Applied Developmental Science

ISSN: 1088-8691 (Print) 1532-480X (Online) Journal homepage: http://www.tandfonline.com/loi/hads20

Does mixing age groups in early childhood

education settings support children’s language
development?

Laura M. Justice, Jessica A. Logan, Kelly Purtell, Dorthe Bleses & Anders
Hogden

To cite this article: Laura M. Justice, Jessica A. Logan, Kelly Purtell, Dorthe Bleses & Anders
Hogden (2018): Does mixing age groups in early childhood education settings support children’s
language development?, Applied Developmental Science, DOI: 10.1080/10888691.2017.1386100

To link to this article: https://doi.org/10.1080/10888691.2017.1386100

Published online: 09 Jan 2018.

Submit your article to this journal

Article views: 28

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=hads20
APPLIED DEVELOPMENTAL SCIENCE
https://doi.org/10.1080/10888691.2017.1386100

none defined

Does mixing age groups in early childhood education settings support

children’s language development?
Laura M. Justicea, Jessica A. Logana, Kelly Purtella, Dorthe Blesesb, and Anders Hogdenb
a
The Ohio State University; bAarhus University

ABSTRACT
As early childhood education programming expands across the globe, there is an increased need
to understand how features of these programs influence children’s development. The composition
of children’s age within a classroom is one such feature, although it is much less studied than other
features. Theoretical and empirical evidence suggests that children’s development may be
influenced by the age range of their classmates. This study examines the relations between
classroom age variability on children’s vocabulary development for 2,743 children between the
ages of two years, nine months, and six years, 11 months enrolled in early childhood education
settings in Denmark. Findings indicate a significant nonlinear relationship between the range of
child age within a classroom and children’s vocabulary development, such that classrooms with a
maximum age range of 24 months were associated with the greatest gains in vocabulary growth.
Results give direction to policy efforts focused on expansion of early childhood education
programming.

Early childhood education (ECE) programs exist rates of ECE participation (94% for 3–5-year olds) for
across the world to serve several important societal all OECD countries, with provision of universal child
goals, including supporting the labor-market care being an important feature of the welfare state
participation of parents of young children, especially (Landersø & Heckman, 2017).
women (Kamerman, 1991), and fostering the early Of relevance to the research reported here, when
developmental competencies of young children ECE programs are created and expanded, there are
(Campbell, Ramey, Pungello, Sparling, & Miller- certain structural features of such programs that can
Johnson, 2002). Across North America, there are be directly affected by the policies that create them.
currently numerous efforts at the local, state, and federal These features include, for instance, teacher-child ratio,
levels to improve access to ECE programs for young class size, and teacher credentials. For example,
children to enhance labor-market participation among state-sponsored support of ECE programs might
parents (Lefebvre, Merrigan, & Verstraete, 2009) and stipulate a 1:9 teacher-child ratio with a maximum class
to improve children’s development of key cognitive size of 18 children, and provision of programming by
and noncognitive skills as well as readiness for formal teachers with at least a bachelor’s degree. An additional
schooling at age 5 years (Winsler et al., 2008). Presently, structural feature often affected by policies is the age
in the United States, about 50% of 3- to 5-year-olds composition of ECE classrooms, as creation/expansion
participate in formal full-day center-based ECE policies often stipulate the age range of children who
(National Center for Education Statistics, 2017), and can participate in subsidized programming. For
this number is likely to increase as many cities and instance, the Commonwealth of Virginia’s long-running
states press for expanded access to ECE for young Virginia Preschool Initiative (VPI) provides funding
children. In some countries within the European Union, for more than 10,000 4-year-olds across the state to
universal participation in ECE is the norm, including participate in center-based ECE each year (Huang,
France, Spain, and Belgium, which each have near Invernizzi, & Drake, 2012). An indirect result of this
saturation of children in ECE by three years of age, with policy, by stipulating participation only to 4-year-old
an average participation rate of about 80% among the children, is that it leads to creation of same-age (SA)
EU 27 [Organisation for Economic Co-operation and ECE classrooms rather than mixed-age (MA)
Development (OECD), 2016]. Denmark, in which the classrooms, as VPI-supported classrooms enroll only
current study was completed, has among the highest 4-year-old children in a SA context. In this research,

CONTACT Laura M. Justice justice.57@osu.edu 175 E 7th, Ohio State University, Columbus, OH 43201, US.
© 2017 Taylor & Francis
2 L. M. JUSTICE ET AL.

we view policy decisions regarding the age range of sociocultural learning, Vygotsky focuses on the role of
children who can participate in expanding peers in children’s learning and notes that interacting
programming as a potentially crucial mechanism for with children more- and less-skilled than oneself is ben-
ensuring that ECE expansion meet its desired goals of eficial for learning and developing, as it provides oppor-
enhancing children’s development, as these decisions tunities for modeling more advanced behavior and
directly affect the classroom age composition of scaffolding learning for younger peers. From this, MA
programs with respect to whether available programs classrooms may be advantageous because they increase
will feature SA or MA programming. the range of peers’ skills than a child may be exposed
Decisions about SA and MA classrooms also occur at to in their classroom. These theoretical underpinnings
more micro-levels. In the United States, many private- have supported the widespread use of mixed-age class-
and public (e.g., Head Start) centers have the option rooms, and in particular, guide the Montessori method
of offering either SA or MA classrooms. Although these of schooling (Lillard & Else-Quest, 2006). However,
decisions are sometimes made for practical reasons, some work has suggested that the benefits of mixed-
such as meeting class size requirements, center directors age programming may vary for older and younger
and teachers likely also have preferences for which children. Winsler and colleagues (2002) described the
environments they believe are most optimal for results of a natural experiment, in which two SA
children’s development. Similarly, parents have classrooms (one serving 3-year-olds and the other
preferences and beliefs about what they believe is the serving 4-year-olds) were converted to two MA
best early childhood education program for their classrooms in a single early childhood program. The
child and these beliefs shape their child care decisions authors used this opportunity to examine a range of
(Crosnoe, Purtell, Davis-Kean, Ansari, & Benner, proximal processes as a function of the SA-MA contrast
2016). Parental beliefs about the utility of MA (Winsler et al., 2002). The authors observed that within
classrooms may, in part, determine the type of early MA settings, older children become more like younger
education center in which they enroll their child. Thus, children, and younger children become more like older
understanding how MA and SA environments shape children. Specifically, 3-year-olds in MA settings exhib-
children’s academic development is critical to informing ited goal-directed behaviors similar to 4-year-olds in SA
these decisions made at the policy, center, and settings, whereas 4-year-olds in MA settings exhibited
parent level. social affiliations more like 3-year-olds in SA settings.
Thus, the MA programming appeared beneficial for
some but not all children, with younger children
Age composition in early childhood
seeming to benefit more than older children.
education
Such findings lend appeal to having classrooms serve
There has long been interest in the relative advantages a narrow age range of children, such as only infants or
and disadvantages of SA versus MA programming in only 4-year-olds, which also may be beneficial to
ECE. SA classrooms enroll children who are relatively teachers. For instance, a teacher may provide a better
similar in age, as is the case of 4-year-old programs quality of instruction and care if children were relatively
that serve only children between 48 and 60 months similar in their ability to attend and learn during
(e.g., Virginia’s VPI), whereas MA classrooms enroll a common classroom routines. Less effort may need to
broader range of children in terms of age. Potentially, be directed toward differentiating instruction and
ECE programs could enroll children from soon after managing children’s behaviors than might occur if
birth until kindergarten entry, which typically occurs children varied significantly in age range.
when children are five or even six years of age, for
children whose parents delay kindergarten enrollment.
Age composition and children’s achievement
Programs can utilize a variety of approaches for
organizing children into classrooms as a function of Only a handful of studies have sought to estimate the
age: children could be organized into SA classrooms, effects of various classroom-age compositions on
creating homogenous settings in which classrooms children’s development in such settings. Referenced
serve only children of a given age, or MA classrooms, previously, Winsler and colleagues (2002) examined
characterized by greater heterogeneity in children’s ages. how MA classrooms influenced children’s classroom
These approaches to age arrangements have histori- behaviors separately for the younger (3-year olds) and
cally been grounded in developmental theory. Most older (4-year olds) children. Their work showed that
notably, supporters of MA classrooms often point to younger children in mixed age classrooms derived
the work of Vygotsky (1978). In his theory of benefits from being in a classroom environment with
 APPLIED DEVELOPMENTAL SCIENCE 3

