the proportion of
students who are bused to achieve racial balance, and school context e.g.,
salience of racial conict.
The individual estimate for an outcome among blacks is another level 1
coefcient that becomes an outcome variable at level 2. Here, we ask whether
schools differ in the extent to which blacks and whites reach similar post
high school outcomes. Differences among schools can be modeled as a function of school characteristics, including composition and policy indicators.
Overall, three elements of the multilevel model are key to our analyses: the
difference in outcomes between blacks and whites within schools, the contribution of school composition and school integration policies to average
school outcomes, and the contribution of school composition and school
policies to the black-white difference in outcomes. Note that composition
and policy effects draw on data for all schools, while results for black-white
gaps within schools actually draw only on schools with at least one black
and at least one white student, even though all the data are used in the
1134


analyses. A similar logic pertains in analyses of students from other minority
groups.8
We treat three of the outcomes as continuous occupational status, log
income, and proportion same-race coworkers and one as binary employed
more than 20 weeks. We use appropriate multilevel models for linear and
binary outcomes, respectively. For ease of interpretation, all predictor variables other than race/ethnicity are centered on their grand means.
Propensity Models and Sensitivity Analyses to Address
Nonrandom Selection
The problem of nonrandom selection is more difcult to address. We
cannot randomly assign students to schools with different racial compositions, so we use statistical approaches to approximate the experimental
comparison. The rst step is to include a rigorous set of controls that helps
take account of nonrandom differences among students who attend different schools. Prior research with national survey data indicates that, for
at least some selection processes notably educational tracking, the rich
background data provide adequate controls Gamoran and Mare 1989.
One way to go beyond covariate controls is to estimate propensity score
models that compare students who have similar likelihoods of attending
schools of a given racial composition but whose actual racial compositions
varied, to approximate an experimental comparison of randomly assigned
treatment and control groups. Morgan 2001 implemented a similar
8
Following the notation of Raudenbush and Bryk 2002, the equations for the basic
models are specied as follows:
level 1:
yij 5 b0j 1 b1j Black 1 b2j Other 1 b3j SES 1 b4j Male
1 b5j Prior Achievement 1 ij ;
level 2:
b0j 5 g00 1 g01 Proportion White 1 g02 Sector 1 g03 Urbanicity
1 g04 Region 1 g05 Sch SES 1 u0j ;
b1j 5 g10 1 g11 Proportion White 1 u1j ;
b2j 5 g20 1 g21 Proportion White 1 u2j ;
where y represents the linear dependent variables and Urbanicity and Region each
stand for a set of dummy variables. Missing data indicators as described in the text are
also included. For the categorical dependent variable, a multilevel logistic specication
is used. The key parameters are g01, g11, and g21, which represent the racial composition
effect for whites g01 and the difference between the effect for whites and that for blacks
g11 and others g21.
1135


model to assess Catholic school effects, and Harding 2003 followed this
strategy to examine neighborhood effects. The primary advantage of the
propensity model is its focus on a balanced comparison in which cases that
are similar on the control variables but differ on the treatment can be
directly compared Morgan and Winship 2007; Gangl 2010. In addition,
propensity models focus the investigators attention on cases that are similar enough in their covariate distributionsreferred to as a region of
common supportso as to make estimates of treatment effects reasonable. Such diagnostics are rarely examined in linear regression analysis, but
they are routine in matching estimation.
Despite their advantages, neither propensity models nor covariate adjustment approaches control for selection on unobserved characteristics. Hence,
we assessed the robustness of the causal conclusions from our propensity score
analysis to possible unobserved confounders. We followed the Rosenbaum
2002 bounds procedure see also DiPrete and Gangl 2004 to assess the
sensitivity of the propensity matching results to unobserved variables whose
omission may distort the treatment effect. The bounds procedure simulates
how powerful an omitted predictor of selection into the treatment would
need to be assuming it is perfectly correlated with the outcome in order to
undermine the treatment effect estimated in the matching model. Comparison of the hypothetical confounder with observed variables involved in
the selection process enables the analyst to judge the vulnerability of causal
conclusions to potential unobserved confounders.
We relied on the NELS eighth-grade sample for the propensity model and
sensitivity analyses because it is the only sample that began prior to high
school and thus best captures the effects of racial composition over the entire
high school period. We also focused exclusively on blacks. We narrowed our
focus to blacks for two reasons: First, our main concern was to assess the
robustness of estimates of the effects of racial composition for blacks. Second, selectivity factors that affect the racial composition of schools attended
by blacks, whites, and others are likely to be very different, so using the same
propensity model for all groups would be inappropriate. We used a logistic
regression model to predict the likelihood that a black student would attend
a school that was 60% or more white versus a school with a smaller proportion white. A propensity model is ordinarily run with a dichotomous
rather than a linear treatment, and we chose the 60% white threshold to
preserve a reasonable number of treatment cases, although we obtained very
similar results with other thresholds e.g., 50% and 80% white.
Predictors used to identify propensities were drawn from the base year
survey and included sex, students locus of control, students self-concept,
students composite SES, and student composite score on the NELS eighthgrade math and reading tests; composite variables indicating participation at school in sports, music, subject clubs, academic activities such as
1136


debate clubs, the schools yearbook, or the student newspaper; and others. The model also included information about the students eighthgrade school, such as urbanicity, sector public or private, region Midwest, South, West, or Northeast, and school resources such as clubs and
other activities for students as reported by the school administrator, as well
as interactions of test scores by region and sector. The initial sample of
eighth-grade black students had 950 cases, of which over 90% 880 students could be matched across treatment conditions. Such matching within
a region of common support is a strength of the propensity approach, in
that it ensures that treatment effects reect cases that are otherwise similar. However, the results pertain only to cases within the matching frame.
Since we found common support for over 90% of cases, our results pertain
to the large majority of black students, although those at the tails of the
distribution on their likelihood of attending schools with many or few whites
may not be represented. Treatment effects from the propensity model are
thus estimated from 880 black students, of whom 320 about 36% of the
cases studied in schools with at least 60% whites and 560 64% studied
in schools with less than 60% whites.
We used a kernel matching procedure to match treatment and control
cases and obtain estimates of the average treatment effect for the treated
ATT. In kernel matching, the average outcome for treated cases is compared to the weighted average outcome of control cases, where control cases
are weighted so that those with the closest propensities to a given treated
case receive the most weight and those farthest away receive the least weight
Heckman et al. 1998; Sianesi 2001; Morgan and Harding 2006. Kernel
matching is often chosen over other matching techniques because it uses all
available data and because the weighting process improves the quality of
the matches over other approaches Becker and Ichino 2002; Brand and
Thomas 2014. In simulations, Morgan and Harding 2006 found that kernel matching provided the least biased estimates when the propensity model
was well specied. As a further check on our results, we also estimated our
models using nearest neighbor matching with replacement, in which each
treatment case is matched to the control case with the most similar propensity drawing from the full pool of control cases for each match. Nearest
neighbor matching yielded ATT estimates very similar to those of kernel
matching and no differences in substantive conclusions.
Propensity models rely on a balanced distribution of observed covariates
across treatment and control conditions to remove bias from estimated
outcome differences Morgan and Harding 2006. In building the propensity model, we checked to ensure that the average propensity to enroll
in a school that was at least 60% white did not differ for treatment and
control cases, that is, among students who actually did or did not attend
such schools, and that the mean of each predictor variable in the propensity
1137


model was not signicantly different across the treatment and control groups.
After building the propensity model, we performed three balancing tests to
ascertain the quality of the matches Stuart 2010; Austin 2011; Lee 2013.
First, we examined the standardized mean differences between treatment
cases and their kernel-matched controls.9 We found that the standardized
mean differences for 15 of the 22 covariates were less than 0.05, 21 of 22
were less than 0.10, and all 22 were less than 0.15 as contrasted with nine
standardized mean differences over 0.15 in the unmatched sample. This indicated very small remaining bias as far as can be determined from observed data and a large reduction in bias as a result of the matching procedure. Second, we examined variance ratios i.e., the ratios of control to
treatment variances, where a ratio of close to 1 would indicate that the
covariates have similar distributions. We found 21 of 22 variance ratios
between 0.75 and 1.25 including 13 between 0.90 and 1.10 and all 22
between 0.75 and 1.50, indicating similar variances.10 Third, we performed
Hotellings test for joint signicance of mean differences between treatment
and matched control covariates and found that the means did not differ. On
the basis of these assessments, we concluded that treatment and control
cases were reasonably well matched on observed covariates, and we proceeded to estimate ATTs and sensitivity bounds.
RESULTS
First, we report results from our comparison of the HSB and NELS cohorts.
These analyses begin with samples that represent the national populations
of tenth graders in 1980 and 1990 and followed up in 1992 and 2000, respectively. Next, we consider the policy context and, with HSB, examine the
role of internal resegregation. Then, we revisit NELS by focusing on the base
year 1988 sample of eighth graders captured before high school entry, again
relying on the 2000 follow-up for outcome data. Finally, we report the results
of the propensity score analysis and bounds procedure.
Cross-Cohort Comparisons
Table 2 displays the labor market consequences of high school proportion
white, with comparisons of the HSB class of 1982 and NELS tenth-grade
class of 1992 cohorts. Note rst that the ndings are remarkably stable
across the two cohorts. Few of the coefcients are signicantly different
from one another across cohorts. This not only suggests that social processes
9
The standardized mean difference the difference between the means divided by the
square root of half the sum of the squared SDs is commonly used to assess bias reduction
in matching models Harding 2003; Lee 2013; Brand and Thomas 2014.
10
Stuart 2010 suggests that variance ratios should fall between 0.50 and 2.0 and
standardized mean differences should be less than 0.25.
1138


