ArticlePDF Available

Family, Neighborhood, and School Settings Across Seasons: When Do Socioeconomic Context and Racial Composition Matter for the Reading Achievement Growth of Young Children?

May 2010
Teachers College Record 112(5):1338-1390

May 2010
112(5):1338-1390

DOI:10.1177/016146811011200505

Authors:

Arizona State University

Background/Context Seasonal researchers have developed a theory and hypotheses regarding the importance of neighborhood and school contexts for early childhood learning but have not possessed nationally representative data and precise contextual measures with which to examine their hypotheses. Purpose/Research Questions This empirical study employs a seasonal perspective to assess the degree to which social context and race/ethnic composition—in neighborhoods and schools—affect the reading achievement growth of young children. The authors ask, Were there specific seasons when context and/or composition were particularly salient for reading achievement? Also, did accounting for context and composition challenge established appraisals of the relationship between family factors and achievement? Population Data for a nationally representative sample of students proceeding through kindergarten and first grade came from the Early Childhood Longitudinal Study-Kindergarten Cohort (ECLS-K). Neighborhood social and race/ethnic measures came from the 2000 Census. Research Design: This quantitative study employs a three-level model that assesses reading achievement at school entry and during three subsequent seasons. The model represents reading achievement as a time-varying process at level 1, conditional upon family socio/demographic factors at level 2, and dependent on social context and race/ethnic composition at level 3. Findings/Results Neighborhood social context mattered substantially for students’ reading achievement levels at school entry and for their reading achievement growth during the summer. The proportion of neighborhood residents from minority race/ethnic groups was not associated with reading achievement at school entry or during the summer season. During the school year, school social context was associated with reading growth during kindergarten, and school social context and race/ethnic composition were associated with reading growth during first grade. Conclusions/Recommendations The magnitude and frequency of contextual effects found in this national sample have considerable implications for achieving educational equality in the United States. The authors recommend that policy makers attend to the quality of neighborhood and school settings as a means of promoting literacy development for young children.

Three-Level Models of Reading Achievement Growth for Students Nested Within Schools, by Season (SE), N=4180 students; 290 schools.

…

Three-Level Models of Reading Achievement Growth for Students Nested Within Schools, by Season (SE) (continued)

…

Figures - uploaded by Geoffrey D. Borman

Content may be subject to copyright.

Content uploaded by Geoffrey D. Borman

Content may be subject to copyright.

Family, Neighborhood, and School

Settings Across Seasons: When Do

Socioeconomic Context and Racial

Composition Matter for the Reading

Achievement Growth of Young Children?

JAMES BENSON

GEOFFREY D. BORMAN

University of Wisconsin–Madison

Background/Context: Seasonal researchers have developed a theory and hypotheses regard-

ing the importance of neighborhood and school contexts for early childhood learning but

have not possessed nationally representative data and precise contextual measures with

which to examine their hypotheses.

Purpose/Research Questions: This empirical study employs a seasonal perspective to assess

the degree to which social context and race/ethnic composition—in neighborhoods and

schools—affect the reading achievement growth of young children. The authors ask, Were

there specific seasons when context and/or composition were particularly salient for reading

achievement? Also, did accounting for context and composition challenge established

appraisals of the relationship between family factors and achievement?

Population: Data for a nationally representative sample of students proceeding through

kindergarten and first grade came from the Early Childhood Longitudinal Study-

Kindergarten Cohort (ECLS-K). Neighborhood social and race/ethnic measures came from

the 2000 Census. Research Design: This quantitative study employs a three-level model that

assesses reading achievement at school entry and during three subsequent seasons. The

model represents reading achievement as a time-varying process at level 1, conditional upon

family socio/demographic factors at level 2, and dependent on social context and

race/ethnic composition at level 3.

Teachers College Record Volume 112, Number 5, May 2010, pp. 1338–1390

0161-4681

Socioeconomic Context and Racial Composition 1339

Findings/Results: Neighborhood social context mattered substantially for students’ reading

achievement levels at school entry and for their reading achievement growth during the sum-

mer. The proportion of neighborhood residents from minority race/ethnic groups was not

associated with reading achievement at school entry or during the summer season. During

the school year, school social context was associated with reading growth during kinder-

garten, and school social context and race/ethnic composition were associated with reading

growth during first grade.

Conclusions/Recommendations: The magnitude and frequency of contextual effects found

in this national sample have considerable implications for achieving educational equality

in the United States. The authors recommend that policy makers attend to the quality of

neighborhood and school settings as a means of promoting literacy development for young

children.

The skills that young children possess when they enter elementary school

and the pace at which they learn while in elementary school matter a

great deal for subsequent academic outcomes and life experiences.

Differences in elementary school learning progress typically become dif-

ferences in high school graduation, college attendance rates, and even-

tual adult status (Entwisle & Alexander, 1999; Kerckhoff, 1993). During

the early primary school years, seemingly small month-to-month achieve-

ment differences accrue across seasons, resulting in substantial differ-

ences as students prepare to enter high school.

By the time students in the United States enter high school, socially dis-

advantaged students and students from minority race/ethnic groups lag

substantially behind their more advantaged and majority counterparts in

fundamental reading skills. Among U.S. eighth graders in 2005, only 57%

of students eligible for free lunch were reading at or above the basic level,

whereas 81% of those not eligible for free lunch were exceeding this level

(U.S. Department of Education, 2005). Similar gaps exist between White

students and Hispanic students, and even wider gaps separate White and

African American students. Although the presence of these gaps is com-

mon knowledge among education researchers, their source has been the

subject of several theories and considerable debate. Disagreement has

extended to questions of when, where, and how these gaps emerge.

Seasonal researchers have identified a blind spot in much of the edu-

cational research on achievement gaps. Research studies based on

annual achievement testing have conflated processes that take place out-

side the school year with those taking place during the school year

(Alexander, Entwisle, & Olson, 2004). In contrast, seasonal researchers

have emphasized that participation in formal schooling varies decisively

across seasons—especially in the United States1—and thus shined a light

on processes taking place at times other than the school year and in

1340 Teachers College Record

places other than school. This approach has led to a fundamental reap-

praisal of the relationship between families, schools, and early childhood

learning. Seasonal researchers have concluded that the processes by

which young students from socially disadvantaged families often fall

alarmingly behind in their academic development occur, in large part if

not entirely, during the summer season, when schools are typically not in

session (Downey, von Hippel, & Broh, 2004; Entwisle, Alexander, &

Olson, 1997; Heyns, 1978). The summer setback disproportionately

affects African American and Hispanic students because they are more

likely than White students to come from low-SES families.

The seasonal perspective has substantial implications for educational

research. It is a call for precise testing and data collection schedules that

facilitate indentifying the periods and locations of achievement gaps

(Heyns, 1978). In particular, the perspective emphasizes the distinction

between school-year and summer-season processes and points to families

and schools as locations with potentially divergent effects—especially for

low-SES students—on achievement inequality. We adopt this analytical

perspective and extend it beyond the family setting, taking into account

the social context and race/ethnic composition of neighborhoods and

schools. We refer to social context as the socioeconomic character of neigh-

borhoods and schools. Race/ethnic composition refers to the makeup of

neighborhoods and schools in terms of the proportion of members from

majority and minority race/ethnic groups. When speaking broadly, we

use the term context to refer to both social context and racial composi-

tion. By adding consideration of context to the seasonal perspective, we

can ask questions regarding the relative contributions of families and

context to achievement inequality during nonschool periods and the

contributions of families, schools, and context during the school year.

To identify the characteristics of specific contexts in which young stu-

dents grow up and attend school, we constructed a unique data set that

links measures of neighborhood social characteristics from the 2000

Census—via residential zip codes—to individual students from the

kindergarten cohort of the Early Childhood Longitudinal Study (ECLS-

K), sponsored by the National Center for Education Statistics (NCES).

We drew on the detailed accounting of test times and school-year begin

and end dates within the ECLS-K database to decompose precisely

school-year and summer-season growth patterns. Using data from both

census and NCES sources, we used an identical coding process to con-

struct comparable contextual measures for neighborhoods and schools.

Using three-level growth models, we combined our accounting of sea-

sonal growth patterns based on NCES achievement tests and testing

schedules, accurate measures of student and family characteristics, and

Socioeconomic Context and Racial Composition 1341

contextual measures for neighborhoods and schools. We focused on

reading achievement because reading skills are the primary focus of

instruction during the first 2 years of elementary school.

Our findings indicate that neighborhood social context mattered sub-

stantially for students’ reading achievement levels at school entry and for

their reading achievement growth during the summer. On the other

hand, we found that the proportion of neighborhood residents from

minority race/ethnic groups was not associated with reading achieve-

ment at school entry or during the summer. During the school year, our

findings generally indicate that school context was more salient than

neighborhood context. School social context was associated with reading

growth during kindergarten, and school social context and race/ethnic

composition were associated with reading growth in first grade. These

contextual effects tended to reinforce longstanding patterns of socioeco-

nomic and race/ethnic achievement inequality and thus have consider-

able implications for educational equality in the United States.

BACKGROUND

THE SEASONAL PERSPECTIVE ON FAMILY SES AND ACHIEVEMENT

INEQUALITY

Seasonal researchers have reconceptualized early childhood learning for

low-SES children as a seasonally varying process in which lower status

families do not possess the resources necessary to continue academic

learning when school is not in session. This seasonal perspective differs

fundamentally from a more pessimistic view in which the families of low-

SES children are seen as impeding the process of academic learning dur-

ing the school year. Unfortunately, this more pessimistic view followed

somewhat rationally from studies such as the Coleman report (Coleman

et al., 1966). The annual achievement tests on which the Coleman study

was based prevented researchers from disentangling the relative contri-

butions of families and schools to achievement growth. Using more

recent semiannual testing data, seasonal researchers have found that stu-

dents’ participation in formal schooling produces a much more egalitar-

ian distribution of achievement growth than does the immersion of

students in their nonschool environments during the summer season

(Downey et al., 2004; Entwisle et al., 1997; Heyns, 1978).

Seasonal researchers have been unanimous in their conceptualization

of schooling as an equalizing force, although they have disagreed some-

what on the extent to which it overcomes the influence of family socioe-

conomic factors. The pioneer of seasonal learning research, Barbara

1342 Teachers College Record

Heyns (1978), conceptualized schools as a countervailing force to family

socioeconomic resources while maintaining that “the socioeconomic

background and general family conditions of children exert an influence

on the achievement of children, whether or not school is in session” (p.

186). Entwisle et al. (1997) went a distance further when arguing that

schooling overcomes limitations posed by home environments.

According to their faucet theory,2“Home resources do not ‘add on’ to

school resources in winter, because poor children do as well as those who

are well off. Only in summer do home resources come into play” (p. 38).

Systematic variation in school quality may also impede the role of schools

as an equalizing force. Downey et al. (2004) addressed this issue by assert-

ing that even though schools may be unfair in their allocation of learn-

ing resources, they still serve as “equalizers” because schools, as a whole,

are fairer than nonschool learning environments.

Findings from seasonal studies reveal greater socioeconomic disper-

sion in achievement scores during the summer than the school year.

Heyns (1978) found that during the summer season, each of her socioe-

conomic indicators—family income, parental education, and household

size—significantly predicted achievement, whereas during the school

year, only parental education significantly predicted achievement, and

much less so than during the summer.3Using categorical measures of stu-

dents’ families’ SES when examining test results from Grades 1–5,

Entwisle et al. (1997) found that the reading achievement level of high-

SES students exceeded that of low- and medium-SES students entirely

because high-SES students continued to grow in reading achievement

during the summer. While low- and middle-SES students did not grow in

reading achievement during the summer, they grew at least as quickly as

high-SES students during the school year; in 2 of the 5 years, middle-SES

students grew faster than high-SES students (Entwisle et al., table 3.1).

Using data from the ECLS-K database, Downey et al. (2004) found that

family SES was positively associated with rates of school-season learning

in reading but that the relationship was much weaker than during the

summer season.

RACE/ETHNICITY AND ACHIEVEMENT INEQUALITY

The Persistent Black-White Achievement Gap

When addressing race/ethnic disparities in achievement within the

United States, researchers have tended to focus on the achievement gap

between African American and White students. National high school sur-

veys have consistently revealed a gap on the order of one standard devia-

Socioeconomic Context and Racial Composition 1343

tion between composite—reading and math—achievement scores for

Black and White students (Hedges & Nowell, 1998). During the period

from 1965 to 1992, this gap narrowed in most subjects. For example, in

reading achievement, the Black-White gap narrowed from -1.0 standard

deviation units to -0.70 units (Hedges & Nowell). Given the size and per-

sistence of this gap, researchers have devoted a great deal of considera-

tion to exactly when and how the Black-White gap emerges (Jencks &

Phillips, 1998). With few exceptions, researchers have noted consistent

disparities in achievement test scores between Black and White students

at school entry (Phillips, Brooks-Gunn, Duncan, Klebanov, & Crane,

1998). The picture of the gap that emerges after school entry is more

complex, and the gap sizes vary by specific subject. In a study that

employed all the national high school surveys along with results from the

National Assessment of Educational Progress (NAEP), Phillips et al.

attributed half of the Black-White gap in high school tests to occurrences

after the beginning of school but before high school. Black students who

entered elementary school with the same scores as White students tended

to fall behind White students on reading and vocabulary measures but

tended to keep up in mathematics. These findings suggest that

researchers should be attuned to the emergence of Black-White achieve-

ment gaps during elementary school, especially in the area of reading

skills.

Race/Ethnicity and Achievement Inequality in Seasonal Research

Seasonal researchers have demonstrated a clear concern for

race/ethnic achievement inequality as evidenced by the conduct of land-

mark seasonal studies in the Atlanta (Heyns, 1978) and Baltimore

(Entwisle et al., 1997) urban core areas, each of which had majorities of

African American students during the study periods. The primary focus

of these studies, however, has been seasonal variation in the relationship

between family SES and achievement. Certainly, achievement inequality

rooted in socioeconomic inequality has implications for minority stu-

dents, who are more likely than White students to come from low-SES

families. However, other researchers have emphasized that a substantial

Black-White achievement gap remains even when accounting for family

SES (Phillips et al., 1998).

When controlling for SES, seasonal researchers have tended to find

that race/ethnic achievement gaps do not widen substantially during the

summer season. In a meta-analysis of 39 previous summer learning stud-

ies, Cooper, Nye, Charlton, Lindsay, and Greathouse (1996) found that

achievement growth rates for Black and White students from families

1344 Teachers College Record

with similar income levels slowed equally during the summer. Similarly,

using data from the ECLS-K and controlling for family SES, Downey et al.

(2004) found that African American, Hispanic, and White students did

not differ significantly in summer-season reading and math achievement

growth. However, also studying students from the ECLS-K, Burkam,

Ready, Lee, and Logerfo (2004) found a summer-season Black-White gap

in math achievement growth. In her seminal study of Atlanta public

school students, Heyns (1978) found that net of family income and

parental education, African American students’ composite achievement

scores grew at a slightly slower pace than those of White students during

the summer.

Seasonal studies have more consistently found school-year race/ethnic

gaps in achievement even when controlling for family SES. Using data

from the ECLS-K, Downey et al. (2004) identified substantial Black-White

gaps in school-year reading growth during kindergarten and first grade,

and a modestly sized Hispanic-White achievement gap during first grade.

Both Entwisle and Alexander (1992), from their Beginning School Study

(BSS) sample of Baltimore students, and Heyns (1978), in her sample of

Atlanta sixth- and seventh graders, found school-year differences

between Black and White students. Taken together, these findings and

the summer-only studies suggest that after school entry, the school year—

rather than the summer—is the primary period during which

race/ethnic dispersion in test scores increases.

CONTEXT AND EARLY CHILDHOOD LEARNING

The Challenge of Defining the Boundaries of Context and Composition

Researchers who have refined social theory concerning the effects of

neighborhood and school context have conceptualized context as perti-

nent to groups of people bounded by geography (such as neighborhood

boundaries) or by organizational location (such as attendance at the

same school). Geographic bounding brings together fairly large4and

potentially diverse groups of individuals. Delineating boundaries have

included school catchment areas, census tracts, and zip code tabulation

areas (Jencks & Mayer, 1990). The conceptualization of neighborhoods

and schools as geographic areas that include all members within them

differs from the notion of a social network, which groups more specific

sets of people who regularly interact with each other (Wasserman &

Faust, 1994). Even poor neighborhoods provide a range of potential rela-

tionships (Jencks & Phillips, 1998). When working at the neighborhood

or school level of aggregation, the researcher cannot know if all of a per-

Socioeconomic Context and Racial Composition 1345

son’s important and immediate social relations are contained within his

or her neighborhood or school.

Despite the potential for considerable variation in social interactions

within contexts, contextual research has proceeded according to a set of

reasonable assumptions. Within geographic areas with a fair degree of

socioeconomic homogeneity, processes of social interaction and access to

social resources take on patterns that influence the lives of individuals

and families. Depending on the quality and content of these patterns,

context can positively or negatively influence individual development.

Empirical findings from this body of research have tended to validate this

hypothesis. Researchers have found that contexts—in ways that can vary

from person to person, group to group, and place to place—significantly

influence the actions and development of individuals within them, inde-

pendent of one’s immediate family environs (Brooks-Gunn, Duncan, &

Aber, 1997; Brooks-Gunn, Duncan, Klebanov, & Sealand, 1993; Jencks &

Mayer, 1990).