older peers, presumably who could model positive larger range of ages in ECE settings may have negative
behavior and expose them to more cognitively- effects on the development of older children in these
challenging material. The authors proposed that this classrooms. If this is the case, it would suggest that SA
MA environment resulted in a “stretching” effect for programming is more desirable than MA programming,
the younger children. The older children in MA at least for 3- and 4-year-old children.
classrooms, however, were found to “act younger,” There are several important limitations of these
perhaps because they spend more time interacting with studies, which helped to motivate the present
children who are less developed behaviorally and investigation. First, all have focused exclusively on
cognitively. investigating classroom age composition for programs
More recent studies involving larger samples of serving only children from low-income backgrounds.
classrooms have sought to estimate the actual effects This detracts from the potential generalizability of
of these age-grouping contrasts on children’s findings to classrooms serving a socioeconomically
achievement, largely focusing on children’s language diverse range of students. Further, it can negatively
development (Ansari, Purtell, & Gershoff, 2016; Bell, affect internal validity in that children in such studies
Greenfield, & Bulotsky-Shearer, 2013; Guo, Tompkins, may show a more limited range in skills (and
Justice, & Petscher, 2014; Moller, Forbes-Jones, & development over time) than would occur for a more
Hightower, 2008). Language development, primarily diverse sample. Second, the age range of children within
children’s vocabulary growth, is often studied in relation classrooms studied in these works provided only limited
to classroom composition (e.g., Justice, Logan, Lin, & age variability within classrooms. For instance, Bell et al
Kaderavek, 2014), because children’s language skills (2013) examined the relations between classroom age
are greatly affected by the characteristics of the composition and child outcomes in Head Start
linguistic input they experience (Huttenlocher, classrooms. On average, children within each classroom
Vasilyeva, Cymerman, & Levine, 2002). The ages of ranged in age only up to about 12 months. The children
the children in a classroom seem to shape the linguistic in this study were also relatively narrow in their ages,
environment of the ECE classroom. For example, a with a mean of 48 months and a standard deviation of
child in a classroom with a wide age range of peers only seven months. As with the Ansari et al. (2016)
may be exposed to children with both greater and lower study, which also relied on a Head Start sample, it is
vocabulary skills, which may facilitate or impede his or difficult to draw firm conclusions on the effects of
her vocabulary growth. A recent study showed direct MA versus SA programming on ECE participants using
pathways between the language skills of children’s samples that do not represent the full range of
classmates on their language gains over a year in classroom age composition possibilities nor the
preschool, and that these operate independently of plausible age range of possible participants in ECE
teachers’ instructional pathways (Yeomans-Maldonado, programs.
Justice, & Logan, in press).
The body of work that has sought to determine the
Danish ECE context
extent to which variability in classroom age composition
may affect children’s language development or similar Denmark is a welfare state that provides universal
outcomes provide minimal clarity with respect to the childcare (ECE) to children soon after birth, with an
possible effects of MA versus SA groupings. Two studies enrollment rate of 85% of 1–2 year olds and 94% of
(Bell et al., 2013; Moller et al., 2008) operationalized all 3–5 year olds (OECD Family Database, 2016). The
variability in classroom age composition by using ECE facilities are organized in public, semi-public,
the standard deviation (SD) of children’s ages in a and private centers or in publicly organized nonparental
classroom as the independent variable, whereas Ansari home-based care. All types of ECE facilities are audited
et al. (2016) used the proportion of 3-year-olds in by the municipality and are regulated by national laws
classrooms and Guo et al. (2014) used SD and age range concerning educational quality and safety. Costs are
of children in a classroom. Two studies found minimal highly subsidized, with a minimum of 75% of the cost
to no association between variability in classroom age covered by the municipality and larger subsidies pro-
composition and children’s development (Bell et al., vided for low-income households (for further discussion
2013; Guo et al., 2014) and two reported a negative of this context, see Gupta & Simonsen, 2010).
association between increased variability in classroom Danish ECE programs are situated in most
age composition (as would occur in MA classrooms) residential areas across Denmark; hence, their
and children’s outcomes (Ansari et al., 2016; Moller socioeconomic composition is solely determined by
et al., 2008). The latter work suggests that having a the composition of the neighborhood. Although
4 L. M. JUSTICE ET AL.