were stable over time but also lends condence to the reliability of our
results, effectively providing a consistent replication. Whether school
desegregation policies were waxing or waning, the effects of school composition were essentially constant.
Second, the effects of the school- and student-level control variables conform to well-established patterns: students in private schools, those with
higher test scores, and those from higher-SES backgrounds are typically
privileged, while males are disadvantaged in occupational status but advantaged in employment and earnings. Third, the main effects for racial
gaps show some signs of change over time, but the differences across cohorts are generally nonsignicant.
Turning to our main interest, the proportion of white students in a high
school is unrelated for both whites and blacks to occupational status, employment status, or income, in both cohorts. Interestingly, for students of
other ethnic backgrounds, high school proportion white is negatively related to occupational status in both cohorts, suggesting a possible ethnic
enclave effect: young persons in environments with fewer whites may have
better chances for occupational advancement in same-ethnic niches. The
combination of signicant effects for occupational status and the absence
of effects for earnings is particularly indicative of an ethnic enclave effect,
as earnings may be lower for a given occupation within an ethnic enclave
compared to the same occupation in the wider society.
Only in the case of coworker homogeneity in the workplace does the
proportion of white students in the high school consistently have long-term
consequences for whites and blacks as well as for others. Overall, whites
tend to report higher proportions of coworkers of the same race than minority students; the coefcients for blacks and others are negative, indicating that fewer of their coworkers are same-race as compared to whites, as
one might expect given their lower proportions in the population. In these
equations, the intercepts of .76 HSB and .69 NELS indicate that the
average white student was employed in workplaces that were 76% and
69% white, respectively, for the two cohorts. The negative coefcients for
blacks and others should be interpreted in combination with these intercepts; for example, the coefcient of 2.39 for blacks in HSB indicates that
the average black student was employed in a workplace that was 36.9%
black .76 2 .39 5 .37.
The effects of high school proportion white go in opposite and therefore
complementary directions for whites versus blacks and others. Whites who
attend high schools with higher proportions of whites tend to have more
coworkers of the same race HSB 5 .19, NELS10 5 .18, both signicant,
cross-cohort difference not signicant, while blacks who attend high schools
with higher proportions of whites tended to have fewer coworkers of the
same race HSB 5 2.31, NELS10 5 2.39, both signicant, cross-cohort
1139

1140

Rural . . . . . . . . . . . . . . . . . . .
School-level controls:
Private . . . . . . . . . . . . . . . . . .
Other . . . . . . . . . . . . . . . . . .
Black . . . . . . . . . . . . . . . . . .
Slopes:
School proportion white:
Reference group white . . . .
Other . . . . . . . . . . . . . . . . . . .
Black . . . . . . . . . . . . . . . . . . .
Intercepts:
Reference group white . . . . . .
2.33**a
.59
.06
.47
2.63
1.22
.41
1.84
23.89**
1.48
35.07**a
.25
1.77**
.62
1.00*
.42
HSB
.12a
.70
2.75
.52
2.18
.96
.04
2.09
24.24*
1.80
39.61**a
.25
.29
.80
.17
.63
NELS10
OCCUPATIONAL STATUS
Linear
.04
.04
2.04
.04
.09
.11
2.26a
.17
2.11
.14
9.92**a
.02
2.05
.06
2.03
.04
HSB
.04
.05
2.03
.04
.00
.06
.10a
.13
.11
.13
9.99**a
.02
2.07
.04
2.05
.04
NELS10
LOG OF INCOME
Linear
.00a
.16
.12
.15
.22
.38
2.29
.55
2.25
.45
2.20**a
.08
.13
.19
.03
.13
HSB
2.45**a
.16
.05
.13
.33
.23
.35
.29
.04
.37
3.05**a
.06
2.30**
.08
2.11
.12
NELS10
EMPLOYED > 20
WEEKS Binary
TABLE 2
Labor Market Consequences of School Proportion White in HSB and NELS 10
2.02
.01
.05**
.01
.19**
.03
2.31**
.05
.02a
.04
.75**a
.01
2.39**
.02
2.21**a
.01
HSB
.01
.02
.06**
.01
.18**
.03
2.39**
.05
2.29**a
.05
.69**a
.01
2.35**
.02
2.32**a
.02
NELS10
PROPORTION
COWORKERS SAME
RACE Linear
1141

21.53**a
.24
2.29**
.21
.42**a
.02
870
9,300
.38
.44
2.87
.45
2.99*
.45
21.97**
.51
1.65**
.48
23.69**a
.33
2.59**
.30
.34**a
.02
1,070
10,080
2.32
.51
2.38
.57
2.54
.54
21.53*
.61
.89
.53
.43**
.02
.14**
.02
.01**
.00
870
8,620
.07
.04
2.10**
.03
2.09*
.04
2.12**
.04
.02
.04
.37**
.02
.06**
.02
.005**
.00
1,070
9,200
.05
.04
2.03
.03
2.06
.03
2.13**
.04
2.03
.04
1.27**a
.09
.20**a
.07
.01*
.01
870
9,020
.07
.13
2.04
.14
2.02
.14
2.27a
.16
.11
.15
LOG OF INCOME
Linear
.67**a
.08
2.14**a
.05
.01*
.00
1,070
9,360
.06
.13
.18
.14
.02
.12
.29*a
.13
2.01
.10
.00
.01
.00a
.00
.002**a
.00
870
9,300
.00
.01
.02
.01
2.02
.01
2.06**
.01
.00
.01
EMPLOYED > 20
WEEKS Binary
2.01
.01
2.01a
.01
.00a
.00
1,070
10,120
.01
.01
.04**
.01
2.01
.01
2.04**
.02
.01
.01
NOTE.In accordance with NCES security procedures for restricted data, unweighted sample sizes have been rounded to the nearest 10. Nos. in
parentheses are SEs.
a
The difference between NELS and HSB is signicant at P < .05.
* P < .05.
** P < .01.
N schools . . . . . . . . . . . . . . . . . .
N students . . . . . . . . . . . . . . . . .
Student test . . . . . . . . . . . . . . .
Student SES . . . . . . . . . . . . . .
Student-level controls:
Male . . . . . . . . . . . . . . . . . . . .
School SES . . . . . . . . . . . . . . .
West . . . . . . . . . . . . . . . . . . . .
South . . . . . . . . . . . . . . . . . . .
Midwest . . . . . . . . . . . . . . . . .
Suburban . . . . . . . . . . . . . . . .
OCCUPATIONAL STATUS
Linear
SAME
RACE
Linear
KERS
COWOR-
TION
PROPOR-

difference not signicant. For other minority students, the coefcient in
HSB .02 indicates no difference compared to whites, but in NELS the coefcient is negative 2.29 like that for blacks the cross-cohort difference is
signicant. For whites and blacks, high school racial composition is consistently tied to workplace racial composition: whites who attend more segregated white high schools tend to work in more segregated white
workplaces, while blacks who have more white peers in high school tend to
have more nonblack coworkers. By the year 2000, the same pattern is evident
for other minority students as well. In the lone area of school and workplace
integration, we nd support for perpetuation theory.11
To explore the substantive meaning as opposed to statistical signicance
of these results, we conducted simulations by setting all values other than
race/ethnicity and school racial composition at their sample means. In HSB,
for example, coming from a school composed of 20% white students, a white
respondent would tend to work with 65% white coworkers, while an otherwise similar black respondent would work with 43% black coworkers. In
a school composed of 50% white students, 71% of a white respondents
coworkers would tend to be white, while 40% of a black respondents
coworkers would be black. For a school that is 80% white, 77% of white
respondents coworkers would be white, while 36% of black respondents
coworkers would tend to be black.12 The patterns for NELS10 are quite
similar. If a desegregation policy moved a school from 20% to 50% white,
our results would lead us to expect blacks to nd jobs in workplaces that
were about 3 percentage points less black. A black student who transferred
from a school that was 20% white to a school that was 80% white would,
according to our ndings, end up in a workplace that is seven percentage
points less black. Substantively, these appear to be modest effects.
Does Variation across Contexts Explain the Absence
of Racial Composition Effects?
Why do we nd virtually no long-term effects of school racial composition,
except in the case of workplace composition? One possibility is that at11
In early versions of this work, we also examined postsecondary attainment outcomes. As
with earnings, employment, and occupational status, we found no benets of attending a
high school with a higher percentage of white students for postsecondary attainment.
12
These values are computed as follows note that values for proportion white subtract
the mean of .72 because the data are centered: whites in schools .20 white 5 .755 1
.1932.52 5 .65; blacks in schools .20 white 5 .755 1 .1932.52 2 .386 1
2.3132.52 5 .43; whites in schools .50 white 5 .755 1 .1932.22 5 .71; blacks in
schools .50 white 5 .755 1 .1932.22 2 .386 1 2.3132.22 5 .40; whites in schools
.80 white 5 .755 1 .193.08 5 .77; blacks in schools .80 white 5 .755 1 .193.08 2
.386 1 2.313.08 5 .36.
1142


tending school with high proportions of white peers does not have the
benets attributed to it by perpetuation theory, perhaps because of countervailing social processes also linked to attending such schools. Before
accepting this explanation, however, we may also consider the possibility
that variation in the context of school racial composition has obscured its
effects. Although we failed to nd differences over time, effects of composition may vary across schools within cohorts. For example, minority students who attend schools with high proportions of white students may have
different experiences when the student population is created through busing or court orders than when it derives from a racially mixed neighborhood. Another possibility is that internal resegregation may forestall the
realization of potential benets of racially mixed schools. Resegregation has
been given as an explanation for why desegregation has limited effects on
short-term outcomes e.g., Mickelson 2001; Clotfelter 2002; Wells et al.
2004, so the same point may hold for long-term outcomes.
The context of high school racial composition.To assess the impact of
contextual variation within cohorts, we explored interactions for high school
racial composition with whether or not schools used busing and were under
court orders in HSB and NELS10. These analyses not shown yielded no
evidence that compositional effects were related to the presence or absence
of integration policies. Our ndings thus replicate Rivkins 2000 results
with the full sample of NELS schools and with HSB. The only consistent
signicant effects were for nonblack minority students, for whom busing
was positively related to employment, but only if the school did not have a
high proportion of white students. Nonblack minority students who were
bused to schools with high proportions of white students received no benet
to their employment prospects.
The absence of effects for busing also provides evidence against an
alternative explanation for our positive ndings on workplace integration.
To the extent that schools draw their students from a neighborhood catchment area, it can be difcult to disentangle the effects of school composition
from those of neighborhood composition. The nding that school composition effects on same-race coworkers are present whether the school composition reects busing or neighborhood assignments suggests that school
composition, rather than neighborhood composition, is the operating mechanism see Stearns 2010 for further evidence supporting this interpretation.
Internal resegregation.To determine whether internal resegregation
has obscured the impact of racial composition, we reestimated our HSB
models with controls for segregation across tracks and extracurricular
activities. We also included interaction terms for the internal resegregation
indices by proportion white. We focused on the 2,660 students in the 240
schools in HSB that sampled both white and black students see app. ta-
1143