Social Context and Early Childhood Learning

Two main theoretical perspectives have evolved to explain the relation-

ship between context and the cognitive development of young children.

The collective socialization perspective has conceptualized contextual

effects as consisting of parental actions relevant to other parents and chil-

dren in a neighborhood. Affluent parents are seen as promoting good

parenting among other parents and as exerting a degree of social control

over their children’s peers by discouraging undesirable behavior

(Ainsworth, 2002; Brooks-Gunn et al., 1993). This perspective empha-

sizes the positive role modeling provided by more affluent neighbors. For

this reason, research informed by this perspective tends to include vari-

ables intended to capture the positive effects of socially advantaged

neighbors or classmates.

An alternative conceptualization suggests that institutional resources,

especially the human resources provided by middle-class service person-

nel, form the crux of important content present in a context. Concern

for capturing the scope of institutional resources available to residents

has dovetailed with concern about their social isolation and has led to

measures indicating the proportions of disadvantaged residents in neigh-

borhoods (Wilson, 1987). Within this resource-based approach, large

proportions of disadvantaged residents are seen as limiting the amount

of economic support available for hiring well-qualified teachers, main-

taining storefront businesses, attracting quality health care facilities, and

encouraging respectful policing strategies. A feedback relationship takes

1346 Teachers College Record

place such that the presence of disadvantaged residents leads to the

absence of advantaged residents and the institutions that they sustain.

Because of this linkage, analyses that rely on separate measures of contex-

tual advantage and disadvantage have tended to produce smaller coeffi-

cients than those that rely on one or the other. However, multicategory

analyses have allowed researchers (e.g., Brooks-Gunn et al., 1997) to

compare effect sizes for low- and high-SES residents and thus have pro-

vided a clearer picture of the sources of observed contextual effects.

Findings From Neighborhood and School Social Context Studies

Findings from quasi-experimental studies have revealed significant and

modestly sized associations between social context and the cognitive test

scores of young children. Brooks-Gunn et al. (1993) found a substantial

positive relationship between the proportion of higher income residents

in a neighborhood and early childhood test scores. Using the Stanford-

Binet measure of IQ5administered to children at 36 months of age, the

Brooks-Gunn research team found that a one-standard-deviation increase

in the proportion of affluent neighbors was associated with an increase of

one quarter of a standard deviation in individual IQ test scores. The

authors found no negative effects associated with the proportion of poor

neighbors. The multisample study (Brooks-Gunn et al., 1997) of which

this study was a part created a consensus that neighborhoods matter for

early childhood cognitive development. In a national study of high

school students, Ainsworth (2002) found a significant positive relation-

ship between the neighborhood proportion of affluent residents and

achievement test scores.6

Regarding school social context, Jencks and Mayer’s (1990) reappraisal

of data from the Coleman report (1966) found that mean school SES was

positively associated with achievement levels in core subjects during the

sixth grade. The authors estimated the effect of moving from a poor

school to a middle-class school as worth an increase of approximately one

sixth of a standard deviation in achievement, and this effect held for

Black and White students. Results from this study should be interpreted

with caution; Coleman relied on student-reported SES measures and may

have inadequately controlled for prior achievement. More recently,

researchers using greatly improved measures of family SES and school

context have not detected a relationship between school social context

and achievement growth for high school students (see Gamoran, 1987).

These findings suggest that school social context may be most conse-

quential for achievement growth during early childhood.

Socioeconomic Context and Racial Composition 1347

Findings from Residential Mobility Studies

Evaluations of residential mobility programs have produced mixed find-

ings regarding the relationship between improved neighborhood social

context and educational outcomes. In a quasi-experimental evaluation of

the Gautreaux residential mobility program, Kaufman and Rosenbaum

(1992) found modest achievement benefits associated with relocation of

poor families to suburban neighborhoods. In an experimental evaluation

of the Moving to Opportunity (MTO) program, Sanbonmatsu, Kling,

Duncan, and Brooks-Gunn (2006) found no achievement benefits associ-

ated with relocation of poor families from high-poverty to low- and

medium-poverty neighborhoods. Although they operated by different

rules, these programs sought to improve education and labor market out-

comes by relocating families from high-poverty, largely African American

public housing locations to less poor and less segregated locations in

urban metro areas.7

Between 1976 and 1998, the Gautreaux program moved more than

7,000 families from Chicago public housing to privately leased apart-

ments and houses in the Chicago city area and throughout the Chicago

metro area. When comparing origin and placement addresses, neighbor-

hood poverty dropped from 42% to 17%, and neighborhood percent

African American dropped from 83% to 28% (Duncan & Zuberi, 2006).

However, neighborhood social context and race/ethnic composition dif-

fered markedly depending on whether families relocated to city or sub-

urban areas. Researchers have exploited this difference to compare

outcomes for parents and children in these different settings. Location in

suburban versus city areas was associated with equal marks in school8and

with a greater likelihood of placement in the college-bound track in high

school (Kaufman & Rosenbaum, 1992). Criticism of findings from the

Gautreaux quasi-experimental evaluations centered on the inability,

posed by the program design, to identify control groups not tainted by

selection bias (see Duncan & Zuberi).

The MTO program was designed to remedy this selection problem

through random assignment of enrollees to experimental, Section 8, and

control group conditions (see Orr et al., 2003). The program offered

public housing residents and applicants in five cities the opportunity to

enroll in a lottery, with a two-thirds chance of moving to a low-poverty

neighborhood. On average, mobile families relocated from neighbor-

hoods with more than 50% poverty to neighborhoods with 11% and 28%

poverty for the experimental and Section 8 groups, respectively (Orr et

al.). A comprehensive evaluation of achievement outcomes for 5,000 chil-

dren who changed neighborhoods as part of the MTO program found no

1348 Teachers College Record

significant improvements in test scores (Sanbonmatsu et al., 2006). This

finding held even for the youngest children in the program, who were

preschoolers at the time that their families moved.

Findings from Gautreaux and MTO suggest several necessary condi-

tions for changes in context to make a difference for students. The

Gautreaux evaluation clearly suggests that changes in settings must be

substantial to make a difference. The Gautreaux students who demon-

strated better marks and track placements had relocated to suburban

neighborhoods and schools remarkably different from their original

locations. Placements within the MTO program reduced the poverty lev-

els of participant families’ neighborhood settings but often did not

change students’ school settings. Some students continued attending

their inner-city schools of origin,9and others moved to schools with only

modestly lower rates of poverty than their origin schools (Duncan &

Zuberi, 2006). Thus, the MTO findings suggest that relocation programs

that generate only modest contextual improvements or that do not

address school context and quality are unlikely to improve achievement.

Social Context and Achievement in Seasonal Research

Heyns (1978) noted important differences between the test scores of stu-

dents from poor and nonpoor schools10 in her sample of sixth- and sev-

enth graders from Atlanta. She found less social dispersion of

school-season learning rates among students in schools with majorities of

nonpoor students, as well as an interaction between family income and

school context. The reading skills of lower income11 White students grew

at a considerably faster pace when they attended nonpoor schools. Black

students in nonpoor schools also appeared to learn at faster rates than

those in poor schools, but these differences—when comparing students

from equal income groups—were not statistically significant. Heyns cau-

tiously concluded that the nonpoor schools in her sample appeared to be

doing a better job of equalizing learning growth for students from differ-

ent social backgrounds.

Entwisle et al. (1997) found very little evidence of school contextual

effects in the BSS data. Like Heyns, they used schools’ meal subsidy rate

as an indicator of school social context. To capture the effect of school

social context, they computed partial correlations, net of family SES,

between their contextual indicator and achievement growth for reading

and math during each of the 10 periods (over 5 years) in their survey.

These analyses revealed only one instance (out of 10 possible instances)

of a contextual school effect during the school year. The authors con-

cluded that school-year achievement gains were independent of the

Socioeconomic Context and Racial Composition 1349

socioeconomic composition of schools (Entwisle et al.). Paradoxically,

regarding the effects of school contexts, the authors found a cross-season

effect of school meal subsidy rate. For reading growth, higher school

meal subsidy rates were more consistently related to slower summer sea-

son reading achievement growth than the neighborhood poverty rate.12

Regarding neighborhood context, Entwisle et al. (1997) found that

neighborhood median income was positively related to reading achieve-

ment growth during the summer season, net of students’ family SES.

Similarly, neighborhood family poverty rates were negatively related to

achievement growth during the summer season. These associations held

for both factors in two of the three summers analyzed. Entwisle et al. con-

cluded that neighborhood social context exerted a consistent effect on

summer-season learning growth. Neighborhood context, however, was

judged not salient during the school season (Entwisle et al.). Data were

not available in the BSS to examine the relationship between neighbor-

hood context and achievement at school entry.

Racial Composition and Early Childhood Learning

The race/ethnic composition of neighborhoods and schools may affect

achievement through several different causal pathways, and social theo-

rists continue to debate the validity and strength of these pathways.

Perhaps the simplest reason that settings with large proportions of mem-

bers from minority race/ethnic groups can affect achievement is through

the relationship between race/ethnic composition and poverty. The con-

nection between segregation and the concentration of poverty is well

established (see Massey & Fischer, 2000). If neighborhoods or schools

with large proportions of poor (low-SES) residents and attendees slow

achievement growth, then it follows, given the connection between seg-

regation and poverty, that researchers will find connections between the

proportion of minority residents in a setting and achievement.

Segregation may directly affect the availability of resources for learn-

ing. For example, as recently as the 1980s, pupil-teacher ratios were sub-

stantially higher in schools primarily attended by African American

students in the 17 states that had once had de facto systems of school seg-

regation (see Boozer, Krueger, Wolkon, Haltiwanger, & Loury, 1992).

Although there has been an apparent convergence in the relative avail-

ability of such basic learning resources to Black and White students, evi-

dence suggests that this convergence does not extend to more

contemporary needs such as the availability of computers in classrooms

(Boozer et al.). However, resource measures account for only a small pro-

portion of variation in school quality (Rivkin, 2000). Thus, school com-

1350 Teachers College Record

parisons that use standard school-quality measures may mask important

differences in the quality of schools attended by White and non-White

students. This distinction is important because Black students perform

better on achievement tests when they attend schools deemed higher

quality based on their value-added effects for White students (Rivkin).13.

Race/ethnic composition may also affect achievement outcomes

through peer effects. Using fixed-effects models to control for school

quality, Hanushek, Kain, and Rivkin (2002) found that the proportion of

Black students in a school was negatively related to achievement test

scores for Black students but not for White and Hispanic students.

Looking at test scores for elementary school students, Hoxby (2000)

found that classroom proportion Black had the largest negative relation-

ship to reading scores for Black students, although it also depressed

scores for Hispanic and White students. Similarly, classroom proportion

Hispanic was negatively associated with reading scores for Hispanic stu-

dents and had a weaker negative association with reading scores for

White students. Taken together, these findings suggest that increases in

school and classroom proportions of students from specific minority

groups exert the strongest effects on students from within these minority

groups.

LIMITATIONS TO CONSIDERATION OF CONTEXT AND

COMPOSITION IN SEASONAL STUDIES

Despite the instances noted above, seasonal researchers have not fully

taken into account the relationships among socioeconomic context,

race/ethnic composition, and achievement inequality. For the most part,

seasonal researchers have forgone comprehensive inquiries into this area

because of limitations in the scope of data available to them. In addition,

limitations inherent in the analytical tools within statistical software pack-

ages have, until recently, prevented researchers from simultaneously

modeling students’ achievement growth trajectories while taking account

of extrafamilial context and race/ethnic composition. Thus, although

seasonal researchers have demonstrated a keen awareness of the poten-

tial contributions of neighborhood and school context, they have not

made context a central focus of their research agendas.

This article addresses this shortcoming in the extant literature through

use of a unique data set consisting of extensive socioeconomic and

race/ethnic measures from the 2000 Census and the ECLS-K, and

employment of an advanced statistical technique for analyzing the data.

Through this combination of data and analytical technique, we were able

to address several questions previously unaddressed by seasonal

Socioeconomic Context and Racial Composition 1351

researchers. The extensive set of precise measures in the census and

ECLS-K databases meant that the preliminary part of our research pro-

ject could accurately assess the contributions of family-level factors to

reading achievement inequality. The main stage of our project assessed

the analytically distinct contributions of social context and race/ethnic

composition to reading achievement inequality. We wanted to know what

our measures of context and composition could tell us about when and

where achievement inequality accrued between early childhood and the

end of first grade. Were there specific seasons when context was particu-

larly salient for reading achievement? Was neighborhood context most

salient during nonschool periods? Was school context most salient dur-

ing the school year? Was social context or race/ethnic composition more

strongly associated with reading achievement? Finally, did accounting for

context change our appraisal of the relationship between family factors

and achievement?

METHOD

DATA AND ANALYTIC SAMPLE

The sample for our analysis includes 4,180 students, all of whom were

attending one of 290 schools and living in one of 700 neighborhoods dur-

ing the period of our study, which spanned from the beginning of kinder-

garten until the end of first grade. Data on student test scores,

demographic characteristics, and family background came from the

ECLS-K, which was administered by the National Center for Educational

Statistics (NCES, 2002). The ECLS-K provides nationally representative

data for U.S. students who began kindergarten in fall 1998 and for sub-

sequent time points (for this cohort) through fifth grade. The most

intensive sampling took place during the kindergarten and first-grade

years. Data collected during the fall of first grade (hereafter FFG [fall first

grade]) included a survey supplement aimed at allowing researchers to

“measure the extent of summer learning loss and the factors that con-

tribute to such loss and to better disentangle school and home effects on

children’s learning” (NCES, 2002, p. 1-1). In the fall of 1999, NCES

administered achievement test batteries to all students in the kinder-

garten cohort who attended a random subsample of 30% of schools from

the base year survey. A total of 27% of kindergarten cohort students

attended the schools included in this subsample. Because NCES ran-

domly sampled schools rather than students for the FFG data collection,

this subsample includes sufficiently sized clusters of students within

schools for analysis of school effects. Our sample includes an average of

1352 Teachers College Record

14 students per school. The FFG sample served as the foundation for our

analytic sample. Our primary change to the FFG sample was to eliminate

1,060 students who changed schools or neighborhoods during the period

of this study. An additional 330 students were eliminated because they

had no test scores or parent reports available.14

To measure the social and race/ethnic characteristics of neighbor-

hoods, we drew data from the 2000 Census small area files. This data

source, comprising answers to questions in the “long form” of the 2000

decennial census, provides a rich assortment of measures concerning the

demographic and socioeconomic characteristics of neighborhoods.

Because of the inclusive sweep of long-form data collection, these data

are collected with less measurement error than the small-area microdata

collected between decennial censuses. Collection of the long-form data,

which pertain to neighborhood characteristics during 1999, coincided

with the kindergarten and first-grade years of the ECLS-K survey. We

linked the census data to the ECLS-K sample using the zip code of each

student’s home address. When a student’s zip code was not available, we

linked census measures pertaining to the zip code area of the student’s

school to best approximate the student’s neighborhood environment.

MEASURES

Reading Achievement

To measure reading achievement, we relied on information from one-on-

one child assessments conducted with the sampled children during the

fall of 1998, the spring of 1999, the fall of 1999, and the spring of 2000.

These assessments included cognitive, psychomotor, and physical compo-

nents. The reading assessments included questions designed to measure

basic reading skills (i.e., print familiarity, letter recognition, beginning

and ending sounds, rhyming sounds, word recognition), vocabulary

(receptive vocabulary), and reading comprehension skills (i.e., listening

comprehension, words in context). Comprehension items were targeted

to measure skills in initial understanding, interpretation, personal reflec-

tion, and critical stance. The ECLS-K database provides achievement out-

comes expressed as both scale scores and criterion-referenced measures

of students’ skills. We used the scale score measures—derived using an

item response theory (IRT) model—to model achievement growth.

Student-Level Characteristics

To construct the student-level measures for this study, we relied on

Socioeconomic Context and Racial Composition 1353

responses from the questionnaires administered to parents of ECLS-K

students during the fall of 1998 and spring of 1999 base-year parent sur-

veys. When responses were entirely missing from the base year, we

imputed them from the first-grade parent surveys, preferring responses

from the FFG survey and then moving to the spring of first-grade

responses when necessary.

Family SES. We used the measure of family SES computed by NCES as

an average of component variables indicating the guardian parents’ lev-

els of education, occupational statuses, and total household income

(NCES, 2002). Burkam et al. (2004) noted nonlinearities within the

ECLS-K data in the relationship between family SES and summer achieve-

ment, and in preliminary analyses, this pattern was clear to us as well. For

this reason, we employed the categorical (quintile) measures of SES con-

structed by NCES—computed from the original continuous measure.

Rather than computing quintiles within our analytic sample, we pre-

served the nationally representative character of the NCES quintiles.

Thus, because low-SES students were more likely to leave the ECLS-K sur-

vey, the low-SES student category constitutes less than 20% of our sam-

ple, whereas the high-SES category constitutes more than 20% (see Table

A1).

Race/ethnicity. Our race/ethnicity indicators were constructed from par-

ent-reported race/ethnicity information, which included the following

eight categories: White non-Hispanic, Black non-Hispanic, Hispanic,

Asian, Pacific Islander, American Indian, Alaska Native, and multiracial.

We collapsed categories to classify students into five categories: African

American, Hispanic, Asian-Pacific Islander, Native American (including

Alaska natives), and White. Students listed as multiracial were assigned to

the race of their mother.