programs with a high proportion of at-risk children do as noted previously. For infants and toddlers (up to 3
exist, Danish ECE programs typically serve a more years of age), about 70% of children are in formal child
socioeconomically diverse group of children as care (the highest rate of participation among OECD
compared to subsidized programs in the United States, countries), and about 94% for 3- to 5-year-olds (within
which often feature targeted enrollment of only the top 10 of OECD countries) (OECD Family Data-
low-income children. base, 2016). With data in this study representing 13
Danish ECE programs are often considered to (out of 98) municipalities across the country, the sample
relatively high quality in terms of structural characteris- in this study more closely approximates a population
tics (Esping-Andersen et al., 2012). For instance, the base from which to derive estimates of classroom-age
average children-to-teacher ratio for preschoolers is composition effects.
7:1. Approximately 60% of teachers have a 3.5-year Second, children in Denmark participate in publicly
pedagogical Bachelor degree and the other 40% are funded child-care programs irrespective of the family’s
mainly teaching assistants who do not have a university socioeconomic status or ability to pay for such services,
degree. Beyond these structural characteristics, observa- thereby mitigating the selection of families into parti-
tional work suggests that the observed classroom quality cular programs as occurs in the United States. Relatedly,
is similar to the United States. Specifically, the programs are not aligned to a specific funding source,
emotional quality of classrooms is relatively high, and such as public-preK initiatives or Head Start, as is com-
the instructional quality of classrooms is fairly low mon in the United States. As a result, the sample
(Bleses et al., 2017). Emotional quality captures the included in the present study is more socioeconomically
general climate of the classroom in terms of warmth diverse than that a number of studies conducted within
and sensitivity, whereas instructional quality captures the United States that focus on children in targeted
the climate of the classroom in terms of cognitive enrollment preschool programs.
challenge. The low levels of the latter may reflect to Third, publicly funded child-care in Denmark is avail-
the focus of the ECE curriculum that is common to able to families when their children are as young as three
many European countries. This curriculum typically months of age, resulting in a system in which a large
emphasizes children’s social-emotional development, number of infants, toddlers, and preschoolers attend
as opposed to a cognitive or academic orientation, and centers together. Consequently, many classrooms in these
is characterized by a holistic and strong play-oriented centers enroll mixed-aged groupings, with children
curriculum with little time spent on formal instruction potentially aging in range from birth to six years. As a
(Sylva, Ereky-Stevens, & Aricescu, 2015). Legislation result of these phenomena, the classrooms included in
implemented in 2004 requires all programs to formulate this study enrolled children who ranged in age, on
‘learning plans’ focused on children’s development in average, 24 months. This distinguishes the present study
the areas of social competence, language, body and from those conducted in ECE programs in the United
movement, nature and natural phenomena, and cultural States, in which children in a given classroom typically
expressions and values (Sylva et al., 2015). ranged in age only up to about 12 months (e.g., Bell
et al., 2013). This allows us to more rigorously examine
the effects of classroom age composition, especially the
Present study
effects of range, on children’s language outcomes.
The present study was conducted to improve our Two questions were addressed in this study. The first
understanding of the effects of classroom age question asked: To what extent is variability in the age
composition on children’s language development, parti- composition of ECE programs associated with chil-
cularly their vocabulary growth, drawing upon a large, dren’s vocabulary growth? We addressed this question
population-based sample of children attending ECE in two ways. First, we predicted children’s vocabulary
programs in Denmark during the 2013 and 2014 years. growth from the standard deviation of children’s age
Involving the largest and most socioeconomically in months at the start of the year, controlling for
diverse sample of children to date in research on this relevant covariates, similar to Bell et al. (2013). Because
topic, involving 3,340 children between the ages of 2 we found that the within-classroom age variability was
years, 9 months, and 6 years, 11 months, the present considerably wider than has been found in other similar
study exhibits several compelling strengths. First, the work, we also predicted children’s vocabulary growth
Danish child-care system offers an advantageous con- from the range of ages within a classroom at the start
text to address questions regarding classroom age com- of the year, also controlling for relevant covariates. In
position, as formal child-care participation rates in both approaches, we examined the linear relationship
Denmark are among the highest of all OECD countries, between classroom age composition and children’s
 APPLIED DEVELOPMENTAL SCIENCE 5

outcomes. Second, given the age range present in the analyses as individuals. The final analyzed dataset
classrooms involved in this study, which averaged 24 included 2,743 students enrolled in 227 different
months between the youngest and oldest child, we also classrooms.
examined whether the relationship between classroom Examination of the patterns of missingness did show
age composition and children’s vocabulary growth that missing data was associated with certain child- and
approximated a curvilinear relationship. We examined family-level variables. Specifically, missing data from
whether the relation between age composition and lan- recruitment to posttest was higher for immigrant
guage gains was dependent on the degree of variability children (p < .001) as well as children with low maternal
in age composition. In addressing both of these ques- education, and children from low-income families (both
tions, we also examined whether there was a significant p ¼ .004). Again, while these children’s data were
interplay between children’s age and classroom age included in the classroom-based measures of age
composition, as some prior studies have suggested that composition, it was not included in the individual-level
MA programming is beneficial only for younger chil- analysis; thus, the results may not generalize to all
dren (Moller et al., 2008). students in a classroom, but rather only to those who
remain in class the whole year.
To contextualize the sample in terms of
Methods socioeconomic status, we used maternal education
(highest level attained) as a proxy; these data were
Participants
provided by Statistics Denmark. Fourteen percent of
There were 3,340 2- to 6-year-old children (54% male) children’s mothers had primary school as their highest
who participated in this study. The children were, on level attained; 35% had high school and/or vocational
average, 54 months of age (SD ¼ 10.3 months), but education; 31% had a four-year college degree; and
ranged in age from 33 months (2 years, 9 months) to 20% had advanced university education. Thus, there
83 months (6 years, 11 months). About 10% of the was a near normal distribution of children’s socioeco-
children were considered immigrants, the majority of nomic status represented in this study, corresponding
whom come from eastern European (e.g., Poland) or to the universal ECE system in which children were
middle-eastern countries (e.g., Turkey). The children enrolled. In terms of ethnic diversity, information from
were enrolled in publicly funded Danish child care cen- Statistics Denmark was used to identify the number of
ters from across the country; their centers were involved children who come from families considered
in one of two large-scale effectiveness trials supported immigrants. These may be first or second generation
by the Ministry of Education or the Strategic Research immigrants. In this sample, 10% of children were
Council of Denmark. Specifically, centers across the considered immigrants.
country were invited to participate in one of two parallel
effectiveness studies, with each involving three experi-
Procedure
mental conditions plus a control. Random assignment
was used for allocation of centers to conditions. Results Children’s child-care centers were enrolled in one of
of those trials are described in several prior reports two parallel effectiveness trials, as noted. As part of
(Bleses et al., in press). those studies, children’s centers were randomly assigned
For our purposes, we used only children who were to one of four arms of the trial in which they were
enrolled in the control condition of either trial, given enrolled, with one arm comprising a true control
that our outcome of interest (i.e., language develop- condition. All children included in the present study
ment) was a focus within the effectiveness trials. Of were in a control condition, corresponding to
those children, 27 were missing data on vocabulary “business-as-usual instruction” in their centers. Thus,
measures at the pre-test or both the pre- and post-test any effects of classroom age composition in this study
and were therefore excluded from analyses (0.8% of correspond to those occurring under typical educational
the total sample). An additional 554 were missing data circumstances.
at the posttest due primarily to matriculation outside As part of the larger study, children were assessed
of the school system, and another 16 participants were prior to and following the intervention period,
missing data on one or more demographic character- with an approximately 30-week interval between
istic. Information about these 570 children contributed assessments. Assessments were conducted by children’s
to the classroom-based measurement (including the teachers using a familiar instrument that is part of a
main predictors of interest: age composition within each national language screening program already in place.
classroom), but were excluded from the inferential Teachers receive a training manual and videos to
6 L. M. JUSTICE ET AL.