ble A1. If resegregation prevented black students from beneting from
racially mixed schools, then school proportion white should exert positive
effects on outcomes for blacks when levels of internal segregation are low.
Our analyses revealed no such results: none of the coefcients for blacks
and resegregation is statistically signicant, and the coefcients for proportion white are unaffected compared to our main analyses. To illustrate these ndings, appendix table A3 reports selected coefcients from
our basic model table 2 with the addition of track segregation and the
interaction of track segregation by proportion white. The panel of effects for
school proportion white closely replicates the HSB coefcients in table 2,
and the track segregation effects are insignicant. We found similar results
for the two measures of extracurricular segregation the black-white gap in
participation and black-white dissimilarity in extracurricular activities.
Thus, we nd no evidence that resegregation accounts either for the lack of
composition effects on attainment or earnings or for the presence of high
school composition effects on workplace composition.
So far we have ruled out two possible explanations for the near absence
of high school composition effects on outcomes other than workplace
composition. However, busing and internal resegregation are not the only
aspects of the racial context of schools. NELS, unlike HSB, offers direct
survey measures of school racial climate, which may better allow us to
understand the results we have obtained. Moreover, because it begins with
eighth graders, NELS captures effects that may occur throughout the high
school period. Suppose that high school racial composition affects job
prospects, but those effects occur in the rst two years of high school and
are perhaps reected in higher test scores or in not dropping out before
grade 10. Effects such as these would be hidden in data that began with
tenth graders. Also, owing to oversampling of Asian Americans and the
population growth of Hispanics, NELS offers a means to disaggregate
students of other ethnicities into separate groups. For these reasons, we
next turn to NELS to examine contextual conditions more thoroughly.
A Closer Look at NELS
We use the NELS eighth-grade sample to consider additional possible
moderating conditions, to disaggregate nonblack minority ethnic groups,
to explore nonlinear indicators of racial composition, and to apply propensity models that may better adjust for selectivity bias.
Moderating conditions.Table 3 presents results for conditions that
potentially exacerbate or lessen the effects of racial composition on labor
market outcomes: whether busing was used to achieve racial balance in the
school, whether racial conict was a moderate to serious problem according
1144


to the principal, and whether students reported positive race relations.13
First, note that the baseline results model 1 are the same as reported in the
NELS tenth-grade sample compare to table 2.14 Second, as in HSB and
NELS10, we nd no evidence that busing alters the pattern of racial composition effects although busing boosts the employment of nonblack minority students if they attend schools with low proportions of whites. Third,
for all outcomes other than percentage of same-race coworkers, the inclusion of indicators of racial conict also fails to change our conclusion that
the percentage of white students in a high school makes little difference for
students in the long term.
Percentage of same-race coworkers again presents the exception to the
general pattern. While none of the moderating conditions exerts signicant
effects on its own and busing and the quality of race relations as perceived
by students are similarly inconsequential for composition effects, the ndings for racial conict as reported by the principal are different. For students other than whites and blacks, the interaction of school proportion
white by racial conict is positive and signicant, indicating that moderate
or serious racial conict reduces the benets of attending schools with high
proportions of white students for nding employment with persons of other
ethnicities. Thus, the analysis of racial conict reveals a condition under
which the one consistent benet of mixed-race schoolspromoting workplace integrationmay not hold. Even considering this nding, it seems
clear that contextual conditions such as busing and racial conict have not
obscured the potential benets of high school racial composition for longterm status outcomes.
Disaggregation of ethnic groups.Table 4 repeats the results for model 1
for the continuous outcomes only, but with ethnic groups disaggregated
into black, Asian American, Hispanic, and Native American, all compared
to non-Hispanic whites. These ndings reveal differences compared to table 3 in some of the main effects of ethnicity intercepts, but the interactions with high school proportion white slopes are nearly all consistent
The indicator of whether students were assigned to the school to achieve a desired
racial composition was highly correlated with the busing indicator and yielded identical results.
14
As a check on our approach to missing data, we also ran this analysis using multiple
model-based imputation and obtained near-identical results. For example, the effects of
racial composition on the percentage of same-race coworkers were .18, 2.42, and 2.24 all
signicant for whites, blacks, and others, respectively, in the multiply imputed sample n 5
10,530 students, 1,240 schools, compared to .19, 2.45, and 2.24 in table 3, model 1 n 5
10,060 students, 1,070 schools. For nonblack other minority students the effect of school
racial composition on occupational status was a statistically signicant 26.58 in the multiply imputed sample compared to 25.62 in table 3, model 1. All other effects of school racial
composition were insignicant in both approaches to missing data.
13
1145

1146

Other . . . . . . . . . . . . . . . . . . . .
Black . . . . . . . . . . . . . . . . . . . .
Bused:
Other . . . . . . . . . . . . . . . . . . . .
Black . . . . . . . . . . . . . . . . . . . .
Slopes:
Race relations . . . . . . . . . . . . . . .
Other . . . . . . . . . . . . . . . . . . . . .
Black . . . . . . . . . . . . . . . . . . . . .
Intercepts:
Reference group white . . . . . . . .
2.51
1.05
2.02
2.12
25.62*
2.25
39.34**
.30
.19
.94
.01
.80
.35
2.91
.44
5.16
21.14
4.00
2.50
1.09
1.34
2.15
25.41*
2.29
39.34**
.29
.62
.87
2.03
.80
Model
2
.20
2.90
1.54
4.35
21.00
4.01
24.24
6.73
7.64
13.19
213.30
13.27
.09
.07
.01
.13
2.06
.14
2.05
.18
.10
.30
.58
.31
.09
.08
.02
.14
2.03
.16
Model
1
2.04
.18
2.04
.30
.54
.31
2.11
.40
1.28
.97
21.36
.73
.61
.31
2.34
.36
2.32
.40
2.09
.56
2.48
.75
3.05**
.68
.63*
.32
2.43
.37
2.29
.42
3.13**
.07
2.51**
.09
2.18
.13
Model
2
Model
1
Model
2
Model
3
PROPORTION COWORKERS
SAME RACE Linear
2.12
.55
2.56
.75
3.18**
.66
2.64
1.69
2.69
2.08
21.08
2.22
.08
.05
2.12
.12
2.04
.12
.08
.05
2.18
.12
2.04
.11
.19**
.19**
.33*
.03
.03
.14
2.45** 2.46**
.21
.06
.06
.37
2.24** 2.23** 2.85**
.05
.05
.27
3.13**
.69**
.69**
.69*
.07
.01
.01
.01
2.41** 2.35** 2.34** 2.35**
.02
.09
.02
.02
2.14
2.32** 2.32** 2.32**
.12
.02
.01
.01
2.06
2.01
.06
.01
Model
3
EMPLOYED > 20 WEEKS

Binary
9.97** 3.13**
.02
.07
2.04
2.57**
.05
.09
2.03
2.29
.04
.13
2.04
.02
Model
3
INCOME Linear
39.37** 9.97** 9.97**

.29
.02
.02
.53
2.07
2.07
.92
.05
.05
.10
2.05
2.04
.71
.04
.04
2.27
.28
Model
3
Model
1
Model
2
Model
1
OF
LOG
OCCUPATIONAL STATUS Linear
TABLE 3
Busing and Racial Conict as Potential Moderating Conditions for Effects of School Proportion White
NELS Eighth-Grade Sample
1147

21.02
4.03
212.92
9.64
3.34
7.02
.06
.11
2.36
.27
.37
.21
1,060
9,140
1.04
1.80
22.01
3.77
2.21
3.51
1,070
10,020
.09
.27
2.26
.43
2.75
.51
2.06
.09
.37*
.18
2.15
.14
.12
.26
2.52
.41
2.86
.52
21.49
1.47
1.72
2.53
2.22
2.05
21.10
4.03
214.69
8.39
3.63
7.23
.34
.47
.12
.58
.24
.61
1,070
9,300
2.34
.37
.33
.44
.41
.46
.92
.91
.88
.85
2.57
2.28
1.12
1.11
22.85** 22.82**
.96
.94
2.07
.07
.14
.20
.03
.21
2.04
.04
2.19
.11
.17*
.08
1,070
10,060
.03
.03
.13
.07
2.03
.05
2.08
.07
.21
.20
.09
.20
NOTE.Models include school- and student-level controls as in table 2. In accordance with NCES security procedures for restricted data, unweighted sample
sizes have been rounded to the nearest 10. Nos. in parentheses are SEs.
* P < .05.
** P < .01.
N schools . . . . . . . . . . . . . . . . . . . .
N students . . . . . . . . . . . . . . . . . . .
Other . . . . . . . . . . . . . . . . . . . .
Black . . . . . . . . . . . . . . . . . . . .
School proportion white

racial conict:
Other . . . . . . . . . . . . . . . . . . . .
Black . . . . . . . . . . . . . . . . . . . .
Racial conict:
Other . . . . . . . . . . . . . . . . . . . .
Black . . . . . . . . . . . . . . . . . . . .
School proportion white bused:

1148

Native American . . . . . . . . . . . . . . .
Latino . . . . . . . . . . . . . . . . . . . . . . .
Asian . . . . . . . . . . . . . . . . . . . . . . . .
Black . . . . . . . . . . . . . . . . . . . . . . . .
Slopes:
Reference group white . . . . . . . . . .
Racial relations . . . . . . . . . . . . . . . . . .
Native American . . . . . . . . . . . . . . . . .
Latino . . . . . . . . . . . . . . . . . . . . . . . . .
Asian . . . . . . . . . . . . . . . . . . . . . . . . .
Black . . . . . . . . . . . . . . . . . . . . . . . . .
Intercepts:
Reference group white . . . . . . . . . . . .
2.37
1.14
2.02
2.19
24.48
2.90
26.90*
2.94
.47
3.63
39.29**
.31
.29
.95
.82
.88
2.74
1.16
1.38
1.61
Model 1
2.24
1.20
1.38
2.23
23.70
3.00
26.86*
2.97
2.54
3.67
39.29**
.31
.67
.87
1.18
.86
2.73
1.13
2.57
1.89
Model 2
24.56
7.03
9.56
13.40
2.87
13.96
222.61
15.89
12.65
24.13
39.33**
.30
.59
.97
1.09
.86
2.37
.91
3.14*
1.58
2.27
.27
Model 3
OCCUPATIONAL STATUS
.00
.07
.09
.13
.56*
.28
2.18
.15
.58
.32
9.99**
.02
2.10
.05
2.08
.06
2.08
.06
2.25**
.09
Model 1
OF
INCOME
.01
.08
.09
.14
.64*
.31
2.17
.15
.56
.32
9.99**
.02
2.09
.05
2.06
.06
2.08
.06
2.07
.08
Model 2
LOG
2.01
.41
.98
.96
2.29
1.20
21.99*
.80
21.98
1.96
9.99**
.02
2.07
.05
2.06
.06
2.06
.05
2.01
.08
2.04
.02
Model 3
.18**
.03
2.45**
.06
2.11
.08
2.21**
.06
2.43**
.11
.69**
.01
2.35**
.02
2.41**
.02
2.27**
.02
2.31**
.05
Model 1
.19**
.03
2.46**
.07
2.12
.09
2.20**
.06
2.48**
.11
.69**
.01
2.34**
.02
2.41**
.02
2.27**
.02
2.28**
.05
Model 2
.24
.13
.31
.38
2.31
.43
2.88**
.30
1.49
.87
.69**
.01
2.35**
.02
2.40**
.02
2.27**
.02
2.28**
.05
2.01
.01
Model 3
PROPORTION OF COWORKERS
SAME RACE
TABLE 4
Labor Market Consequences of School Proportion White, Disagreggated by Ethnic Group
1149

Native American . . . . . . . . . . . . . . .
Latino . . . . . . . . . . . . . . . . . . . . . . .
Asian . . . . . . . . . . . . . . . . . . . . . . . .
Black . . . . . . . . . . . . . . . . . . . . . . . .
Racial conict:
Native American . . . . . . . . . . . . . . .
Latino . . . . . . . . . . . . . . . . . . . . . . .
Asian . . . . . . . . . . . . . . . . . . . . . . . .
Black . . . . . . . . . . . . . . . . . . . . . . . .
School proportion white bused:

Native American . . . . . . . . . . . . . . .
Latino . . . . . . . . . . . . . . . . . . . . . . .
Asian . . . . . . . . . . . . . . . . . . . . . . . .
Black . . . . . . . . . . . . . . . . . . . . . . . .
Bused:
21.77
4.21
214.89
8.32
217.30
13.84
11.23
7.55
50.02*
21.40
21.45
4.21
213.24
9.32
219.07
14.52
9.88
7.44
61.51*
30.23
21.59
1.55
1.79
2.53
.60
2.59
2.49
2.19
25.93
4.85
.63
3.01
1.83
4.17
12.20
8.73
24.27
4.09
217.54**
5.80
.60
3.02
.97
4.81
13.83
9.16
24.12
4.05
218.83**
7.25
OCCUPATIONAL STATUS
.15
.26
2.41
.43
21.07
.67
2.87
.68
9.77**
.81
2.08
.17
.08
.32
.88
.54
.51
.33
23.02**
.38
OF
INCOME
2.02
.09
.27
.18
2.15
.28
2.21
.12
2.46
.35
.13
.26
2.23
.45
21.02
.66
2.60
.66
11.87**
1.39
2.07
.17
2.03
.33
.86
.53
.39
.33
23.83**
.50
LOG
2.09
.07
.17
.19
.20
.39
.03
.21
1.83**
.57
.09
.05
2.14
.11
2.11
.21
2.07
.12
2.74**
.14
.01
.03
.15*
.07
.04
.09
2.02
.06
.36**
.14
2.10
.07
.21
.19
.15
.37
.11
.21
2.06**
.59
.09
.05
2.19
.11
2.03
.20
2.09
.12
2.77**
.11
PROPORTION OF
COWORKERS
SAME RACE

Model 2
1.20
1.89
22.53
3.89
21.85
4.01
4.49
4.18
23.83
6.49
1,070
10,020
Model 3
Model 1
OF
INCOME
Model 2
LOG
.01
.11
2.26
.27
.26
.37
.51*
.21
.68
.55
1,060
9,140
Model 3
Model 1
Model 2
2.01
.04
2.22*
.11
.05
.12
.19*
.09
2.53*
.23
1,070
10,060
Model 3
PROPORTION OF COWORKERS
SAME RACE
NOTE.Models include school- and student-level control variables as in table 2. In accordance with NCES security procedures for restricted data,
unweighted sample sizes have been rounded to the nearest 10. Nos. in parentheses are SEs.
* P < .05.
** P < .01.
N schools . . . . . . . . . . . . . . . . . . . . . . . .
N students . . . . . . . . . . . . . . . . . . . . . . .
Native American . . . . . . . . . . . . . . .
Latino . . . . . . . . . . . . . . . . . . . . . . .
Asian . . . . . . . . . . . . . . . . . . . . . . . .
Black . . . . . . . . . . . . . . . . . . . . . . . .
School proportion white racial conict:

Model 1
OCCUPATIONAL STATUS
TABLE 4 (Continued )

with the earlier results. In the prediction of income, Native Americans are
disadvantaged compared to whites, while Asian Americans and Hispanics
are not net of background controls. Among the slopes, the only meaningful
difference from those in table 3 is that Asians and Native Americans in
schools with higher proportions of white students are likely to obtain higher
incomes though the coefcients are signicant only for Asians. Otherwise,
the disaggregation of ethnic groups reinforces initial conclusions about composition. The nding in table 4 that the negative interaction between proportion white and occupational status holds for the two immigrant groups
Hispanics and Asian Americans but not for blacks or Native Americans
supports the earlier ethnic enclave interpretation. In addition, the coefcients
for same-race coworkers are negative for all groups and signicant for all but
Asian Americans, indicating that members of all minority groups who attend
high school with more white peers are likely to work in environments with
lower concentrations of persons of their own ethnic backgrounds. The
inclusion of busing again fails to overturn the general conclusions, although
busing works to the disadvantage of Native Americans, unless they are bused
to schools with high proportions of white students, in which case the disadvantage is erased. These coefcients are exceptionally large, and in light of the
small numbers of Native Americans included in NELS, the coefcients
should be interpreted with caution.
The disaggregation of ethnic groups sheds more light on the role of racial
conict as reported by the principal. Once again, signicant effects are
largely restricted to the nal outcome, percentage of same-race coworkers. In
the disaggregated analysis, it appears that moderate or serious racial conict
in a school diminishes the benets for workplace integration of attending
school with a high proportion of white students for Hispanics and, to a lesser
and nonsignicant degree, for Asian Americans. For blacks and Native
Americans, the results are more complicated. Racial conict has signicant
positive main effects recall that a positive coefcient means that students
are more likely to end up in work environments with greater concentrations
of their own ethnic groups. However, the interaction of school proportion
white by racial conict is negative for both groups. Taken together, these
results indicate that when blacks and Native Americans attend schools that
are high in proportion white, they have fewer same-race coworkers, even
when racial conict is present. Only when they attend schools that have racial conict but relatively few white students are their workplace integration
prospects diminished. Overall, then, racial conict reduces the benets of
attending school with high proportions of white students for workplace integration for Hispanics and possibly for Asian Americans, but not for
blacks or Native Americans.
Categorical models of racial composition.To allow for possible nonlinear effects, we reestimated the basic model using a categorical version of
1151


the variable school proportion white with ve categories: 0%20% white,
21%40% white, 41%60% white, 61%80% white, and 81%100%
white. The omitted category is 41%60% white, that is, the integrated
school. The results appear in appendix table A4. As in the previous analysis,
we found signicant effects of school proportion white on workplace racial
composition but not on most other outcomes. One exception is an interesting
nonlinearity in the employment analysis: blacks in schools with very few
whites 0%20% and very many whites 80%100% were employed more
than those in other schools. The linear model obscured these divergent ndings. Otherwise, the ndings mirror the linear results.
Propensity models.The previous models have persistently shown signicant results of school proportion white for the proportion of same-race
coworkers and virtually no effects for the other outcomes. However, either
the presence of effects or their absence might still be due to selection bias
originated from observed characteristics of the students that inuence their
tendency to study in a school with a certain racial composition. To further
adjust for this potential bias, we estimated a model with only black students, comparing the effects among students with the same propensity to
attend a school with a given proportion white. The results of the propensity
models are shown in table 5. The treatment variable is the propensity
to attend a high school that is 60% white or more. The ATT between
matched cases in the model was nonsignicant for every dependent variable except proportion of coworkers of the same race, which corroborates
the previous models described in tables 2 4. These results strengthen the
TABLE 5
Average Treatment on the Treated Effects from Propensity Score Kernel
Matching Models for Schools at Least 60% White
Blacks Only, NELS Eighth-Grade Sample
DEPENDENT VARIABLE
Occupational status . . . . . . . . . . . . .
Employed > 20 weeks . . . . . . . . . . .
Log of income . . . . . . . . . . . . . . . . .
Proportion coworkers same race . . . .
AVERAGE
TREATMENT EFFECT
.55
1.14
2.01
.01
2.08
.07
2.10**
.02
NUMBER
OF
CASES
Treated
Control
320
560
300
530
280
490
320
560
NOTE.Data are not weighted. Number of cases varies because of missing data on the dependent variables; analysis with a common case base yields identical results. In accordance with
NCES security procedures for restricted data, unweighted sample sizes have been rounded to
the nearest 10. Nos. in parentheses are SEs.
* P < .05.
** P < .01.
1152


conclusion that high school racial composition is linked to the racial
compositions of pathways students follow after high school. The size of the
treatment effect in the propensity model is smaller than the coefcient from
a multilevel model with a categorical predictor 2.100 vs. 2.134, suggesting that our original estimates may have been inated by selectivity,
but the signicant nding holds.
Sensitivity analysis.The Rosenbaum 2002 bounds procedure evaluates the robustness of a causal inference in the presence of an unobserved
predictor of selection into the treatment. We applied the procedure to help
judge whether the effect of school racial composition on percentage of
same-race coworkers warrants a causal interpretation. We considered a
hypothetical unobserved confounder with effects on enrolling in a highpercent-white school that ranged in our simulation from an odds ratio of 1
no change in enrollment pattern to an odds ratio of 2 a 100% increase in
the chances of enrolling in a high-percent-white school. The simulation
provides upper- and lower-bound point estimates and condence intervals
for the ATT at each level of unobserved confounder. When the condence
intervals expand far enough to include 0, the hypothetical confounder would
render the point estimate insignicant. The simulation results, displayed in
table 6, reveal that an unobserved predictor of school racial composition
would need to have an odds ratio of 1.8that is, it would have to increase
the likelihood of enrolling in a high-proportion-white school by 80%to
reduce the effect of proportion white on workplace racial composition to
insignicance. This would be a more powerful predictor of school composition than all but three of the 22 variables in our selection model private
TABLE 6
Sensitivity Analysis for Proportion of Same-Race Coworkers
ODDS RATIO FOR