Additional covariates. To control for students’ family structure, we used

a dummy variable to distinguish between two-biological-parent families

and all others,15 and a continuous variable to account for their number of

siblings. To account for school readiness, we included a measure of stu-

dents’ age (in months) at the beginning of school during the fall of 1998.

Because 4% of students were not beginning school for the first time in

fall 1998, we included a dummy variable that flagged these students who

were repeating kindergarten.

Measures of Context and Composition

We assumed that neighborhood and school contextual characteristics

remained stable during the period of the study. At the neighborhood

level, accounting for changing neighborhood characteristics would have

1354 Teachers College Record

required moving to much less reliable data than the Census 2000 data.

Our examination of school-level data indicated that in fact school char-

acteristics changed very little during kindergarten and first grade. For

these reasons, we feel confident that our contextual measures accurately

represent those neighborhoods and schools inhabited by students during

the course of this study.

Our measures of school context are aggregate, school-average mea-

sures that reflect the SES and race/ethnicity of individual students within

the ECLS-K sample. Our measures of neighborhood context come

directly from neighborhood averages computed by the Bureau of the

Census from responses to the 2000 decennial census long-form surveys.

Because survey respondents were asked to answer questions regarding

their statuses during 1999, our neighborhood measures most accurately

indicate the neighborhood context of students’ families during this year.

The overlap between the timing of the Census 2000 survey and the ECLS-

K study is fortuitous because 1999 was the year during which students

attended the second half of kindergarten, had their summer break, and

attended the first half of first grade.

To facilitate comparisons across neighborhoods and schools, we coded

separate contextual measures using identical procedures. Because the

2000 Census does not provide a detailed occupational scale at the zip

code level,16 we relied on income and education components to construct

our measures of neighborhood and school social context. We began by

determining the median income (in dollars) for each neighborhood and

school in our study. After computing the natural log of each unit’s

median income, we converted these figures to zscores. Next, we ascer-

tained the mean years of education for adults in each neighborhood and

school. Within neighborhoods, we computed the mean years of educa-

tion for neighborhood residents from categorical attainment data for all

male and female adults over the age of 25.18 Within schools, we relied on

parent reports and computed the mean years of education based on

available categorical educational attainment measures for each child’s

parents (or parent)17 and then converted these means to zscores. Next,

we computed continuous measures of neighborhood and school SES as

the arithmetic averages of the two component zscores. Preliminary analy-

ses using the continuous social context measures yielded two fundamen-

tal insights: (a) The relationship between social context and achievement

was not linear, and (b) the strongest relationships between social context

and achievement were located at the tails of the socioeconomic context

distribution. Thus, using the continuous social context measures, we

divided the 700 neighborhoods and 290 schools into equally sized socio-

economic quintiles. Because contextual effects were concentrated in the

Socioeconomic Context and Racial Composition 1355

highest and lowest socioeconomic categories, we collapsed the middle

three categories into one for the regression analysis.19

To account for the race/ethnic composition of neighborhoods and

schools, we again constructed identical measures for the two contexts. We

summed the numbers of residents from all minority groups within neigh-

borhoods, and the numbers of students from all minority groups within

schools, and divided these figures by the unit populations. In essence,

these measures captured the proportions of residents and students other

than non-Hispanic Whites.20

ANALYTICAL STRATEGY

The main methodological innovations in this study are the linking of

2000 Census data to indicate the socioeconomic character of neighbor-

hoods (see the preceding section), and the application of a third model-

ing level to measure the relationships between neighborhood and school

context, and student achievement. The modeling strategy used for this

study is quite similar to that used in other studies that have assessed

achievement growth for ECLS-K students over the first 2 years of elemen-

tary school (see Downey et al., 2004, and Reardon, 2003). In the termi-

nology of hierarchical linear modeling (HLM), the models employed in

this study are random intercept models (Raudenbush & Bryk, 2002). In

contrast to fixed-effects modeling strategies, the random coefficients

approach facilitates explaining variation in the outcome across units. We

employed that option in our models to explain variation in achievement

across two sets of contextual units—neighborhoods and schools. Before

discussing our modeling strategy for assessing the relationship between

context and achievement, we introduce the functions of level 1 and level

2 in our three-level growth models.

The Level 1 Model

Level 1 is a time-based model that employs repeated within-student obser-

vations to accomplish two main objectives. This level partitions achieve-

ment during the study period into a school-entry (intercept) parameter

and three growth parameters for subsequent seasons. The level 1 model

accommodates substantial season-to-season differences in reading

achievement growth by separately computing growth rates for kinder-

garten, summer, and first grade. Unique season-specific growth parame-

ters represent growth rates for each of these three periods. A seasonal

time variable activates each seasonal growth parameter, indicating the

number of months that each individual student had spent in the corre-

1356 Teachers College Record

sponding season, as of the date of each test administration. To account

for four separate test administrations, the level 1 model contains a sepa-

rate observation for each date at which a student was tested.21 Averages

for students’ time spent in each season are presented in panel 3 of Table

1. These measures account for temporal misalignments between the

school calendar and test administration dates. For a detailed description

of time accounting within the level 1 model, please see the methodolog-

ical appendix.

Table 1. Mean Reading Achievement at School Entry and Mean Growth Rates in Subsequent Seasons

(reference category), and Deviations from Means and Mean Growth Rates (others); Duration (in

Months) of Seasons as of Each Test Administration

Student (n) School Entry K Summer First

All 4,180 19.410*** +1.669*** -0.016 +2.527***

(standard errors) (0.291) (0.028) (0.050) (0.032)

Family SES

Low 670 -3.424*** -0.188** -0.248* -0.262***

(SE) (0.352) (0.056) (0.121) (0.062)

Lower middle 790 -1.011* -0.052 +0.052 -0.074

(SE) (0.392) (0.054) (0.109) (0.052)

Middle (reference) 850 18.687*** +1.681*** -0.060 +2.588***

(SE) (0.297) (0.045) (0.083) (0.044)

Upper middle 930 +1.864*** +0.038 +0.157 +0.028

(SE) (0.405) (0.054) (0.106) (0.052)

High 950 +5.138*** +0.110* +0.203 -0.037

(SE) (0.513) (0.052) (0.112) (0.062)

Race/Ethnicity

White (reference) 2,520 19.959*** +1.743*** -0.072 +2.594***

(SE) (0.309) (0.032) (0.060) (0.036)

African American 600 -1.370** -0.273*** +0.159 -0.221**

(SE) (0.503) (0.057) (0.117) (0.062)

Hispanic 330 -2.278*** -0.187** +0.070 -0.136*

(SE) (0.486) (0.056) (0.117) (0.060)

Asian/Pacific Islander 620 +1.839** -0.022 +0.449** -0.139

(SE) (0.611) (0.070) (0.148) (0.075)

Native American 110 -3.969** -0.088 -0.308 -0.045

(SE) (1.182) (0.133) (0.258) (0.147)

Season Duration at Test

Fall kindergarten 3,580 n/a 2.189 00

(standard deviations) (0.527)

Spring kindergarten 4,040 n/a 8.338 00

(SD) (0.530)

Fall first grade 4,100 n/a 9.389 2.611 1.394

(SD) (0.325) (0.325) (0.501)

Spring first grade 4,150 n/a 9.397 2.603 8.219

(SD) (0.349) (0.349) (0.555)

Note. Estimates from models computed for students nested within schools.

*p< 0.05. **p< 0.01. *** p< 0.001.

Socioeconomic Context and Racial Composition 1357

The Level 2 Model

The level 2 model resembles a traditional sociodemographic model of an

educational outcome and has two primary, interrelated functions. For

one, it accurately measures the multivariate associations among family

SES, student race/ethnicity, and reading achievement. Equally as impor-

tant, it computes unit-average (neighborhood or school) estimates of

achievement at school entry and during the three subsequent seasons.

This second objective requires identical22 level 2 models for each of the

four periods during which we estimated growth parameters. Within each

period, the intercept represents unit-average achievement level (at

school entry) or unit-average achievement growth (during kindergarten,

summer, and first grade). Parameter estimates for the social background

and demographic variables capture the associations between these vari-

ables and achievement during each specified period. Thus, our models

separately estimate unit-average achievement and covariate effects dur-

ing each study period. For a more detailed description of the level 2

model, please see the methodological appendix.

The Level 3 Model

The level 3 model accounts for the extent of achievement variability

across neighborhoods and schools, net of the student-level covariates pre-

sent in level 2. The model accomplishes this objective by including ran-

dom parameters for each level 2 intercept, each of which captures the

extent to which neighborhoods and schools departed from the unit-aver-

age achievement estimates computed in the level 2 portion of the model.

The presence of these unit-specific random parameters at level 3 (a)

accounted for the clustered nature of data within specific neighborhoods

and schools, (b) allowed for an accurate decomposition of variance

between the individual (level 2) and contextual (level 3) portions of the

overall model, and (c) reduced the possibility of bias in our parameter

estimates at both levels. Model 1 is a simple model that includes only the

random parameters—and no predictors—at level 3. Our subsequent

level 3 models represent variability in the level 2 intercepts as a function

of the socioeconomic context and race/ethnic composition of neighbor-

hoods and schools. Thus, Models 2–4 introduce contextual SES and

race/ethnic composition measures and assess the sizes of fixed effects

associated with these measures. Because of the careful partitioning of

variance into appropriate levels of analysis, these models produce accu-

rate estimates of the relationship between contextual measures and

1358 Teachers College Record

achievement. For a more detailed description of the level 3 model, please

see the methodological appendix.

FOUR EXPLANATORY MODELS

Preliminary Models: Unconditional, SES Only, and Race/Ethnicity Only

We began by computing an unconditional three-level model with no pre-

dictors other than the seasonal time variables at level 1. This model

allowed us to calculate a school-year average achievement growth rate

while making the necessary corrections for test-scheduling misalign-

ments. The seasonal averages from our unconditional model are pre-

sented in row 1 of Table 1. Next, we computed a three-level SES-only

model that included only the seasonal time measures and the four fam-

ily-SES quintile dummy variables. As in all subsequent models, the mid-

dle-SES quintile served as the reference categor y. Without exploring

causes, this model demonstrates the full extent of socioeconomic

achievement inequality during each of the four periods of our study, for

students nested within schools.23 The results from this model are pre-

sented in rows 2–6 of Table 1. The final preliminary model was a three-

level race/ethnicity-only model that included the seasonal time variables

and four categorical race/ethnic dummy variables. As in all subsequent

models, White race/ethnicity served as the excluded categor y. Results

from this model depict how race/ethnic achievement inequality looks

before accounting for socioeconomic status and demographic factors.

The results from this model are included in rows 7–11 of Table 1.

Model 1: Sociodemographic Model

Model 1 includes our full set of sociodemographic variables: the family

SES and race/ethnicity dummy variables (see preceding text); two family

structure measures (biological two-parent family and number of sib-

lings); two demographic measures indicating the child’s gender and age

at the beginning of school in fall 1998; and a final measure indicating

whether the child was repeating kindergarten as of fall 1998. This model

facilitates examining socioeconomic inequality in reading achievement

net of family factors that correlate with SES. In addition, race/ethnic

achievement inequality can be observed net of socioeconomic and fam-

ily factors that often correlate with race/ethnicity. Because level 1 of the

model estimates an intercept and separate growth parameters for each

of the three subsequent seasons, the four subsections of Model 1

portray how the relationship between sociodemographic factors and

Socioeconomic Context and Racial Composition 1359

achievement differs over the four periods included in this study.

Although Model 1 did not include contextual measures, it did include

random parameters that captured unit-to-unit variations among neigh-

borhood and school contexts. Separate versions of Model 1 were esti-

mated for students nested within neighborhoods and schools, and we

present these models in column 1 of Tables 2 and 3.

Model 2: Race/Ethnic Composition

Model 2 adds the measure for proportion minority to the full vector of

sociodemographic measures included in Model 1. To control for partici-

pation in academic programs that may be correlated with school or

neighborhood race/ethnic composition, Model 2 also includes a mea-

sure indicating full-day kindergarten and a measure indicating whether

the student attended an academic summer school.24 The students-within-

neighborhoods model (Table 2) differs from the students-within-schools

model (Table 3) because the full-day kindergarten measure is entered at

the student level (level 2) rather than the contextual level (level 3).

Model 2 facilitates a clear appraisal of the association between

race/ethnic composition and achievement before accounting for the

relationship between social context and achievement. Separate versions

of Model 2 were estimated for students within neighborhoods and

schools, and these models are presented in column 2 of Table 2 and

Table 3.

Model 3: Social Context

Model 3 adds measures for low- and high-SES contexts to the full vector

of sociodemographic measures included in Model 1 and the program

participation measures included in Model 2. This model assesses not only

the disadvantages associated with low-SES neighborhoods and schools

but also the advantages associated with high-SES contexts. Separate ver-

sions of Model 3 for students within neighborhoods and schools are pre-

sented in column 3 of Tables 2 and 3.

Model 4: Social Context and Race/ethnic Composition

Model 4 is the full model, which adds the measure of race/ethnic com-

position—included in Model 2—to the social context measures included

in Model 3. This model effectively assesses the relative strength of social

context and race/ethnic composition as influences on achievement

inequality across seasons. Model 4 is a barometer for determining

1360 Teachers College Record

Table 2. Three-Level Models of Reading Achievement Growth for Students Nested Within Neighborhoods, by Season (SE), N=4180 students; 700 neighborhoods.

Model 1 Model 2 Model 3 Model 4

Variable name Coefficient SE Coefficient SE Coefficient SE Coefficient SE

Initial status (intercept) 19.998*** (0.389) 20.070*** (0.397) 19.772*** (0.398) 19.629*** (0.405)

Neighborhood SES – low -0.906* (0.418) -1.110* (0.456)

Neighborhood SES – high 2.110*** (0.419) 2.191*** (0.420)

Proportion minority -0.235 (0.705) 0.981 (0.767)

Low SES -3.339*** (0.358) -3.308*** (0.358) -3.098*** (0.368) -3.059*** (0.368)

Lower middle SES -1.215** (0.379) -1.194** (0.378) -1.043** (0.379) -1.026** (0.381)

Upper middle SES 2.055*** (0.396) 2.039*** (0.395) 1.826*** (0.408) 1.816*** (0.409)

High SES 5.555*** (0.521) 5.481*** (0.519) 5.181*** (0.519) 5.104*** (0.520)

African American 0.186 (0.406) 0.235 (0.480) 0.491 (0.410) 0.167 (0.478)

Hispanic -1.326** (0.467) -1.300** (0.492) -1.055** (0.463) -1.328** (0.492)

Asian or Pacific Islander 2.273** (0.765) 2.336** (0.740) 2.313** (0.772) 2.100** (0.740)

Native American -3.633*** (0.671) -3.579*** (0.740) -3.057*** (0.656) -3.468*** (0.725)

Biological two-parent family 1.282*** (0.268) 1.283*** (0.267) 1.180*** (0.267) 1.184*** (0.267)

Number of siblings -0.830*** (0.116) -0.829*** (0.116) -0.817*** (0.116) -0.822*** (0.116)

Gender (male = 1) -1.662*** (0.256) -1.670*** (0.256) -1.672*** (0.256) -1.676*** (0.256)

Age, school entry (fall 1998) 0.333*** (0.035) 0.332*** (0.035) 0.342*** (0.035) 0.346*** (0.035)

Repeating kindergarten 0.907 (0.676) 0.922 (0.677) 0.822 (0.675) 0.757 (0.680)

Kindergarten slope 1.787*** (0.054) 1.685*** (0.060) 1.706*** (0.062) 1.733*** (0.063)

Neighborhood SES – low -0.009 (0.058) 0.030 (0.062)

Neighborhood SES – high -0.175** (0.058) -0.189** (0.058)

Proportion minority -0.088 (0.086) -0.160 (0.093)

Full-day kindergarten (level 2) 0.209*** (0.041) 0.196* (0.041) 0.202*** (0.042)

Low SES -0.124* (0.058) -0.122* (0.058) -0.128* (0.059) -0.133* (0.059)

Lower middle SES -0.048 (0.054) -0.048 (0.053) -0.055 (0.054) -0.057 (0.054)

Upper middle SES 0.003 (0.052) 0.005 (0.052) 0.019 (0.052) 0.020 (0.052)

High SES 0.064 (0.052) 0.058 (0.051) 0.088 (0.052) 0.093 (0.052)

African American -0.176** (0.052) -0.196** (0.058) -0.203*** (0.052) -0.192** (0.058)

Socioeconomic Context and Racial Composition 1361

Hispanic -0.106 (0.056) -0.092 (0.058) -0.122 (0.055) -0.090 (0.057)

Asian or Pacific Islander 0.041 (0.078) 0.046 (0.078) 0.037 (0.077) 0.063 (0.078)

Native American -0.054 (0.073) -0.069 (0.079) -0.100 (0.080) -0.057 (0.082)

Biological two-parent family 0.108** (0.034) 0.112** (0.034) 0.119** (0.034) 0.120** (0.034)

Number of siblings -0.024 (0.014) -0.022 (0.014) -0.023 (0.014) -0.022 (0.014)

Gender (male = 1) -0.112** (0.034) -0.107** (0.034) -0.107** (0.034) -0.106** (0.034)

Age, school entry (fall 1998) 0.007 (0.004) 0.006 (0.004) 0.005 (0.004) 0.005 (0.004)

