Socioeconomic Differences in Reading Trajectories: The Contribution of Family, Neighborhood, and School Contexts

Journal of Educational Psychology Copyright 2008 by the American Psychological Association
2008, Vol. 100, No. 2, 235–251 0022-0663/08/$12.00 DOI: 10.1037/0022-0663.100.2.235
Socioeconomic Differences in Reading Trajectories: The Contribution of

Family, Neighborhood, and School Contexts
Nikki L. Aikens Oscar Barbarin

Mathematica Policy Research, Inc. University of North Carolina at Chapel Hill
In the present study, the authors use the Early Childhood Longitudinal Study, Kindergarten Cohort of
1998 –1999, to examine the extent to which family, school, and neighborhood factors account for the
impact of socioeconomic status (SES) on children’s early reading. Through the use of hierarchical linear
modeling techniques, growth curve models were estimated to depict children’s reading trajectories from
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
kindergarten to 3rd grade. Family characteristics made the largest contribution to the prediction of initial
This document is copyrighted by the American Psychological Association or one of its allied publishers.
kindergarten reading disparities. This included home literacy environment, parental involvement in
school, and parental role strain. However, school and neighborhood conditions contributed more than
family characteristics to SES differences in learning rates in reading. The association between school
characteristics and reading outcomes suggests that makeup of the student population, as indexed by
poverty concentration and number of children with reading deficits in the school, is related to reading
outcomes. The findings imply that multiple contexts combine and are associated with young children’s
reading achievement and growth and help account for the robust relation of SES to reading outcomes.
Keywords: socioeconomic status, reading, achievement trajectories
Socioeconomic status (SES) differences in children’s reading interactions encountered by children across settings. Functioning
and educational outcomes are ubiquitous, stubbornly persistent, and development are also a reflection of how settings interact (see,
and well documented (Arnold & Doctoroff, 2003; Duncan, Yeung, for example, Early Child Care Research Network, National Insti-
Brooks-Gunn, & Smith, 1998; Lee & Burkam, 2002; McLoyd, tute of Child Health and Human Development [NICHD], 2004).
1998; Yeung, Linver, & Brooks-Gunn, 2002). Economically dis- Specifically, these frameworks account for influences of social
advantaged children acquire language skills more slowly, exhibit environment on reading development in terms of (a) the qualities
delayed letter recognition and phonological sensitivity, and are at of environments, such as climate, activities, resources, and strains,
risk for reading difficulties (Whitehurst & Lonigan, 1998). Spec- and (b) the quality of social relations within and across these
ulation about the mechanisms through which economic status settings. Through their resources, experiences, and interactions,
affects reading achievement often concerns family life, and more families, schools, and neighborhoods create auspicious or risky
recently, contexts like schools and neighborhoods are promising environments for children’s reading development.
sources of influence. In the present study, we examine the contri- Social relationships include parent– child, teacher– child, and
bution of these multiple settings to SES differences in early read- peer–peer interactions. The nature and quality of interactions with
ing. This study is guided by ecological and developmental systems important adults are important to children’s academic and social–
theories, in which it is recognized that children’s lives unfold emotional development (NICHD, 1998, 2000). Maternal warmth
within multiple settings and in relationships with multiple others. and mother– child interaction patterns have been cited as important
Each of these settings functions in a dynamic, reciprocal process to children’s language outcomes (Hess, Holloway, Dickson, &
between the settings and the individual child (Bronfenbrenner, Price, 1984; Murray & Hornbaker, 1997).
1979, 1989; Magnusson & Cairns, 1996). Under such perspectives, Bronfenbrenner’s (1979, 1989) work recognized that systems
functioning and development are not merely reflections of children differ in their proximity to the child, with more distal settings
themselves but also of the nature of experiences, resources, and emanating outward from the child. The framework also captures
the interrelations of systems. For example, parents’ reports of
involvement with the school represent the interaction of two set-
Nikki L. Aikens, Mathematica Policy Research, Inc.; Oscar Barbarin, tings—the family and the school. Here, we focus on those settings
School of Social Work, University of North Carolina at Chapel Hill. that children have direct contact with, including family, neighbor-
This research was supported by a grant from the American Educational hood, and school characteristics. We recognize that each of these
Research Association (AERA), which receives funds for its AERA Grants settings may produce unique and cumulative effects on early
Program from the National Science Foundation and the National Center for development. Questions remain about the relative contribution of
Education Statistics of the Institute of Education Sciences (U.S. Depart-
these contexts to the development of reading and to the relation of
ment of Education) under National Science Foundation Grant REC-
9980573. Opinions stated are those of Nikki L. Aikens and Oscar Barbarin
SES to early literacy outcomes. Ecological and developmental
and are not necessarily those of the granting agencies. systems frameworks also suggest ways environments may interact
Correspondence concerning this article should be addressed to Nikki L. with one another and may thus attenuate or amplify their effects on
Aikens, Mathematica Policy Research, Inc., 600 Maryland Avenue, S. W., development. How, for instance, do family, neighborhood, and
Suite 550, Washington, DC 20024. E-mail: naikens@mathematica-mpr.com school contexts combine to channel the effects of SES on reading
235
236 AIKENS AND BARBARIN
development? Because of the relation between early reading out- Gunn, 2000; Pianta & Walsh, 1996). For example, there is a strong
comes and later academic success, research exploring the contri- association among school quality, classroom quality, and family
bution of multiple settings to reading outcomes is critical in the SES (Entwisle & Alexander, 1993; Pianta, LaParo, Payne, Cox, &
effort to deconstruct and identify the processes that give rise to Bradley, 2002), particularly in regard to teacher and peer quality,
socioeconomic differences in achievement. The theory also recog- instructional methods, and physical and material resources (Evans,
nizes that the relative salience of contexts may shift over time. 2004; Kozol, 1991; Lee & Burkam, 2002; Rothstein, 2004). Child
Although ecological and developmental system frameworks under- and school poverty have also been linked to kindergarten teachers’
score the need to examine multiple contexts on developmental out- positive interactions, classroom’s instructional climate, and class-
comes (Bronfenbrenner, 1979, 1989; Magnusson & Cairns, 1996), room’s child-centered climate (Pianta et al., 2002). Differences in
family environment has most often been explored because it is con- classroom quality as early as preschool are connected to early
sidered to be the principal contributor to differences in early literacy reading-related outcomes (Bryant, Burchinal, Lau, & Sparling,
and language development associated with SES. Developmental re- 1994; Dickinson, 2001; NICHD Early Child Care Research Net-
search is replete with studies examining the relation between child work, 2000). A plethora of studies have assessed the relation of
outcomes and family climates of low-SES children. For example, this
preschool programs, elementary school reading instruction, and

body of work has demonstrated that children in low-SES households peers to reading competence (see NICHD, 2004; Snow, Burns, &
have less exposure to books at home (Evans, 2004; Lee & Burkam, Griffin, 1998; Whitehurst & Lonigan, 1998; Xue & Meisels,
2002; Vernon-Feagans, Hammer, Miccio, & Manlove, 2002; White- 2004). This work suggests that classroom environments that are
hurst & Lonigan, 1998) and have parents who are less involved in rich in literacy materials, that have teachers with high expectations
their schooling (Evans, 2004). These children are less likely to be of students and with adequate preparation to teach reading, that
regularly read to by parents (Federal Interagency Forum on Child and provide opportunities for dialogic reading or for children to be
Family Statistics, 2005; Lee & Burkam, 2002; Whitehurst & Lonigan, involved in the book reading experience, that provide support and
1998). Each of these family climate indicators has been cited as a opportunities for writing, and that promote stimulating teacher–
factor underlying disparities in early literacy and language outcomes. child conversations enhance early language and literacy skills.
With respect to relationship indicators, maternal sensitivity also has This work also suggests that peers play a role in influencing early
been implicated in the association between SES and children’s early
reading proficiency, with children in schools with larger concen-
language abilities (Raviv, Kessenich, & Morrison, 2004). In addition,
trations of less skilled, lower SES, and minority peers exhibiting
evidence suggests that differences in the quality of parents’ behaviors
lower gains in reading during the kindergarten year (Xue &
during joint book reading (Whitehurst & Lonigan, 1998) and in the
Meisels, 2004). However, remarkably few studies have sought to
frequency and the quality of language interaction with parents in the
identify the contribution of school characteristics to the relations
home (Hart & Risley, 1995; National Research Council, 2000) con-
between SES and children’s reading achievement.
tribute to disparities in early reading-related outcomes.
Compared with more advantaged peers, economically disadvan-
Only recently has there been an investigation of the influence of
taged children are more likely to reside in poorer quality neighbor-
more distal settings, including those outside the home, on such
hoods (Evans, 2004; Lee & Burkam, 2002). Although research has
reading-related differences. More important, few studies have con-
highlighted mechanisms by which neighborhoods and communities
currently examined the effects of multiple contexts on children’s
learning outcomes (see NICHD, 2004; Weigel, Martin, & Bennett, impact adolescent development, especially behavioral problems, very
2005, for recent exceptions). Exploring the contribution of these few of these studies have explored reading outcomes, and research on
additional settings is important because interpreting SES effects as very young children is rare. What has been done links neighborhood
emanating exclusively from the family or the child means that quality, children’s verbal ability, IQ scores, and school achievement
policy and program interventions may focus too narrowly as they among young and early school age children (Leventhal & Brooks-
attempt to improve the educational outcomes of low-SES children. Gunn, 2000). Again, however, in this work children’s early reading
In addition, by neglecting the interactions of multiple contexts, one outcomes have not been explicitly examined. These researchers have
risks missing the dynamic and contextual nature of development also not sought to understand the neighborhood characteristics that
and its outcomes (Cairns, Elder, & Costello, 1996). mediate the relations between socioeconomic background and chil-
Important differences in classroom, teacher, school, and com- dren’s reading achievement.
munity resources emerge along socioeconomic lines (Evans, 2004; The current research assessed, among a nationally representative
Kozol, 1991; Lee & Burkam, 2002; Rothstein, 2004), highlighting sample of children, the relative contribution of families, schools,
the fact that socioeconomic disadvantage exposes poor children to and neighborhoods to the relation between SES and children’s
multiple risky extrafamilial environments, including low quality early reading achievement and growth between kindergarten and
child care, poor and distressed schools, and economically de- third grade. In addition, we sought to identify the predictors that
pressed neighborhoods. Given the confluence of negative factors account for this association across contexts and to identify how
associated with socioeconomic disadvantage, research examining these relations vary over the first 4 years of school. This research
the influence of multiple, nested settings may capture more accu- assessed the extent to which the effects of SES on young children’s
rately and more fully the influences that shape children’s reading reading development occur through what happens in the school,
and academic growth than do single factor explanations. the neighborhood, and the home. Variables were selected to reflect
Factors that put children at risk for learning and school diffi- those that have been previously linked in the literature to reading
culty, particularly those related to socioeconomic background, are or academic achievement, with an emphasis on those that may
distributed across environments and are not restricted to a single differ across class lines. Accordingly, the present study is designed
setting (Evans, 2004; Goldenberg, 2002; Leventhal & Brooks- to answer the following questions:
SOCIOECONOMIC DIFFERENCES IN READING TRAJECTORIES 237
1. What is the relation between SES and children’s reading Prior to administering the cognitive assessment, the Oral Lan-
development from the fall of kindergarten to the spring of guage Development Scale (Duncan & DeAvila, 1986) was admin-
the third grade school year? istered to those children identified from their school records (or by
their teacher, if no school records were available) as coming from
2. To what extent do resources, experiences, and interac- a household whose primary language was not English. This screen-
tions within family, neighborhood, and school contexts ing test determined whether a child was able to understand and
independently and cumulatively explain socioeconomic respond to the cognitive assessment items in English. Children
differences in early reading outcomes and growth? How who passed the language screener received the full ECLS-K direct
do these relations vary over the first 4 years of school? assessment battery. Children who did not pass an established cut
score on the language screener did not receive the standard reading
Method assessment.
The direct reading assessment consisted of a set of two-stage
The data for the present study are from the Early Childhood assessments: a first-stage routing section, followed by several
Longitudinal Study, Kindergarten Class of 1998 –1999 (ECLS-K; alternative second-stage forms. First, a screening test consisting of
U.S. Department of Education, National Center for Education 20 items with a broad range of difficulty was used to assign each
Statistics, 2004). The ECLS-K study followed a nationally repre- child to one of three extensive batteries in the second stage of the
sentative sample of American children from kindergarten through assessment, in which items varied in difficulty. This was done to
fifth grade, providing a comprehensive picture of children’s early reduce testing time and to ensure that children would be required
family, neighborhood, and school experiences over time. The total to respond to questions of moderate difficulty. Testing on this
sample consists of 21,260 children recruited from 1,277 kinder- second-stage battery continued until the child reached a ceiling of
garten classrooms in public, Catholic, and non-Catholic private three consecutive items missed. Item response theory (IRT) was
schools nationwide. A subsample of 17,401 children was retained used to generate standard scores based on an analysis of the item
in the longitudinal kindergarten to third grade sample. The study difficulty and the pattern of right and wrong responses. IRT scale
design involved multistage probability sampling to achieve a na- scores allow for longitudinal measurement of achievement gain
tionally representative sample of children eligible to attend kin- over time because assessments at each time are measured in a
dergarten in the 1998 –1999 school year. common metric and are therefore comparable across waves. IRT
scores from each wave were used in the study’s analyses as
Instrumentation repeated measures of children’s reading outcomes.
Child variables. The ECLS-K provides information on child
During kindergarten, first-, and third-grade school years, the gender, race, age at first assessment, and SES. Field staff identified
field staff conducted direct child assessments, parental interviews, child gender during the direct child assessment, and this informa-
and observations of the school environment and the community tion was subsequently confirmed during the parent interview (1 ⫽
surrounding the school. Teachers and school administrators com- female, 0 ⫽ male). A child composite for race provided eight
pleted questionnaires during the spring data collection periods in categories for race and/or ethnicity. Children of all races are
kindergarten, first grade, and third grade. Data for the current included in the current analyses, but a dichotomous, dummy vari-
analyses come from child, parent, teacher, school administrator, able for White children is modeled as a predictor of reading
and field staff responses gathered during this time. A description of development. Child age reflects the child’s age in months at the
the items used in the current project follows. time of the fall kindergarten assessment. SES is a composite of five
Child individual reading assessment. Reading assessments in- pieces of information: father’s or male guardian’s education, moth-
volved a multiple choice or an open-ended format to measure basic er’s or female guardian’s education, father’s or male guardian’s
skills (including print familiarity, letter recognition, mastery of occupation, mother’s or female guardian’s occupation, and house-
beginning and ending sounds and rhyming sounds, word recogni- hold income. The index conceptualizes SES in quintiles, with
tion), vocabulary (receptive vocabulary), and comprehension (lis- lower scores indicating lower SES. SES is used in the current
tening comprehension, words in context). Test specifications were analyses because it tends to be less volatile than family income and
based on the 1992 and 1994 National Assessment of Educational encompasses several dimensions of economic conditions and pa-
Progress framework. The National Assessment of Educational rental capital beyond income.
Progress framework was developed to begin at fourth grade; Family variables. Parent interviews were completed primarily
consequently, the framework focuses on skills that are just emerg- via telephone, and provisions were made to interview parents
ing in early readers. For this reason, early elementary school speaking Spanish, Lakota, Hmong, and Chinese. Parents provided
educators and literacy curriculum specialists helped modify the information on child characteristics, family practices and re-
assessment to be suitable for kindergarten and first grade. Pools of sources, parent well-being, and home learning environment. Par-
items for the reading assessment were borrowed or adapted, with ents also reported the number of books the child had in the home
permission, from published tests, including the Peabody Individual (including library books). Data on whether the child had attended
Achievement Test–Revised (Markwardt, 1989), the Peabody Pic- a day care center, nursery school, preschool, or prekindergarten
ture Vocabulary Test–Revised (Dunn & Dunn, 1981), the Primary program on a regular basis the year before he or she started
Test of Cognitive Skills ( Huttenlocher & Levine, 1990), the Test kindergarten (1 ⫽ yes, 0 ⫽ no) were used as a measure of whether
of Early Reading Ability (Reid, Hresko, & Hammill, 1981), and the child had received center-based care.
the Woodcock-Johnson Tests of Achievement–Revised (Wood- Three variables for the current analyses were derived from
cock & Bonner, 1989). parents’ responses. Home literacy environment is a sum score of
the frequency with which parents engaged in joint book reading boarded-up buildings, and people congregating (1 ⫽ none, 2 ⫽ a
with the child, the frequency with which children read (or pre- little, 3 ⫽ some, 4 ⫽ a lot). The average across the four items was
tended to read) books outside of school, and the frequency with used as the index of bad conditions near the school. The average
which household members visited the library with the child. Par- of kindergarten and first-grade data was estimated for third grade
ents’ reports of the frequency of joint reading with the child and (kindergarten ␣ ⫽ .95; first-grade ␣ ⫽ .93). Finally, the ECLS-K
the frequency with which the child read (or pretended to read) had defines school urbanicity on the basis of census tract region:
the same 4-point scale: 1 ⫽ not at all, 2 ⫽ once or twice a week, central city; large town, urban fringe; and small town, rural. The
3 ⫽ three to six times a week, and 4 ⫽ everyday. Library visits current analyses define schools located in the central city as urban;
with the child within the last month was a dichotomous variable: schools in large towns and urban fringe as suburban; and schools
1 ⫽ yes, 0 ⫽ no. in small towns and rural areas as rural. A dichotomous, dummy
Involvement in child’s school is based on the sum of parents’ variable for suburban is modeled as a predictor of reading devel-
reports of their own or another adult household member’s partic- opment.
ipation in six activities reflecting involvement in the child’s school School variables. Classroom teachers completed self-
(1 ⫽ yes, 0 ⫽ no), including attending a parent–teacher confer-
administered questionnaires, providing reports on children’s class-