UNOBSERVED CONFOUNDER
1 .
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
2 .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
POINT ESTIMATE
CONFIDENCE INTERVAL
Upper
Lower
Upper
Lower
2.12
2.13
2.15
2.16
2.17
2.18
2.18
2.19
2.20
2.20
2.21
2.12
2.11
2.11
2.09
2.08
2.07
2.06
2.05
2.04
2.03
2.02
2.16
2.17
2.18
2.19
2.20
2.21
2.21
2.22
2.23
2.24
2.25
2.09
2.07
2.06
2.05
2.04
2.03
2.02
2.01
.00
.01
.02
NOTE.Bounds are estimated on the basis of 320 matched pairs. In accordance with NCES
security procedures for restricted data, unweighted sample sizes have been rounded to the
nearest 10.
1153


school sector, urbanicity, and school SES; see app. table A2. Moreover, the
unobserved confounder would need to be as strong a predictor of attending a
high-proportion-white high school as a move in eighth grade from a middleSES to a high-SES school, or from a low-SES to a middle-SES school, to
change the conclusion about effects on workplace composition. In light of
the rich set of predictors in the propensity model, as well as the long list of
controls that are balanced for the outcome analysis, the presence of such a
powerful unobserved confounder seems unlikely.
DISCUSSION AND CONCLUSIONS
Our results provide limited support for Braddocks 1980, 2009 perpetuation theory. We nd that African-American and other minority students
who enrolled in high schools with higher proportions of white students
tend to work in environments with more persons from dissimilar racial
backgrounds. At the same time, white students who attended schools with
lower proportions of whites found themselves in less white-dominated
workplacesa perpetuation of interracial contact for whites as well. The
results were largely unaffected by the context in which high school racial
compositions occurred. They were stable over time and impervious to
busing and internal resegregation. The presence of racial conict did not
change the overall outcomes for blacks, although Hispanic students who
attended schools characterized by moderate to serious levels of racial
conict did not obtain the benets for workplace integration that otherwise ensued.
While the effects for workplace integration were statistically signicant
and robust to model specication, they appeared modest in substantive
terms. For example, a policy that increased the proportion of whites in a
school from 20% to 50% would reduce the proportion of same-race coworkers for blacks by only about 3 percentage points. Moreover, blacks
occupational status, employment, and earnings were all unaffected by high
school racial composition, whether that composition was achieved by busing or other means.
Despite their modest size, the effects on workplace composition have
meaning in their own terms. They offer a direct test of perpetuation theory,
which in its simplest form states that racial isolation tends to perpetuate
itself across the life course Braddock 1980. Moreover, racially mixed work
environments are an indicator of social cohesion and, along with schools
and communities, are central to efforts to reduce racial exclusion. With that
said, our larger conclusion about high school racial composition must be tempered by the lack of impact across the range of social stratication outcomes.
Racially mixed high schools may contribute modestly to racially mixed workplaces, but they have little power to promote economic equality.
1154


Contrary to the ndings for black and white students, for others the
effect of school proportion white was signicant and negative for occupational status; that is, students of other ethnicities notably, Hispanics and
Asian Americans have better job prospects when they come out of more
minority-dominated schools. This suggests the existence of an ethnic enclave effect whereby children of immigrants nd support in a relatively
homogeneous community, increasing their occupational opportunities. It
is noteworthy that this effect is not reproduced for blacks. An explanation
may be that the black urban environment not only is poor but is a place
where work is rare and episodic and engages few residents, limiting the effects of solidarity bounds Wilson 1978, whereas immigrant communities
are rich in social capital and exchanges and have regulations and rhythm
given by labor markets Waldinger and Feliciano 2004.
The general absence of effects among African-Americans may not indicate that school composition is irrelevant for stratication outcomes, but
rather that the relationship is complex. Whereas racially mixed schools
may indeed promote interracial contact and more experience in navigating
integrated environments, they pose other challenges to the formation of
cultural dispositions toward a readiness to succeed in contexts in which
whites constitute a large fraction of the population. Opportunities for social networks that bridge across ethnic groupsan advantage of racially
mixed schoolsmay not be helpful if young persons hesitate to activate
their network ties so as to avoid acting whitea circumstance that is
apparently more prevalent in schools with higher percentages of white
students.
As the nation turns to solutions other than school desegregation to
address the profound racial and ethnic inequalities that still exist, it is
important to reect on what has been gained and lost through ve decades
of desegregation efforts. While desegregation likely contributed to the narrowing of the black-white achievement gap, the gap remained wide even at
the best of times i.e., the late 1980s, and the pace of improvement was
slow. Desegregation may have yielded long-term economic benets, but
these likely reected changes in resource allocation such as smaller classes
and increased per pupil expenditures rather than changes in racial composition. By contrast, the results of this study indicate that variation in
racial composition yielded little toward the reduction of black-white inequality in occupational attainment, employment, and earnings. Consequently, turning away from desegregation probably will not cause much
change in these social stratication outcomes. Desegregation also afforded
an opportunity to promote interracial contact in the school setting, contact
that, to some degree, carried over to the workplace. By giving up on desegregation, we have relinquished that policy tool. Past ndings are no
guarantee of future effects, but it seems likely on the basis of this study that
1155


the end of school desegregation means a modest increase in workplace
segregation.
It remains to be seen whether alternative strategies, such as the current
federal emphasis on accountability through high-stakes testing, will fare
any better at promoting equity goals in the school years and beyond. The
answer no doubt depends on whether or not accountability policies actually result in enhanced learning opportunities for African-American and
other minority students relative to whites. At the high school level, sociological research has consistently shown that learning opportunities are the
most powerful school-related inuences on student outcomes, both shortand long-term Gamoran 2001. In that sense, reformers are right to argue
that the quality of schooling matters more for a students future than the
composition of a students school. Two lessons of history, however, must
temper any optimism. The rst is that racial inequalities will not resolve
themselves on their own, that improvement is not inevitable, and that a
laissez-faire approach to inequality will fail to yield improvement
Gamoran 2001. As one policy tool desegregation is lost, therefore, the
reduction in inequality that reected that tool will continue only if another
can be found. Whether accountability is that tool is the question of the day.
The second lesson of history is that whereas learning opportunities matter more than composition, opportunities themselves are often inuenced
by composition. Whether a matter of more qualied teachers in schools
with higher proportions of white students or richer academic content in
classes with higher-achieving students, the compositions of schools and
classrooms have been salient factors in the unequal distribution of opportunities for learning. The link between composition and resources is an
important justication for school integration, even if direct composition
effects are minimal Johnson 2011. In principle, a strategy of accountability that relies on spotlighting inequality and directing resources in
response could reduce achievement gaps and, if the impact were powerful
enough, could yield the long-term effects on status attainment that were
absent in this study Gamoran 2007. If this aim is to be accomplished,
however, policy mechanisms will have to overcome long-standing patterns
in which more resources ow to schools and classrooms with the most
advantaged populations.
1156

APPENDIX
TABLE A1
Descriptive Statistics for Internal Resegregation HSB Only
Mean
SD
Student level:
Workplace % same race . . . . . . . . . . . . . . . . . . . . .
Student test scores . . . . . . . . . . . . . . . . . . . . . . . . .
Student SES . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Male . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Black . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Other . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
N students . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
School level:a
School proportion white . . . . . . . . . . . . . . . . . . . . .
Busing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Proportion white busing . . . . . . . . . . . . . . . . . . .
Court order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Proportion white court order . . . . . . . . . . . . . . . .
Academic track segment . . . . . . . . . . . . . . . . . . . . .
Proportion white academic track segment . . . . . .
Extracurricular activities segment . . . . . . . . . . . . . .
Proportion white extracurricular activities segment
Dissimilarity index . . . . . . . . . . . . . . . . . . . . . . . . .
Proportion white dissimilarity index . . . . . . . . . .
School SES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Private . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Midwest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
South . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
West . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Rural . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Suburban . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
N schools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Other dependent variables:
Occupational status N 5 2,660 . . . . . . . . . . . . . . .
Log of income N 5 2,470 . . . . . . . . . . . . . . . . . .
Employed > 20 weeks N 5 2,560 . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.65
49.54
2.07
.47
.23
.13
2,660
.32
8.91
.75
.50
.42
.34
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.70
.14
.08
.33
.20
.29
.21
.20
.14
.06
.04
2.03
.15
.20
.48
.08
.21
.50
240
.23
.34
.22
.47
.31
.24
.20
.16
.13
.06
.04
.45
.36
.40
.50
.28
.41
.50
........
........
........
35.67
9.89
.90
12.64
.89
.30
.
.
.
.
.
.
.
.
.
.
NOTE.Student-level statistics are weighted using FU4WT. In accordance with NCES

security procedures for restricted data, unweighed sample sizes have been rounded to the
nearest 10.
a
For cases with data on workplace % same race.
1157