Repeating kindergarten -0.373*** (0.083) -0.395*** (0.081) -0.397*** (0.082) -0.389*** (0.082)

Summer slope -0.046 (0.108) -0.064 (0.109) -0.088 (0.110) -0.134 (0.108)

Neighborhood SES – low -0.066 (0.123) -0.125 (0.136)

Neighborhood SES – high 0.272* (0.126) 0.295* (0.126)

Proportion minority 0.054 (0.190) 0.221 (0.212)

Low SES -0.263* (0.121) -0.261* (0.121) -0.243* (0.122) -0.238+ (0.122)

Lower middle SES 0.038 (0.111) 0.040 (0.111) 0.053 (0.111) 0.057 (0.111)

Upper middle SES 0.180 (0.107) 0.183 (0.107) 0.161 (0.109) 0.161 (0.109)

High SES 0.249* (0.113) 0.251* (0.113) 0.200 (0.117) 0.194 (0.116)

African American 0.249* (0.112) 0.234 (0.128) 0.272* (0.114) 0.225 (0.127)

Hispanic 0.124 (0.130) 0.115 (0.135) 0.148 (0.129) 0.108 (0.135)

Asian or Pacific Islander 0.505** (0.164) 0.497** (0.168) 0.496** (0.165) 0.466** (0.169)

Native American -0.276* (0.127) -0.298 (0.160) -0.240 (0.148) -0.292 (0.166)

Biological two-parent family 0.032 (0.080) 0.031 (0.080) 0.020 (0.080) 0.019 (0.080)

Number of siblings -0.057 (0.030) -0.057 (0.030) -0.054 (0.030) -0.055 (0.030)

Gender (male = 1) -0.071 (0.075) -0.072 (0.075) -0.072 (0.074) -0.073 (0.075)

Age, school entry (fall 1998) 0.020* (0.009) 0.020* (0.009) 0.020* (0.008) 0.021* (0.009)

Repeated kindergarten -0.237 (0.184) -0.251 (0.184) -0.253 (0.183) -0.268 (0.183)

Academic summer school 0.091 (0.110) 0.088 (0.110) 0.092 (0.110)

1362 Teachers College Record

Table 2. Three-Level Models of Reading Achievement Growth for Students Nested Within Neighborhoods, by Season (SE) (continued)

Model 1 Model 2 Model 3 Model 4

Variable name Coefficient SE Coefficient SE Coefficient SE Coefficient SE

First-grade slope 2.621*** (0.059) 2.688*** (0.062) 2.625*** (0.060) 2.686*** (0.062)

Neighborhood SES – low -0.059 (0.066) 0.024 (0.072)

Neighborhood SES – high 0.036 (0.066) 0.006 (0.065)

Proportion minority -0.304** (0.091) -0.314** (0.100)

Low SES -0.244*** (0.064) -0.237*** (0.064) -0.236*** (0.064) -0.241*** (0.064)

Lower middle SES -0.048 (0.053) -0.048 (0.053) -0.045 (0.053) -0.050 (0.053)

Upper middle SES 0.044 (0.055) 0.042 (0.055) 0.042 (0.056) 0.042 (0.056)

High SES -0.021 (0.060) -0.021 (0.060) -0.029 (0.063) -0.020 (0.063)

African American -0.278*** (0.066) -0.202** (0.070) -0.269*** (0.066) -0.200** (0.070)

Hispanic -0.136* (0.061) -0.062 (0.065) -0.125 (0.063) -0.062 (0.065)

Asian or Pacific Islander -0.225* (0.079) -0.180* (0.082) -0.223** (0.079) -0.176* (0.082)

Native American -0.072 (0.125) 0.045 (0.126) -0.042 (0.133) 0.038 (0.129)

Biological two-parent family 0.051 (0.039) 0.049 (0.039) 0.049 (0.040) 0.050 (0.040)

Number of siblings 0.013 (0.016) 0.015 (0.016) 0.013 (0.016) 0.015 (0.016)

Gender (male = 1) -0.028 (0.036) -0.027 (0.036) -0.028 (0.036) -0.027 (0.036)

Age, school entry (fall 1998) -0.016*** (0.004) -0.017*** (0.004) -0.016*** (0.004) -0.017*** (0.004)

Repeated kindergarten -0.178 (0.111) -0.165 (0.112) -0.181 (0.111) -0.161 (0.112)

Random effects Estimate 2Estimate 2Estimate 2Estimate 2

Level 1and 2 variance (df) (1837) (1842) (1830) (1830)

Initial status 49.757 11427.3*** 49.780 11485.1*** 49.988 11347.4*** 49.964 11409.2***

Kindergarten slope 0.638 6526.2*** 0.635 6513.8*** 0.636 6482.0*** 0.635 6499.9***

Summer slope 2.286 4251.5*** 2.283 4261.3*** 2.283 4228.8*** 2.284 4244.8***

First-grade slope 1.008 10312.6*** 1.009 10325.3*** 1.008 10280.2*** 1.010 10313.4***

Level 3 variance (df) (275) (276) (275) (278)

Initial status 2.322 314.2 2.428 314.7 1.783 301.2 1.844 299.3

Kindergarten slope 0.106 494.8*** 0.094 496.9*** 0.086 481.0*** 0.086 474.5***

Summer slope 0.296 401.8*** 0.293 403.5** 0.282 386.1*** 0.277 391.1***

First-grade slope 0.108 491.2*** 0.100 475.5*** 0.108 489.6*** 0.100 467.7***

Deviance (parameters) 105632 (81) 105591 (87) 105568 (91) 105554 (95)

***P-value < 0.001. **P-value < 0.01. *P-value < 0.05. +P-value < 0.055. All estimates computed using HLM 6.06.

Socioeconomic Context and Racial Composition 1363

Table 3. Three-Level Models of Reading Achievement Growth for Students Nested Within Schools, by Season (SE), N=4180 students; 290 schools.

Model 1 Model 2 Model 3 Model 4

Variable name Coefficient SE Coefficient SE Coefficient SE Coefficient SE

Initial status (intercept) 19.768*** (0.380) 20.307*** (0.410) 19.580*** (0.402) 19.750*** (0.437)

School SES – low -1.672** (0.470) -1.448** (0.533)

School SES – high 2.404*** (0.460) 2.338*** (0.462)

Proportion minority -1.819* (0.693) -0.573 (0.788)

Low SES -3.138*** (0.363) -3.056*** (0.366) -2.740*** (0.380) -2.746*** (0.380)

Lower middle SES -1.022** (0.380) -0.981** (0.375) -0.854* (0.379) -0.854* (0.378)

Upper middle SES 1.902*** (0.398) 1.882*** (0.395) 1.619*** (0.398) 1.622*** (0.398)

High SES 4.734*** (0.500) 4.700*** (0.499) 4.190*** (0.508) 4.192*** (0.508)

African American 0.307 (0.454) 0.825 (0.536) 0.690 (0.482) 0.792 (0.541)

Hispanic -1.312** (0.474) -0.868 (0.530) -1.019* (0.470) -0.927 (0.534)

Asian or Pacific Islander 2.041** (0.764) 2.438** (0.764) 2.152** (0.766) 2.242** (0.763)

Native American -3.256*** (0.729) -2.582** (0.771) -2.399** (0.742) -2.304** (0.782)

Biological two-parent family 1.319*** (0.288) 1.286*** (0.286) 1.216*** (0.286) 1.212*** (0.285)

Number of siblings -0.798*** (0.116) -0.790*** (0.116) -0.771*** (0.116) -0.773*** (0.116)

Gender (male = 1) -1.608*** (0.256) -1.597*** (0.256) -1.601*** (0.256) -1.598*** (0.256)

Age, school entry (fall 1998) 0.326*** (0.037) 0.320*** (0.037) 0.325*** (0.037) 0.324*** (0.038)

Repeating kindergarten 0.970 (0.636) 1.029 (0.638) 0.916 (0.657) 0.928 (0.659)

Kindergarten slope 1.749*** (0.056) 1.636*** (0.062) 1.706*** (0.060) 1.677*** (0.066)

School SES – low -0.130 (0.066) -0.163* (0.073)

School SES – high -0.198** (0.072) -0.188* (0.073)

Proportion minority 0.039 (0.093) 0.089 (0.104)

Full-day kindergarten 0.189*** (0.051) 0.170** (0.049) 0.175** (0.050)

Low SES -0.113* (0.057) -0.114* (0.057) -0.106 (0.058) -0.105 (0.058)

Lower middle SES -0.032 (0.054) -0.034 (0.054) -0.034 (0.054) -0.034 (0.054)

Upper middle SES 0.025 (0.054) 0.025 (0.054) 0.037 (0.054) 0.036 (0.054)

High SES 0.091 (0.053) 0.088 (0.053) 0.124* (0.054) 0.123* (0.054)

1364 Teachers College Record

Table 3. Three-Level Models of Reading Achievement Growth for Students Nested Within Schools, by Season (SE) (continued)

Model 1 Model 2 Model 3 Model 4

Variable name Coefficient SE Coefficient SE Coefficient SE Coefficient SE

African American -0.206*** (0.056) -0.246*** (0.064) -0.226*** (0.058) -0.245*** (0.064)

Hispanic -0.126* (0.056) -0.137* (0.060) -0.117* (0.055) -0.132* (0.060)

Asian or Pacific Islander 0.008 (0.081) -0.005 (0.083) 0.007 (0.080) -0.008 (0.084)

Native American -0.031 (0.080) -0.088 (0.090) -0.043 (0.083) -0.062 (0.089)

Biological two-parent family 0.106** (0.035) 0.109** (0.035) 0.112** (0.035) 0.114** (0.035)

Number of siblings -0.013 (0.014) -0.012 (0.014) -0.012 (0.014) -0.011 (0.014)

Gender (male = 1) -0.122*** (0.034) -0.121** (0.034) -0.122** (0.034) -0.122** (0.034)

Age, school entry (fall 1998) 0.004 (0.004) 0.004 (0.004) 0.004 (0.004) 0.004 (0.004)

Repeating kindergarten -0.344*** (0.085) -0.359*** (0.083) -0.344*** (0.084) -0.348*** (0.084)

Summer slope -0.032 (0.108) 0.050 (0.112) -0.060 (0.110) 0.014 (0.115)

School SES – low -0.101 0.147 -0.001 (0.164)

School SES – high 0.174 0.138 0.147 (0.138)

Proportion minority -0.340 (0.184) -0.304 (0.206)

Low SES -0.270* (0.125) -0.253* (0.125) -0.241+ (0.125) -0.246+ (0.126)

Lower middle SES 0.040 (0.110) 0.049 (0.109) 0.052 (0.110) 0.052 (0.110)

Upper middle SES 0.160 (0.104) 0.160 (0.104) 0.147 (0.106) 0.150 (0.106)

High SES 0.215+ (0.111) 0.211 (0.111) 0.177 (0.118) 0.183 (0.118)

African American 0.252* (0.119) 0.365* (0.135) 0.271* (0.124) 0.363** (0.135)

Hispanic 0.148 (0.133) 0.253 (0.140) 0.166 (0.130) 0.248 (0.141)

Asian or Pacific Islander 0.489** (0.154) 0.578*** (0.157) 0.488** (0.153) 0.572*** (0.157)

Native American -0.238 (0.137) -0.091 (0.164) -0.179 (0.152) -0.088 (0.167)

Biological two-parent family 0.015 (0.078) 0.006 (0.079) 0.007 (0.078) 0.002 (0.079)

Number of siblings -0.053 (0.031) -0.052 (0.031) -0.051 (0.031) -0.052 (0.031)

Gender (male = 1) -0.074 (0.074) -0.072 (0.074) -0.074 (0.074) -0.072 (0.074)

Age, school entry (fall 1998) 0.017* (0.008) 0.016* (0.008) 0.017* (0.008) 0.016* (0.008)

Repeated kindergarten -0.228 (0.173) -0.227 (0.172) -0.241 (0.172) -0.233 (0.172)

Academic summer school 0.100 (0.106) 0.083 (0.106) 0.101 (0.106)

Socioeconomic Context and Racial Composition 1365

Table 3. Three-Level Models of Reading Achievement Growth for Students Nested Within Schools, by Season (SE) (continued)

First-grade slope 2.635*** (0.057) 2.780*** (0.063) 2.610*** (0.060) 2.725*** (0.067)

School SES – low -0.144 (0.078) 0.010 (0.088)

School SES – high 0.245* (0.084) 0.203* (0.082)

Proportion minority -0.461*** (0.088) -0.410*** (0.101)

Low SES -0.221** (0.064) -0.201** (0.063) -0.198** (0.064) -0.202** (0.064)

Lower middle SES -0.058 (0.052) -0.051 (0.051) -0.049 (0.051) -0.049 (0.051)

Upper middle SES 0.013 (0.051) 0.009 (0.051) -0.001 (0.052) 0.001 (0.052)

High SES -0.035 (0.061) -0.043 (0.061) -0.085 (0.060) -0.080 (0.060)

African American -0.209** (0.067) -0.103 (0.070) -0.186** (0.067) -0.105 (0.070)

Hispanic -0.096 (0.064) 0.004 (0.066) -0.074 (0.064) 0.001 (0.067)

Asian or Pacific Islander -0.126 (0.074) -0.043 (0.074) -0.120 (0.073) -0.046 (0.075)

Native American -0.020 (0.128) 0.123 (0.123) 0.033 (0.127) 0.114 (0.126)

Biological two-parent family 0.041 (0.039) 0.035 (0.039) 0.034 (0.039) 0.031 (0.039)

Number of siblings 0.002 (0.014) 0.002 (0.014) 0.002 (0.014) 0.002 (0.014)

Gender (male = 1) -0.007 (0.036) -0.004 (0.036) -0.007 (0.036) -0.004 (0.036)

Age, school entry (fall 1998) -0.013** (0.005) -0.014** (0.005) -0.013** (0.004) -0.014** (0.004)

Repeated kindergarten -0.248* (0.101) -0.238* (0.101) -0.254* (0.100) -0.225* (0.099)

Random effects Estimate 2Estimate 2Estimate 2Estimate 2

Level 1and 2 variance (df) (2362) (2364) (2360) (2362)

Initial status 50.916 14912.2*** 50.888 14896.9*** 50.786 14875.3*** 50.790 14874.6***

Kindergarten slope 0.596 7855.1*** 0.595 7853.2*** 0.595 7852.2*** 0.595 7851.5***

Summer slope 2.242 5119.3*** 2.238 5112.9*** 2.239 5115.7*** 2.238 5111.8***

First-grade slope 0.930 11072.6*** 0.929 11062.0*** 0.929 11067.2*** 0.928 11060.9***

Level 3 variance (df) (206) (204) (205) (203)

Initial status 3.588 324.8*** 3.389 328.1*** 2.942 312.9*** 2.916 313.1***

Kindergarten slope 0.143 663.7*** 0.129 633.6*** 0.121 605.8*** 0.120 601.5***

Summer slope 0.333 400.9*** 0.323 397.2*** 0.324 392.9*** 0.319 392.2***

First-grade slope 0.155 531.6*** 0.141 480.8*** 0.141 502.5*** 0.134 479.8***

Deviance (parameters) 105250 (87) 105192 (93) 105168 (97) 105144 (1010)

***P-value < 0.001. **P-value < 0.01. *P-value < 0.05. +P-value < 0.055. All estimates computed using HLM 6.06.

1366 Teachers College Record

whether significant relationships observed in Models 2 and 3 persist

when considering social context and race/ethnic composition together

in the same model.

INTERPRETING COEFFICIENTS IN THREE-LEVEL GROWTH MODELS

We employed two metrics for interpreting results from the growth mod-

els. When considering reading achievement inequality at school entry, we

relied on standard deviations in scale score points for the test adminis-

tered in fall of kindergarten. As indicated in Table A1, this standard devi-

ation was 9.074. When considering growth rates in subsequent seasons,

we interpreted significant coefficients and gaps (between coefficients) in

relationship to an average monthly rate of achievement growth during

the school year. Averaging the school-year growth rates from the uncon-

ditional model (Table 1, row 1) produced a monthly school-year growth

average of 2.1 points per month. We found it helpful to evaluate seasonal

gaps in reading growth in relationship to this average. The regression

models described previously easily facilitate such calculations and com-

parisons. Because 1 month is the unit of time at level 1 (in all models),

all seasonal growth slopes (presented at the top of each seasonal sub-

model) represent monthly growth rates, and all growth coefficients (asso-

ciated with the model covariates) represent monthly changes to the

growth slopes.

RESULTS

FAMILY SES, RACE/ETHNICITY, AND READING ACHIEVEMENT AT

SCHOOL ENTRY

Nowhere was the relationship between family SES and reading achieve-

ment more apparent than at school entry. Students from low-SES families

entered school roughly one standard deviation below their counterparts

from high-SES families. The initial status portion of Model 1 (Table 2)

shows the significant relationship between family SES and achievement at

school entry. Using the high- and low-SES coefficients from Model 1, we

computed a social achievement gap of 0.98 standard deviation units,

(5.555 – (-3.339) / 9.074.