ence, attending a parent–teacher association (PTA) meeting, at- room peers, classroom literacy instruction, and teacher background
tending an open house, volunteering, participating in fundraising, and beliefs. School administrators also completed questionnaires
and attending a school event (kindergarten ␣ ⫽ .58; first-grade reporting on the percentage of students in the school who were
␣ ⫽ .61; third-grade ␣ ⫽ .61). Parental role strain is represented eligible to receive free or reduced price lunch. School poverty
by the average of parents’ responses to nine items that assessed status was determined by supplementing this information with
degree of difficulty and strain that the respondent experienced in school-wide Title I status (at least 50% of the student body is
his or her functioning as a parent (1 ⫽ not at all true, 2 ⫽ poor), as reported by school administrators. In the case of missing
somewhat true, 3 ⫽ mostly true, 4 ⫽ completely true). Only four data on the free or reduced price lunch variable, schools imple-
items were asked in the first grade, so an average of kindergarten menting school-wide Title I programs were identified as high-
and third grade data was estimated for first grade (kindergarten poverty schools. Consequently, a dichotomous variable reflecting
␣ ⫽ .69; third-grade ␣ ⫽ .56). Conversely, parental warmth is high- (above 50%) and low-poverty (below 50%) schools was
represented by the average of parents’ responses to four items generated. School administrators also indicated whether the child’s
assessing the warmth and closeness experienced in their relation- school was public or private. From this, a dichotomous, dummy
ship with the child (1 ⫽ not at all true, 2 ⫽ somewhat true, 3 ⫽ variable for private school was generated for the current analyses.
mostly true, 4 ⫽ completely true). Again, an average of kinder- Peers reading below grade is based on teachers’ reports of the
garten and third grade data was estimated for first grade (kinder- number of children reading below grade level in the classroom.
garten ␣ ⫽ .54; third-grade ␣ ⫽ .68) Teacher experience is the sum of the number of years the teacher
Neighborhood variables. Appraisals of the neighborhood set- had taught at that school and the number of years the teacher had
ting were provided by three independent sources: parents, school taught at that grade level. Teacher preparation is a variable derived
administrators, and field staff. Parent reports of home neighbor- from the sum of five teacher responses concerning their prepara-
hood safety were assessed with a single item in which parents were tion to teach: the number of courses taken on early education,
asked about how safe it was to play in the neighborhood (1 ⫽ not elementary education, and child development; the highest degree
at all safe, 2 ⫽ somewhat safe, 3 ⫽ very safe). Home neighbor- earned; and the number of courses taken on teaching reading
hood problems is a derived variable created from the sum of (kindergarten ␣ ⫽ .62; first-grade ␣ ⫽ .61; third-grade ␣ ⫽ .60).
parents’ responses about five qualities of their neighborhood, Teachers reported the frequency of children’s participation in sev-
including the degree to which the following activities were a eral literacy-related activities in the classroom weekly, including
problem: garbage or litter in the street, individuals selling or using working on learning the names of letters, practicing reading aloud and
drugs in the street, burglary or robbery in the area, violent crime in silently, reading a variety of texts, engaging in writing activities, and
the area, or vacant homes in the area (0 ⫽ no problem, 1 ⫽ working on phonics. The number of activities reported varied at each
somewhat of a problem, 2 ⫽ big problem). The average of the grade level from 10 activities in kindergarten to 14 activities and 6
kindergarten and third-grade data was estimated for the first grade activities in first and third grades, respectively. Given differences in
(kindergarten ␣ ⫽ .76; third-grade ␣ ⫽ .75). the scale across waves, items were rescaled to create a common scale:
School administrator reports on whether the community the school 0 ⫽ never or hardly ever, 1 ⫽ monthly, 2 ⫽ weekly, 3 ⫽ daily or
served was supportive of its goals and activities (1 ⫽ strongly dis- almost daily. A sum score of these items was calculated across waves
agree, 2 ⫽ disagree, 3 ⫽ neither agree nor disagree, 4 ⫽ agree, 5 ⫽ and represents literacy instruction (kindergarten ␣ ⫽ .61; first-grade
strongly agree) served as an index of community support for learning. ␣ ⫽ .73; third-grade ␣ ⫽ .69).
A sum score of the degree to which school administrators viewed
eight issues in the community surrounding the school as problems
(tension from differences, unkempt areas, substance abuse, gangs, Analytic Sample
heavy traffic, violent crime, vacant buildings, and crime in the area;
0 ⫽ no problem, 1 ⫽ somewhat of a problem, 2 ⫽ big problem) Exactly 10,998 of the children sampled in the fall kindergarten
represents school neighborhood problems (kindergarten ␣ ⫽ .82; frame and followed to the spring third grade frame had positive,
first-grade ␣ ⫽ .68; third-grade ␣ ⫽ .76). nonzero values on the appropriate sampling weight and were
Field staff observed and reported the presence of four conditions retained in the analytic sample. Thus, the present analyses are
in the community near the school: litter and trash, graffiti, based on a weighted sample size of 3,842,954 children. Table 1
Table 1 in time beyond what can be accounted for by children’s underlying