support their use of this tool twice annually with their for the vocabulary subtest with standardized
children. Teachers downloaded all testing materials assessments of receptive and expressive vocabulary are
and stimuli as well as explicit instructions in how to .55 and .42, respectively.
administer the tool and upload data to a central site.
Other studies of children’s development in ECE settings Classroom age composition
have relied on teacher-implemented direct assessments Classroom age composition was calculated in two ways:
(Hamre et al., 2010; Huang et al., 2012). (1) using the within-classroom standard deviation of
children’s ages at the start of the study (Bell et al.,
2013), and (2) as the within-classroom range of ages
Child language assessment at the start of the study. To test the curvilinear effect,
The vocabulary skills of children were assessed pre- and squared products of each age composition variable were
postintervention using one subtest of the standardized, also calculated. Note that all key study variables were
norm-referenced assessment, Language Assessment of examined for outliers and none were identified.
Children: 3–6 (LA; Bleses et al., 2010). The LA is
administered in the vast majority of Danish child-care
centers as part of a national early language and literacy Results
screening program, and was already in use in all but one
Preliminary analyses
set of centers involved in the present study. The LA
comprises seven subtests, one of which captures chil- Table 1 provides basic descriptions and inter-correla-
dren’s vocabulary skills (max score 76). The vocabulary tions among the key study variables at both the child
measure was chosen because vocabulary growth has level and classroom level. In examining these data, a
most often been studied in relation to classroom few points warrant note regarding the classroom-level
compositional factors in ECE settings (Guo et al., compositional variables. First, the mean age of children
2014; Justice, Petscher, Schatschneider, & Mashburn, in a classroom is not associated with the age range of
2011) and is influenced by classroom characteristics. children in the classroom, using either standard
The other measures focus on phonological awareness, deviation of age (r ¼ −.03) or classroom age range
alphabet knowledge, and general communication (r ¼ .03). This shows that there is no systematic
strategies. For the vocabulary subtest, children are pre- relationship between the age of children in a classroom
sented with a picture and are asked to provide a word and their composition of that classroom; classrooms
that describes that picture; stimuli included nouns serving older children are no more heterogeneous in
(e.g., squirrel), verbs (e.g., writing), and adjectives age range than classrooms serving younger children.
(e.g., dirty). An individual child was presented with a Second, classroom size is associated with the age range
maximum of 40 age-dependent items. of children in the classroom, using either standard devi-
Psychometric quality of the tool is adequate, ation of age (r ¼ −.24) or classroom age range (r ¼ .39).
including internal consistency (alpha ranging from This suggests that classrooms that serve a broader range
.75–.91 for individual subtests). Concurrent validity of children in terms of age tend to serve larger numbers

Table 1. Means and correlations for key study variables.

Child Classroom
1 2 3 4 5 6 7 8 9 10 11
Correlations
1 Child Pretest Vocabulary 1 0.86 −0.03 0.62 −0.29 0.41 0.00 0.29 0.28 −0.02 0.00
2 Child Posttest Vocabulary 0.86 1 −0.03 0.56 −0.31 0.36 −0.01 0.29 0.22 −0.05 −0.04
3 Child Male −0.03 −0.03 1 0.05 −0.02 0.00 0.03 0.01 −0.01 −0.02 −0.02
4 Child Age 0.62 0.56 0.05 1 0.02 0.63 0.07 0.48 0.47 −0.03 0.01
5 Child Immigrant −0.29 −0.31 −0.02 0.02 1 0.01 −0.06 0.03 0.04 0.08 0.07
6 Class Mean Age 0.41 0.36 0.00 0.63 0.01 1.00 0.12 0.76 0.76 −0.03 0.03
7 Class Size 0.00 −0.01 0.03 0.07 −0.06 0.12 1 −0.09 0.37 0.24 0.39
8 Class Min Age 0.29 0.29 0.01 0.48 −0.03 0.76 −0.09 1 0.29 −0.59 −0.57
9 Class Max Age 0.28 0.22 −0.01 0.47 0.04 0.76 0.37 0.29 1 0.53 0.62
10 Class SD Age −0.02 −0.05 −0.02 −0.03 0.08 −0.03 0.24 −0.59 0.53 1 0.93
11 Class Age Range 0.00 −0.04 −0.02 0.01 0.07 0.03 0.39 −0.57 0.62 0.93 1
Descriptives
Mean 0.08 0.55 0.54 53.95 0.10 54.00 19.97 41.91 66.55 7.77 24.64
SD 0.97 0.87 0.50 10.28 0.30 6.46 11.63 8.04 8.45 3.04 9.81
Minimum −2.10 −2.10 0.00 33.00 0.00 38.60 4.00 33.00 42.00 1.51 3.00
Maximum 1.74 1.74 1.00 83.00 1.00 68.00 61.00 65.00 83.00 14.82 44.00
Note. Child pretest and posttest vocabulary scores are z-scores, based on a mean of 0 and standard deviation of 1.
 APPLIED DEVELOPMENTAL SCIENCE 7

of children. Third, we conceptualize age composition in of examining the first research question, and because
two ways in this investigation: The within-classroom the classrooms in the present study demonstrated far
standard deviation and the within-classroom range of wider ranges than have been seen in previous studies,
ages. The two methods are strongly correlated (r ¼ .93, we also examined the same question conceptualizing
variables 10 and 11 in Table 1), and a close investigation age composition as the range of ages of children within
of these variables demonstrates the correlation between the classroom.
the two is heteroschedastic, strongest at the lower end of
the distribution. Thus while similar to one another, Age by composition interactions
these two measures are not identical. To answer the first research question, a hierarchical
To investigate predictive relations among the linear regression model was fit to the data, predicting
constructs of interest, hierarchical linear models children’s vocabulary scores from their pretest scores,
(i.e., multilevel models) were fit to the data using the the child’s age, whether the child was male, and whether
Mixed procedure in SAS 9.4. Models predicting the child was an immigrant. We next included the
children’s vocabulary gains (children’s posttest classroom-factors, including the number of children
predicted from their pretest scores) were fit as a prelimi- per classroom and the mean age of all children in the
nary step to examine the variance components within classroom. Of primary interest, we also included both
and between classrooms. This model indicated that there a variable representing the classroom composition,
was significant variance at the between-classroom level and an interaction of the child’s age by composition,
(17%) on children’s vocabulary scores. This model also which allowed us to determine whether classroom
indicated that there was significant between-classroom composition was a more salient predictor of children’s
variability on the relation of pretest with the posttest language gains for older or younger children (Bell
(9%), suggesting that the model providing the best fit et al., 2013). The resulting coefficients are reported in
to the data allowed each classroom to have a unique Table 2, with within-classroom standard deviation on
autoregressive slope. This model was used as the baseline the left, and within-classroom age range on the right.
for all future models described in this section. Prior to For both outcomes, none of the included classroom-
entering any predictors, all independent variables specific factors were predictive of children’s gains. Nor
were z-scored (recalculated to have a mean of zero and did we find evidence of an interplay between classroom
a standard deviation of one) to allow for ease of composition and child’s age (age standard deviation
interpretation and comparison between models. coefficient ¼ .0003, p ¼ .982, Age range coefficient ¼
−.004, p ¼ .758).
Classroom composition and children’s
Nonlinear contribution of age composition
vocabulary gains
Recall that the second research question sought to
The first research question sought to examine the extent examine the extent to which the relation between
to which variability in the age composition of ECE class- within-classroom age composition and vocabulary gain
rooms was associated with children’s vocabulary was nonlinear. To do so, we next added the squared
growth. For this question, age composition was first product of the within-classroom age composition
represented by the within-class standard deviation of variable. The results of this model are presented in
children’s ages at the time of pretest. As another method Table 3, and show a small and nonsignificant quadratic