African American students did not begin school behind White students

from similar family backgrounds. When controlling for family SES, as we

did in this and all subsequent models, we found no evidence that African

American race/ethnicity was associated with children’s reading achieve-

ment at school entry.25 The absence of a Black-White gap in reading

Socioeconomic Context and Racial Composition 1367

achievement at school entry, net of socioeconomic and family character-

istics, is consistent with other research that has explored Black-White

achievement differences among students in the ECLS-K (see Fryer &

Levitt, 2004). Hispanic race/ethnicity, however, was associated with a

somewhat lower level of reading achievement at school entry. Using the

Hispanic coefficient from Model 1, Table 2 (-1.326), the Hispanic-White

gap amounted to -0.15 standard deviation units.26

CONTEXT, COMPOSITION, AND READING ACHIEVEMENT AT

SCHOOL ENTRY

After accounting for family-level differences in reading achievement, our

measures of neighborhood social context explained a substantial addi-

tional portion of variation in reading scores at school entry. Most notable

was the positive increment associated with residence in a high-SES neigh-

borhood. Using the coefficient for high-SES neighborhood in Model 3

(2.110), we computed a contextual advantage of 0.232 standard deviation

units for students growing up in high-SES neighborhoods. The contex-

tual disadvantage for students growing up in low-SES neighborhoods

amounted to a smaller figure of 0.10 standard deviation units.27One way

of interpreting the size of these effects is to compute a total contextual

achievement gap and compare it with the size of the social gap for family

SES (see preceding text). The contextual achievement gap totals 0.33

standard deviation units, 2.110 – (-0.906) / 9.074, roughly one third the

size of the social gap (0.98 SD units).

There was no apparent relationship between the proportion of resi-

dents from minority race/ethnic groups in neighborhoods and reading

achievement at school entry. Model 2 presented the strongest test of the

hypothesis that such a relationship existed, and the coefficient for pro-

portion minority (-0.235; SE = 0.705) did not approach statistical signifi-

cance. Our findings are consistent with the neighborhood study of early

childhood learning by Brooks-Gunn et al. (1993), who found larger

effect sizes for affluent than poor neighborhoods and no additional dele-

terious effects for largely minority neighborhoods.

FAMILY SES, RACE/ETHNICITY AND SUMMER SEASON ACHIEVE-

MENT

Family SES was clearly associated with reading achievement growth dur-

ing the summer season. On average, students from middle-SES families

did not grow in reading achievement during this period; they simply

maintained their end-of-kindergarten reading level until the beginning

1368 Teachers College Record

of first grade. In contrast, students from high-SES families continued to

grow, whereas students from low-SES families lost ground. The third sec-

tion of Model 1 (Table 2) presents coefficients for summer reading

achievement growth. The family-SES coefficients for the summer growth

slope indicate that high-SES students grew 0.25 points per month,

whereas low-SES students lost 0.26 points per month of reading achieve-

ment. To interpret these results, we computed a social achievement gap

between high- and low-SES students and compared it with our average

school year reading growth rate of 2.1 points per month (see preceding

text). The social achievement gap of 0.51 points per month, 0.249 – (-

0.263), summed to 1.35 points over the course of an average summer (2.6

months), resulting in a noticeable difference of 0.64 months of school-

year growth between low- and high-SES students. In contrast, we found

that lower middle SES students did not differ from the static growth tra-

jectory of middle-SES students. Upper middle SES students appeared to

grow at a slightly faster pace (0.18 points per month), but their coeffi-

cient was not statistically significant. Thus, we conclude that differences

in summer-season achievement were concentrated among students with

families whose SES differed significantly from that of middle-SES stu-

dents, a finding consonant with prior summer research using the ECLS-

K (see Burkam et al., 2004).

Race/ethnicity was related to summer reading achievement growth in

an unexpected way. Controlling for a host of sociodemographic factors,

as in all the explanatory models, we found that African American stu-

dents learned at a faster pace than (similar) White students during the

summer. Model 1 (Table 2) indicates a monthly summer learning growth

increment of 0.249 points for African American students relative to

White students. The size of this coefficient is slightly larger than the neg-

ative monthly Black-White gap of roughly 0.2 points that persisted

throughout kindergarten and first grade. However, over the course of an

average summer, this monthly increment summed to 0.31 months of

school-year reading growth, less than one quarter the size of the Black-

White gap that accumulated during the school year (see the section titled

“Family SES, Race/Ethnicity, and School-Year Achievement”).

CONTEXT, COMPOSITION, AND SUMMER READING ACHIEVEMENT

Neighborhood social context was salient during the summer season,

whereas neighborhood race/ethnic composition was not. The associa-

tion between social context and achievement was concentrated in high-

SES neighborhoods. As indicated in the summer portion of Model 3

(Table 2), students residing in high-SES neighborhoods grew in reading

Socioeconomic Context and Racial Composition 1369

achievement by an additional 0.272 points per month (relative to middle-

SES students). Over the course of an average summer, this monthly incre-

ment summed to 0.34 months of school-year reading growth. The

coefficient for low-SES neighborhoods was negative, and the coefficient

for the proportion minority residents was positive, but neither

approached statistical significance. Thus, high-SES neighborhoods were

the only context significantly related to summer season achievement

growth.

Neighborhood social context was linked to the summer season reading

growth patterns for students from high-SES families. Comparing across

Models 1 and 3 (Table 2), we found that the coefficient for high SES fell

in size from 0.249 to 0.200 points per month, thereby dropping below the

level of statistical significance. Thus, accounting for social context

explained one quarter of the summer social advantage for students in

high-SES families, which indicates that some of the summer advantage

attributed to high-SES families in Model 1 was a consequence of the

neighborhoods in which these families tended to live. Looking across

Models 3 and 4, we also noted that, when accounting for both neighbor-

hood SES and proportion minority, the negative coefficient for low-SES

family dropped below the statistical significance level of p< 0.05.

However, the drop in coefficient size was slight; thus, accounting for

neighborhood social context does not change our finding that, regard-

less of context, students from low-SES families tended to lose ground in

reading achievement during the summer. There was, however, some evi-

dence that accounting for neighborhood proportion minority explained

a portion of the previously noted summer advantage for African

American students. Comparing across Models 3 and 4, we found that

accounting for neighborhood proportion minority reduced the African

American coefficient size from 0.272 to 0.225, thus pushing it below the

threshold for statistical significance. Although positive, the coefficients

for proportion minority in Models 2 and 4 were not statistically signifi-

cant. Thus we simply conclude that neighborhood proportion minority

was not associated with slower rates of summer reading growth for

African American students, or students in general.

FAMILY SES, RACE/ETHNICITY, AND SCHOOL-YEAR ACHIEVEMENT

We found strong evidence that low-SES students grew in reading achieve-

ment at a slower pace than students from all other SES groups during

first grade, and weaker evidence for the same trend during kindergarten.

Coefficients from Model 1, Table 3, indicate that low-SES students’ read-

ing achievement grew 0.221 points per month slower than middle-SES

1370 Teachers College Record

students in first grade. In kindergarten, low-SES students grew 0.113

points per month slower than middle-SES students. Over the course of an

average school year, the monthly decrement in first-grade growth cumu-

lated to approximately 1 month, (9.4 * 0.221) / 2.1, and the kinder-

garten decrement cumulated to approximately half a month, (9.4 *

0.113) / 2.1, of school-year reading growth.

The school year was also a period of dispersion in reading achievement

growth by race/ethnicity, with African American students growing at a

substantially slower pace than similar White students in both school years.

The size of the monthly Black-White gap was large and roughly equal in

both school years. Model 1 of Table 3 indicates that Black students grew

0.206 points slower than comparable White students in kindergarten,

and 0.209 points slower in first grade. Over the course of the two school

years, these monthly gaps summed to 3.9 points, or almost 1.9 months of

school year reading growth. The Hispanic-White gap in school-year

growth was smaller than the Black-White gap and not as consistent. The

monthly Hispanic-White gap in kindergarten (0.126 points per month)

summed to a modest difference of 1.18 points over the school year, or just

over half a month of school-year reading growth. In first grade, however,

only the students-within-neighborhoods models (Table 2) identified a sig-

nificant Hispanic-White gap (0.1 points per month). We conclude that

Hispanic-White gaps were present but considerably smaller than Black-

White gaps in both kindergarten and first grade. Once school began,

African American students were quickly falling behind White students.

CONTEXT, COMPOSITION, AND SCHOOL-YEAR READING

ACHIEVEMENT

School social context exerted a strong influence on reading achievement

during kindergarten and first grade. However, in kindergarten, the rela-

tionship between school social context and reading achievement

reflected an unexpected pattern. Both low-SES and high-SES contexts

were associated with slower rates of achievement growth during kinder-

garten. Thus, although there was a social gap in reading growth during

kindergarten, there was not a contextual gap. To point, Model 4 of Table

3 indicates that, compared with students in middle-SES schools, students

in low-SES schools grew 0.163 points per month slower, and students in

high-SES schools grew 0.188 points per month slower.

The relationship between school social context and reading growth

during first grade took on a pattern similar to that noted for neighbor-

hood social context at school entry and during the summer. There was

an advantage associated with attending high-SES schools, relative to

Socioeconomic Context and Racial Composition 1371

middle-SES schools, whereas there was no significant disadvantage asso-

ciated with attending a low-SES school. This advantage is indicated by the

monthly coefficient for high-SES schools of 0.245 in Model 3, and 0.203

for high-SES schools in Model 4 (Table 3). The coefficient for low-SES

schools in Model 3 (-0.144) approached statistical significance. However,

once we accounted for school proportion minority in Model 4, the coef-

ficient for low-SES schools sank to 0, indicating that the slower growth

rates observed in some low-SES schools were more strongly associated

with the proportion of minority students than school social context.

The proportion of minority students was not associated with kinder-

garten reading growth but was strongly associated with first-grade reading

growth. The coefficient for proportion minority in first grade (-0.410)

indicates that for each standard deviation (0.36)28 increase in the propor-

tion of minority students in a school, the average reading growth rate for

students in the school fell by 0.15 points per month, 0.36 * (-0.41). In

addition to its relationship to average school achievement growth, school

percent minority was related to the first-grade growth rates of African

American students. The span of Models 1–4 illustrates that as we

accounted for school proportion minority in Models 2 and 4, the Model

1 coefficient for African American students shrank to half its original

size, from - 0.209 in Model 1 to - 0.103 in Model 2, and likewise to -0.105

in Model 4. In both instances, this change in coefficient size rendered the

smaller coefficient for African American students nonsignificant. We did

not uncover a similar pattern in the kindergarten findings. Nevertheless,

our first-grade finding indicates that school race/ethnic composition was

strongly linked to reading achievement, especially for African American

students. Moreover, this finding suggests that a substantial portion of the

first-grade reading disadvantage for Black students was a consequence of

location in schools with large concentrations of minority students, rather

than rearing in African American families.

DISCUSSION

FAMILY, RACE/ETHNICITY, AND ACHIEVEMENT ACROSS SEASONS

Family SES was most striking in its relationship to achievement at school

entry. As noted previously for Model 1 of the students-within-neighbor-

hoods analysis, students from high- and low-SES families entered school

almost a standard deviation from each other in levels of reading achieve-

ment. Similar dispersion was present in Model 1 of the students-within-

schools models. With controls for family background, both of these

models reflect the compositional effects of preschools and other commu-

1372 Teachers College Record

nity institutions and resources that are designed to enhance students’

readiness for school. The students-within-schools model, rather than

implicating schools as causes of inequality, provides a more general por-

trait of the inequalities that schools inherit from their neighborhood con-

texts. The vast majority of students attend neighborhood schools, and, as

a result, the makeup of the student body mirrors the socioeconomic,

racial, and ethnic makeup of the surrounding neighborhood (Panel on

High-Risk Youth, 1993). Relative to middle- and high schools, elementary

schools have tighter neighborhood-based attendance zones that are espe-

cially prone to school segregation by poverty, race, and ethnicity.

Therefore, our first model shows that schools are faced with substantial

inequalities at the beginning of the kindergarten year that are rooted in

the larger social contexts of the neighborhoods in which they are situated.

Our findings validate previous seasonal studies that have identified

social achievement gaps during the summer season; we also find cause

for concern with the slower pace of reading growth for low-SES children

during first grade. Because the school year lasted longer than 9 months,

whereas the summer break was less than 3 months, the implication of the

monthly first-grade achievement decrement for low-SES children was

substantial. Whereas the summer decrement of 0.26 points (Model 1,

Table 2) per month summed to 0.32 [(2.6 * 0.26) / 2.1] months of

school-year achievement growth, the monthly first-grade decrement of

0.21 points per month (Model 1, Table 3) summed to 0.94 [(9.4 * 0.21)

/ 2.1] months of school-year growth.

Our finding of a school-year disadvantage in reading growth for low-

SES students contradicts previous seasonal research by Entwisle and

Alexander (1997), who did not find significant dispersion of reading

achievement growth by family SES during elementary school. Our find-

ing is more consistent with Heyns’s (1978) view of school as a countervail-

ing, but not entirely compensatory, force in relationship to families.

However, in a pattern consistent with Entwisle and Alexander’s faucet

theory, lower middle SES and upper middle SES students’ school-year

growth rates did not differ significantly from those of middle-SES stu-

dents. Thus, it could be said that schooling helped to equalize achieve-

ment for the broad middle-SES population of students. Nevertheless, the

magnitude of the disadvantage for low-SES students—especially in first

grade—stands out as a significant problem, especially when considering

that these students entered school one third of a standard deviation

behind their middle-SES peers.

Our findings regarding race/ethnicity and achievement over the first 2

years of formal schooling are also cause for concern. Of particular inter-

est is our finding that race/ethnic inequalities in achievement emerged

Socioeconomic Context and Racial Composition 1373

almost entirely during the school year. The Black-White achievement gap

was large in kindergarten and first grade, yet African American students

grew at a faster pace than similar White students during the summer sea-

son. Hispanic students did not fall behind similar White students during

the summer season either, yet a Hispanic-White gap was clearly present in

kindergarten and, in the neighborhood models, in first grade. Moreover,

African American students entered school at achievement levels equal to

White students from similar social backgrounds, and Hispanic students

entered school only moderately behind similar White students. Our find-

ings, consistent with prior seasonal research, indicate that the school sea-

son is the main time period during which race/ethnic achievement gaps

widen.

The Seasonal Relevance of Neighborhood and School Context

Our modeling strategy facilitates a clear set of inferences concerning the

relative importance of neighborhood and school context for achieve-

ment at school entry and during the summer season. We treated neigh-

borhood context, rather than school context, as relevant to the

prekindergarten achievement reflected in test scores at school entry. For

the summer season, the possibility existed that schools could have influ-

enced summer season reading achievement. The strongest test of this

hypothesis is to evaluate the strength of association between school con-

text and summer achievement growth without controlling for neighbor-

hood context. This is exactly the test posed by our students-within-schools

model of summer reading growth (Table 3). As noted, none of the school

context measures reached statistical significance when predicting sum-

mer reading growth. Thus, we conclude that neighborhood context was

more relevant than school context to reading achievement during the

summer season, as well as during the preschool period. Our analysis did

not produce evidence of a cross-season interaction as noted by Entwisle

et al. (1997).

The weight of evidence from this study indicates that school context

was more strongly associated with school-year achievement growth than

neighborhood context. We drew this inference by comparing each of the

school-year contextual coefficients across the neighborhood and school

models for kindergarten and first grade. Inference from the two

instances in which coefficients did not match is fairly straightforward.

Low-SES schools were negatively associated with kindergarten reading

growth, whereas low-SES neighborhoods were not. High-SES schools

were positively associated with first-grade reading growth, whereas high-

SES neighborhoods were not. For each of these instances, we conclude

1374 Teachers College Record

that school context was more salient than neighborhood context. For

proportion minority during kindergarten and low-SES during first grade,

we found that neither school nor neighborhood context was significantly

associated with reading growth. Inference for the remaining two

instances is more difficult.

In first grade, neighborhood and school proportion minority were neg-

atively associated with overall reading growth. Comparing across neigh-

borhood (Table 2) and school models (Table 3) suggests that the

school-reading linkage was stronger than the neighborhood-reading link-

age because the coefficient for proportion minority was larger in the

school model (-0.410 vs. -0.314). The school-reading linkage also appears

stronger for African American students, as evidenced by comparing

across the neighborhood and school models. Accounting for school pro-

portion minority explained roughly half of the Black-White reading

growth gap (Model 1 to Model 2 of Table 3), and accounting for neigh-

borhood proportion minority explained one quarter of the Black-White

gap (Model 1 to Model 2 of Table 2). For Hispanic students, however, this

pattern was reversed. Accounting for neighborhood proportion minority

reduced the Hispanic-White gap of - 0.136 points per month to less than

half that size (Models 2–4 in Table 2). In the students-within-schools

models (Table 3), the Hispanic-White gap was not statistically significant

in any of the models. Thus we conclude that school context was more

salient for African American students and neighborhood context was

more salient for Hispanic students during first grade. Although, on the

whole, school race/ethnic composition appeared more salient than

neighborhood race/ethnic composition during the school year, we find

it entirely plausible that neighborhood proportion minority contributed

to slower rates of growth during first grade.

The final instance for the school-neighborhood comparison concerns

the relative salience of high-SES neighborhood and school contexts to

kindergarten reading growth. Because high-SES contexts were negatively

related to reading growth at the same magnitude, our findings indicate

that either or both contexts could have been causally related to kinder-

garten reading growth. Summarizing across each of the six school-year

comparisons of neighborhood and school contexts, we conclude that our

findings are consistent with the foundational hypothesis of seasonal

learning theory: When school is in session, it supersedes the influence of

other contexts. However, we cannot rule out the possibility that neighbor-

hood context remains salient during the school year, albeit to a lesser

extent than school context, or that neighborhood context continues in

importance for specific student groups (such as Hispanic students) dur-

ing the school year.