Percentage Distributions for Selected Characteristics of the trajectories. The HLM 6 software (Raudenbush, Bryk, & Congdon,
Analytic Sample 2005) was used to fit the multilevel model. The HLM program
allows for the use of sampling weights that are associated with a
Population Total complex sampling design and permits the estimation of models
Child race or ethnicity with random intercepts and slopes. It also permits nonequidistant
White 58% times of observation, which is relevant in the ECLS-K data.
Black 16% The data of all models were weighted at the child level in an effort
Hispanic 19% to account for unequal probabilities of selection and nonresponse. The
Asian 4%
present HLM model has a general three-level nested structure: time at
Other 5%
Child sex Level 1; child, family, and neighborhood characteristics at Level 2;
Male 52% and school characteristics at Level 3. Initial examination of the
Female 48% smoothed reading trajectories suggested that reading growth was not
Primary home language linear. As exhibited in Figure 1, many children exhibited a spike or
English 89%
Non-English 12% growth spurt in reading scores between the spring of kindergarten and
Household income the spring of first grade, with slower reading growth exhibited before
Below poverty line 21% and after this period. Accordingly, a three-level model with intercept,
At or above poverty line 79% three linear slope terms (SLOPE K: fall of kindergarten to spring of
kindergarten, SLOPE 1: spring of kindergarten to spring of first grade,
Note. N ⫽ 3,842,954. Detail sums may not equal totals because of
rounding. SLOPE 3: spring of first grade to spring of third grade), and predictors
at each of the three levels was modeled. This model provided superior
fit to the data as compared with a model including a single linear slope
presents the demographic characteristics of those whose data are term (i.e., significantly smaller deviance statistic), and it provided
represented in the current analyses. greater interpretability. The Level 1 model represents how each child
changes over time; the Level 2 model represents variation in growth
parameters among children within the same school; the Level 3 model
Analysis Overview
represents variation in growth parameters across schools. The follow-
In the present study, AM (American Institutes of Research & ing equations represent each of these models:
Cohen, 2005; http://am.air.org/default.asp), a data analysis pack-
age designed to analyze complex sample survey data, was used for Level 1: Y tij ⫽ ␲ 0ij ⫹ ␲ 1ij共SLOPE K兲 ⫹ ␲ 2ij共SLOPE 1兲
estimation of frequency and descriptive information. Because the
⫹ ␲ 3ij共SLOPE 3兲 . . . ⫹ ␲ pij共apij兲 ⫹ ε ij,
ECLS-K data are not simple random samples, the analyses took
into account the stratification of the survey design, and the study’s where Ytij represents child i in school j’s reading performance at
sampling weights were used when computing frequency and de- time t; ␲0ij represents the intercept or the reading performance of
scriptive information as well as tests of statistical significance. child i in school j at the initial assessment; ␲1ij represents the linear
Standard errors were adjusted with a Taylor series approach. Use slope or the monthly learning rate in reading of child i in school j
of sampling weights accounts for unequal probabilities of selection between the fall and the spring of kindergarten; ␲2ij represents the
and nonresponse, and it also allows inferences from the analyses to monthly learning rate in reading of child i in school j between the
be extrapolated to the larger population which the data were meant spring of kindergarten and the spring of first grade; ␲3ij represents
to represent. For the present analyses, longitudinal child sample the monthly learning rate in reading of child i in school j between
weights from the ECLS-K kindergarten to third grade data file the spring of first grade and the spring of third grade; ␲pij repre-
were used (C15PW0; this included children with parent interview sents the strength and the direction of association between predic-
data across the waves of interest in the analyses). tor variables apij and child i in school j’s performance at time t; and
In addition, because the present study concerned the influence of εij represents the time-specific error of child i in school j at time t.
family, neighborhood, and school variables on children’s reading
trajectories, growth curve models were estimated with hierarchical Level 2: ␲ 0ij ⫽ ␤ 00j . . . ⫹ ␤ 0pj共X0pj兲 ⫹ r ij,
linear modeling (HLM) techniques. Growth curve modeling allows
for the estimation of achievement trajectories that model change ␲ 1ij ⫽ ␤ 10j . . . ⫹ ␤ 1pj共X 1pj兲 ⫹ r ij,
over time in relation to multiple influences. These models support
␲ 2ij ⫽ ␤ 20j . . . ⫹ ␤ 2pj共X 2pj兲 ⫹ r ij, and
investigation of covariates or predictors that underlie initial read-
ing achievement, growth over time, and achievement outcomes at ␲ 3ij ⫽ ␤ 30j . . . ⫹ ␤ 3pj共X 3pj兲 ⫹ r ij,
specific points in time during the period of interest. In the present
study, four repeated measures of reading skills are modeled as a where ␤00j represents the mean intercept or the reading perfor-
developmental trajectory. Specific child demographic characteris- mance within school j at the initial assessment; ␤10j represents the
tics (e.g., gender, SES, race, age) influence children’s reading mean linear slope or the learning rate in reading within school j
status in the fall of kindergarten and rates of reading growth between the fall and the spring of kindergarten; ␤20j represents the
through third grade. Additional family, neighborhood, and school linear slope or the learning rate in reading within school j between
characteristics help account for variation in children’s initial read- the spring of kindergarten and the spring of first grade; ␤30j
ing, growth through third grade, and performance at specific points represents the linear slope or the learning rate in reading within
School 1 School 2
Mean Reading Scale Score

144.76 144.76
110.59 110.59
76.43 76.43
42.26 42.26
8.09 8.09
1998 1999 2000 2001 2002 1998 1999 2000 2001 2002
Time of Assessment Time of Assessment
School 3 School 4

144.76 144.76
110.59 110.59
76.43 76.43
42.26 42.26
8.09 8.09
1998 199 1998 1998 1998 1998 1999 2000 2001 2002
Time of Assessment Time of Assessment
Figure 1. Smoothed plots of reading trajectories of a random subsample of children in four schools.
school j between the spring of first grade and the spring of third grade; varying variables include those that may change over time and that are
␤0pj represents the strength and the direction of association between measured at multiple time points, whereas time invariant predictors
predictor variables X0pj and child i in school j’s reading performance reflect data that do not change over time. For example, child’s gender
at the initial assessment, ␲0ij; ␤Xpj represents the strength and the and race and child’s attendance (or nonattendance) at center-based
direction of association between predictor variables XXpj and child i in care the year prior to kindergarten are time invariant. Parents in the
school j’s linear learning rate in reading during the three periods of ECLS-K reported on aspects of the home environment across waves,
interest, ␲1ij, ␲2ij, ␲3ij; and rij represents school j’s deviation from the making these variables time varying. Time-varying predictors pre-
reading score predicted by the model. dicted time-specific performance, whereas time-invariant predictors
predicted the intercept (i.e., initial performance) and slope (i.e.,
Level 3: ␤ 00j ⫽ ␥ 000 . . . ⫹ ␥ 0p0 共X0p0 兲 ⫹ u 00j, growth over time) parameters. School-level variables, whether time
␤ 10j ⫽ ␥ 100 . . . ⫹ ␥ 1p0 共X 1p0 兲 ⫹ u 10j, varying or time invariant, predicted the intercept and slope parame-
ters. Because time-varying predictors were entered at Level 2, they
␤ 20j ⫽ ␥ 200 . . . ⫹ ␥ 2p0 共X 2p0 兲 ⫹ u 20j, and also predicted the intercept and slope parameters.
Because HLM cannot fit a four-level model, classroom and
␤ 30j ⫽ ␥ 300 . . . ⫹ ␥ 3p0 共X 3p0 兲 ⫹ u 30j, teacher characteristics were aggregated and used as covariates
where ␥000 represents the mean intercept or the reading perfor- at the school level. Neighborhood variables that were associated
mance at the initial assessment across schools; ␥100 represents the with school environment were entered as covariates at Level 3.
mean linear slope or the learning rate in reading across schools In HLM models, regression coefficients were interpreted as the
between the fall and the spring of kindergarten; ␥200 represents the value of the dependent variable when the contextual predictor
mean learning rate in reading across schools between the spring of and all the other variables in the model were equal to zero or the
kindergarten and the spring of first grade; ␥300 represents the mean mean. Accordingly, with the exception of dummy codes, all
learning rate in reading across schools between the spring of first contextual predictors were grand mean centered at zero to
grade and the spring of third grade; ␥0p0 represents the strength improve interpretability (Raudenbush & Bryk, 2002; Singer &
and the direction of association between predictor variables X0p0 Willet, 2003).
and mean initial reading performance within school j, ␤00j; ␥Xp0
represents the strength and the direction of association between Results
predictor variables XXp0 and mean linear learning rate within Descriptive Statistics of Analytic Sample
school j during the three periods of interest, ␤10j, ␤20j, ␤30j; and uij
equals overall deviation across schools from the reading score To characterize family, neighborhood, and school characteris-
predicted by the model. tics of participating children, we present descriptive information
All time varying and time invariant covariates were entered into the for the analytic sample in kindergarten, first grade, and third grade
models per recommendations by Singer and Willett (2003). Time in Table 1. As might be expected in a representative sample of U.S.
children, White children made up the majority of the sample (56%), month in reading between the spring of first grade and the spring
and 44% of the children were non-White (16% were reported by of third grade ( p ⬍ .001). Thus, the rate of children’s reading
parents as Black, 19% were reported as Hispanic; 4% were reported growth differed across the three time periods, with the most rapid
as Asian, and 5% were reported as other). English was the main acquisition of skills taking place between the spring of kindergar-
language in about 9 of 10 homes, and about 1 in 5 families had ten and the spring of first grade. See Figure 1 for examples of
household incomes that place them below the federal poverty line. children’s smoothed reading trajectories in a random subsample of
Table 2 presents children’s mean reading achievement. Overall, four schools. The figure illustrates the spike or the heightened
children’s reading achievement and reading achievement variabil- reading skill acquisition occurring between the spring of kinder-
ity increased over time. Significant socioeconomic differences of garten and the spring of first grade. In addition, as noted, there was
11.1 points were observed between the lowest and the highest significant variation around intercept and linear slope terms, sug-
quintiles in children’s reading. IRT scores were observed in the fall gesting that reading trajectories varied across children.
of kindergarten. These differences increased over the first 4 years Next, a conditional model was estimated with child demo-
of schooling. The reading gap between the lowest and the highest graphic characteristics included as predictors of children’s initial
quintiles grew by 2.8 points to 13.9 points from the fall to the reading scores and monthly rates of growth in reading across the
three time periods.1

spring of kindergarten. By the spring of first grade, the gap grew

by 9.7 points to 23.6 points. By the end of third grade, it grew
another 3.6 points to 27.2 points. This is an overall increase of 16.1 Level 1: READ tij ⫽ ␲ 0ij ⫹ ␲ 1ij共SLOPE K兲 ⫹ ␲ 2ij共SLOPE 1兲
points in the reading score gap between the poorest children and
⫹ ␲ 3ij共SLOPE 3兲 ⫹ ε ij.
the most affluent children from kindergarten to third grade, though
most of the increase occurred in the first grade year. In the
Level 2: ␲ 0ij ⫽ ␤ 00j共 AGE兲 ⫹ ␤ 01j共FEMALE兲
following section, we more fully discuss the meaning of these
reading point differences. ⫹ ␤ 02j共WHITE兲 ⫹ ␤ 03j共SES兲 ⫹ r ij,
␲ 1ij ⫽ ␤ 10j共 AGE兲 ⫹ ␤ 11j共FEMALE兲

Reading Trajectories
⫹ ␤ 12j共WHITE兲 ⫹ ␤ 13j共SES兲,
Initially, an unconditional model of reading development was
estimated to establish the functional form of reading growth in the ␲ 2ij ⫽ ␤ 20j共 AGE兲 ⫹ ␤ 21j共FEMALE兲
full sample. Reading achievement was modeled as a linear func-
tion of children’s initial reading (intercept), growth per month in ⫹ ␤ 22j共WHITE兲 ⫹ ␤ 23j共SES兲, and
reading between the fall and the spring of kindergarten (linear
slope), monthly reading growth between the spring of kindergarten ␲ 3ij ⫽ ␤ 30j共 AGE兲 ⫹ ␤ 31j共FEMALE兲
and the spring of first grade (linear slope), monthly reading growth ⫹ ␤ 32j共WHITE兲 ⫹ ␤ 33j共SES兲.
between the spring of first grade and the spring of third grade
(linear slope), and measurement error. The estimated uncondi- Level 3: ␤ 00j ⫽ ␥ 000 ⫹ u 00j,
tional model is represented by the following equations.
␤ 01j ⫽ ␥ 010 ,
Level 1: READ tij ⫽ ␲ 0ij ⫹ ␲ 1ij共SLOPE K兲 ⫹ ␲ 2ij共SLOPE 1兲
␤ 02j ⫽ ␥ 020 ,
⫹ ␲ 3ij共SLOPE 3兲 ⫹ ε ij.
␤ 03j ⫽ ␥ 030 ,
Level 2: ␲ 0ij ⫽ ␤ 00j ⫹ r ij,
␤ 10j ⫽ ␥ 100 ,
␲ 1ij ⫽ ␤ 10j,
␤ 11j ⫽ ␥ 110 ,
␲ 2ij ⫽ ␤ 20j, and
␤ 12j ⫽ ␥ 120 ,
␲ 3ij ⫽ ␤ 30j.
Level 3: ␤ 00j ⫽ ␥ 000 ⫹ u 00j, 1