TABLE A2
Results for the Propensity Score Model: OutcomeStudents School
in Grade 10 Was at Least 60% White
Intercept . . . . . . . . . . .
Male . . . . . . . . . . . . . .
Locus of control . . . . .
Self-concept . . . . . . . . .
Student SES . . . . . . . .
Student test . . . . . . . . .
Missing test . . . . . . . . .
Sports . . . . . . . . . . . . .
Art . . . . . . . . . . . . . . .
Academic . . . . . . . . . .
Other . . . . . . . . . . . . .
School resources . . . . .
Private . . . . . . . . . . . .
Rural . . . . . . . . . . . . .
Urban . . . . . . . . . . . . .
North . . . . . . . . . . . . .
Midwest . . . . . . . . . . .
West . . . . . . . . . . . . . .
School SES . . . . . . . . .
Student test North . .
Student test Midwest
Student test West . . .
Student test private .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Coefcient
SE
Odds Ratio
2.51
2.25
.03
2.35**
2.12
.021
.44
.52**
2.03
2.19
.12
2.041
21.50
2.03
21.53**
.42
.23
2.30
1.65**
.01
2.01
2.00
.03
.72
.17
.12
.11
.12
.01
.44
.19
.18
.18
.18
.02
1.78
.21
.22
1.25
1.35
1.99
.25
.03
.03
.04
.04
.60
.78
1.03
.71
.89
1.02
1.55
1.68
.97
.82
1.13
.96
.22
.97
.22
1.53
1.26
.74
5.23
1.01
.99
1.00
1.03
NOTE.N 5 890 students blacks only. School-level predictors refer to the students school
in eighth grade, and student-level predictors are from the base year eighth-grade survey. In
accordance with NCES security procedures for restricted data, unweighted sample sizes have
been rounded to the nearest 10.
1
P < .10.
* P < .05.
** P < .01.


Other . . . . . . . . . . . . . . . . . . . . . . . . .
Black . . . . . . . . . . . . . . . . . . . . . . . . .
School proportion white slopes:

Other . . . . . . . . . . . . . . . . . . . . . . . . .
Black . . . . . . . . . . . . . . . . . . . . . . . . .
Intercepts:
3.61
.46
1.63
2.73
24.70
3.00
36.23**
.46
1.12
.84
.50
.85
3.53
3.11
3.49
3.82
27.95
5.00
36.21**
.46
1.08
.86
.05
.86
.09
.20
2.17
.28
2.34
.30
9.97**
.03
2.10
.07
2.07
.08
.34
.30
2.79
.47
2.39
.51
9.97**
.03
2.09
.06
2.07
.08
LOG OF INCOME
Linear
OCCUPATIONAL
STATUS Linear
.64
.58
2.24
.72
2.97
.94
2.23**
.09
.08
.17
2.09
.22
.76
.89
2.82
1.23
2.96
1.71
2.24**
.09
.12
.17
2.09
.22
EMPLOYED > 20
WEEKS Binary
TABLE A3
Coefcients for Segregation in the Academic Track HSB Only
.19**
.05
2.30**
.08
.02
.11
.75**
.01
2.37**
.03
2.22**
.11
.21**
.07
2.39**
.12
.01
.19
.75**
.01
2.37**
.03
2.22**
.03
PROPORTION
COWORKERS SAME
RACE Linear
1160

1.27
8.13
28.83
11.73
9.42
11.33
240
2,660
23.12
6.54
6.92
8.61
25.27
8.44
2.58
1.69
2.02
2.48
1.02
2.95
2.10
2.19
1.93
3.46
2.48
3.98
240
2,560
2.81
.74
2.12
1.27
.20
1.08
240
2,470
.58
.61
21.10
.92
.01
.82
EMPLOYED > 20
WEEKS Binary
1
2.08
.18
.32
.32
.07
.41
240
2,660
.05
.15
2.30
.23
2.09
.28
PROPORTION
COWORKERS SAME
RACE Linear
NOTE.Models include school- and student-level control variables as in table 2. In accordance with NCES security procedures for restricted data,
unweighted sample sizes have been rounded to the nearest 10. Nos. in parentheses are SEs.
* P < .05.
** P < .01.
N schools . . . . . . . . . . . . . . . . . . . . . . . .
N students . . . . . . . . . . . . . . . . . . . . . . .
Other . . . . . . . . . . . . . . . . . . . . . . . . .
Black . . . . . . . . . . . . . . . . . . . . . . . . .
School proportion white segregation

in academic track:
Other . . . . . . . . . . . . . . . . . . . . . . . . .
Black . . . . . . . . . . . . . . . . . . . . . . . . .
Segregation in school academic track slopes:

LOG OF INCOME
Linear
OCCUPATIONAL
STATUS Linear
TABLE A3 (Continued )
TABLE A4
Basic Model with Categorical Values for School Proportion White
Outcomes
Intercepts:
Reference group white . . . . .
Black . . . . . . . . . . . . . . . . . .
Other . . . . . . . . . . . . . . . . . .
Slopes:
School 0%20% white:
Reference group white . . .
Black . . . . . . . . . . . . . . . . .
Other . . . . . . . . . . . . . . . . .
School 21%40% white:
Black . . . . . . . . . . . . . . . . .
Other . . . . . . . . . . . . . . . . .
Black . . . . . . . . . . . . . . . . .
Other . . . . . . . . . . . . . . . . .
Black . . . . . . . . . . . . . . . . .
Other . . . . . . . . . . . . . . . . .
School-level controls:
Private . . . . . . . . . . . . . . . . .
Rural . . . . . . . . . . . . . . . . . .
Suburban . . . . . . . . . . . . . . .
Midwest . . . . . . . . . . . . . . . .
South . . . . . . . . . . . . . . . . . .
West . . . . . . . . . . . . . . . . . . .
Occupational Log of
Status
Income
Linear
Linear
39.34**
.28
.12
.89
2.08
.85
9.96**
.02
2.07
.05
2.03
.04
Employed >
20 Weeks
Binary
Proportion
Coworkers
Same Race
Linear
3.14**
.07
2.52**
.08
2.25
.13
.69**
.01
2.34**
.02
2.32**
.02
2.03
.04
.18**
.06
.10
.06
2.45
1.47
21.64
2.88
.90
2.13
2.12
.11
.14
.15
.18
.17
21.23**
.34
1.56**
.46
1.71**
.46
2.76
2.05
23.31
3.58
23.59
2.82
.20
.11
2.26
.21
2.13
.17
.28
.52
.12
.62
2.56
.73
2.03
.03
2.04
.06
.00
.06
21.20
1.29
22.22
3.14
2.96
1.92
.17
.09
2.10
.14
2.09
.14
2.18
.24
.58
.34
.64
.38
.04
.02
2.07
.05
2.05
.05
2.81
1.19
21.78
2.87
24.55
2.33
.10
.08
2.05
.12
2.03
.14
2.20
.21
.83**
.31
.75*
.35
.10**
.02
2.20**
.05
2.08
.05
.39
.78
21.23*
.60
2.67
.58
2.61
.62
2.53
.59
21.64*
.69
.04
.07
2.03
.05
.05
.04
2.06
.04
2.10*
.04
2.15**
.04
2.45**
.17
.10
.15
.15
.14
.22
.14
2.02
.13
.11
.15
.03
.02
.05**
.02
.00
.01
.04
.02
.00
.02
2.04*
.02
1161


TABLE A4 (Continued )
Outcomes
School SES . . . . . . . . . . . . . .
Student-level controls:
Male . . . . . . . . . . . . . . . . . . .
Student SES . . . . . . . . . . . . .
Student test . . . . . . . . . . . . . .
N schools . . . . . . . . . . . . . . . . .
N students . . . . . . . . . . . . . . . .
Occupational Log of
Status
Income
Linear
Linear
.76
.61
23.92**
.37
3.17**
.35
.29**
.02
1,070
10,020
2.10*
.04
.38**
.02
.08**
.02
.01**
.00
1,060
9,140
Employed >
20 Weeks
Binary
Proportion
Coworkers
Same Race
Linear
.08
.13
2.01
.02
.57**
.08
2.17**
.06
.02**
.01
1,070
9,300
2.01
.01
2.01
.01
.00
.00
1,070
10,060
NOTE.In accordance with NCES security procedures for restricted data, unweighted
sample sizes have been rounded to the nearest 10. Nos. in parentheses are SEs.
* P < .05.
** P < .01.
REFERENCES
Alba, Richard. 2009. Blurring the Color Line: The New Chance for a More Integrated
America. Cambridge, Mass.: Harvard University Press.
Allison, Paul. 2001. Missing Data. Thousand Oaks, Calif.: Sage.
Allport, Gordon. 1954. The Nature of Prejudice. Reading, Mass.: Addison-Wesley.
An, Brian P., and Adam Gamoran. 2009. Trends in School Racial Composition in the
Era of Unitary Status. Pp. 19 47 in From the Courtroom to the Classroom: The
Shifting Landscape of School Desegregation, edited by C. E. Smrekar and E. B.
Goldring. Cambridge, Mass.: Harvard Education Press.
Anderson, Elijah. 1990. Streetwise: Race, Class, and Change in an Urban Community.
Chicago: University of Chicago Press.
Armor, David. 1972. The Evidence on Busing. Public Interest Summer: 90126.
Ashenfelter, Orley, William J. Collins, and Albert Yoon. 2006. Evaluating the Role of
Brown v. Board of Education in School Equalization, Desegregation, and the Income
of African Americans. American Law and Economics Review 8:213 48.
Austin, Peter. 2011. An Introduction to Propensity Score Methods for Reducing Effects
of Confounding in Observational Studies. Multivariate Behavioral Research
46:399424.
Bailey, Thomas, and Roger Waldinger. 1991. Primary, Secondary, and Enclave Labor
Markets: A Training Systems Approach. American Sociological Review 56:432 45.
Bates, Percy. 1990. Desegregation: Can We Get There from Here? Phi Delta Kappan
72 September: 817.
Becker, Sascha O., and Andrea Ichino. 2002. Estimation of Average Treatment Effects
Based on Propensity Scores. Stata Journal 2:35877.
Billings, Stephen B., David J. Deming, and Jonah E. Rockoff. 2014. School Segregation, Educational Attainment, and Crime: Evidence from the End of Busing in
Charlotte-Mecklenburg. Quarterly Journal of Economics 129 1: 43576.
BLS Bureau of Labor Statistics. 2013. The Employment SituationNovember 2013.
Washington, D.C.: Bureau of Labor Statistics, U.S. Department of Labor. http://www.bls
.gov/news.release/pdf/empsit.pdf.
1162