Socioeconomic Context and Racial Composition 1375

The Seasonal Relevance of Social Context and Race/Ethnic Composition

Our findings indicate that social context was relevant to achievement in

all seasons, whereas race/ethnic composition was only relevant during

the school year. The social contextual gap in achievement at school entr y

was substantial, and the clear pattern of advantage accruing to students

located in high-SES social contexts continued during the summer season

and first grade. Although we did not find a social contextual gap in read-

ing growth during kindergarten, the appearance of a gap between stu-

dents in low- and middle-SES schools still merits further attention to the

factors producing that gap. With the exception of this kindergarten find-

ing, our work is consonant with that of Brooks-Gunn et al. (1993), who

found that the positive benefits of socially advantaged neighborhoods

often outweighed the negative effects associated with poor neighbor-

hoods.

Regarding race/ethnic composition, our findings indicate that neigh-

borhood proportion minority was not salient for students’ reading

achievement in the period prior to school entry, during kindergarten, or

during the summer season, but it appeared important, especially for

Hispanic students, during first grade. School race/ethnic composition

was clearly associated with reading achievement growth during first

grade. Despite this strong relationship in first grade, however, school

minority composition was not related to reading growth during kinder-

garten. There is no conclusive way to reconcile these two findings. The

first-grade association could signal the beginning of a trend with a

strengthening relationship between school proportion minority and

achievement growth. This pattern would be consistent with the findings

of Hanushek et al. (2002), who found substantial school compositional

effects for middle school students, especially African American students.

Or, the association could be an aberration, with no such relationship

occurring in subsequent grades. More research based on students within

the ECLS-K is needed to draw conclusions about the consistency of this

relationship across later grades.

The Independent and Additive Nature of Contextual and Compositional Effects

Our analysis clearly demonstrates that social context and race/ethnic

composition were related to reading achievement in ways that go beyond

the contributions of family social background and race/ethnicity to

inequality. We construe these contextual effects as largely additive to fam-

ily-based differences in achievement. We note strong linkages between

family SES and social context, and between student race/ethnicity and

1376 Teachers College Record

school race/ethnic composition. Thus, for large proportions of low-SES,

lower middle-SES, African American, and Hispanic students in this study,

the combination of family origins and context constituted a double dis-

advantage, with respect to reading achievement, relative to higher SES

and White students. As noted, there were instances in which accounting

for context explained a portion of achievement inequality first noted at

the family level. Even in these instances, it is safe to say that context

tended to exacerbate persistent patterns of family-based inequality in

achievement.

POLICY IMPLICATIONS

Our findings complicate possible strategies for ameliorating achievement

inequality among young children and help to clarify the broader picture

of educational inequality in U.S. society. With some important exceptions

(e.g., Entwisle et al., 1997), much previous educational research has

tended to ignore neighborhoods as a key nonschool environment that

shapes children’s cognitive development. Our findings suggest that con-

temporary policy reforms aimed at equalizing achievement among stu-

dents from different social and racial/ethnic backgrounds may not

accomplish their aims unless they extend to addressing the vastly differ-

ent neighborhood conditions in which children are growing up and

learning.

In tandem with implementation and evaluation of the Gautreaux and

MTO residential mobility programs, substantial attention has been given

to the potential benefits of moving low-SES and African American fami-

lies to more affluent and integrated neighborhoods. Findings from eval-

uations of these programs seem to indicate that only moves to

substantially improved contexts boost achievement and educational out-

comes for students. Among students in the ECLS-K, we find scant inci-

dence of such moves, and thus, there is no way to evaluate their benefits

for students. However, using an additional set of slopes-as-outcomes

growth models,29 we have evaluated the benefits for low-SES and lower

middle SES students who were located in middle-SES neighborhoods and

schools. At school entry, we saw no advantage conferred on lower SES stu-

dents located in middle-SES neighborhoods. When examining the bene-

fits for lower SES students attending middle-SES schools, there was some

evidence that low-SES students grew faster in reading during first grade

when located in middle-SES schools, but there were so few low-SES stu-

dents (only 7% of all low-SES students) located in middle-SES schools

that the parameter estimates are unreliable. In summary, the narrow

scope of movement of families to contexts that did not match their own

Socioeconomic Context and Racial Composition 1377

SES makes it impossible to estimate the effects of moving socially disad-

vantaged students to more advantaged contexts. Our findings are consis-

tent with the conclusion, from mobility studies, that modest changes in

context do not stimulate significant differences in achievement.

Our findings also raise questions about the degree to which schools are

functioning as equalizers for low-SES and minority students. Although

our kindergarten findings were mixed, our first-grade findings point to a

clear achievement disadvantage for low-SES students relative to all their

peers, and a clear advantage for students attending high-SES schools.

Moreover, our findings from both years demonstrate a clear disadvantage

for African American students, and our first-grade findings show a disad-

vantage for students attending schools with large percentages of minor-

ity students. Each of these disadvantages reinforces patterns of

educational inequality that have been the focus of extensive policy

debate and intervention.

If policy makers seek to address inequality, they must look beyond sim-

ply improving high-poverty and segregated neighborhoods and schools

to replicating the advantages present in socially advantaged neighbor-

hoods and schools.

CONCLUSIONS

Because of differing perspectives on how one should interpret and mea-

sure contextual effects, the causal mechanisms and the actual magni-

tudes of these effects continue to be matters of scholarly debate

(Gamoran, 1992; Jencks & Mayer, 1990). We have taken several steps to

address methodological challenges to identifying contextual effects in

observational studies. We have accounted for many of the family back-

ground characteristics that may explain achievement differences across

contexts. After taking these factors into consideration and after modeling

the attributes of neighborhoods and schools at the proper level of aggre-

gation, we are confident that the observed relationships represent the

contextual effects of neighborhoods and schools above and beyond the

effects of family SES and other background characteristics. We see in our

analysis clear evidence that socially advantaged contexts support early

childhood learning. The possibility remains that families differ consider-

ably in their preferences for, and their abilities to benefit from, the con-

texts they inhabit. New research is needed to identify the conditions

under which contextual effects can generalize to families relocated

through residential mobility programs and school integration programs.

Our primary task in this article, however, has been to isolate contextual

from family effects.

1378 Teachers College Record

Looking beyond methodological challenges to identifying contextual

effects, we find theoretical and substantive reasons to believe that we have

accurately captured the contextual effects of neighborhoods and schools.

The idea that social structure affects an individual’s outcomes, indepen-

dent of his or her characteristics and background, is one that is funda-

mental to sociological investigations. Schooling, which affects each

student individually but is also a collective enterprise occurring within

classrooms, schools, and neighborhoods, is a prime example of this phe-

nomenon. Duncan and Raudenbush (2001) have stated that “research

on the organization of preschool and elementary school would lead into

exciting and as yet unexplored territory” (p. 368). However, to date,

models of educational outcomes and context have been confined mainly

to the secondary school years. Our work supports the alternate perspec-

tive, offered by Entwisle and Alexander (1999), which emphasizes that

the context in which children’s early development occurs is crucial to

understanding the trajectories that take young children to their adult

social positions. Our study of context, inequality, and educational out-

comes during the early elementary school years helps to address a rela-

tively large void in the literature. This study has extended previous

seasonal research using the ECLS-K by bringing into focus the dimen-

sions of social context and race/ethnic composition. We hope that our

fine-grained decomposition of the “when and where” of achievement

gaps can serve as a guiding complement to other research agendas that

seek to explain exactly “how” gaps emerge within each context and how

they can be ameliorated.

Acknowledgments

The research reported in this article was supported by a grant from the American Educational Research

Association, which receives funds for its AERA Grants Program from the National Science Foundation

and the National Center for Education Statistics of the Institute of Education Sciences (U.S.

Department of Education) under NSF Grant No. REC-0310268. The research reported here was also

supported by the Institute of Education Sciences, U.S. Department of Education, through Award #

R305C050055 to the University of Wisconsin-Madison. Any opinions, findings, or conclusions

expressed in this article are those of the authors and do not represent views of the funding agencies,

including the U.S. Department of Education.

Notes

1. The long summer session is peculiar to the United States, a holdover from the

agrarian necessities for child labor during the 18th and 19th centuries (see Wiseman &

Baker, 2004).

2. The term faucet theory derives from the proposition that the school resource “faucet”

is turned on during the school year and turned off during the summer (Entwisle et al.,

1997).

Socioeconomic Context and Racial Composition 1379

3. In her regression analysis of achievement growth for seventh graders, the size of the

coefficient for parental education dropped by 40% when moving from the summer season

to school year analysis (Heyns, 1978, table 4.6, p. 72).

4. The number of people in a census tract varies from 1,500 to 8,000 people; the aver-

age size is approximately 4,000 people. Census zip code tabulation areas (ZCTAs) often

include more than one census tract, and thus they tend to exceed tracts in size and vary

more in size. On average, 8,600 individuals reside in a ZCTA; among all ZCTAs, the num-

ber of residents varies from 0 to 113,600. In the sample for this study, the number of resi-

dents in a ZCTA varied from 58 to 101,571, averaging 26,465 persons.

5. The Stanford-Binet IQ Test was the most commonly administered test of cognitive

ability at the time of this study. Form L-M, third edition, was used.

6. Using zip codes, Ainsworth matched neighborhood measures from the 1990 U.S.

Census to student data from the National Educational Longitudinal Study of 1988

(NELS:88).

7. For more detailed coverage of the Gautreaux program, see Keels, Duncan, DeLuca,

Mendenhall, and Rosenbaum, 2005; for more detailed coverage of Moving to Opportunity,

see Orr et al., 2003.

8. The authors argued that equal grades meant higher achievement in suburban than

city schools.

9. Additionally, some students were located in schools outside the catchment areas for

their residences of origin prior to the beginning of the MTO program.

10. Heyns referred to these schools as low SES and high SES, but the distinction between

the two was based exclusively on whether the school had 45% or more of its students

enrolled in the free and reduced price lunch program. If the school met this criterion, it

was deemed low SES; all other schools were considered high SES.

11. Heyns used two low-income categories: (a) low-income students came from house-

holds with incomes between $4,000 and $7,999 per year, and (b) extremely low-income students

came from families with incomes of less than $4,000 per year (Heyns, 1978).

12. The school meal subsidy rate was negatively related to reading growth during each

of three summers considered, whereas the neighborhood poverty rate was negatively

related to reading growth during two of the three summers (Entwisle et al., 1997, table 3.8).

13. Rivkin computed value-added statistics for White students and used them as a mea-

sure of school quality for Black students attending the same schools.

14. See the methodological appendix for further details of the sample construction.

15. Preliminar y analyses indicated no substantial differences in the effects associated

with single- and alternative two-parent families; thus, we collapsed these two categories in

the final models.

16. At the zip code level of aggregation, the Bureau of the Census provides only a six-

category division of occupational sectors, each of which contains occupations with widely

varying statuses.

17. NCES categorized parents’ educational attainment according to nine categories.

Within two-parent families, attainment responses were sought, and frequently obtained,

from both parents. For two-parent cases, we converted each parent’s attainment category to

a continuous measure and computed the average to produce a measure of years of educa-

tion on a scale of 0 to 20 years. For one-parent cases, we simply used the continuous mea-

sure for the single parent.

18. The census provided attainment data, by gender, for 16 categories of attainment; we

converted each category to an appropriate continuous value before computing the mean

years of education within each neighborhood.

19. Alternate analyses (not included) using four categorical SES variables (and an

1380 Teachers College Record

excluded middle-SES category) did not produce any significant achievement differences

among neighborhoods and schools in the three middle-SES categories.

20. As with our individual-level measures, the Hispanic categor y comprised all

race/ethnic categories for Hispanics, including Hispanic Whites.

21. Our final models included 15,870 observations at level 1.

22. Two exceptions to the identical nature of our level 2 models were (a) the account-

ing for summer school attendance during the summer season, and (b) the accounting for

full-day kindergarten attendance during the kindergarten season. By necessity, these factors

were represented at different levels in the neighborhood and school models. Neither of

these factors was included in the school-entry and first-grade portions of our models.

23. Table 1 is based entirely on students-within-schools models, so it does not represent

the full extent of summer season inequality.

24. For a detailed explanation of these program participation measures, please see the

methodological appendix.

25. Even considering the race/ethnicity-only model—presented in rows 7–11 of Table

1—the reading disadvantage for African American students relative to White students was

fairly small: -1.370 / 9.074 = -0.15 standard deviation units.

26. The Hispanic-White disadvantage in the race/ethnicity only model was larger, -

2.278 / 9.074 = -0.25 standard deviation units.

27. This figure is based on the low-SES neighborhood coefficient (-0.906) from Model

3, which was divided by the standard deviation in test scores for the fall of kindergarten test

(9.074).

28. The mean proportion minority in schools was 0.413, with a standard deviation of

0.361; the mean proportion minority in neighborhoods was 0.338, with a standard deviation

of 0.303.

29. These models are available from the first author upon request.

References

Ainsworth, J. W. (2002). Why does it take a village? The mediation of neighborhood effects

on educational achievement. Social Forces, 81, 117–152.

Alexander, K. L., Entwisle, D. R., & Olson, L. S. (2004). Schools, achievement, and inequal-

ity: A seasonal perspective. In G. D. Borman & M. Boulay (Eds.), Summer learning:

Research, policies and programs (pp. 25–51). Mahwah, NJ: Erlbaum.

Boozer, M. A., & Krueger, A. B., & Wolkon, S., & Haltiwanger, J. C., & Loury, G. (1992). Race

and school quality since Brown v. Board of Education. Brookings Papers on Economic

Activity. Microeconomics, 1992, 269–338. Retrieved from http://www.jstor.org/stable/

pdfplus/2534765.pdf

Brooks-Gunn, J., Duncan, G. J., & Aber, J. L. (1997). Contexts and consequences for children

(Vol. 1). New York: Russell Sage Foundation.

Brooks-Gunn, J., Duncan, G. J., Klebanov, P. K., & Sealand, N. (1993). Do neighborhoods

influence child and adolescent development? American Journal of Sociology, 99, 353–395.

Burkam, D. T., Ready, D. D., Lee, V. E., & Logerfo, L. F. (2004). Social class differences in

summer learning between kindergarten and first grade: Model specification and estima-

tion. Sociology of Education, 77, 1–31.

Coleman, J. S., Campbell, E. Q., Hobson, C. J., McPartland, J., Mood, A. M., Weinfeld, F. D.,

et al. (1966). Equality of educational opportunity (No. OE-38001). Washington, DC: U.S.

Department of Health, Education, and Welfare, Office of Education, National Center

for Education Statistics.

Socioeconomic Context and Racial Composition 1381

Cooper, H., Nye, B., Charlton, K., Lindsay, J., & Greathouse, S. (1996). The effects of sum-

mer vacation on achievement test scores: A narrative and meta-analytic review. Review of

Educational Research, 66, 227–268.

Downey, D. B., von Hippel, P. T., & Broh, B. (2004). Are schools the great equalizer?

Cognitive inequality during the summer months and the school year. American

Sociological Review, 69, 613–635.

Duncan, G. J., & Raudenbush, S. W. (2001). Getting context right in quantitative studies of

child development. In A. Thornton (Ed.), The well-being of children and families: Research

and data needs (pp. 356–383). Ann Arbor: University of Michigan Press.

Duncan, G. J., & Zuberi, A. (2006). Mobility lessons from Gautreaux and Moving to

Opportunity. Northwestern Journal of Law and Social Policy, 1, 110–126.

Entwisle, D. R., & Alexander, K. L. (1992). Summer setback: Race, poverty, school compo-

sition, and mathematics achievement in the first two years of school. American Sociological

Review, 57, 72–84.

Entwisle, D. R., & Alexander, K. L. (1996). Family type and children’s growth in reading and

math over the primary grades. Journal of Marriage and the Family, 58, 341–355.

Entwisle, D. R., & Alexander, K. L. (1999). Early schooling and social stratification. In R. C.

Pianta & M. J. Cox (Eds.), The transition to kindergarten (pp. 13–38). Baltimore: Paul H.

Brookes.

Entwisle, D. R., Alexander, K. L., & Olson, L. S. (1997). Children, schools, and inequality.

Boulder, CO: Westview Press.

Fryer, R. G., Jr., & Levitt, S. D. (2004). Understanding the Black-White test score gap in the

first two years of school. Review of Economics and Statistics, 86, 447–464.

Gamoran, A. (1987). The stratification of high school learning opportunities. Sociology of

Education, 60, 135–155.

Gamoran, A. (1992). Social factors in education. In M. C. Alkin (Ed.), Encyclopedia of educa-

tional research (Vol. 6, pp. 1222–1229). New York: Macmillan.

Hanushek, E. A., Kain, J. F., & Rivkin, S. G. (2002). New evidence about Brown v. Board of

Education: The complex effects of school racial composition on achievement. Unpublished work-

ing paper, National Bureau of Economic Research, Cambridge, MA.

Hedges, L. V., & Nowell, A. (1998). Black–White test score convergence since 1965. In C.

Jencks & M. Phillips (Eds.), The Black–White test score gap (pp. 149–181). Washington, DC:

Brookings Institution Press.

Heyns, B. (1978). Summer learning and the effects of schooling. New York: Academic Press.

Hoxby, C. M. (2000). Peer effects in the classroom: Learning from gender and race

variation (Working Paper No. 7867). Cambridge, MA: National Bureau of Economic

Research.