In the equations, READ ⫽ child’s reading IRT score; SUPPORT ⫽
neighborhood support for the school’s goals and activities (as reported by
␤ 10j ⫽ ␥ 100 , school administrators); AGE ⫽ child age; FEMALE ⫽ child is female;
WHITE ⫽ child is White; SES ⫽ child’s socioeconomic status; PRE K ⫽
␤ 20j ⫽ ␥ 200 , and center-based care prior to kindergarten; HOMELIT ⫽ home literacy environ-
ment; BOOKS ⫽ number of books in the home; INVOLVE ⫽ parent in-
␤ 30j ⫽ ␥ 300 . volvement in the school; STRAIN ⫽ parental role strain; WARMTH ⫽
parental warmth; HPROBS ⫽ neighborhood problems; HSAFETY ⫽ neigh-
Results from the unconditional model suggested that across
borhood safety; BAD ⫽ bad conditions near the school; SUBURB ⫽ subur-
children, mean initial reading achievement was 25.72 points ( p ⬍ ban neighborhood; SPROBS ⫽ school neighborhood problems; BELOW ⫽
.001), and children gained an average of 1.86 points per month in number of children reading below grade; INSTRUCT ⫽ classroom instruction
reading between the fall and the spring of kindergarten ( p ⬍ .001), practices; EXPERIENCE ⫽ teacher’s number of years experience teaching;
2.42 points per month in reading between the spring of kindergar- PREPARATION ⫽ teacher’s preparation to teach; POVERTY ⫽ school
ten and the spring of first grade ( p ⬍ .001), and 1.66 points per poverty concentration; and PRIVATE ⫽ private school.
Table 2
Distributions of Variables Used in HLM Models by SES Quintiles
Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 Total
Variable M SE % M SE % M SE % M SE % M SE % M SE %
Reading fall K 21.8 0.31 24.7 0.28 26.7 0.23 28.8 0.26 33.1 0.37 27.5 0.22
Reading spring K 31.7 0.51 35.7 0.45 38.2 0.31 40.5 0.36 45.6 0.49 39.1 0.30
Reading spring 1st 55.3 0.74 63.5 0.64 67.7 0.52 72.2 0.66 78.9 0.62 69.3 0.48
Reading spring 3rd 92.4 0.85 102.2 0.62 107.9 0.57 112.4 0.57 119.6 0.46 108.8 0.55
Home literacy at K 6.2 0.05 6.5 0.04 6.7 0.05 7.0 0.04 7.2 0.03 6.7 0.02
Home literacy at 1st 6.2 0.07 6.6 0.05 6.7 0.03 6.8 0.04 7.2 0.04 6.7 0.02
Home literacy at 3rd 6.1 0.07 6.1 0.05 6.2 0.04 6.4 0.05 6.6 0.04 6.3 0.03
Books in home in K 32.6 1.7 56.7 1.6 74.5 1.8 88.2 1.9 107.7 1.7 72.8 1.5
Books in home in 1st 40.5 2.1 75.1 2.5 100.9 3.5 113.7 3.2 155.2 3.7 96.5 2.3
Books in home in 3rd 57.1 4.7 95.9 4.3 111.0 4.0 140.8 6.0 171.2 4.2 115.8 3.1
Involvement in school at K 2.5 0.07 3.3 0.06 3.7 0.05 4.1 0.04 4.4 0.04 3.6 0.04
Involvement in school at 1st 2.8 0.06 3.6 0.05 4.1 0.05 4.4 0.05 4.8 0.03 3.9 0.04
Involvement in school at 3rd 3.2 0.05 3.7 0.05 4.1 0.04 4.5 0.04 4.9 0.03 4.1 0.03
Parent role strain at K 1.7 0.03 1.6 0.01 1.5 0.02 1.5 0.01 1.5 0.01 1.6 0.01
Parent role strain at 1st 1.7 0.02 1.6 0.02 1.5 0.01 1.5 0.01 1.5 0.01 1.6 0.01
Parent role strain at 3rd 1.7 0.03 1.6 0.03 1.5 0.02 1.5 0.02 1.5 0.02 1.6 0.01
Parent warmth at K 3.7 0.01 3.7 0.01 3.7 0.01 3.7 0.01 3.7 0.01 3.7 0.01
Parent warmth at 1st 3.7 0.01 3.7 0.01 3.7 0.01 3.7 0.01 3.7 0.01 3.7 0.01
Parent warmth at 3rd 3.6 0.02 3.6 0.02 3.6 0.01 3.7 0.01 3.6 0.01 3.6 0.01
Home neighborhood problems
at K 0.81 0.07 0.55 0.04 0.39 0.03 0.31 0.03 0.19 0.01 0.37 0.03
at 1st 0.66 0.04 0.49 0.04 0.33 0.03 0.24 0.02 0.16 0.01 0.33 0.03
at 3rd 0.81 0.08 0.48 0.04 0.32 0.04 0.23 0.03 0.15 0.01 0.35 0.02
Home neighborhood safety at K 2.4 0.03 2.6 0.02 2.7 0.02 2.8 0.02 2.8 0.01 2.7 0.01
Home neighborhood safety at
1st 2.5 0.02 2.7 0.02 2.7 0.02 2.8 0.02 2.9 0.01 2.7 0.01
Home neighborhood safety at
3rd 2.5 0.02 2.7 0.02 2.8 0.02 2.8 0.01 2.9 0.01 2.7 0.01
School neighborhood problems
at K 3.4 0.25 2.7 0.15 2.4 0.16 2.1 0.14 1.4 0.12 2.1 0.14
at 1st 3.6 0.19 2.7 0.15 2.0 0.11 1.7 0.11 1.2 0.09 1.9 0.13
at 3rd 3.4 0.19 2.6 0.13 2.0 0.13 1.4 0.08 1.1 0.08 1.8 0.12
Community support for
learning at K 3.9 0.05 4.0 0.04 4.0 0.04 4.2 0.04 4.3 0.05 4.1 0.03
learning at 1st 3.9 0.08 4.1 0.05 4.1 0.04 4.2 0.04 4.4 0.04 4.1 0.04
learning at 3rd 3.9 0.06 4.1 0.05 4.1 0.04 4.2 0.04 4.4 0.04 4.2 0.04
Bad conditions near school at K 0.8 0.07 0.6 0.05 0.5 0.04 0.4 0.04 0.4 0.04 0.5 0.04
Bad conditions near school at 1st 1.4 0.05 1.3 0.03 1.2 0.02 1.2 0.02 1.1 0.01 1.2 0.02
Bad conditions near school at
3rd 1.1 0.05 0.9 0.04 0.8 0.03 0.8 0.03 0.7 0.02 0.9 0.03
Peers reading below grade at K 4.7 0.19 3.8 0.13 3.3 0.11 3.1 0.10 2.5 0.11 3.3 0.11
Peers reading below grade at 1st 5.8 0.24 4.9 0.15 4.4 0.15 4.0 0.13 3.6 0.11 4.5 0.15
Peers reading below grade at
3rd 6.6 0.32 5.4 0.21 5.0 0.14 4.5 0.15 3.6 0.10 4.9 0.16
Attends high-poverty school in
K 67% 59% 48% 37% 26% 48%
1st 60% 51% 42% 31% 22% 41%
3rd 37% 24% 15% 9% 4% 18%
Literacy instruction at K 11.7 0.06 11.7 0.06 11.7 0.05 11.7 0.05 11.8 0.05 11.7 0.07
Literacy instruction at 1st 18.1 0.18 17.9 0.15 17.7 0.15 17.5 0.18 17.0 0.19 17.6 0.12
Literacy instruction at 3rd 23.4 0.21 23.3 0.19 23.1 0.24 23.3 0.20 23.3 0.21 23.3 0.14
Teacher experience at K 17.4 0.64 18.9 0.81 19.2 0.72 18.9 0.67 18.3 0.64 18.5 0.53
Teacher experience at 1st 14.8 0.65 16.6 0.61 17.2 0.59 16.6 0.54 17.2 0.59 17.2 0.54
Teacher experience at 3rd 15.7 0.75 16.6 0.60 17.5 0.57 16.5 0.42 16.2 0.54 17.3 0.47
Teacher preparation at K 18.3 0.28 18.3 0.25 18.6 0.23 18.8 0.22 18.9 0.27 18.5 0.23
Teacher preparation at 1st 17.8 0.26 17.8 0.22 18.4 0.23 18.3 0.20 18.6 0.23 18.5 0.19
Teacher preparation at 3rd 16.0 0.27 16.5 0.21 16.1 0.22 16.4 0.20 16.4 0.20 16.3 0.17
Note. For Quintile 1, n ⫽ 765,187; for Quintile 2, n ⫽ 767,500; for Quintile 3, n ⫽ 783,296; for Quintile 4, n ⫽ 758,962; for Quintile 5, n ⫽ 758,962.
Total N ⫽ 3,833,907. Numbers in parentheses are standard errors. HLM ⫽ hierarchical linear modeling; SES ⫽ socioeconomic status; K ⫽ kindergarten;
1st ⫽ first grade; 3rd ⫽ third grade.
␤ 13j ⫽ ␥ 130 , reading growth differs across the three time periods, with the
period between the spring of kindergarten and the spring of first
␤ 20j ⫽ ␥ 200 , grade being that of greatest differentiation by SES (see Table 3 for
parameter estimates).
␤ 21j ⫽ ␥ 210 ,
␤ 23j ⫽ ␥ 230 , Family Characteristics, Reading Achievement, and

Growth
␤ 31j ⫽ ␥ 310 ,
Next, family covariates (home literacy environment, school
␤ 32j ⫽ ␥ 320 , and involvement, parental role strain, number of books in home, and
parental warmth) were added to the model to assess the extent to
␤ 33j ⫽ ␥ 330 .
which SES gaps in initial reading and monthly reading growth
According to this piecewise model, children who were older, rates during the three periods of interest could be explained by
female, and from higher SES households had higher reading family characteristics beyond demographic characteristics. This
scores at the initial assessment. In addition, female and high- model revealed the effects of family covariates on initial reading
SES children gained reading skills at a faster rate than did their scores, monthly reading growth during the three periods, and
peers between the fall and the spring of kindergarten; White, time-specific reading performance beyond what could be predicted
female, and high-SES children gained reading skills at a faster by children’s underlying trajectories and demographic character-
rate than did other children between the spring of kindergarten istics.
and the spring of first grade; and younger, White, and high-SES
children gained reading skills at a faster rate between the spring
of first grade and the spring of third grade. Of most importance ⫹ ␲ 3ij共SLOPE 3兲 ⫹ ␲ 4ij共HOMELIT兲 ⫹ ␲ 5ij共BOOKS兲
are the parameter estimates representing the effect of family
SES on children’s initial reading skills and monthly growth ⫹ ␲ 6ij共INVOLVE兲 ⫹ ␲ 7ij共STRAIN兲 ⫹ ␲ 8ij共WARMTH兲 ⫹ ε ij.
during the three periods.
When other child characteristics were controlled for, as SES in- Level 2: ␲ 0ij ⫽ ␤ 00j ⫹ ␤ 01j共 AGE兲 ⫹ ␤ 02j共FEMALE兲
creased, so did children’s reading achievement at the initial assess- ⫹ ␤ 03j共WHITE兲 ⫹ ␤ 04j共SES兲
ment in the fall of kindergarten. In fact, children at 1 standard
deviation above the mean in SES had reading scores at the initial ⫹ ␤ 05j共PRE K兲 ⫹ r ij,
kindergarten assessment that were 2.24 points higher than children of
average SES and 4.48 points higher than children 1 standard deviation ␲ 1ij ⫽ ␤ 10j ⫹ ␤ 11j共 AGE兲 ⫹ ␤ 12j共FEMALE兲
below the mean in SES. Given that 1.86 was the average monthly
⫹ ␤ 13j共WHITE兲 ⫹ ␤ 14j共SES兲
learning rate for the full sample in the kindergarten year, a 2.24-point
advantage in initial reading is a head start, for children 1 standard ⫹ ␤ 15j共PRE K兲,
deviation above average in SES, of more than 1 month relative to
children of average SES. It is a head start of more than 2 months ␲ 2ij ⫽ ␤ 20j ⫹ ␤ 21j共 AGE兲 ⫹ ␤ 22j共FEMALE兲
relative to children 1 standard deviation below mean SES.
⫹ ␤ 23j共WHITE兲 ⫹ ␤ 24j共SES兲 ⫹ ␤ 25j共PRE K兲,
Moreover, on the basis of the model, we expected that children
who differed by 1 standard deviation in SES would differ by 0.10 ␲ 3ij ⫽ ␤ 30j ⫹ ␤ 31j共 AGE兲 ⫹ ␤ 32j共FEMALE兲
points per month in the rate of reading skill acquisition during the
kindergarten year, 0.19 points per month between the spring of ⫹ ␤ 33j共WHITE兲 ⫹ ␤ 34j共SES兲 ⫹ ␤ 35j共PRE K兲,
kindergarten and the spring of first grade, and 0.02 points per
month between the spring of first grade and the spring of third ␲ 4ij ⫽ ␤ 40j,
grade. Thus, higher SES relates to higher initial achievement in the ␲ 5ij ⫽ ␤ 50j,
fall of kindergarten and more rapid reading growth per month
across the three periods of interest. The rate of these advantages in ␲ 6ij ⫽ ␤ 60j,
Table 3
Parameter Estimates and Standard Errors for the Child Model
Independent variable Intercept K linear slope K–1 linear slope 1–3 linear slope
Age at initial assessment 0.42 (0.04)*** ⫺0.00 (0.00) 0.00 (0.00) ⫺0.01 (0.00)***
Female 1.78 (0.29)*** 0.11 (0.03)** 0.15 (0.03)*** ⫺0.00 (0.02)
White 0.60 (0.43) 0.06 (0.04) 0.14 (0.04)** 0.15 (0.02)***
SES 2.24 (0.15)*** 0.097 (0.02)*** 0.189 (0.02)*** 0.019 (0.01)*
Note. Numbers in parentheses are robust standard errors. K ⫽ kindergarten; K–1 ⫽ kindergarten to first grade; 1–3 ⫽ first grade to third grade; SES ⫽
socioeconomic status.
*
p ⱕ .05. ** p ⱕ .01. *** p ⱕ .001.
␲ 7ij ⫽ ␤ 70j, and ␤ 34j ⫽ ␥ 340 ,
␲ 8ij ⫽ ␤ 80j. ␤ 35j ⫽ ␥ 350 ,
Level 3: ␤ 00j ⫽ ␥ 000 ⫹ u 00j, ␤ 40j ⫽ ␥ 400 ,
␤ 01j ⫽ ␥ 010 , ␤ 50j ⫽ ␥ 500 ,
␤ 02j ⫽ ␥ 020 , ␤ 60j ⫽ ␥ 600 ,
␤ 03j ⫽ ␥ 030 , ␤ 70j ⫽ ␥ 700 , and
␤ 04j ⫽ ␥ 040 , ␤ 80j ⫽ ␥ 800 .
␤ 05j ⫽ ␥ 050 , Comparisons of the SES coefficient on the intercept and the
three slope terms from this model with those of the previous
␤ 10j ⫽ ␥ 100 ,
demographic model indicated the extent to which mean differ-