Table 2. Predicting vocabulary growth from age composition and child’s age.
Within-Classroom Age SD Within-Classroom Age Range
Estimate Standard Error t p Estimate Standard Error t p
Intercept 0.585 0.019 31.18 <001 0.585 0.019 31.22 <001
Pretest Vocabulary 0.687 0.015 44.36 <001 0.687 0.016 44.31 <001
Child Male −0.019 0.015 −1.23 218 −0.019 0.015 −1.24 214
Child Bilingual −0.196 0.031 −6.27 <001 −0.194 0.031 −6.24 <001
Child’s Age 0.053 0.014 3.82 <001 0.055 0.014 3.98 <001
Children per Class −0.031 0.020 −1.51 131 −0.026 0.022 −1.18 236
Class mean Age 0.004 0.020 0.20 844 0.001 0.020 0.04 966
Age Composition −0.003 0.016 −0.16 870 −0.010 0.017 −0.60 551
Child Age*Age composition 0.000 0.012 −0.02 982 −0.004 0.012 −0.31 758
Variance Components
Tau (Intercept) 0.034 0.005 6.77 <001 0.033 0.005 6.73 <001
Tau (Pretest Slope) 0.018 0.004 5.00 <001 0.018 0.004 5.01 <001
Sigma Squared (Error) 0.139 0.004 33.98 <001 0.139 0.004 33.98 <001
8 L. M. JUSTICE ET AL.

Table 3. Predicting vocabulary growth from nonlinear classroom age composition.

Within-Classroom Age SD Within-Classroom Age Range
Estimate Standard Error t p Estimate Standard Error t p
Intercept 0.612 0.024 25.06 <001 0.628 0.025 25.57 <001
Pretest Vocabulary 0.686 0.016 44.26 <001 0.685 0.016 44.11 <001
Child Male −0.019 0.015 −1.27 204 −0.019 0.015 −1.27 204
Child Immigrant −0.197 0.031 −6.31 <001 −0.194 0.031 −6.24 <001
Child Age 0.054 0.012 4.50 <001 0.054 0.012 4.52 <001
Class Size −0.034 0.020 −1.64 102 −0.023 0.021 −1.07 287
Class Mean Age 0.005 0.017 0.33 744 0.004 0.016 0.24 807
Age Composition −0.008 0.016 −0.47 638 −0.025 0.018 −1.43 153
Squared Age Composition −0.026 0.015 −1.67 095 −0.041 0.016 −2.64 008
Variance Components
Tau (Intercept0 0.033 0.005 6.74 <001 0.032 0.005 6.67 <001
Tau (Pretest Slope) 0.018 0.004 5.03 <001 0.018 0.004 5.06 <001
Sigma Squared (Error) 0.139 0.004 34.00 <001 0.139 0.004 34.01 <001

relation between within-classroom standard deviation the asymptote of the function is located at 24 months,
and children’s language gain (estimate ¼ −.026, which shows that each additional month of age range
p ¼ .095), and a slightly larger but statistically signifi- does not change the predicted estimate (expected
cant effect of within classroom range (estimate ¼ −.041, additional change per month ¼ −.03, p ¼ .15). At two
p ¼ .008). standard deviations above the mean of age range
To better understand this effect, we graphed the (42 months), age range again shows a small relation
fitted relation between class range and language gain (estimate ¼ .05), but each additional month of age range
(Figure 1). In Figure 1, the x-axis is the classroom age would correspond to a negative change (−.21, p ¼ .01).
range in months, and the y-axis is the expected It is clear from this depiction the interaction manifests
additional gain in language scores for each additional such that some age-range is beneficial to children, but
month of student age range. For example, at two stan- too much age range can be detrimental to language
dard deviations below the mean of classroom age range gains.
(6 months), the expected gain in language is .10 points,
and the positive slope of the line suggests that one Exploratory interactions
additional month of age range would expect to result In addition to the a-priori tests of classroom
in an additional 0.15 points of gain from fall to spring composition and children’s vocabulary gain, we also
(through a centering procedure, we find that this differ- tested an additional exploratory interaction effect. First,
ence is significantly different from zero, p ¼ .01). At the whether the quadratic effect of classroom age
mean (two-year age range; 24 months), we expect to see composition was dependent on children’s age, again
.20 points of gain. Examining the function on Figure 1, testing whether classroom composition is more or less

Figure 1. Graphical representation of results depicted in Table 3: Additional expected language gain for each observed value of
within-classroom age range in months.
 APPLIED DEVELOPMENTAL SCIENCE 9

important for some ages than others. We did so by Table 4. Same age (SA) and mixed age (MA) classrooms to
adding the three-way interaction term with child age predict children’s vocabulary gains.
interacting with the squared age composition quadratic Estimate Standard Error t p
Intercept 0.221 0.137 1.62 108
term. In each case the standardized coefficients were Pretest 0.700 0.016 44.33 <001
near zero and nonsignificant (estimates ¼ .002, p ¼ .16 Child Male −0.019 0.015 −1.29 198
and −.001, p ¼ .31, respectively). The remaining model Child Immigrant −0.195 0.031 −6.26 <001
Child Age 0.005 0.001 4.45 <001
was unchanged. Class Size −0.002 0.002 −1.27 206
Class Mean Age 0.002 0.003 0.59 557
SA vs WMA 0.002 0.043 −0.04 972
Curvilinear relations SA vs MMA 0.086 0.042 2.06 040