Jencks, C., & Mayer, S. (1990). The social consequences of growing up in a poor neighbor-

hood. In L. E. Lynn, Jr., & M. G. H. McGeary (Eds.), Inner-city poverty in the United States

(pp. 111–186). Washington, DC: National Academy Press.

Jencks, C., & Phillips, M. (1998). The Black-White test score gap. Washington, DC: Brookings

Institution Press.

Kaufman, J. E., & Rosenbaum, J. (1992). The education and employment of low-income

Black youth in White suburbs. Educational Evaluation and Policy Analysis, 14, 229–240.

Keels, M., Duncan, G. J., DeLuca, S., Mendenhall, R., & Rosenbaum, J. (2005). Fifteen years

later: Can residential mobility programs provide a long-term escape from neighborhood

segregation, crime, and poverty? Demography, 42, 51–73.

Kerckhoff, A. C. (1993). Diverging pathways: Social structure and career deflections. New York:

Cambridge University Press.

1382 Teachers College Record

Massey, D. S., & Fischer, M. J. (2000). How segregation concentrates poverty. Ethnic and

Racial Studies, 23, 670–691.

National Center for Education Statistics. (2002). User’s manual for the ECLS-K first grade

restricted-use data files and electronic codebook (No. NCES 2002-128). Washington, DC: U.S.

Department of Education, Office of Educational Research and Improvement.

Orr, L., Feins, J. D., Jacob, R., Beecroft, E., Sanbonmatsu, L., Katz, L. F., et al. (2003). Moving

to Opportunity for Fair Housing Demonstration Program: Interim impacts evaluation Retrieved

from U.S. Department of Housing and Urban Development Web site:

http://www.huduser.org/Publications/pdf/MTOFullReport.pdf

Panel on High-Risk Youth, National Research Council. (1993). Losing generations: Adolescents

in high-risk settings. Washington, DC: National Academy Press.

Phillips, M., Brooks-Gunn, J., Duncan, G. J., Klebanov, P., & Crane, J. (1998). Family back-

ground, parenting practices, and the Black-White test score gap. In C. Jencks & M.

Phillips (Eds.), The Black-White test score gap (pp. 103–148). Washington, DC: Brookings

Institution Press.

Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models: Applications and data

analysis methods (2nd ed.). Thousand Oaks, CA: Sage.

Raudenbush, S. W., Bryk, A. S., & Congdon, R. (2008). HLM: Hierarchical linear and nonlin-

ear modeling (Version 6.06). Lincolnwood, IL: Scientific Software International.

Reardon, S. F. (2003). Sources of educational inequality: The growth of racial/ethnic and socioeco-

nomic test score gaps in kindergarten and first grade. Unpublished working paper, Population

Research Institute, The Pennsylvania State University.

Rivkin, S. G. (2000). School desegregation, academic attainment, and earnings. Journal of

Human Resources, 35, 336–346.

Sanbonmatsu, L., Kling, J. R., Duncan, G. J., & Brooks-Gunn, J. (2006). Neighborhoods and

academic achievement: Results from the Moving to Opportunity experiment. Journal of

Human Resources, 41, 649–691.

U.S. Department of Education. (2005). The nation’s report card: Reading 2005 (No. NCES

2006-451). Washington, DC: U.S. Government Printing Office.

Wasserman, S., & Faust, K. (1994). Social network analysis: Methods and applications. New York:

Cambridge University Press.

Wilson, W. J. (1987). The truly disadvantaged: The inner city, the underclass and public policy.

Chicago: University of Chicago Press.

Wiseman, A. W., & Baker, D. P. (2004). The American summer learning gap from an inter-

national perspective. In G. D. Borman & M. Boulay (Eds.), Summer learning: Research, poli-

cies, and programs (pp. 53–70). Mahwah, NJ: Erlbaum.

METHODOLOGICAL APPENDIX

ANALYTIC SAMPLE CONSTRUCTION

In keeping with other contextual studies, we removed students who

changed schools or neighborhoods during the study period. We began

with the full FFG subsample—5,470 students. We eliminated 940 students

because they changed schools between the beginning of kindergarten

and the end of first grade, and 130 students because they changed neigh-

borhoods between the fall of first grade and the spring of first grade.1

Next, to ensure a minimum level of achievement data for each student in

Socioeconomic Context and Racial Composition 1383

the analytic sample, we eliminated 130 students who did not have at least

one test score available. Finally we eliminated an additional 100 cases

from our analytic sample because they had no family background data

available from either year. Table A.1 presents descriptive statistics for the

full and analytic samples, and for movers excluded from the analytic sam-

ple. Results of t-tests and chi-square tests indicate the presence (or

absence) of differences between movers and students in the analytic sam-

ple. Movers tended to have slightly lower test scores, were more likely to

be from low-SES and non-traditional families, and were more likely to be

African American and Hispanic.

Because item nonresponse rates were low and data—when missing for

any background variable—tended to be missing for all background mea-

sures at once, we did not see a benefit to multiply imputing background

data. Rather, when background measures were missing from the kinder-

garten rounds of parental interviews, we imputed data from parallel

questions in the first-grade interviews. This process allowed us to pre-

serve many cases in the dataset that other wise would have been elimi-

nated.

Program Participation Measures

Our measures of program participation reflect seasonal researchers’

emphasis on the importance of exposure to schooling (Entwisle et al.,

1997; Heyns, 1978). This line of research has tended to focus on the

effectiveness of summer school programs for boosting achievement and

attenuating socially induced achievement inequalities (Cooper et al.,

1996), but it has also considered school-year enhancements such as full-

day kindergarten (Lee, Burkam, Ready, Honigman, & Meisels, 2006).

Our analysis includes summer school and full-day kindergarten mea-

sures. We included these measures in the analysis because availability of

these programs may vary across contexts, and we wanted to assess contex-

tual effects net of program participation.

Our program participation measures were designed to capture

increased academic training. With regard to summer school, researchers

using ECLS-K data have tended to find negative but non-significant

effects of summer school enrollment (Burkam et al., 2004; Downey et al.,

2004). These results may reflect issues of selection into summer school

programs and were obtained without controls for the academic content

of the programs. Our measure of academic summer school captured

exposure to academic material during the summer season, as reported by

parents during the FFG survey. When parents indicated that their chil-

dren had attended a summer school program that included both reading

1384 Teachers College Record

and math instruction, we coded our dichotomous measure as one; for all

other cases, we coded this measure as zero. With regard to the school

year, researchers using ECLS-K data have found positive effects of full-day

kindergarten programs (Lee et al., 2006). We coded a dichotomous mea-

sure as one when a school provided full-day kindergarten classes to a

majority of its students, as indicated by both the ECLS-K field manage-

ment data report and individual student reports.2

THREE-LEVEL MODELING STRATEGY

The Level-1 Model

Accurately estimating seasonal growth parameters for kindergarten, sum-

mer, and first grade requires knowing the exact durations of time that

students spent in each season. Because students were not tested on the

first and last days of the school year, it was necessary to correct for mis-

alignments between the school calendar and testing schedule, which

required precisely measuring the duration of time between school begin

(and end) dates and test administration dates. Our seasonal time vari-

ables accurately apportion time spent in each season to the tenth of a

month, for each student, as of each testing date. For the time variables in

the level-1 model, the beginning of kindergarten during fall 1999 is point

zero. From there, we counted forward in months to the first testing date,

and—for the observation at testing point one—allotted the number of

months spent in kindergarten—as of that date—to the kindergarten time

variable. To illustrate, “Fall kindergarten”, in the “Season Duration at

Test” panel of Table 1, indicates that the average student had spent 2.2

months in kindergarten prior to the first assessment. Thus, in the obser-

vation for the test conducted at time point 1 (fall of kindergarten) the

kindergarten time variable in level 1 was coded as 2.2. For the observa-

tion at test date two—we added the number of months between test 1 and

test 2 to the number of months between point zero and test 1, and allot-

ted the total (indicated by “Spring kindergarten” in Table 1) to the

kindergarten time variable. We continued this process until we had

accounted for the exact time spent in each season as of each testing date.

To correct all misalignments between school calendars and testing

schedules, we relied on responses from the school administrator reports

indicating the exact begin and end dates for the first-grade school year.

NCES did not query school administrators during kindergarten regard-

ing these dates. Under the reasonable assumption that school calendars

did not change significantly between the 2 years, we imputed dates—for

specific schools—from first grade to kindergarten. For schools in which

school administrators failed to report begin or end dates for first grade,

we imputed the average dates for similar schools.3

Our individual, level-1 growth model is as follows:

The outcome measure, Ytij, is the test score at time tfor child iin unit

j(neighborhood or school). The intercept, 0ij, represents achievement

at school entry, and the three subsequent parameters indicate growth

rates during kindergarten, summer, and first grade, respectively. The let-

ters K, S, and F denote the seasonal time variables—beginning with Ktij—

which specify the exact durations (in months) spent in each season as of

a test administration at time t. Level-1 observations occur for each testing

point (1,2, 3 and 4) for each student, except for time points when a stu-

dent’s test score was missing. The subscript iindicates the student-specific

nature of measures and parameters at level 1, and the subscript jcon-

notes membership of students within specific neighborhoods or schools.

Because reading growth rates varied substantially across seasons, the

models required separate growth parameters for each season, and this

number of parameters could not be identified given the limited number

of testing dates. Thus, we applied an error estimation strategy developed

by seasonal researchers (Downey et al. 2004) who have fitted growth

models to ECLS-K data. This strategy makes the level-1 error variance a

known quantity—thus eliminating the need for estimating this additional

parameter—by computing this variance from the published test reliabili-

ties and observed sample variance in test scores among students. The

error term at level 1, etij, represents the difference between the student’s

true performance on the test and the scale score measured by the test

instrument. This measurement error (sigma-squared) can be computed

as the product of the test unreliability [1.0 minus the instrument reliabil-

ity] multiplied by the observed variance in test scores. The NCES assess-

ment validation report (Rock & Pollack, 2002) contains the assessment

reliabilities for the reading achievement tests administered to ECLS-K

students at each testing point. Because reliabilities and score variances

varied slightly across tests, we computed an average of the measurement

error variances across the four test administrations, weighted by the pro-

portion of students taking the test at each time point. The error variance

ranged from 5.15 points for the fall 1998 test to 6.42 points for the fall

1999 test, and our computed figure for measurement error variance was

5.89. When estimating our models in HLM 6.06 (Raudenbush, Bryk, &

Congdon, 2008), we constrained the value of sigma-squared to this value.

Socioeconomic Context and Racial Composition 1385

tijtijijtijijtijijijtjj eFSKY ++++= )()()( 3210



1386 Teachers College Record

The Level-2 Model

Level-2 comprises 4 submodels which separately explain the intercept

and growth parameters from level 1 as functions of a unit mean (the

level-2 intercept) and unit-average parameters corresponding to the

sociodemographic characteristics of individual students. The full level-2

model explaining variation among students in kindergarten reading

growth was as follows:



10j represents the unit-average growth rate during kindergarten. Each

subsequent beta parameter captures alterations to the unit average

growth rate associated with student characteristics represented by the

sociodemographic variables. Thus,



11j through



14j capture deviations

associated with each of the four family SES quintile categories denoted by

SESQUINTS—the vector of 4 socioeconomic quintiles (low, low-middle,

upper-middle, and high)—compared to the excluded category (middle-

SES).



15j through



18j capture differences in achievement growth associ-

ated with each race/ethnicity category contained within RACE—the

vector of four dummy variables for student race/ethnicity—compared to

White race/ethnicity.



19j captures the benefit associated with 2-biologi-

cal parent family structure—indicated by the dummy variable 2BIOPAR—

compared to all other family formations.



10j captures the effect

associated with adding each sibling to a student’s family, and corresponds

to NUMSIBS, which specifies the number of siblings in each student’s

household.



111j captures the increment or decrement associated with

each one month deviation from the mean age at kindergarten entry rep-

resented by AGE, the student’s age in months at the beginning of the

1999 school year.



111j refers to departures from the mean age at kinder-

garten entry because AGE was centered around the sample (grand)

mean.



112j estimates the achievement decrement for male students com-

pared to females, corresponding to the dummy variable for gender, GEN-

DER. Finally,



113j captures the deviation from the unit average growth

rate associated with a student having repeated kindergarten, as indicated

by the dummy variable REPEAT, which indicates whether the student was

repeating kindergarten as of the beginning of the 1999 school year. The

level-2 error term, r1ij, represents the student-specific error term for stu-

dent iin unit j. Identical level-2 submodels predict the achievement para-

meters for school entry and first grade; the model predicting summer

season achievement growth differs slightly because it includes a dummy

ijjijjjijj

ijjjjjijjjjjjij

rREPEATGENDERAGENUMSIBS

BIOPARRACESESQUINTS

113112111110

1918,17,16,1514,13,12,11101

+++++

+++=





Socioeconomic Context and Racial Composition 1387

variable and corresponding parameter for academic summer school

participation.

The Level-3 Model

The level-3 equations serve as a random intercept model (Raudenbush &

Bryk, 2002), in which separate level-3 equations explain the level-2 (unit-

average) intercepts for achievement at school entry,



00j, and each of the

unit-average seasonal growth parameters,



10j,



20j, and



30j. Each level-3

model (Models 1 through 4) includes random parameters that capture

achievement variation across units (neighborhoods and schools). The

average sizes of these parameters are presented in the bottom panels of

Tables 2 and 3, under the heading “Level 3 variance.” Subsequent to

Model 1—which includes only random parameters at level 3—Models 2

through 4 estimate fixed effects for the seasonal averages as well as the

contextual measures. For example, the level-3 equation, which modeled

unit-to-unit variation in the kindergarten growth parameter, was as

follows:

The intercept,



100j, represents the average kindergarten growth rate

for all units in the sample, net of the student-level predictors at level 2.

The random parameter,



10j, represents the deviation of unit jfrom this

unit-mean growth rate. In Model 1 (which contains submodels for each

season),



10j is the only level-3 parameter accompanying the intercept. In

Model 2, PROPMIN is a continuous variable indicating the percentage of

minority students in a unit, and



103 estimates the deviation from the

unit-mean growth rate associated with each one-unit (100 percentage

point) increase in the minority composition of a unit. In Model 2 and

subsequent models, the dichotomous variable FDKIND indicates whether

the school implemented a full-day kindergarten schedule, and



104 esti-

mates the fixed effect of this schedule on kindergarten achievement

growth. In Models 3 and 4, the dummy variable SES_LOW indicates

whether a unit was located in the bottom socioeconomic quintile of

neighborhoods (or schools), and



101 captures the average achievement

decrement (or increment) associated with a unit’s inclusion in this quin-

tile. Similarly, SES_HIGH indicates whether a unit was located in the top

socioeconomic quintile of units, and



102 captures the average achieve-

ment increment (or decrement) associated with a unit’s membership in

this quintile. Similar level-3 models explained achievement at school

entry and growth during summer and first grade, with the exception

jjjjj

FDKINDPROPMINHIGHSESLOWSES

1010410310210110010



+++++=

1388 Teachers College Record

that these models did not include a measure or parameter for full-day

kindergarten.

Fitting Models

Model fit was evaluated for all models by comparing random-intercept

models (specified above) to more complex random-coefficients models

including additional random error terms for level-2 covariates

(Raudenbush and Bryk, 2002). We compared our simpler models to

more complex ones using chi-square tests which employed test statistics

computed by dividing the difference in deviance statistics (numerator)

by the degrees of freedom (denominator) for each pair of models. The

deviance statistics and numbers of parameters estimated are included at

the bottom of Tables 2 and 3. We evaluated the appropriateness of ran-

dom coefficients models by allowing the slopes of level-2 covariates to

vary across units. We evaluated whether each level-2 covariate varied sig-

nificantly across units, and in cases where one did, we compared model

fit statistics between the simpler random-intercept model and the ran-

dom-coefficients model. We found few instances in which level-2 covari-

ate estimates differed significantly across units. After evaluating whether

freeing the covariate slopes (for each of these instances) produced better

fitting models using the chi-square test, we concluded that additional

error terms were prudent for only one level-2 covariate, in a limited num-

ber of seasons. In the students-within-neighborhoods analysis, we

improved model fit by freeing the slope for high-SES students in the ini-

tial status (intercept) portion of each model. In the students-within-

schools analysis, we freed the slope for high-SES students for the school

entry and first grade portions of each model. Thus, the best fitting mod-

els for the analysis were random-intercept models with the minor adjust-

ments just noted.

Notes

1. NCES did not provide students’ home zip codes during kindergarten, and thus it

was impossible to check whether students changed residences during kindergarten.

2. The great majority of schools were homogeneous in the length of their kinder-

garten offerings, with only 10% offering both half-day and full-day classes.

3. To estimate dates for similar schools, we computed average begin and end dates for

schools in each state by school sector (public or private). When we imputed a date, we used

the sector-specific date for the state in which the school was located.

Socioeconomic Context and Racial Composition 1389

Note

1. The descriptive statistics reflect the cases for which data were present. 230 students

were eliminated from the analytic sample because they were missing test scores for all test-

ing dates or were missing parent data, or were missing both.

JAMES BENSON has recently completed a PhD in the Department of

Sociology at the University of Wisconsin–Madison. His research focuses

on the roles of high school curriculum and postsecondary finance policy

in promoting postsecondary attainment for community college students.