␤ 11j ⫽ ␥ 110 , ences in initial reading scores and learning rates in reading were
attributable to family characteristics. In fact, with the addition
␤ 12j ⫽ ␥ 120 , of family variables, the SES-intercept coefficient was reduced
by 16%, though the three slope parameters were not reduced.
␤ 13j ⫽ ␥ 130 , Thus, characteristics associated with the family context helped
account for SES differences in children’s initial reading scores,
␤ 14j ⫽ ␥ 140 ,
but they did not account for many of the SES differences in the
␤ 15j ⫽ ␥ 150 , rate at which children gained reading skills during the three
periods of interest. Notably, children with richer home literacy
␤ 20j ⫽ ␥ 200 , environments (T ⫽ 6.20, p ⬍ .001, effect size [es] ⫽ .07), more
books at home (T ⫽ 2.10, p ⬍ .05, es ⫽ .02), and parents who
␤ 21j ⫽ ␥ 210 ,
were less strained in the parenting role (T ⫽ ⫺3.28, p ⬍ .001,
␤ 22j ⫽ ␥ 220 , es ⫽ .04) had enhanced time-specific reading performance
during the first 4 years of school. On the basis of the model,
␤ 23j ⫽ ␥ 230 , children who attended center-based care the year prior to kin-
dergarten were expected to have higher initial reading scores
␤ 24j ⫽ ␥ 240 , (T ⫽ 4.46, p ⬍ .001, es ⫽ .09), though receipt of such care was
␤ 25j ⫽ ␥ 250 , not expected to relate to reading growth during any of the three
time periods through third grade (see Table 4).
␤ 30j ⫽ ␥ 300 ,
␤ 31j ⫽ ␥ 310 , Neighborhood Characteristics, Reading Achievement, and

Growth
␤ 32j ⫽ ␥ 320 ,
Next, neighborhood covariates were added to the model to allow
␤ 33j ⫽ ␥ 330 , assessment of the extent to which SES gaps in initial reading and
Table 4
Parameter Estimates and Standard Errors for the Family Model
Time-specific
Independent variable Intercept K linear slope K–1 linear slope 1–3 linear slope performance
Female 1.42 (0.29)*** 0.10 (0.03)** 0.16 (0.04)*** ⫺0.00 (0.02)
White 0.15 (0.42) 0.05 (0.04) 0.13 (0.04)** 0.16 (0.02)***
SES 1.88 (0.16)*** 0.104 (0.03)*** 0.188 (0.02)*** 0.016 (0.01)
Center-based care 1.49 (0.33)*** ⫺0.07 (0.05) ⫺0.01 (0.04) 0.01 (0.02)
Home literacy environment 0.47 (0.08)***
Books in home 0.00 (0.00)*
Involvement in school 0.17 (0.10)
Parental role strain ⫺1.25 (0.33)***
Parental warmth 0.31 (0.36)
Note. Numbers in parentheses are robust standard errors. K ⫽ kindergarten; K–1 ⫽ kindergarten to first grade; 1–3 ⫽ first grade to third grade; SES ⫽
socioeconomic status.
*
p ⱕ .05. ** p ⱕ .01. *** p ⱕ .001.
monthly reading growth rates could be explained by neighborhood ␤ 20j ⫽ ␥ 200 ⫹ ␥ 201 共SUBURB兲 ⫹ ␥ 202 共SUPPORT兲
characteristics beyond children’s demographic and family charac-
teristics. ⫹ ␥ 203 共SPROBS兲 ⫹ ␥ 204 共BAD兲,
Level 1: READ tij ⫽ ␲ 0ij ⫹ ␲ 1ij共SLOPE K兲 ⫹ ␲ 2ij共SLOPE 1兲 ␤ 21j ⫽ ␥ 210 ,
⫹ ␲ 3ij共SLOPE 3兲 ⫹ ␲ 4ij共HOMELIT兲 ⫹ ␲ 5ij共BOOKS兲 ␤ 22j ⫽ ␥ 220 ,
⫹ ␲ 6ij共INVOLVE兲 ⫹ ␲ 7ij共STRAIN兲 ⫹ ␲ 8ij共WARMTH兲 ␤ 23j ⫽ ␥ 230 ,
⫹ ␲ 9ij共HPROBS兲 ⫹ ␲ 10ij共HSAFETY兲 ⫹ ε ij. ␤ 24j ⫽ ␥ 240 ,
Level 2: ␲ 0ij ⫽ ␤ 00j ⫹ ␤ 01j共 AGE兲 ⫹ ␤ 02j共FEMALE兲 ␤ 25j ⫽ ␥ 250 ,
⫹ ␤ 03j共WHITE兲 ⫹ ␤ 04j共SES兲 ⫹ ␤ 05j共PRE K兲 ⫹ r ij, ␤ 30j ⫽ ␥ 300 ⫹ ␥ 301 共SUBURB兲 ⫹ ␥ 302 共SUPPORT兲
␲ 1ij ⫽ ␤ 10j ⫹ ␤ 11j共 AGE兲 ⫹ ␤ 12j共FEMALE兲 ⫹ ␥ 303 共SPROBS兲 ⫹ ␥ 304 共BAD兲,

⫹ ␤ 13j共WHITE兲 ⫹ ␤ 14j共SES兲 ⫹ ␤ 15j共PRE K兲, ␤ 31j ⫽ ␥ 310 ,
␲ 2ij ⫽ ␤ 20j ⫹ ␤ 21j共 AGE兲 ⫹ ␤ 22j共FEMALE兲 ␤ 32j ⫽ ␥ 320 ,
␲ 3ij ⫽ ␤ 30j ⫹ ␤ 31j共 AGE兲 ⫹ ␤ 32j共FEMALE兲 ␤ 34j ⫽ ␥ 340 ,
␲ 4ij ⫽ ␤ 40j, ␤ 40j ⫽ ␥ 400 ,
␲ 5ij ⫽ ␤ 50j, ␤ 50j ⫽ ␥ 500 ,
␲ 6ij ⫽ ␤ 60j, ␤ 60j ⫽ ␥ 600 ,
␲ 7ij ⫽ ␤ 70j, ␤ 70j ⫽ ␥ 700 ,
␲ 8ij ⫽ ␤ 80j, ␤ 80j ⫽ ␥ 800 ,
␲ 9ij ⫽ ␤ 90j, and ␤ 90j ⫽ ␥ 900 , and
␲ 10ij ⫽ ␤ 100j. ␤ 100j ⫽ ␥ 1000 .
Level 3: ␤ 00j ⫽ ␥ 000 ⫹ ␥ 001 共SUBURB兲 ⫹ ␥ 002 共SUPPORT兲 With the addition of neighborhood variables, the SES-intercept
coefficient was reduced by 2%, the slope parameter for the period
⫹ ␥ 003 共SPROBS兲 ⫹ ␥ 004 共BAD兲 ⫹ u 00j, between the spring of kindergarten and the spring of first grade was
reduced by 1%, and the slope parameter for the period between the
␤ 01j ⫽ ␥ 010 , spring of first grade and the spring of third grade was reduced by 2%.
The kindergarten slope parameter was unchanged with the addition of
␤ 02j ⫽ ␥ 020 ,
neighborhood covariates. Thus, this model suggested that beyond
␤ 03j ⫽ ␥ 030 , family and demographic characteristics, neighborhood conditions did
not account much for expected differences in children’s initial reading
␤ 04j ⫽ ␥ 040 , achievement in the fall of kindergarten, but they did account for
expected differences in rates of children’s monthly reading growth as
␤ 05j ⫽ ␥ 050 ,
children matured. In addition, schools with poor conditions in the
␤ 10j ⫽ ␥ 100 ⫹ ␥ 101 共SUBURB兲 ⫹ ␥ 102 共SUPPORT兲 surrounding area (T ⫽ ⫺2.19, p ⬍ .05, es ⫽ .03) had constrained
reading growth during the kindergarten year and during the period
⫹ ␥ 103 共SPROBS兲 ⫹ ␥ 104 共BAD兲, between the spring of kindergarten and the spring of first grade (T ⫽
⫺2.07, p ⬍ .05, es ⫽ .02). Community support for learning was
␤ 11j ⫽ ␥ 110 , related to higher school mean reading scores at the initial kindergarten
␤ 12j ⫽ ␥ 120 , assessment (T ⫽ 2.79, p ⬍ .01, es ⫽ .1890; see Table 5).
␤ 13j ⫽ ␥ 130 , School Characteristics, Reading Achievement, and

Growth
␤ 14j ⫽ ␥ 140 ,
To assess the extent to which SES gaps in initial reading and
␤ 15j ⫽ ␥ 150 , monthly reading growth rates could be explained by school char-
Table 5
Parameter Estimates and Standard Errors for the Neighborhood Model
Time-specific
Independent variable Intercept K linear slope K–1 linear slope 1–3 linear slope performance
Female 1.41 (0.29)*** 0.10 (0.03)** 0.16 (0.04)*** ⫺0.00 (0.02)
White 0.60 (0.41) 0.05 (0.05) 0.07 (0.05) 0.15 (0.03)***
SES 1.85 (0.15)*** 0.105 (0.02)*** 0.177 (0.02)*** 0.016 (0.01)
Home neighborhood problems ⫺0.17 (0.18)

Home neighborhood safety 0.02 (0.25)