As an additional way to understand the curvilinear Note. SA ¼ same age classrooms; MMA ¼ moderate mixed age classrooms;
WMA ¼ wide mixed age classrooms.
relationship between classroom age composition and
children’s outcomes, we next created a categorical
representation of whether a given class was SA or children’s ages interacted with any main effects of SA
MA. To do so, we examined the distribution of vs. MA classrooms on children’s vocabulary growth,
within-classroom age ranges (see Figure 2), and but we did not find any evidence to support these ideas
identified three equally-spaced age groupings (each 13 (absolute value of all estimates �.001, p-values >.80).
months) that characterized the classroom sample.
Classrooms in which the age range was within 13
months between the oldest and youngest child were Discussion
coded as same age classrooms (SA; n ¼ 543); classrooms
Many initiatives across North America seek to increase
in which the age range between the oldest and youngest
young children’s access to and participation in ECE. For
child was between 13 and 26 months were coded as
instance, Quebec legislation provided publicly funded
moderately mixed age classrooms (MMA; n ¼ 927);
ECE programs to all 4-year-old children across the
and classrooms in which the age range exceeded 26
Canadian province at a very low cost to parents in
months between the oldest and youngest child
1997, with the program expanded to include all
were coded as widespread mixed age classrooms
children from birth to age 5 by 2000 (Haeck, Lefebvre,
(WMA; n ¼ 1300). These categorical codes were entered
& Merrigan, 2014). Importantly, the nature of policies
into the same multiple regression analysis previously
enacted regarding ECE expansion have direct effects
described, with the results presented in Table 4.
on the nature of classroom composition, including the
In Table 4, we see that there were no differences in
age range of children served within a given program.
the expected vocabulary gains for the SA group vs. the
For instance, state-funded programs that provide ECE
WMA group, but there was a small but significant
to only 4-year-olds yield classrooms serving children
difference in vocabulary gains favoring the moderate
who range in age relatively narrowly (i.e., all are four
mixed age group over the wide mixed age group
years of age), whereas other programs, such as the
(estimate ¼ .84, p ¼ .021). As with the previous
federal Head Start program, yield classrooms that serve
question, we also investigated the extent to which
children ranging in age more broadly, as the program
supports participation of children ranging in age from
3 to 5 years of age. Given that classroom age
composition is a structural characteristic of ECE
classrooms that can be readily affected by public policy,
it is valuable to consider whether the age range of
children in a classroom may have effects on their
development within that setting.
For many years, educational researchers and theorists
have argued the value of MA programming for young
children. This advocacy stems from prominent
developmental theories, including those of Vygotsky
(1978), which assert that children’s development and
learning is contingent on access to more capable peers.
Figure 2. Histogram (n on Y-axis) of classroom age ranges
with vertical lines representing designated cut points for the Two-year-old children who have opportunities to
three groups of classroom age ranges (cut points at 12 and engage with three- and four-year-old children in ECE
26 months). settings, for instance, are allowed significant enrichment
10 L. M. JUSTICE ET AL.

opportunities in which they observe their elder peers in did not find that these relations varied by children’s
engage in more complex play and talk, which “stretches” own age. In other words, both the older children and
their development (Winsler et al., 2002); for older the younger children benefited when the age range
children in these settings, their interactions with was between 14 and 26 months.
younger children provide them crucial opportunities A few other findings of interest warrant note. First,
to model advanced skills and serve as mentors (Whaley study results showed that vocabulary growth for
& Kantor, 1992). children in same-age programming, in which children
Recent, larger-scale developmental research has are no more than 13 months apart in age, was similar
raised questions regarding the developmental benefits to that seen for children in mixed-age classrooms in
of mixed-age groupings, however, especially for older which children varied widely in their ages. In the
children. For instance, Ansari and colleagues (2016) widespread mixed-age classrooms, which enrolled the
found that 4-year-olds in Head Start programs made largest number of children in this study, there were more
fewer gains in language and literacy skills when their than 26 months between the oldest and young children
classrooms contained a relatively high density of in the classroom. While we can only speculate as to
3-year-olds. This finding seems to argue against the why these classroom configurations seem to equivalent
value of mixing ages, at least for 4-year-olds, and would influence on children’s language development, we
provide impetus for advancing ECE programs that serve speculate that in the widespread mixed-age classrooms,
only 4-year-old children. Guo and colleagues’ (2014) children may segregate themselves into smaller
study of classroom age composition for 3-, 4-, and groupings exhibiting age homophily. Studies consistently
5-year-old children showed that mixed-age program- show that children begin to self-segregate into smaller
ming had positive effects on the vocabulary skills of groupings during the preschool years, with segregation
young children, and negative effects on the vocabulary related to such factors as children’s gender, behavioral
of older children. This finding complicates the issue, competence, and skill levels (Chen, Lin, Justice, &
as it would provide impetus for advancing mixed-age Sawyer, 2017; Daniel, Santos, Peceguina, & Vaughn,
ECE programming to the benefit of younger children, 2013). These homophily effects seem to reflect children’s
while recognizing that it could have negative impacts interest in engaging with others who are like them.
for older children. While age has not typically been included in studies of
A limitation of the classroom-age composition social networks in preschool classrooms, children’s do
literature, and any interpretation of results, is that all show homophily in their friendships based on skill levels,
studies have focused only at-risk, low-income children which often correlate closely with age (Lin, Justice, Paul,
enrolled in classrooms with relatively little overall & Mashburn, 2016). It may be that in widely mixed age
variation in classroom-age composition. To this end, classrooms, children interact primarily with those similar
such work has not had the benefit of exploring age in age, thus mirroring what occurs in same-age class-
composition effects with an unselected sample of rooms. If this is the case, children’s language growth
children attending programs that vary widely in the ages may be similar in both types of settings.
of children enrolled, as we did in this study. The Second, the study results did not find there to be an
primary finding of this study, which fills a critical gap interaction between children’s age and the classroom
in the literature, is showing that classroom age composition variables of range and SD, as has been
composition does relate to children’s development. found in several prior studies (Ansari et al., 2016; Moller
Specifically, with respect to children’s vocabulary et al., 2008). Such work, conducted in the United States,
growth, we found that mixed-age programming yielded indicated that having a larger range of ages in ECE
the best outcomes for children when compared to SA settings may have negative effects on the development
programming, but only when the mixed-age of older children in these classrooms, possibly because
programming results in classroom configurations in they begin “act younger” (Winsler et al., 2002). In the
which the youngest and older child differ in age by at present study, we found no evidence that older children
least 14 months but no more than 26 months. This are negatively affected by being in mixed-age
result suggests that children in ECE classrooms in which classrooms, potentially because mixed-age settings allow
classmates range in age from 36 months to 56 months of older children to serve as mentors of younger children;
age, as an example, would show greater vocabulary theoretically, engaging with less-skilled children
gains than children whose classmates range from 36 provides opportunities for older children to learn and
months to 45 months of age (range is too limited) or develop (Vygotsky, 1978).
those whose classmates range from 36 months to 66 Third, the study results suggest that while there are
months of age (range is too broad). Importantly, we benefits to mixing ages within early childhood
 APPLIED DEVELOPMENTAL SCIENCE 11