Bohon, Stephanie. 2001. Latinos in Ethnic Enclaves: Immigrant Workers and the Competition for Jobs. New York: Garland.
Boozer, Michael A., Alan B. Krueger, and Shari Wolkon. 1992. Race and School
Quality since Brown v. Board of Education. Brookings Papers on Economic Activity:
Microeconomics, 269338.
Braddock, Jomills H., III. 1980. The Perpetuation of Segregation across Levels of
Education: A Behavioral Assessment of the Contact-Hypothesis. Sociology of
Education 53:17886.
. 2009. Looking Back: The Effects of Court-Ordered Desegregation. Pp. 317
in From the Courtroom to the Classroom: The Shifting Landscape of School Desegregation, edited by C. E. Smrekar and E. B. Goldring. Cambridge, Mass.: Harvard
Education Press.
Braddock, Jomills H., III., Marvin P. Dawkins, and William Trent. 1994. Why Desegregate? The Effect of School Desegregation on Adult Occupational Desegregation
of African Americans, Whites, and Hispanics. International Journal of Contemporary Sociology 31:27383.
Braddock, Jomills H., III., and James M. McPartland. 1989. Social-Psychological
Processes That Perpetuate Segregation: The Relationship between School and
Employment Desegregation. Journal of Black Studies 19:26789.
Brand, Jennie E., and Julie Simon Thomas. 2014. Job Displacement among Single
Mothers: Effects on Childrens Outcomes in Young Adulthood. American Journal of
Sociology 119:9551001.
Carter, Prudence L. 2010. Race and Cultural Flexibility among Students in Different
Multiracial Schools. Teachers College Record 112:152974.
Charles, Camille Zubrinsky. 2003. The Dynamics of Racial Residential Segregation.
Annual Review of Sociology 29:167207.
Chetty, Raj, Nathaniel Hendren, and Lawrence Katz. 2015. The Effects of Exposure to
Better Neighborhoods on Children: New Evidence from the Moving to Opportunity
Project. Unpublished manuscript, Harvard University, Department of Economics.
Clampet-Lundquist, Susan, and Douglas S. Massey. 2008. Neighborhood Effects on
Economic Self-Sufciency: A Reconsideration of the Moving to Opportunity
Experiment. American Journal of Sociology 114:107 43.
Clotfelter, Charles T. 2002. Interracial Contact in High School Extracurricular
Activities. Urban Review 34:2546.
. 2004. After Brown: The Rise and Retreat of School Desegregation. Princeton,
N.J.: Princeton University Press.
Clotfelter, Charles T., Jacob L. Vigdor, and Helen F. Ladd. 2006. Federal Oversight,
Local Control, and the Specter of Resegregation in Southern Schools. American
Law and Economics Review 8:34789.
Coleman, James S. 1976 1996. Presentation to the Massachusetts Legislature. Pp. 198
210 in Equality and Achievement in Education. Boulder, Colo.: Westview.
Coleman, James S., Ernest Q. Campbell, Carol F. Hobson, James M. McPartland,
Alexander M. Mood, Fredrick D. Weinfeld, and Robert L. York. 1966. Equality of
Educational Opportunity. Washington, D.C.: U.S. Government Printing Ofce.
Crain, Robert L. 1970. School Integration and Occupational Achievement of Negroes.
American Journal of Sociology 75:593606.
. 1971. School Integration and the Academic Achievement of Negroes. Sociology of Education 44:126.
Crain, Robert L., and Rita Mahard. 1978. School Racial Composition and Black
College Attendance and Achievement Test Performance. Sociology of Education
51:81101.
Crain, Robert L., and Jack Strauss. 1985. School Desegregation and Occupational
Attainment: Results from a Long-Term Experiment. Report no. 359. Baltimore: Johns
Hopkins University, Center for Social Organization of Schools.
1163


Curtin, Thomas R., Stephen J. Ingels, Shiying Wu, and Ruth Heuer. 2002. National
Education Longitudinal Study of 1988: Base-Year to Fourth Follow-up Data File
User s Manual. NCES 2002-323. Washington, D.C.: U.S. Department of Education.
http://nces.ed.gov/pubs2002/2002323.pdf.
Dawkins, Marvin P. 1994. Long-Term Effects of School Desegregation on African
Americans: Evidence from the National Survey of Black Americans. Negro Educational Review 45:415.
DeLuca, Stefanie, and Peter Rosenblatt. 2010. Does Moving to Better Neighborhoods
Lead to Better Schooling Opportunities? Parental School Choice in an Experimental
Housing Voucher Program. Teachers College Record 112:144391.
DiPrete, Thomas A., and Markus Gangl. 2004. Assessing Bias in the Estimation of
Causal Effects: Rosenbaum Bounds on Matching Estimators and Instrumental
Variables with Imperfect Instruments. Sociological Methodology 34:271310.
DiPrete, Thomas A., Andrew Gelman, Tyler McCormick, Julien Teitler, and Tian
Zheng. 2011. Segregation in Social Networks Based on Acquaintanceship and
Trust. American Journal of Sociology 116:123483.
Eaton, Susan. 2001. The Other Boston Busing Story: Whats Won and Whats Lost
across the Boundary Line. New Haven, Conn.: Yale University Press.
Edin, Per-Anders, Peter Fredricksson, and Olof Aslund. 2003. Ethnic Enclaves and the
Economic Success of Immigrants: Evidence from a Natural Experiment. Quarterly
Journal of Economics 118:32957.
Ellison, Christopher G., and Daniel A. Powers. 1994. The Contact Hypothesis and
Racial Attitudes among Black Americans. Social Science Quarterly 75:3854 00.
Entwisle, Doris, and Karl L. Alexander. 1992. Summer Setback: Race, Poverty, School
Composition, and Mathematics Achievement. American Sociological Review 57:7284.
. 1994. Winter Setback: The Racial Composition of Schools and Learning to
Read. American Sociological Review 59:44660.
Eyler, Janice, Valerie J. Cook, and Leslie E. Ward. 1983. Resegregation: Segregation
within Desegregated Schools. Pp. 12662 in The Consequences of School Desegregation, edited by Christine H. Rossell and Willis D. Hawley. Philadelphia: Temple
University Press.
Farley, Reynolds, and Alma F. Taeuber. 1974. Racial Segregation in the Public
Schools. American Journal of Sociology 79:888905.
Fischer, Claude S., Gretchen Stockmayer, Jon Stiles, and Michael Hout. 2004. Distinguishing the Geographic Levels and Social Dimensions of U.S. Metropolitan
Segregation, 19602000. Demography 41:3759.
Fryer, Roland G. 2006. Acting White. Education Next 6 1: 5359.
Gamoran, Adam. 1987. The Stratication of High School Learning Opportunities.
Sociology of Education 60:13555.
. 2001. American Schooling and Educational Inequality: A Forecast for the
21st Century. Sociology of Education 74 extra issue: 13553.
. 2007. Standards-Based Reform and the Poverty Gap: Lessons for No Child Left
Behind. Washington, D.C.: Brookings Institution Press.
Gamoran, Adam, and Brian An. In press. Effects of School Segregation and School
Resources in a Changing Policy Context. Educational Evaluation and Policy Analysis.
Gamoran, Adam, and Daniel A. Long. 2007. Equality of Educational Opportunity:
A 40-Year Retrospective. Pp. 23 47 in International Studies in Educational Inequality: Theory and Policy, vol. 1. Edited by Richard Teese, Stephen Lamb, and
Marie Duru-Bellat. Dordrecht: Springer.
Gamoran, Adam, and Robert D. Mare. 1989. Secondary School Tracking and Educational Inequality: Reinforcement, Compensation, or Neutrality? American Journal
of Sociology 94:114683.
Gangl, Markus. 2010. Causal Inference in Sociological Research. Annual Review of
Sociology 36:21 47.
1164