Suburban neighborhood ⫺0.30 (0.61) ⫺0.09 (0.05) 0.02 (0.04) 0.00 (0.03)
Community support for learning 0.99 (0.36)** 0.03 (0.03) 0.00 (0.03) ⫺0.01 (0.01)
School neighborhood problems ⫺0.16 (0.16) 0.01 (0.01) ⫺0.02 (0.01) ⫺0.01 (0.01)
Bad conditions near school 0.13 (0.39) ⫺0.08 (0.04)* ⫺0.07 (0.03)* ⫺0.02 (0.02)
Note. Number of schools ⫽ 939. Numbers in parentheses are robust standard errors. K ⫽ kindergarten; K–1 ⫽ kindergarten to first grade; 1–3 ⫽ first
grade to third grade; SES ⫽ socioeconomic status.
*
p ⱕ .05. ** p ⱕ .01. *** p ⱕ .001.
acteristics, beyond children’s demographic and family character- ⫹ ␥ 006 共PRIVATE兲 ⫹ u 00j,
istics, we added school covariates to the model.
␤ 01j ⫽ ␥ 010 ,
␤ 02j ⫽ ␥ 020 ,
⫹ ␲ 3ij共SLOPE 3兲 ⫹ ␲ 4ij共HOMELIT兲 ⫹ ␲ 5ij共BOOKS兲
␤ 03j ⫽ ␥ 030 ,
⫹ ␲ 6ij共INVOLVE兲 ⫹ ␲ 7ij共STRAIN兲 ⫹ ␲ 8ij共WARMTH兲 ⫹ ε ij.
␤ 04j ⫽ ␥ 040 ,
Level 2: ␲ 0ij ⫽ ␤ 00j ⫹ ␤ 01j共 AGE兲 ⫹ ␤ 02j共FEMALE兲
⫹ ␤ 03j共WHITE兲 ⫹ ␤ 04j共SES兲 ⫹ ␤ 05j共PRE K兲 ⫹ r ij, ␤ 05j ⫽ ␥ 050 ,
␲ 1ij ⫽ ␤ 10j ⫹ ␤ 11j共 AGE兲 ⫹ ␤ 12j共FEMALE兲 ␤ 10j ⫽ ␥ 100 ⫹ ␥ 101 共BELOW兲 ⫹ ␥ 102 共INSTRUCT兲
⫹ ␤ 13j共WHITE兲 ⫹ ␤ 14j共SES兲 ⫹ ␤ 15j共PRE K兲, ⫹ ␥ 103 共EXPERIENCE兲 ⫹ ␥ 104 共PREPARATION兲
␲ 2ij ⫽ ␤ 20j ⫹ ␤ 21j共 AGE兲 ⫹ ␤ 22j共FEMALE兲 ⫹ ␥ 105 共POVERTY兲 ⫹ ␥ 106 共PRIVATE兲,

␲ 3ij ⫽ ␤ 30j ⫹ ␤ 31j共 AGE兲 ⫹ ␤ 32j共FEMALE兲
␤ 12j ⫽ ␥ 120 ,
⫹ ␤ 33j共WHITE兲 ⫹ ␤ 34j共SES兲 ⫹ ␤ 35j共PRE K兲,
␤ 13j ⫽ ␥ 130 ,
␲ 4ij ⫽ ␤ 40j,
␤ 14j ⫽ ␥ 140 ,
␲ 5ij ⫽ ␤ 50j,
␤ 15j ⫽ ␥ 150 ,
␲ 6ij ⫽ ␤ 60j,
␤ 20j ⫽ ␥ 200 ⫹ ␥ 201 共BELOW兲 ⫹ ␥ 202 共INSTRUCT兲
␲ 7ij ⫽ ␤ 70j, and
␲ 8ij ⫽ ␤ 80j. ⫹ ␥ 203 共EXPERIENCE兲 ⫹ ␥ 204 共PREPARATION兲
Level 3: ␤ 00j ⫽ ␥ 000 ⫹ ␥ 001 共BELOW兲 ⫹ ␥ 002 共INSTRUCT兲 ⫹ ␥ 205 共POVERTY兲 ⫹ ␥ 206 共PRIVATE兲,
⫹ ␥ 003 共EXPERIENCE兲 ␤ 21j ⫽ ␥ 210 ,
⫹ ␥ 004 共PREPARATION兲 ⫹ ␥ 005 共POVERTY兲 ␤ 22j ⫽ ␥ 220 ,

␤ 23j ⫽ ␥ 230 , not reduced with school variables added. This suggested that
although characteristics associated with the school environment
␤ 24j ⫽ ␥ 240 , accounted for a small portion of SES gaps in children’s initial
reading skills, these characteristics accounted for a larger portion
␤ 25j ⫽ ␥ 250 , of differences in the rates of children’s monthly reading growth
during the periods of interest, with the largest effect between the
␤ 30j ⫽ ␥ 300 ⫹ ␥ 301 共BELOW兲 ⫹ ␥ 302 共INSTRUCT兲
spring of kindergarten and the spring of first grade. Notably, in
⫹ ␥ 303 共EXPERIENCE兲 ⫹ ␥ 304 共PREPARATION兲 private schools (T ⫽ 1.92, p ⬍ .05, es ⫽ .13), children had higher
initial reading scores, and in schools with more children reading
⫹ ␥ 305 共POVERTY兲 ⫹ ␥ 306 共PRIVATE兲, below grade level in classrooms (T ⫽ ⫺2.94, p ⬍ .01, es ⫽ .19),
children had constrained reading performance at the initial assess-
␤ 31j ⫽ ␥ 310 , ment in the fall of kindergarten. In these schools, children evi-
denced slower average reading growth in the period between the
␤ 32j ⫽ ␥ 320 , spring of kindergarten and the spring of first grade (T ⫽ ⫺2.25,
p ⬍ .05, es ⫽ .03). Children in high poverty schools evidenced

␤ 33j ⫽ ␥ 330 , slower reading growth between the spring of kindergarten and the
spring of first grade (T ⫽ ⫺2.04, p ⬍ .05, es ⫽ .02). There was no
␤ 34j ⫽ ␥ 340 ,
evidence that teacher experience, preparation, or classroom liter-
␤ 35j ⫽ ␥ 350 , acy instruction influenced initial reading scores or monthly reading
growth rates beyond children’s anticipated trajectories and demo-
␤ 40j ⫽ ␥ 400 , graphic and family characteristics (see Table 6). In other work at
prekindergarten, elementary, and secondary grade levels, research-
␤ 50j ⫽ ␥ 500 , ers have found no relation (see for example, Early et al., 2006), not
even a negative relation (Goldhaber & Brewer, 1997), between
␤ 60j ⫽ ␥ 600 , teacher background and student achievement.
In sum, these models suggest that the family context best ac-
␤ 70j ⫽ ␥ 700 , and
counts for SES disparities in children’s initial reading achievement
␤ 80j ⫽ ␥ 800 . as they enter school. In addition, schools and neighborhoods best
explain SES gaps in children’s rates of monthly reading growth,
With the addition of school variables, the SES-intercept coeffi- particularly as children mature, whereas the families appear to do
cient was reduced by less than 1%, and the kindergarten slope less to account for such gaps. The period between the spring of
parameter was reduced by 4%. The slope parameter for the period kindergarten and the spring of first grade is the one of the most
between the spring of kindergarten and the spring of first grade rapid reading growth for children and the greatest differentiation
was reduced by 13%. Notably, the slope parameter for the period on demographic, school, and neighborhood characteristics for chil-
between the spring of first grade and the spring of third grade was dren. It is important to note, however, that much of the SES gap in
Table 6
Parameter Estimates and Standard Errors for the School Model
Time-specific
Independent variables Intercept K linear slope K–1 linear slope 1–3 linear slope performance
Female 1.47 (0.29)*** 0.10 (0.03)** 0.16 (0.03)*** ⫺0.00 (0.02)
White 0.55 (0.41) 0.04 (0.05) 0.09 (0.05)* 0.16 (0.03)***
SES 1.88 (0.16)*** 0.101 (0.02)** 0.163 (0.02)*** 0.019 (0.01)*
Poor readers in classroom ⫺0.44 (0.15)** ⫺0.02 (0.01) ⫺0.02 (0.01)* 0.00 (0.01)
Literacy instruction 0.06 (0.05) ⫺0.00 (0.00) 0.00 (0.00) ⫺0.00 (0.00)
Teacher experience ⫺0.03 (0.03) ⫺0.00 (0.00) 0.00 (0.00) 0.00 (0.00)
Teacher preparation ⫺0.03 (0.11) ⫺0.00 (0.01) 0.01 (0.01) ⫺0.00 (0.00)
High-poverty school 0.03 (0.66) 0.02 (0.05) ⫺0.09 (0.05)* ⫺0.02 (0.03)
Private school 1.64 (0.85)* ⫺0.05 (0.08) 0.07 (0.06) ⫺0.06 (0.04)
Note. Number of schools ⫽ 939. Numbers in parentheses are robust standard errors. K ⫽ kindergarten; K–1 ⫽ kindergarten to first grade; 1–3 ⫽ first
grade to third grade; SES ⫽ socioeconomic status.
*
p ⱕ .05. ** p ⱕ .01. *** p ⱕ .001.
initial reading achievement and in monthly learning rates remains Although family factors were most strongly associated with
unexplained by the variables included in the model. Table 7 SES differences in children’s reading competence at the initial
includes random effects for all models. assessment, they were less associated with differences in gains
or lack of progress in children’s reading competence up to third
Discussion grade. Conversely, experiences and resources associated with
school and neighborhood conditions were understandably less
In the present study, we investigated the reading trajectories of associated with SES differences in initial reading scores but
children of differing socioeconomic backgrounds, from kindergar- more closely associated with differences in children’s reading
ten to third grade, using a nationally representative sample. In
skills acquisition rate, particularly during the period of most
addition, a special concern was understanding the unique and
rapid reading growth— between the spring of kindergarten and
cumulative role of family, neighborhood, and school contexts in
the spring of first grade. Thus, there is a reversal in the relative
accounting for disparities in reading achievement during this time
relation of these settings once children enter school, such that
period. First, when considering family context, the following were
qualities of school and community are associated with differ-
all significantly related to reading outcomes: home literacy envi-
ences in reading development to a greater extent than is family

ronment, number of books owned, parent distress, and receipt of
life. Accordingly, family life can be viewed as being strongly