classrooms, there is a levelling off of this benefit when program transitioned from SA to MA classrooms. In
children’s age range exceeds a certain point; that is, the SA classrooms, the age range of children in the
when the difference in age between the youngest and classrooms was about 13 months; whereas, it was 18
oldest child in a classrooms exceeds 24 months, the ben- months in the MA classrooms. The authors showed
efits of mixing ages diminishes. We can only speculate there to be significantly more cross-age interaction in
as to why this occurs, but suspect that it reflects age the MA classrooms as compared to SA classrooms,
homophily effects that occur in early childhood class- which presumably provides young children the opport-
rooms, in which children tend to prefer interacting unity to learn important skills from older children.
with children who are similar in age. In a classic study Indeed, 3-year-olds in MA classrooms spent a greater
of 1- to 12-year-old children’s interactions with others amount of time in on-task goal-oriented activities than
at home and in the neighborhood, researchers showed 3-year-olds in SA classrooms. Such work suggests that
a developmental increase in children’s preference to the broadened age range and therefore developmental
interact with others who are closer in age (Ellis, Rogoff, competencies of children that occur in MA classrooms
& Cromer, 1981). Specifically, from the toddler years enhance the learning opportunities available within
into early adolescence, children showed a gradual, sig- such settings. As a result, we would argue that provision
nificant increase in interacting with peers who are closer of MA classrooms is an important policy lever by which
(<1 year age difference) in age to them. It may be that to maximize the benefits of ECE expansion for children.
in preschool settings in which children vary significantly The results of this study should help to inform poli-
in age, children create social networks comprising those cies related to ECE access and expansion, especially
who are relatively close in age. If this is the case, chil- efforts that maximize the potential for MA program-
dren would be clustered in same-age social networks ming. For instance, rather than providing funding to
within these classrooms, with limited opportunities to programs to serve children of only a given age (e.g.,
benefit from interactions with those who are signifi- 4-year-olds), expansion policies can allow ECE funds
cantly younger or older. Research that examines to be layered with other funds to create programs that
micro-level patterns in children’s social networks as a serve children across a range of ages. In Ohio, for
product of same- and mixed-age programming is instance, the state’s effort to expand participation in
needed to confirm this possibility. ECE focuses on providing funding to serve 4-year-olds
It is worthwhile to point out that the contribution of specifically, but ECE programs can layer that funding
classroom age composition on children’s vocabulary into a broader preschool program that serves children
development was relatively small in magnitude (stan- varying widely in ages, thus allowing for MA program-
dardized estimate ¼ .04). This effect can be contextua- ming. An alternative approach is for policies to support
lized in light of a recent integrative analysis of four a broader range of ages, as does Early Head Start/Head
large-scale datasets by Keys et al. (2013), which exam- Start (birth to five years) and early childhood special
ined the relations between a number of ECE classroom education (three to five years). However, while these
characteristics and 6,240 3- to 5-year-old children’s programs allow potential for MA programming, these
development in language, math, and socioemotional also represent targeted enrollment programs that can
skills. Application of meta-analytic techniques indicated result in children being tracked into classrooms that
that the average relation between classroom quality and serve only children who are poor or only children with
children’s language outcomes was .05 (standardized disabilities. It is unclear that the results presented in this
coefficient), which is similar to that reported in this study, which involved a diverse, un-selected sample,
study. While it can take extensive efforts to improve would generalize effectively to targeted-enrollment
classroom quality, including considerable professional MA programming.
development of teachers (Early, Maxwell, Ponder, & There are limitations of this work that should
Pan, 2017), modifying the classroom age composition be recognized. First, this study took place in the well-
of ECE classrooms could take relatively minimal effort established universal child-care system of Denmark.
and yield positive benefits to children. Given universal participation of children in ECE in this
The findings of this work provide empirical support country, important opportunities are available to learn
for the studies that have identified crucial, proximal about how to configure ECE structural characteristics
processes in MA classrooms that would seem to yield in ways to maximize developmental benefits for chil-
better benefits for children. For instance, Winsler and dren, as we do in this study. While we would argue that
colleagues (2002) examined children’s goal-directed the results presented here provide important truths
activities and peer interactions in MA and SA class- regarding how classroom age composition interfaces
rooms in a longitudinal study in which one ECE with children’s development in ECE settings, the Danish
12 L. M. JUSTICE ET AL.

context varies in important ways from other contexts, readiness for children enrolled in Head Start. Early
thus raising questions about generalization. Second, Childhood Research Quarterly, 28(1), 1–10. doi:10.1016/j.
we examined only a single outcome of interest, namely ecresq.2012.06.002
Bleses, D., Hogen, A., Slot, P., & Justice, L. M. (2017).
children’s vocabulary growth. We did not assess Relations between structural quality aspects and process
whether the age-composition effect transcended quality in Danish preschools: Evidence for compensating
different dimensions of development. Relatedly, this factors. Presentation at the International Association for
assessment was completed by teachers rather than the Study of Children Language, Lyon France.
independent research staff. Examining classroom- Bleses, D., Højen, A., Justice, L. M., Dale, P. S., Dybdal, L.,
Piasta, S. B., … & Haghish, E. F. (in press). The Effective-
composition effects on other measures, including those
ness of a Large-Scale Language and Preliteracy Inter-
implemented by research professionals is an important vention: The SPELL Randomized Controlled Trial in
direction for future research. Finally, the research Denmark. Child Development.
presented here is correlational in nature. Experimental Bleses, D., Vach, W., Jørgensen, R. N., & Worm, T. (2010).
research that examines the developmental benefits of The internal validity and acceptability of the Danish SI-3:
SA versus MA programming for children is necessary A language screening instrument for 3-year-olds. Journal
of Speech, Language and Hearing Research, 53, 490–507.
to ensure that the results observed this in and other Campbell, F. A., Ramey, C. T., Pungello, E., Sparling, J., &
studies are not simply a result of selection bias, given Miller-Johnson, S. (2002). Early childhood education:
the nonrandom enrollment of children into ECE Young adult outcomes from the abecedarian project.
classrooms. As ECE programs expand, there should be Applied Developmental Science, 6(1), 42–57. doi:10.1207/
opportunities to insert experimental approaches into S1532480XADS0601_05
Chen, J., Lin, T.-J., Justice, L., & Sawyer, B. (2017). The social
initiatives to assess causal relations between
networks of children with and without disabilities in early
classroom age composition and children’s cognitive childhood special education classrooms. Journal of Autism
and noncognitive development, as has been possible in and Developmental Disorders, 1–16.
other major societal initiatives (Leventhal & Crosnoe, R., Purtell, K. M., Davis-Kean, P., Ansari, A., &
Brooks-Gunn, 2003). Benner, A. D. (2016). The selection of children from low-
In sum, classroom age composition is a malleable income families into preschool. Developmental Psychology,
52(4), 599. doi:10.1037/dev0000101
feature of ECE classrooms, representing one that can Daniel, J. R., Santos, A. J., Peceguina, I., & Vaughn, B. E.
be easily affected through public policies and practical (2013). Exponential random graph models of preschool
initiatives. Despite the importance of this characteristic, affiliative networks. Social Networks, 35(1), 25–30.
there is little research to guide policymakers and doi:10.1016/j.socnet.2012.11.002
practitioners on the topic. Our study uses a large, Early, D. M., Maxwell, K. L., Ponder, B. B., & Pan, Y. (2017).
Improving teacher-child interactions: A randomized
representative sample to shed light on topic and finds
control trial of making the most of classroom interactions
a complex, but practical finding. Children’s develop and my teaching partner professional development models.
vocabulary best when exposed to peers of different ages, Early Childhood Research Quarterly, 38, 57–70.
but not to peers who are too different in age. Taking this doi:10.1016/j.ecresq.2016.08.005
into consideration when designing children’s classroom Ellis, S., Rogoff, B., & Cromer, C. C. (1981). Age segregation
environment in future ECE programming has the in children’s social interactions. Developmental Psychology,
17(4), 399. doi:10.1037/0012-1649.17.4.399
potential to increase children’s vocabulary growth and, Esping-Andersen, G., Garfinkel, I., Han, W.-J., Magnuson, K.,
ultimately, their school readiness. Wagner, S., & Waldfogel, J. (2012). Child care and school
performance in Denmark and the United States. Children
and Youth Services Review, 34(3), 576–589. doi:10.1016/j.
Funding childyouth.2011.10.010
This work was supported by the Denmark National Board of Guo, Y., Tompkins, V., Justice, L. M., & Petscher, Y. (2014).
Social Services [Grant Number x] and This work was Classroom age composition and vocabulary development
supported by the The Danish Council for Strategic research among at-risk preschoolers. Early Education and
[Grant Number x]. Development, 25(7), 1016–1034. doi:10.1080/10409289.
2014.893759
Gupta, N. D., & Simonsen, M. (2010). Non-cognitive child
References outcomes and universal high quality child care. Journal of
Public Economics, 94(1), 30–43. doi:10.1016/j.jpubeco.
Ansari, A., Purtell, K., & Gershoff, E. (2016). Classroom age 2009.10.001
composition and the school readiness of 3-and 4-year-olds Haeck, C., Lefebvre, P., & Merrigan, P. (2014). The
in the Head Start Program. Psychological Science, 27(1), distributional impacts of a universal school reform
53–63. doi:10.1177/0956797615610882 on mathematical achievements: A natural experiment
Bell, E. R., Greenfield, D. B., & Bulotsky-Shearer, R. J. (2013). from Canada. Economics of Education Review, 41,
Classroom age composition and rates of change in school 137–160. doi:10.1016/j.econedurev.2014.03.004
 APPLIED DEVELOPMENTAL SCIENCE 13