Goldsmith, Pat Antnio. 2004. Schools Racial Mix, Student Opinions, and the BlackWhite and Latino-White Achievement Gaps. Sociology of Education 77:121 47.
. 2009. Schools, Neighborhoods, or Both? Educational Attainment and Race
and Ethnic Segregation. Social Forces 87:1913 43.
Granovetter, Mark. 1986. The Micro-structure of School Desegregation. Pp. 81110 in
School Desegregation Research: New Directions in Situational Analysis, edited by
Jeffrey Prager, Douglas Longshore, and Melvin Seeman. New York: Plenum.
Grissmer, David W., Sheila Nataraj Kirby, Mark Berends, and Stephanie Williamson.
1994. Student Achievement and the Changing American Family. Santa Monica,
Calif.: RAND.
Grogger, Jeffrey. 1996. Does School Quality Explain the Recent Black/White Wage
Trend? Journal of Labor Economics 14:23153.
Guryan, Jonathan. 2004. Desegregation and Black Dropout Rates. American Economic Review 94:919 43.
Hallinan, Maureen T., and Richard A. Williams. 1987. The Stability of Students
Interracial Friendships. American Sociological Review 52:65364.
. 1989. Interracial Friendship Choices in Secondary Schools. American
Sociological Review 54:6778.
Harding, David. 2003. Counterfactual Models of Neighborhood Effects: The Effect
of Neighborhood Poverty on Dropping Out and Teenage Pregnancy. American Journal of Sociology 109:676719.
Hauser, Robert M., and John Robert Warren. 1997. Socioeconomic Indexes for
Occupations: A Review, Update, and Critique. Sociological Methodology 27:
177298.
Heckman, James, Hidehiko Ichimura, Jeffrey Smith, and Petra Todd. 1998. Characterizing Selection Bias Using Experimental Data. Econometrica 66:101798.
Hedges, Larry V., and Amy Nowell. 1999. Changes in the Black-White Gap in Test
Scores. Sociology of Education 72:11135.
Hochschild, Jennifer, and Nathan Scovronick. 2003. The American Dream and the
Public Schools. New York: Oxford University Press.
Johnson, Rucker C. 2011. Long-Run Impacts of School Desegregation and School
Quality on Adult Attainments. NBER Working paper no. 16664. National Bureau of
Economic Research, Cambridge, Mass.
Kaufman, Julie E., and James E. Rosenbaum. 1992. The Education and Employment
of Low-Income Black Youth in White Suburbs. Educational Evaluation and Policy
Analysis 14:229 40.
Konstantopoulos, Spyros. 2006. Trends of School Effects on Student Achievement:
Evidence from NLS:72, HSB:82, and NELS:92. Teachers College Record 108:2550
81.
Lee, Hedwig. 2007. The Effects of School Racial and Ethnic Composition on Academic
Achievement during Adolescence. Journal of Negro Education 76:15472.
Lee, Wang-Sheng. 2013. Propensity Score Matching and Variations on the Balancing
Test. Empirical Economics 44:4780.
Ludwig, Jens, Greg J. Duncan, Lisa A. Gennetian, Lawrence F. Katz, Ronald C.
Kessler, Jeffrey R. Kling, and Lisa Sanbonmatsu. 2013. Long-Term Neighborhood
Effects on Low-Income Families: Evidence from Moving to Opportunity. American
Economic Review Papers and Proceedings 103:22631.
Massey, Douglas S., and Nancy A. Denton. 1993. American Apartheid: Segregation and
the Making of the Underclass. Cambridge, Mass.: Harvard University Press.
Mickelson, Roslyn M. 2001. Subverting Swann: First- and Second-Generation Segregation in the Charlotte-Mecklenburg Schools. American Educational Research Journal 38:21552.
Moody, James. 2001. Race, School Integration, and Friendship Segregation in America. American Journal of Sociology 107:679716.
1165


Morgan, Stephen L. 2001. Counterfactuals, Causal Effect Heterogeneity, and the
Catholic School Effect on Learning. Sociology of Education 74:34174.
Morgan, Stephen L., and David J. Harding. 2006. Matching Estimators of Causal
Effects: Prospects and Pitfalls in Theory and Practice. Sociological Methods and
Research 35:360.
Morgan, Stephen L., and Christopher Winship. 2007. Counterfactuals and Causal
Inference: Methods and Principles for Social Research. Cambridge: Cambridge
University Press.
Mouw, Ted, and Barbara Entwisle. 2006. Residential Segregation and Interracial
Friendship in Schools. American Journal of Sociology 112:394 441.
National Urban League. 2010. The State of Black America 2010. Washington, D.C.:
National Urban League.
Oreld, Gary. 2001. Schools More Separate: Consequences of a Decade of Resegregation.
Cambridge, Mass.: Harvard University, Civil Rights Project.
Pettigrew, Thomas F. 1965. Continuing Barriers to Desegregated Education in the
South. Sociology of Education 38:99111.
Pettigrew, Thomas F., and Linda R. Tropp. 2006. A Meta-Analytic Test of Intergroup
Contact Theory. Journal of Personality and Social Psychology 90:75183.
Portes, Alejandro, and Patricia Fernndez-Kelly. 2008. No Margin for Error: Educational and Occupational Achievement among Disadvantaged Children of Immigrants. Annals of the American Academy of Political and Social Sciences 620:1236.
Portes, Alejandro, and Linxin Hao. 2004. The Schooling of Children of Immigrants:
Contextual Effects on the Educational Attainment of the Second Generation. Proceedings of the National Academy of Sciences 101:1192027.
Portes, Alejandro, and Leif Jensen. 1989. The Enclave and the Entrants: Patterns of
Ethnic Enterprise in Miami Before and After Mariel. American Sociological Review
54:929 49.
Portes, Alejandro, and Dag McLeod. 1996. Educational Progress of Children of Immigrants: The Roles of Class, Ethnicity, and School Context. Sociology of Education
69:25575.
Portes, Alejandro, and Ruben Rumbaut. 2001. Legacies: The Story of the Immigrant
Second Generation. Berkeley and Los Angeles: University of California Press.
Portes, Alejandro, and Min Zhou. 1993. The New Second Generation: Segmented
Assimilation and Its Variants. Annals of the American Academy of Political and
Social Science 530:74 96.
Raudenbush, Stephen W., and Anthony S. Bryk. 2002. Hierarchical Linear Models:
Applications and Data Analysis Methods. Thousand Oaks, Calif.: Sage.
Reardon, Sean F., Elena T. Grewal, Demetra Kalogrides, and Erica Greenberg. 2012.
Brown Fades: The End of Court-Ordered Desegregation and the Resegregation of
American Public Schools. Journal of Policy Analysis and Management 31:876904.
Rivkin, Steven. 2000. School Desegregation, Academic Attainment, and Earnings.
Journal of Human Resources 35:333 46.
Rosenbaum, Paul. 2002. Obser vational Studies. New York: Springer-Verlag.
Sampson, Robert J. 2008. Moving to Opportunity: Neighborhood Effects and Experiments Meet Social Structure. American Journal of Sociology 114:189231.
Schmidt, Peter. 1994. Plan to Return Boston to Neighborhood Schools Is Proposed.
Education Week, March 1.
Schoeld, Janet. 1995. Review of Research on School Desegregations Impact on
Elementary and Secondary School Students. Pp. 597616 in Handbook of Research
on Multicultural Education, edited by J. A. Banks and C. A. M. Banks. New York:
Macmillan.
Sianesi, Barbara. 2001. Implementing Propensity Score Matching Estimators with
Stata. Presented at the U.K. Stata Users Group, London. http://fmwww.bc.edu
/RePEc/usug2001/psmatch.pdf.
1166


Stearns, Elizabeth. 2010. Long-Term Correlates of High School Racial Composition:
Perpetuation Theory Revisited. Teachers College Record 112:1654 78.
Stuart, Elizabeth. 2010. Matching Methods for Causal Inference: A Review and a Look
Forward. Statistical Science 25:121.
Trent, William T. 1997. Outcomes of School Desegregation: Findings from Longitudinal Research. Journal of Negro Education 66:25577.
Tyson, Karolyn. 2011. Integration Interrupted: Tracking, Black Students, and Acting
White after Brown. New York: Oxford University Press.
Tyson, Karolyn, William Darity, Jr., and Domini R. Castellino. 2005. Its Not a Black
Thing: Understanding the Burden of Acting White and Other Dilemmas of High
Achievement. American Sociological Review 70:582605.
Vigdor, Jacob, and Jens Ludwig. 2008. Segregation and the Test Score Gap. Pp. 181
211 in Steady Progress and Stalled Gains: Inequality and the Black-White Test-Score
Gap, edited by Katherine A. Magnuson and Jane Waldfogel. New York: Russell Sage
Foundation.
Waldinger, Roger, and Cynthia Feliciano. 2004. Will the New Second Generation
Experience Downward Assimilation? Segmented Assimilation Re-assessed. Ethnic
and Racial Studies 27:376 402.
Wells, Amy S., and Robert L. Crain. 1994. Perpetuation Theory and the Long-Term
Effects of School Desegregation. Review of Educational Research 64:53155.
Wells, Amy S., Jennifer J. Holme, Anita T. Revilla, and Awo K. Atanda. 2004. How
Society Failed Desegregation Policy: Looking Past the Schools to Understand Them.
Review of Research in Education 28:4799.
Wilson, Kenneth L. 1979. The Effects of Integration and Class on Black Educational
Attainment. Sociology of Education 52:84 98.
Wilson, Kenneth L., and W. Allen Martin. 1982. Ethnic Enclaves: A Comparison of the
Cuban and Black Economies in Miami. American Journal of Sociology 88:13560.
Wilson, William J. 1978. The Declining Signicance of Race. Chicago: University of
Chicago Press.
. 1987. The Truly Disadvantaged. Chicago: University of Chicago Press.
Wood, Peter B., and Nancy Sonleiter. 1996. The Effect of Childhood Interracial
Contact on Adult Antiblack Prejudice. International Journal of Intercultural Relations 20:117.
Zahs, Daniel, Steven Pedlow, Marjorie Morrissey, Patricia Marnell, and Bronwyn
Nichols. 1995. High School and Beyond Fourth Follow-Up Methodology Report.
Washington, D.C.: U.S. Department of Education. http://nces.ed.gov/pubs95/95426
.pdf.
1167


Ga Moran 2016

Diunggah oleh

Informasi Dokumen

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Ga Moran 2016

Diunggah oleh

Hak Cipta:

Format Tersedia

Does Racial Isolation in School Lead

to Long-Term Disadvantages? Labor

Ana Cristina Collares

2016 by The University of Chicago. All rights reserved.

AJS Volume 121 Number 4 (January 2016): 111667

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

Racial Isolation in School

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

American Journal of Sociology

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

Racial Isolation in School

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

American Journal of Sociology

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

Racial Isolation in School

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

American Journal of Sociology

Despite continued interest in perpetuation theory, recent studies offer at

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

Racial Isolation in School

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

American Journal of Sociology

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

Racial Isolation in School

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

American Journal of Sociology

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

Racial Isolation in School

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

Employed > 20 weeks . . . . . . . . . . . . . . .

American Journal of Sociology

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

Racial Isolation in School

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

American Journal of Sociology

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

Racial Isolation in School

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

American Journal of Sociology

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

Racial Isolation in School

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

American Journal of Sociology

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

Racial Isolation in School

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

American Journal of Sociology

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

Racial Isolation in School

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

American Journal of Sociology

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

Racial Isolation in School

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

American Journal of Sociology

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

Racial Isolation in School

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

This content downloaded from 128.252.067.066 on July 25, 2016 23:09:01 PM

EMPLOYED > 20 WEEKS

39.37 9.97 9.97**