center-based care. This result suggests that both family resource
associated with the starting point of children’s reading compe-
and/or investment models (e.g., books in home; Yeung, Linver, &
tence, with other ecological settings being more strongly asso-
Brooks-Gunn, 2002) and family process models (e.g., parent dis-
ciated with children’s reading progress. It is important to note
tress; Yeung et al., 2002) help explain reading outcomes. Of
notable importance to reading outcomes are social relationships that even in the presence of family characteristics, neighbor-
and processes (e.g., parental role strain). hood and school characteristics were associated with children’s
Next, each of the contexts was associated with SES gaps in reading performance. This reinforces the need to focus on
initial reading, reading achievement at specific points in time, characteristics and qualities of family life in preschool children
and/or monthly learning rates. However, resources, experiences, and to widen the emphasis to include schools and neighbor-
and relationships associated with the family context were most hoods as children enter school. It is important to note that these
closely associated with reading gaps at the initial kindergarten findings are consistent with the theoretical framework guiding
assessment. The importance of the experiences within the family the study—specifically that the relative salience of contexts
context for early reading is hardly surprising because most chil- may shift over time. As Bronfenbrenner (1989) noted, not only
dren had experienced little schooling by the first assessment. does the child change over time but the systems with which the
Specifically, the results suggest that the relation between SES and child interacts change as well. Consequently, the nature of the
children’s initial reading competence is mediated by home literacy child’s interactions with these systems and their effect on the
environment, number of books available within the home to the child may change over time. This is consistent with work by
child, parental involvement in the school, parental role strain and others (Aber, Gephart, Brooks-Gunn, & Connell, 1997) that
warmth, and provision of center-based care prior to kindergarten. recognized the differential effects of environmental contexts
This is a useful finding. Although it may be difficult to alter family across developmental epochs.
SES quickly enough to make a difference for young children, Few studies, if any, have examined the relation between
interventions can reduce the adverse effects of family mediators. neighborhood context and reading development of the very
The results of this study underscore the wisdom of programs that young. This study provides preliminary evidence that neighbor-
direct resources to strengthening family literacy environment, en- hood context may in fact be related to children’s growth in
couraging parental involvement in schools, and reducing parental reading. Specifically, community support for the school and
role strain. poor physical conditions surrounding the school were associ-
Table 7
Random Effects for All Models
Child model Family model Neighborhood model School model
Random effect Variance ␹2 df Variance ␹2 df Variance ␹2 df Variance ␹2 df
Levels 1 and 2
variance
components
intercept (r0) 84.77 135,806.99 8017 81.91 47,537.35*** 7901 82.20 157,332.28*** 7889 82.00 90,094.94 7901
Level 3 variance
components
intercept (u00) 35.52 3,184.22*** 936 35.50 3,170.02*** 925 32.55 3,057.63*** 920 31.52 2,976.21*** 919
Deviance 216,069.11 (23) 207,251.43 (32) 206,276.11 (50) 207,092.46 (56)
Note. In the last row, the numbers in parentheses are parameters. For the chi squares, N ⫽ df ⫹ 1.
***
p ⱕ .001.
ated with children’s reading. Future studies should explore in vantage does not stop there. These children will often enter schools
greater detail the nature and the role of neighborhood context on that have a higher proportion of poor children and those with low
cognitive outcomes and on young children. Exploring the role reading skills (Lee & Burkam, 2002; Phillips & Chin, 2004). Each
of various schooling characteristics to reading outcomes and of these factors is associated with poorer reading outcomes (Xue &
SES disparities in reading is relatively novel. Peers were a Meisels, 2004). Schools may be overwhelmed in trying to serve
critical component of school context associated with children’s concentrations of children who require considerable attention and
reading outcomes in the present study. The number of children resources. The present analyses support the policy that some
reading below grade and the presence of low-income peers were schools have adopted of dispersing low-SES children across mul-
consistently associated with initial achievement and growth tiple schools so that there is not a high concentration of low-
rates. The present findings, therefore, speak to the importance income children in a single school. Because school attendance
of grouping and segregation that occurs within schools. Nota- areas correspond to neighborhoods and neighborhoods are often
bly, they are consistent with other work (e.g., Coleman, 1966) economically segregated, this policy often works against the cher-
that suggested that “the social composition of the student body ished notion of neighborhood schools. As a consequence, stiff
is more highly related to achievement, independent of the resistance to busing has arisen from parents who are unwilling to
student’s own social background, than is any school factor” pay the price of dislocation necessary to implement these pro-
(Coleman, 1966, p. 325). Thus, those designing educational grams.
policy to eliminate reading disparities must be aware of the
relation of the presence and the concentration of peers with Limitations
limited skills or fewer economic resources to students’ skills
and achievement. Teachers’ experience, preparation, and class- Several limitations mark the present study. First, a considerable
room literacy instruction were not consistently related to chil- portion of the SES gap remains unexplained by the models pre-
dren’s reading outcomes beyond their expected trajectories. sented in the current analyses. Across models, we were able to
However, previous findings examining teacher effects have account for only 17% of the initial SES gap and up to 16% of the
been mixed (Early et al. 2006). With a few notable exceptions, growth rate gaps. Related to this, the selection of covariates for the
this work has not focused on the reading achievement of very growth models was not exhaustive of the ECLS-K dataset or of the
young children. Taken together, the current findings highlight set of contextual variables that may be related to reading achieve-
the ecological and systemic nature of development and the fact ment or SES. Certainly, there are other factors that may be linked
that resources and experiences across multiple contexts are to reading achievement and SES that were not explored in the
associated with children’s reading development. Morever, these analyses presented here. The inability to explain a larger propor-
findings underscore the influence of SES. Consistent with the tion of the SES gap may be a result of the variables selected; that
study’s theoretical framework, in which developmental out- is, other variables may better capture contributing factors to the
comes are a function of dynamic settings and contexts that reading achievement gap. This difficulty also may be due to the
interact with one another (Bronfenbrenner, 1979, 1989; Mag- limited measurement and the inherent weaknesses of relying on
nusson & Cairns, 1996), contexts produce unique and cumula- self-report measures of family practices and environmental condi-
tive influences on reading outcomes and disparities. tions. Self-reports may be biased and affected by the social desir-
In addition to describing the relation between family, school, ability of respondents. Others have similarly noted that parents
and neighborhood characteristics to SES reading, our analyses with different educational backgrounds might vary in their ability
also uncovered important phenomena in children’s development to provide accurate reports (Dickinson & DeTemple, 1998). In
of early reading competence. The SES-related differences in addition, observational techniques or qualitative methods would be
reading were stark at the initial kindergarten assessment fol- especially beneficial in capturing the true nature of children’s
lowing children’s entry to school. Furthermore, the SES gap in contexts. For example, important aspects of teacher– child and
reading between the poorest and the most affluent children grew parent– child interactions and social characteristics of those envi-
as children progressed through school. The first grade year ronments cannot be fully captured by a large-scale dataset like the
appears to be a critical time, as the SES gap in reading achieve- ECLS-K. The researchers recognize that the advantages associated
ment grew the widest during this period. This may be due to the with using a large, nationally representative sample are tempered
coming together of phonological processing, rapid accurate by the preclusion of other techniques that offer important insights
decoding, and comprehension by skilled readers during this in the contexts affecting children.
time while the less skilled readers are still struggling with Finally, perhaps a more fruitful analysis of school versus
letter-sound combinations. It is important to note that this nonschool factors would split gains into seasons and examine
period was associated with the greatest differentiation across school learning versus summer learning. That is, by examining
children on the basis of demographic characteristics and other students’ reading gains across years, as is done in the present
ecological factors. For example, high-SES children started off analyses, it is more difficult to truly distinguish school effects
as better readers and had more rapid progress than did low-SES from nonschool ones. We also may have underestimated school
children, across the three time periods, with the gap growing the effects by not partitioning gains by seasons. Analyses by
most between the spring of kindergarten and the spring of first Downey, von Hippel, and Broh (2004) suggested that learning
grade. during the summer is more variable than is learning during the
Our analyses also suggest a compounding effect of low quality school year and that summer experiences are more responsible
environments. Children from low-SES homes grow up in home for the achievement gap. Others (see Rothstein, 2004) have
environments poor in literacy experiences. However, their disad- made similar claims that children’s summertime experiences
(and afterschool hours) play a role in contributing to disparities shaped by social, political, and economic forces occurring
in achievement. In similar work, Burkam, Ready, Lee, and outside of the home. Policies and interventions that fail to
LoGerfo (2004) found that social stratification occurs during recognize such intricacies may not ultimately succeed. Any
the summer months, with higher SES children learning more policy seeking to improve conditions within the family system
during these periods. However, highly interpretable models in should seek to produce change in the contexts that families
the current analyses mimic children’s overall reading and operate within as well.
growth experiences, as losses and gains that children experience
are cumulative across seasons. Conclusion
The present research suggests that no one solution or effort
Strengths targeted to any single context will ameliorate the reading achieve-
The current study has several strengths. First, the use of a ment gap. Ultimately, those involved in policy and intervention
nationally representative sample of children eligible to attend must recognize the ecological, dynamic nature of development and
kindergarten in the 1998 –1999 school year greatly benefited the functioning. Children’s development is multidetermined and em-
bedded in dynamic, interconnected systems. As Farmer and

present analyses. Unlike previous studies in which convenience

samples were used or only poor children were focused on, the Farmer (2001) contended, improvement due to interventions is
present data provide a portrait of children from varying socioeco- likely to be short-lived if those involved in the interventions fail to
nomic backgrounds and from across the nation. The data are not understand the interconnection among systems and the ways in
limited to low-income children living in urban environments, as which multiple risks constrain developmental trajectories. Thus, as
often occurs in research on the effects of economic disadvantage. Rothstein (2004) argued, the answer to disparities in reading
Such characteristics add to the external validity of this research. In achievement lies across systems and in the amelioration of the
addition, the current study benefits from the longitudinal nature of disparate conditions that children and their families experience
the ECLS-K. Rather than traditional cross-sectional designs, which there.
often do not recognize the volatility in family characteristics linked
to disadvantage and adversity, the current research may provide a References
more accurate portrait of the influence of the family environment Aber, J. L., Gephart, M. A., Brooks-Gunn, J., & Connell, J. P. (1997).
on reading achievement over time. Moreover, the present study Development in context: Implications for studying neighborhood ef-
benefits from the comprehensive nature of the ECLS-K data on fects. In J. Brooks-Gunn, G. J. Duncan, & J. L. Aber (Eds.), Neighbor-
family and school contexts. hood poverty: Vol. 1. Context and consequences for children (pp. 44 –
Finally, the current findings provide several implications for 61). New York: Russell Sage Foundation.
policy. To reduce the reading gap, one cannot place emphasis American Institutes of Research, & Cohen, J. (2005). AM Statistical
solely on improving experiences, resources, and interactions in Software (Version 0.06.03) [Computer software]. Available from Amer-
any single context. The disadvantages that low-SES children ican Institutes for Research: http://am.air.org/default.asp
Arnold, D. H., & Doctoroff, G. L. (2003). The early education of socio-
face across contexts are great, and all contexts are associated
economically disadvantaged children. Annual Review of Psychology, 54,
with disparities in early reading. However, the present models 517–545.
suggest that families and characteristics associated with the Bronfenbrenner, U. (1979). The ecology of human development: Experi-
home environment are most closely associated with SES gaps in ments by nature and design. Cambridge, MA: Harvard University Press.
children’s reading achievement on entering school. Accord- Bronfenbrenner, U. (1986). Ecology of the family as a context for human
ingly, efforts to improve children’s home and family experi- development: Research perspectives. Developmental Psychology, 22,
ences, particularly before children arrive at kindergarten, may 723–742.
prove most important in addressing early disparities in reading Bronfenbrenner, U. (1989). Ecological systems theory. In R. Vasta (Ed.),
achievement. For example, it may be beneficial to provide Annals of child development: Six theories of child development: Revised
support services to reduce the effects of role strain unduly formulations and current issues (pp. 1–103). Greenwich, CT: JAI Press.
Bryant, D. M., Burchinal, M., Lau, L. B., & Sparling, J. J. (1994). Family
experienced by low-SES parents before children enter formal
and classroom correlates of Head Start children’s developmental out-
schooling. In addition, policies that increase parents’ abilities to comes. Early Childhood Research Quarterly, 9, 289 –309.
provide books in the home and to improve the home literacy Burkam, D. T., Ready, D. D., Lee, V. E., & LoGerfo, L. F. (2004).
environment for their children may also benefit young children. Social-class differences in summer learning between kindergarten and
Similarly, the provision of prekindergarten childcare may pre- first grade: Model specification and estimation. Sociology of Education,
pare children for the school environment and provide founda- 77, 1–31.
tional skills. Cairns, R. B., Elder, G. H., & Costello, E. J. (1996). Developmental
Researchers and policymakers should take care not to locate science. New York: Cambridge University Press.
the source of reading disparities in children and families with- Coleman, J. S. (1966). Equality of educational opportunity (ICSPR Study
out also recognizing the constraints on families or the other No. 6389). Washington DC: U.S. Government Printing Office.
Dickinson, D. K. (2001). Book reading in preschool classrooms: Is rec-
contexts that may be at work. Like children themselves, family
ommended practice common? In D. K. Dickinson & P. O. Tabors (Eds.),
functioning exists within a larger context (Bronfenbrenner, Beginning literacy with language: Young children learning at home and
1986) so that parent behaviors, practices, and resources are a school (pp. 175–204). Baltimore: Brookes Publishing.
reflection of the experiences parents face outside of the home. Dickinson, D. K., & DeTemple, J. (1998). Putting parents in the picture:
For example, parents’ behaviors and beliefs are not independent Maternal reports of preschool literacy as a prediction of early reading.
systems separate from the outside world; instead, they are Early Childhood Research Quarterly, 13, 241–261.
Downey, D. B., von Hippel, P. T., & Broh, B. A. (2004). Are schools the National Research Council. (2000). From neurons to neighborhoods: The
great equalizer? Cognitive inequality during the summer months and the science of early childhood development. Washington, DC: National
school year. American Sociological Review, 69, 613– 635. Academy Press.
Duncan, S. E., & DeAvila, E. (1986). Preschool Language Assessment NICHD Early Child Care Research Network. (1998). Early child care and
Survey 2000 examiner’s manual. Monterey, CA: CTB/McGraw-Hill. self-control, compliance, and problem behavior at 24 and 36 months.
Duncan, G. J., Yeung, W. J., Brooks-Gunn, J., & Smith, J. R. (1998). How Child Development, 69, 1145–1170.
much does childhood poverty affect the life chances of children? Amer- NICHD Early Child Care Research Network. (2000). The relation of child
ican Sociological Review, 63, 406 – 423. care to cognitive and language development. Child Development, 71,
Dunn, L. M., & Dunn, L. M.. (1981). Peabody Picture Vocabulary Test, 960 –980.
revised edition. Circle Pines, MN: American Guidance Services. NICHD Early Child Care Research Network. (2004). Multiple pathways to
Early, D. M., Bryant, D. M., Pianta, R. C., Clifford, R. M., Burchinal, early academic achievement. Harvard Educational Review, 74, 1–29.
M. R., Ritchie, S., et al. (2006). Are teachers’ education, major, and Phillips, M., & Chin, T. (2004). School inequality: What do we know? In
credentials related to classroom quality and children’s academic gains in K. Neckerman (Ed.), Social inequality. New York: Russell Sage Foun-
pre-kindergarten? Early Childhood Research Quarterly, 21, 174 –195. dation.
Entwisle, D. R., & Alexander, K. L. (1993). Entry into school: The Pianta, R. C., LaParo, K. M., Payne, C. C., Cox, M. J., & Bradley, R. H.
beginning school transition and educational stratification in the United (2002). The relation of kindergarten classroom environment to teacher,
States. Annual Review of Sociology, 19, 401– 423. family, and school characteristics and child outcomes. Elementary
Evans, G. W. (2004). The environment of childhood poverty. American School Journal, 102, 225–238.
Psychologist, 59, 77–92. Pianta, R. C., & Walsh, D. (1996). High-risk children in the schools:
Farmer, T. W., & Farmer, E. M. Z. (2001). Developmental science, Creating sustaining relationships. New York: Routledge.
systems of care, and prevention of emotional and behavioral problems in Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models:
youth. American Journal of Orthopsychiatry, 71, 171–181. Applications and data analysis methods (2nd ed.). Thousand Oaks, CA:
Federal Interagency Forum on Child and Family Statistics. (2005). Amer- Sage.
ica’s children: Key indicators of well-being, 2005. Washington, DC: Raudenbush, S. W., Bryk, A. S., & Congdon, R. (2005). HLM 6 (Version
U.S. Government Printing Office. 6.02) [Computer software]. Lincolnwood, IL: Scientific Software Inter-
Goldenberg, C. (2002). Making schools work for low-income families in national.
the 21st century. In S. B. Neuman & D. K. Dickinson (Eds.), Handbook Raviv, T., Kessenich, M., & Morrison, F. J. (2004). A mediational model
of early literacy research (pp. 211–231). New York: Guilford Press. of the association between socioeconomic status and three-year old
Goldhaber, D. D., & Brewer, D. J. (1997). Why don’t schools and teachers language abilities: The role of parenting factors. Early Childhood Re-
seem to matter? Assessing the impact of unobservables on educational search Quarterly, 19, 528 –547.
productivity. Journal of Human Resources, 32, 505–523. Reid, D. K., Hresko, W .P., & Hammill, D. D. (1981). Test of Early
Hart, B., & Risley, T. (1995). Meaningful differences in the everyday experi- Reading Ability (TERA-2). Austin, TX: PRO-ED.
ences of young American children. Baltimore: Brookes Publishing. Rothstein, R. (2004). Class and schools: Using social, economic, and
Hess, R. D., Holloway, S. D., Dickson, W. P., & Price, G. G. (1984). educational reform to close the Black–White achievement gap. Wash-
Maternal variables as predictors of children’s school readiness and later ington, DC: Economic Policy Institute.
achievement in vocabulary and mathematics in sixth grade. Child De- Singer, J., & Willett, J. B. (2003). Applied longitudinal data analysis:
velopment, 55, 1902–1912. Modeling change and event occurrence. Oxford, England: University
Huttenlocher, J., & Levine, S. C. (1990). The Primary Test of Cognitive Press.
Skills (PTCS). New York: CTB/McGraw Hill. Snow, C. E., Burns, S. M., & Griffin, P. (Eds.). (1998). Preventing reading
Kozol, J. (1991). Savage inequalities: Children in America’s schools. New difficulties in young children. Washington, DC: National Academy
York: Harper Perennial. Press.
Lee, V. E., & Burkam, D. T. (2002). Inequality at the starting gate: Social Vernon-Feagans, L., Hammer, C. S., Miccio, A., & Manlove, E. (2002).
background differences in achievement as children enter school. Wash- Early language and literacy skills in low-income African American and
ington, DC: Economic Policy Institute. Hispanic children. In S. B. Neuman & D. K. Dickinson (Eds.), Hand-
Leventhal, T., & Brooks-Gunn, J. (2000). The neighborhoods they live in: book of early literacy research (pp. 192–210). New York: Guilford
The effects of neighborhood residence on child and adolescent out- Press.
comes. Psychological Bulletin, 126, 309 –337. Weigel, D. J., Martin, S. S., & Bennett, K. K. (2005). Ecological influences
Magnusson, D., & Cairns, R. B. (1996). Developmental science: Toward a of the home and the child-care center on preschool-age children’s
unified framework. In R. B. Cairns, G. H. Elder, & E. J. Costello (Eds.). literacy development. Reading Research Quarterly, 40, 205–233.
Developmental science (pp. 7–30). New York: Cambridge University Whitehurst, G. J., & Lonigan, C. J. (1998). Child development and emer-
Press. gent literacy. Child Development, 69, 848 – 872.
Markwardt, Jr., F. C., (1989). Peabody Individual Achievement Test– Woodcock, R. W., & Bonner, M. (1989). Woodcock-Johnson III Tests of
Revised (PIAT-R). Circle Pines, MN: American Guidance Services. Achievement–Revised. Itasca, IL: Riverside.
McLoyd, V. C. (1998). Socioeconomic disadvantage and child develop- Xue, Y., & Meisels, S. J. (2004). Early literacy instruction and learning to
ment. American Psychologist, 53, 185–204. read in kindergarten. American Educational Research Journal, 41, 191–
Murray, A. D., & Hornbaker, A. V. (1997). Maternal directive and facil- 229.
itative interaction styles: Associations with language and cognitive de- Yeung, W. J., Linver, M. R., & Brooks-Gunn, J. (2002). How money
velopment of low risk and high risk toddlers. Development and Psycho- matters for young children’s development: Parental investment and
pathology, 9, 507–516. family processes. Child Development, 73, 1861–1879.
National Center for Education Statistics. (2004). ECLS-K longitudinal
kindergarten-third grade public-use child file, CD-ROM and users’ Received August 24, 2006
manual (NCES 2004 – 089). Washington, DC: U.S. Department of Ed- Revision received August 28, 2007
ucation. Accepted September 24, 2007 䡲