Hamre, B., Justice, L. M., Pianta, R. C., Kilday, C., Sweeney, Lillard, A., & Else-Quest, N. (2006). The early years: Evaluat-
B., Downer, J. T., & Leach, A. (2010). Implementation ing Montessori education. Science, 313(5795),
fidelity of my teaching partner literacy and language 1893–1894. doi:10.1126/science.1132362
activities: Association with preschoolers’ language and Lin, T.-J., Justice, L. M., Paul, N., & Mashburn, A. J. (2016).
literacy growth. Early Childhood Research Quarterly, Peer interaction in rural preschool classrooms:
25(3), 329–347. doi:10.1016/j.ecresq.2009.07.002 Contributions of children’s learning-related behaviors,
Huang, F. L., Invernizzi, M. A., & Drake, E. A. (2012). The language and literacy skills, and problem behaviors. Early
differential effects of preschool: Evidence from Virginia. Childhood Research Quarterly, 37, 106–117. doi:10.1016/j.
Early Childhood Research Quarterly, 27(1), 33–45. ecresq.2016.04.001
doi:10.1016/j.ecresq.2011.03.006 Moller, A. C., Forbes-Jones, E., & Hightower, A. D. (2008).
Huttenlocher, J., Vasilyeva, M., Cymerman, E., & Levine, S. Classroom age composition and developmental change in
(2002). Language input and child syntax. Cognitive 70 urban preschool classrooms. Journal of Educational
Psychology, 45, 337–374. doi:10.1016/S0010-0285(02) Psychology, 100(4), 741. doi:10.1037/a0013099
00500-5 National Center for Education Statistics. (2017). The
Justice, L. M., Logan, J. A., Lin, T.-J., & Kaderavek, J. N. Condition of Education: Preschool and Kindergarten
(2014). Peer effects in early childhood education Enrollment. Washington, DC: National Center for
testing the assumptions of special-education inclusion. Education Statistics.
Psychological Science, 25(9), 1722–1729. DOI: 10.1177/ Organisation for Economic Co-operation and Development.
0956797614538978. (2016). Enrolment in childcare and pre-school. Retrieved
Justice, L. M., Petscher, Y., Schatschneider, C., & Mashburn, from https://www.oecd.org/els/soc/PF3_2_Enrolment_
A. (2011). Peer effects in preschool classrooms: Is children’s childcare_preschool.pdf.
language growth associated with their classmates’ skills? Sylva, K., Ereky-Stevens, K., & Aricescu, A. (2015). Overview
Child Development, 82(6), 1768–1777. doi:10.1111/j.1467- of European ECEC curricula and curriculum template:
8624.2011.01665.x CARE, Utrecht, The Netherlands.
Kamerman, S. B. (1991). Child care policies and programs: An Vygotsky, L. S. (1978). Mind in society: The development of
international overview. Journal of Social Issues, 47(2), higher mental process.
179–196. doi:10.1111/j.1540-4560.1991.tb00294.x Whaley, K., and Kantor, R. (1992). Mixed-age grouping
Keys, T. D., Farkas, G., Burchinal, M. R., Duncan, G. J., in infant-toddler childcare: Enhancing developmental
Vandell, D. L., Li, W., & Howes, C. (2013). Preschool center processes. Child and Youth Care Forum, 21, 369–384.
quality and school readiness: Quality effects and variation Winsler, A., Caverly, S. L., Willson-Quayle, A., Carlton, M. P.,
by demographic and child characteristics. Child Howell, C., & Long, G. N. (2002). The social and behavioral
Development, 84(4), 1171–1190. doi:10.1111/cdev.12048 ecology of mixed-age and same-age preschool classrooms:
Landersø, R., & Heckman, J. J. (2017). The Scandinavian A natural experiment. Journal of Applied Developmental
fantasy: The sources of intergenerational mobility in Psychology, 23(3), 305–330. doi:10.1016/S0193-3973(02)
Denmark and the US. The Scandinavian Journal of 00111-9
Economics, 119(1), 178–230. Winsler, A., Tran, H., Hartman, S. C., Madigan, A. L., Manfra,
Lefebvre, P., Merrigan, P., & Verstraete, M. (2009). Dynamic L., & Bleiker, C. (2008). School readiness gains made by
labour supply effects of childcare subsidies: Evidence from ethnically diverse children in poverty attending
a Canadian natural experiment on low-fee universal child center-based childcare and public school pre-kindergarten
care. Labour Economics, 16(5), 490–502. doi:10.1016/j. programs. Early Childhood Research Quarterly, 23(3),
labeco.2009.03.003 314–329. doi:10.1080/03004430.2011.622755
Leventhal, T., & Brooks-Gunn, J. (2003). Moving to Yeomans-Maldonado, G., Justice, L. M., & Logan, J. (In
opportunity: An experimental study of neighborhood Press). The mediating role of classroom quality on peer
effects on mental health. American Journal of Public Health, effects and language gain in pre-kindergarten ECSE
93(9), 1576–1582. classrooms. Applied Developmental Science.