Recent publications include “A Randomized Field Trial of the Fast

ForWord Language Computer-Based Training Program” with Geoffrey D.

Borman (first author) and Laura Overman in Educational Evaluation and

Table A.1. Table of Means and Standard Deviations for Full and Analytic Samples, and Students Excluded

from the Analytic Sample

Full sample Analytic (non-mover) Movers (excluded)

(N=5,470) sample (n=1290)1

(n=4,180)

Mean SD Mean SD Mean SD

Individual-level variables

Fall kindergarten reading scale score 23.467 9.143 23.608 9.074 22.943 9.384

Spring kindergarten reading scale score 33.676 11.426 33.887 11.266 32.840* 12.007

Fall first-grade reading scale score 38.803 12.971 39.017 12.785 37.880* 13.710

Spring first-grade reading scale score 55.898 14.143 56.218 13.967 54.299** 14.896

Low SES 0.185 0.159 0.269***

Lower middle SES 0.185 0.189 0.173

Middle SES 0.193 0.202 0.164**

Upper middle SES 0.206 0.222 0.157***

High SES 0.211 0.228 0.157***

White 0.574 0.603 0.480***

African American 0.152 0.144 0.176**

Hispanic 0.167 0.147 0.232***

Asian or Pacific Islander 0.077 0.078 0.073

Native American 0.025 0.027 0.017*

Gender (male = 1) 0.508 0.506 0.514

Biological two-parent family 0.628 0.670 0.493***

Number of siblings 1.466 1.180 1.472 1.189 1.508 1.307

Age (in months) at school entry 66.414 4.483 66.464 4.417 66.245 4.695

Repeated kindergarten 0.041 0.040 0.046

Academic summer school 0.077 0.078 0.072

***T-test (two-tailed) or 2test significant at < 0.001. **T-test (two-tailed) or 2test significant at < 0.01. *T-

test (two-tailed) or 2test significant at < 0.05.

1390 Teachers College Record

Policy Analysis, and “Families, Schools and Summer Learning” in The

Elementary School Journal, with Geoffrey D. Borman (first author) and

Laura Overman.

GEOFFREY D. BORMAN is professor of educational leadership and pol-

icy analysis, educational psychology, and educational policy studies at the

University of Wisconsin–Madison. His areas of expertise include experi-

mental and quasi-experimental design, school reform, and social and

educational inequality. Recent publications include “Teacher Attrition

and Retention: A Meta-Analytic and Narrative Review of the Research,”

with N. Maritza Dowling in Review of Educational Research, and “A Multi-

Site Cluster Randomized Field Trial of Open Court Reading,” with N.

Maritza Dowling and Carrie Schneck in Educational Evaluation and Policy

Analysis.

Children’s Health, Wellbeing and Academic Outcomes over the Summer Holidays: A Scoping Review

Article

Full-text available

Feb 2024

Background: The school day provides a supportive and stimulating environment that may protect children and adolescents (5-18 years) from behaviours that are adverse for health and wellbeing. Objective: To review the literature regarding changes in children's academic achievement or overall wellbeing during the extended school summer break and evaluate if the outcomes are different for children experiencing disadvantage. Methods: The peer-reviewed literature was searched across six electronic databases for studies tracking changes in any academic, health or wellbeing outcome in children over the summer holidays. Studies were screened in duplicate for inclusion. Data were extracted using a standardized data extraction form. Outcomes were coded as decline (suggestive or significant), increase (suggestive or significant) or mixed/neutral and then compared to the school year or according to disadvantaged. Results: Seventy-six studies (n = 14,230,846 participants) were included. Strong evidence was found of a decline in academic outcomes and increases in adiposity, sedentary behaviour and screen time. There was moderate evidence of declines in cardiovascular fitness and physical activity. These patterns were magnified for disadvantaged children. Limited data were available on muscular fitness, sleep, diet quality and social, emotional or mental wellbeing. A total of 80% of studies were from the United States. Most data were from children 12 years of age and younger. Conclusions: Over the summer break, children's academic and health outcomes decline. Children experiencing disadvantage display magnified losses that warrant further investigation. The summer holidays present an opportunity to improve children's health and wellbeing.

Communal Socialization: (Re)Imagining the Role of Communal Socialization on African American High Schoolers’ Engagement and Performance in Mathematics

Article

Full-text available

May 2024

Tamika L. McElveen

Studies have primarily identified the positive associations between students’ communal values and teachers’ culturally relevant practice in separate investigations. The current study examined African American students’ perceptions of their mathematics teachers’ practice specifically related to the transmission of communal values, communal socialization. It also explores associations between these perceptions and students’ academic engagement and performance in mathematics. Communal socialization is the transmission of implicit and explicit messages, models, and experiences that teach, honor, and reinforce students’ values of social connectedness, interdependence, and common purpose. African American high school students (N = 308, 81% 14–17 years old, 74% female, 71% subsidized lunch status) participated in this study. Exploratory factor analyses were used to examine the dimensionality of communal socialization. Latent profile analyses yielded profiles to explore the varied experiences of students in secondary mathematics. Multinomial logistic regressions were used to examine the associations between communal socialization and students’ academic (i.e., academic engagement and performance), achievement motivation (i.e., attainment value), and social responsibility outcomes. Results were in support of the multidimensional nature of the Communal Socialization Scale including four factors: personal responsibility, community utility, equitable actions, and global responsibility. In addition, four latent profiles emerged, each with distinct levels in these factors: highly communal, moderately communal, least critical, and least communal. Students in the highly communal socialization profile reported significantly more value for mathematics, academic engagement, and perceived higher social responsibility. As a culturally sustaining practice, communal socialization has the potential to shape mathematics instruction toward fostering and sustaining social and community connections for the benefit of students’ value, engagement, and persistence in mathematics.

Impact of school SES on literacy development

Article

Full-text available

Dec 2023
PLOS ONE

Effective literacy skills are essential to actively participate in today’s society. However, little research has been conducted that examined the impact of contextual variables on literacy development. The present paper addressed whether and how the socioeconomic status of the school (S-SES) children attend affects their literacy achievements. Eight-hundred and seventy-eight 2nd and 4th grade children participated in the study. Data were collected in low-SES (vulnerable) and in mid-high-SES (non-vulnerable) schools. Children completed a large battery of language, cognitive, and literacy tasks in Catalan, a language spoken in a region in Spain where virtually all children are at least bilingual (they also speak Spanish) and it is the main language of instruction. Results showed that children in vulnerable schools were outperformed by children in non-vulnerable schools across all literacy competencies, but particularly affected higher order skills; that is, text quality and reading comprehension. Differences with their non-vulnerable peers remained, even after controlling for context-level covariates, including familial SES. However, S-SES ceased to exert significant influence once children’s cognitive and, especially, linguistic skills were considered. The study adds to previous research claiming that school SES has an effect on students’ literacy skills, above and beyond children’s home SES. However, our findings also suggest that literacy performance is ultimately mostly dependent on educationally actionable, subject-level skills. Educational implications are discussed.

Adjusting for academic preparedness when estimating enrollment disparities in advanced high school coursework

Preprint

Full-text available

Jun 2023

Whether racial disparities in enrollment in advanced high school coursework can be attributed to differences in prior academic preparation is a central empirical question in sociological research, with important implications for education policy. However, the regression-based approaches to this question that are predominant in the literature suffer significant methodological limitations by implicitly assuming that students are similarly prepared if and only if they have similar values on a selected set of academic background measures. Here, we provide a general technique to estimate enrollment disparities in advanced coursework between similarly-prepared students of different races that is less vulnerable to these limitations. We introduce a novel measure of academic preparedness, a student's ex-ante probability of "success" in the course, directly adjust for this single measure in a regression model of enrollment on race, and assess the robustness of estimated disparities to potential unmeasured confounding. We illustrate this approach by analyzing Black-White disparities in AP mathematics enrollment in a large, urban, public school system in the United States. We find that preexisting differences in academic preparation do not fully explain the under-representation of Black students relative to White students in AP mathematics, and contrast our results with those from traditional approaches.

When you are not alone: school composition by family structure and students’ academic achievements in South Korea

Article

Feb 2024

Youngshin Lim

Differences in Academic Preparedness Do Not Fully Explain Black–White Enrollment Disparities in Advanced High School Coursework

Article

Jan 2024

Prioritizing School Integration in the Affirmatively Furthering Fair Housing (AFFH) Process

Research

Full-text available

Feb 2024

A reinstated and expanded AFFH rule is uniquely positioned to promote school integration.

KÜLTÜREL DEĞERLERE DUYARLI SINIF YÖNETİMİ ÖZ YETERLİK ÖLÇEĞİ'NİN TÜRKÇEYE UYARLAMA ÇALIŞMASITHE ADAPTION STUDY OF CULTURALLY RESPONSIVE CLASSROOM MANAGEMENT SELF-EFFICACY SCALE TO TURKISH

Article

Sep 2023

Bu çalışmanın amacı, öğretmen ve öğretmen adaylarının kültürel değerlere duyarlı sınıf yönetimi öz yeterlik düzeylerini belirlemek için Siwatu, Putman, Starker-Glass ve Lewis, (2011) tarafından geliştirilen Kültürel Değerlere Duyarlı Sınıf Yönetimi Öz Yeterlik Ölçeği’ni (KDD-SY-ÖYÖ) Türkçeye uyarlamaktır. Çalışma grubu, farklı branşlardan 534 öğretmen ve çeşitli lisans programlarında öğrenim gören 294 öğretmen adayı olmak üzere toplam 828 kişiden oluşmaktadır. Uyarlama çalışmasının ilk aşamasında Türkçe dil geçerliliğine ilişkin uygulamalar yapılmıştır. Türkçe dil eşdeğerliği sağlandıktan sonra, kapsam geçerliliği ve görünüş geçerliliği için uzman görüşüne başvurulmuş, alınan dönütler doğrultusunda gerekli düzenlemeler yapılmıştır. Ölçeğin geçerlik çalışmaları kapsamında Doğrulayıcı Faktör Analizi (DFA) yoluyla ölçeğin yapı geçerliliği analiz edilmiştir. Sonrasında “Kültürel Değerlere Duyarlı Eğitime Hazırbulunuşluk Ölçeği” ve “Kültürel Değerlere Duyarlı Öğretmen Ölçeği”yle ölçüt geçerliliği test edilmiştir. Güvenirlik çalışmaları kapsamında Cronbach Alpha iç tutarlılık katsayısı hesaplanmış. 94 olarak hesaplanmıştır. Elde edilen bulgular ışığında Türkçeye uyarlaması yapılan tek boyutlu 29 maddelik KDD-SY-ÖYÖ’nün geçerli ve güvenilir bir araç olduğu sonucuna varılmıştır.

School Settings

Chapter

Jan 2020

John Jessel

In this entry the nature of primary education and the availability of an inclusive basic education worldwide are introduced and some key aspects of contemporary primary school settings outlined. These include the organization of compulsory education at national and school levels, school enrollment, class size, and the grouping of students. The cultural background of student populations and the linguistic diversity arising from migration are also considered, as are curriculum and learning approaches, the school year, and the school day. Topics relating to the physical amenities available, the impact of digital technology on school settings, and virtual schools are also introduced.

Neighborhood–School Structures: A New Approach to the Joint Study of Social Contexts

Article

Mar 2023

Robust literatures separately estimate school effects and neighborhood effects on children's educational, economic, health, and other outcomes that measure well-being. A growing body of research acknowledges that both contexts matter and considers neighborhoods and schools jointly. In this review, we synthesize the array of results that emerge from these studies and critique the tendency for researchers to evaluate which matters more, neighborhoods versus schools. We propose a reorientation of this scholarship that incorporates research on neighborhood and school selection and segregation processes. We argue that contextual effects research would be enriched by considering local neighborhood–school structures: the ways that families choose neighborhoods and schools and that neighborhoods and schools mutually and cyclically constitute one another. We conclude with recommendations for bringing neighborhood–school structures to bear on both outcomes-oriented studies of neighborhood and school effects as well as studies of contextual selection and segregation. Expected final online publication date for the Annual Review of Sociology, Volume 49 is July 2023. Please see http://www.annualreviews.org/page/journal/pubdates for revised estimates.

Children, Schools, and Inequality

Book

Mar 2018

Educational sociologists have paid relatively little attention to children in middle childhood (ages 6 to 12), whereas developmental psychologists have emphasized factors internal to the child much more than the social contexts in explaining children’s development. Children, Schools, and Inequality redresses that imbalance. It examines elementary school outcomes (e.g., test scores, grades, retention rates) in light of the socioeconomic variation in schools and neighborhoods, the organizational patterns across elementary schools, and the ways in which family structure intersects with children’s school performance. Adding data from the Baltimore Beginning School Study to information culled from the fields of sociology, child development, and education, this book suggests why the gap between the school achievement of poor children and those who are better off has been so difficult to close. Doris Enwistle, Karl Alexander, and Linda Olson show why the first-grade transition?how children negotiate entry into full-time schooling?is a crucial period. They also show that events over that time have repercussions that echo throughout children’s entire school careers. Currently the only study of this life transition to cover a comprehensive sample and to suggest straightforward remedies for urban schools, Children, Schools, and Inequality can inform educators, practitioners, and policymakers, as well as researchers in the sociology of education and child development.

Equality of Educational Opportunity.

Article

Mar 1967

The Truly Disadvantaged: The Inner City, The Underclass, and Public Policy.

Article

Sep 1989

Hierarchical Linear Models: Applications and Data Analysis Methods.

Article

Mar 1993

PART I THE LOGIC OF HIERARCHICAL LINEAR MODELING Series Editor 's Introduction to Hierarchical Linear Models Series Editor 's Introduction to the Second Edition 1.Introduction 2.The Logic of Hierarchical Linear Models 3. Principles of Estimation and Hypothesis Testing for Hierarchical Linear Models 4. An Illustration PART II BASIC APPLICATIONS 5. Applications in Organizational Research 6. Applications in the Study of Individual Change 7. Applications in Meta-Analysis and Other Cases where Level-1 Variances are Known 8. Three-Level Models 9. Assessing the Adequacy of Hierarchical Models PART III ADVANCED APPLICATIONS 10. Hierarchical Generalized Linear Models 11. Hierarchical Models for Latent Variables 12. Models for Cross-Classified Random Effects 13. Bayesian Inference for Hierarchical Models PART IV ESTIMATION THEORY AND COMPUTATIONS 14. Estimation Theory Summary and Conclusions References Index About the Authors

The Education and Employment of Low-Income Black Youth in White Suburbs

Article

Nov 1993

Early schooling and social stratification

Article

Jan 1999

The Truly Disadvantaged: The Inner City, The Underclass, and Public Policy

Book

Jan 1987

William Julius. Wilson

"The Truly Disadvantagedshould spur critical thinking in many quarters about the causes and possible remedies for inner city poverty. As policy makers grapple with the problems of an enlarged underclass they—as well as community leaders and all concerned Americans of all races—would be advised to examine Mr. Wilson's incisive analysis."—Robert Greenstein,New York Times Book Review "'Must reading' for civil-rights leaders, leaders of advocacy organizations for the poor, and for elected officials in our major urban centers."—Bernard C. Watson,Journal of Negro Education "Required reading for anyone, presidential candidate or private citizen, who really wants to address the growing plight of the black urban underclass."—David J. Garrow,Washington Post Book World Selected by the editors of theNew York Times Book Reviewas one of the sixteen best books of 1987. Winner of the 1988 C. Wright Mills Award of the Society for the Study of Social Problems.

Summer Learning and the Effects of Schooling.

Article

Mar 1981

Family Type and Children's Growth in Reading and Math Over the Primary Grades

Article

May 1996

In a random sample of Baltimore school children over the first 2 years of school, there are no direct effects of parent configuration on marks or test score gains in reading and math with one exception: African American children in single-mother families where other adults are present got higher marks in reading at the beginning of 1st grade than did their counterparts in mother- only or mother-father families. Irrespective of family type, however, children whose families had more economic resources and whose parents had higher expectations for their school performance consistently outperformed other children in reading and math. These findings suggest that the effects of parents' psychological and economic resources that are correlated with family type go far toward explaining previous reports of schooling deficits for children from single-parent homes, especially in the early grades.

The Stratification of High School Learning Opportunities

Article

Jul 1987

Adam Gamoran

This paper suggests that students' opportunities to learn may be stratified both between and within schools: Schools serving a more affluent and able clientele may offer more rigorous and enriched programs of study, and students in college-preparatory curricular programs may have greater access to advanced courses within schools. This notion is tested with a longitudinal, nationally representative sample of public school students from the High School and Beyond data base. The results show few between-school effects of school composition and offerings but important within-school influences of curriculum tracking and coursetaking. In most cases, the difference in achievement between tracks exceeds the difference in achievement between students and dropouts, suggesting that cognitive skill development is affected more by where one is in school than by whether or not one is in school.

Family, Neighborhood, and School Settings Across Seasons: When Do Socioeconomic Context and Racial Composition Matter for the Reading Achievement Growth of Young Children?

Abstract and Figures

Recommended publications

Assessing the Role of Family Structure in Racial/Ethnic Residential Isolation

Comparing residential mobility programs: Design elements, neighborhood placements, and outcomes in M...

Racial Segregation and the Black-White Test Score Gap

Racial and Ethnic Inequality in the Duration of Children's Exposure to Neighborhood Poverty and Affl...

A Quasi-Experimental Study of the Impacts of the Kids Read Now Summer Reading Program