Socioeconomic Differences in Reading Trajectories: The Contribution of Family, Neighborhood, and School Contexts

Uploaded by

Copyright:

Available Formats

You might also like

Socioeconomic Differences in Reading Trajectories: The Contribution of Family, Neighborhood, and School Contexts

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Socioeconomic Differences in Reading Trajectories: The Contribution of Family, Neighborhood, and School Contexts

Uploaded by

Copyright:

Available Formats

Journal of Educational Psychology Copyright 2008 by the American Psychological Association

2008, Vol. 100, No. 2, 235–251 0022-0663/08/$12.00 DOI: 10.1037/0022-0663.100.2.235

Socioeconomic Differences in Reading Trajectories: The Contribution of

Nikki L. Aikens Oscar Barbarin

Keywords: socioeconomic status, reading, achievement trajectories

preschool programs, elementary school reading instruction, and

administered questionnaires, providing reports on children’s class-

Table 1 in time beyond what can be accounted for by children’s underlying

Mean Reading Scale Score

Mean Reading Scale Score

three time periods.1

spring of kindergarten. By the spring of first grade, the gap grew

␲ 1ij ⫽ ␤ 10j共 AGE兲 ⫹ ␤ 11j共FEMALE兲

Level 3: ␤ 00j ⫽ ␥ 000 ⫹ u 00j, 1

Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 Total

␤ 23j ⫽ ␥ 230 , Family Characteristics, Reading Achievement, and

␲ 7ij ⫽ ␤ 70j, and ␤ 34j ⫽ ␥ 340 ,

␲ 8ij ⫽ ␤ 80j. ␤ 35j ⫽ ␥ 350 ,

Level 3: ␤ 00j ⫽ ␥ 000 ⫹ u 00j, ␤ 40j ⫽ ␥ 400 ,

␤ 01j ⫽ ␥ 010 , ␤ 50j ⫽ ␥ 500 ,

␤ 02j ⫽ ␥ 020 , ␤ 60j ⫽ ␥ 600 ,

␤ 03j ⫽ ␥ 030 , ␤ 70j ⫽ ␥ 700 , and

␤ 04j ⫽ ␥ 040 , ␤ 80j ⫽ ␥ 800 .

demographic model indicated the extent to which mean differ-

␤ 31j ⫽ ␥ 310 , Neighborhood Characteristics, Reading Achievement, and

Level 1: READ tij ⫽ ␲ 0ij ⫹ ␲ 1ij共SLOPE K兲 ⫹ ␲ 2ij共SLOPE 1兲 ␤ 21j ⫽ ␥ 210 ,

⫹ ␲ 3ij共SLOPE 3兲 ⫹ ␲ 4ij共HOMELIT兲 ⫹ ␲ 5ij共BOOKS兲 ␤ 22j ⫽ ␥ 220 ,

⫹ ␲ 6ij共INVOLVE兲 ⫹ ␲ 7ij共STRAIN兲 ⫹ ␲ 8ij共WARMTH兲 ␤ 23j ⫽ ␥ 230 ,

⫹ ␲ 9ij共HPROBS兲 ⫹ ␲ 10ij共HSAFETY兲 ⫹ ε ij. ␤ 24j ⫽ ␥ 240 ,

Level 2: ␲ 0ij ⫽ ␤ 00j ⫹ ␤ 01j共 AGE兲 ⫹ ␤ 02j共FEMALE兲 ␤ 25j ⫽ ␥ 250 ,

␲ 1ij ⫽ ␤ 10j ⫹ ␤ 11j共 AGE兲 ⫹ ␤ 12j共FEMALE兲 ⫹ ␥ 303 共SPROBS兲 ⫹ ␥ 304 共BAD兲,

⫹ ␤ 13j共WHITE兲 ⫹ ␤ 14j共SES兲 ⫹ ␤ 15j共PRE K兲, ␤ 31j ⫽ ␥ 310 ,

␲ 2ij ⫽ ␤ 20j ⫹ ␤ 21j共 AGE兲 ⫹ ␤ 22j共FEMALE兲 ␤ 32j ⫽ ␥ 320 ,

⫹ ␤ 23j共WHITE兲 ⫹ ␤ 24j共SES兲 ⫹ ␤ 25j共PRE K兲, ␤ 33j ⫽ ␥ 330 ,

␲ 3ij ⫽ ␤ 30j ⫹ ␤ 31j共 AGE兲 ⫹ ␤ 32j共FEMALE兲 ␤ 34j ⫽ ␥ 340 ,

⫹ ␤ 33j共WHITE兲 ⫹ ␤ 34j共SES兲 ⫹ ␤ 35j共PRE K兲, ␤ 35j ⫽ ␥ 350 ,

␲ 4ij ⫽ ␤ 40j, ␤ 40j ⫽ ␥ 400 ,

␲ 5ij ⫽ ␤ 50j, ␤ 50j ⫽ ␥ 500 ,

␲ 6ij ⫽ ␤ 60j, ␤ 60j ⫽ ␥ 600 ,

␲ 7ij ⫽ ␤ 70j, ␤ 70j ⫽ ␥ 700 ,

␲ 8ij ⫽ ␤ 80j, ␤ 80j ⫽ ␥ 800 ,

␲ 9ij ⫽ ␤ 90j, and ␤ 90j ⫽ ␥ 900 , and

␲ 10ij ⫽ ␤ 100j. ␤ 100j ⫽ ␥ 1000 .

␤ 13j ⫽ ␥ 130 , School Characteristics, Reading Achievement, and

Home neighborhood problems ⫺0.17 (0.18)

Home neighborhood safety 0.02 (0.25)

⫹ ␤ 03j共WHITE兲 ⫹ ␤ 04j共SES兲 ⫹ ␤ 05j共PRE K兲 ⫹ r ij, ␤ 05j ⫽ ␥ 050 ,

⫹ ␤ 13j共WHITE兲 ⫹ ␤ 14j共SES兲 ⫹ ␤ 15j共PRE K兲, ⫹ ␥ 103 共EXPERIENCE兲 ⫹ ␥ 104 共PREPARATION兲

␲ 2ij ⫽ ␤ 20j ⫹ ␤ 21j共 AGE兲 ⫹ ␤ 22j共FEMALE兲 ⫹ ␥ 105 共POVERTY兲 ⫹ ␥ 106 共PRIVATE兲,

␲ 8ij ⫽ ␤ 80j. ⫹ ␥ 203 共EXPERIENCE兲 ⫹ ␥ 204 共PREPARATION兲

⫹ ␥ 003 共EXPERIENCE兲 ␤ 21j ⫽ ␥ 210 ,

⫹ ␥ 004 共PREPARATION兲 ⫹ ␥ 005 共POVERTY兲 ␤ 22j ⫽ ␥ 220 ,

p ⬍ .05, es ⫽ .03). Children in high poverty schools evidenced

ences in reading development to a greater extent than is family

life. Accordingly, family life can be viewed as being strongly

Child model Family model Neighborhood model School model

Random effect Variance ␹2 df Variance ␹2 df Variance ␹2 df Variance ␹2 df

Deviance 216,069.11 (23) 207,251.43 (32) 206,276.11 (50) 207,092.46 (56)