Available via license: CC BY-NC
Content may be subject to copyright.
AERA Open
January-March 2018, Vol. 4, No. 1, pp. 1 –16
DOI: 10.1177/2332858418756598
© The Author(s) 2018. http://journals.sagepub.com/home/ero
Creative Commons Non Commercial CC BY-NC: This article is distributed under the terms of the Creative Commons
Attribution-NonCommercial 4.0 License (http://www.creativecommons.org/licenses/by-nc/4.0/) which permits non-commercial
use, reproduction and distribution of the work without further permission provided the original work is attributed as specified on the SAGE and
Open Access pages (https://us.sagepub.com/en-us/nam/open-access-at-sage).
With scientists and policy decision makers lauding the
promise of early childhood education and care (ECEC) for
raising the achievement of disadvantaged children, there
remains concern about whether large-scale, public-funded
ECEC programs can replicate the results of small-scale
interventions (European Commission, 2011; Farran, 2016;
Magnuson & Shager, 2010; Obama, 2014; Organisation
for Economic Co-operation and Development [OECD],
2006; Yoshikawa et al., 2013). There are also questions
(Barnett, 2010, 2011) about whether scale-up should be
universal (i.e., designed to serve all children) or targeted
(i.e., exclusively serving disadvantaged children). To date,
scale-up efforts in the United States have largely been tar-
geted programs for disadvantaged preschoolers (3- to
5-year-olds), but some nations now provide publicly
funded, universal ECEC beginning in infancy. Norway is
a case in point, having taken universal ECEC, beginning at
age 1, to a national scale. As a complement to growing
evidence about scale-up in the United States, we investi-
gated the consequences of Norway’s scale-up of ECEC for
children’s early language skills. In doing so, we examined
whether the universal scale-up had differential conse-
quences for children from low-income versus middle- and
high-income families.
Estimating the Consequences of Norway’s National Scale-Up of Early
Childhood Education and Care (Beginning in Infancy) for Early
Language Skills
Eric Dearing
Boston College
Norwegian Center for Child Behavioral Development
Henrik Daae Zachrisson
Norwegian Center for Child Behavioral Development
University of Oslo
Arnstein Mykletun
University of Tromsø
Norwegian Institute of Public Health
Haukeland University Hospital
Nordland Hospital Trust
Claudio O. Toppelberg
Judge Baker Children’s Center
Harvard Medical School
While most early childhood education and care (ECEC) programs taken to scale in the United States have served socially
disadvantaged 3- to 5-years-olds, Norway scaled up universal ECEC from age 1. We investigated the consequences of
Norway’s universal ECEC scale-up for children’s early language skills, exploiting variation in ECEC coverage across birth
cohorts and municipalities in a population-based sample (n = 63,350). Estimates from two-stage least squares (i.e., instru-
mental variable) regression and generalized difference-in-differences models indicated the scale-up of universal ECEC led
to improved language outcomes, particularly for low-income children. As preschool programs at scale become increasingly
common in the United States, our results from Norway help inform debate about the merits of universal versus targeted
policies and should provoke discussion about the benefits of beginning ECEC programs as early as infancy.
Keywords: Child Development, Early Childhood, Econometric Analysis, Educational Policy, Longitudinal Studies, Policy
Analysis, Poverty
756598EROXXX10.1177/2332858418756598Dearing et al.Norway’s National Scale-Up
research-article20182018
Dearing et al.
2
Preschool ECEC and the Achievement of Socially
Disadvantaged Children
Given evidence that social inequality in achievement
may be largely attributed to limited developmental stimu-
lation in the home (e.g., Dearing, 2014; Hackman, Farah,
& Meaney, 2010; Hoff, 2006; Shonkoff & Phillips, 2000;
Yoshikawa et al., 2012), researchers across multiple disci-
plines have argued that high-quality ECEC may serve a
compensatory function for children growing up in poverty
(e.g., Dearing, McCartney, & Taylor, 2009; Magnuson &
Shager, 2010). There are, in fact, a relatively large number
of experimental and quasi-experimental studies addressing
the effects of preschool attendance (typically for children
age 3 or older) on achievement and meta-analyses of the
results. For example, two meta-analyses of experimental
and quasi-experimental preschool interventions targeting
disadvantaged children in the United States demonstrate
short-term and longer term achievement gains with effect
sizes (Hedge’s g) of about .20 (Camilli, Vargas, Ryan, &
Barnett, 2010; Duncan & Magnuson, 2013), although often
with some diminishing of the effects as children get older
(Barnett, 2011).
Despite this evidence, it remains less clear whether (or
in what direction) universal ECEC policies may affect
achievement gaps between more and less advantaged chil-
dren. To the extent that enrichment in one early learning
environment compensates for deprivation in another
(Sameroff & Chandler, 1975), universal ECEC coverage
may narrow achievement gaps. Yet to the extent that enrich-
ment in ECEC programs complements enrichment at home
(Ceci & Papierno, 2005), achievement gaps between
socially advantaged and disadvantaged children may be
widened by universal provision. In an evaluation of a state-
funded universal preschool program in the United States,
ECEC had the largest achievement benefits for poor chil-
dren but also important benefits for near-poor and middle-
class children (Gormley, Gayer, Phillips, & Dawson, 2005;
Gormley, Phillips, & Gayer, 2008). Both of these studies
used regression discontinuity designs based on cut-off
birth dates accounting for program admission to address
selection bias.
In a meta-analysis of universal preschool program evalu-
ations using quasi-experimental designs, there was mixed
evidence, with the most consistent effects evident for disad-
vantaged children (van Huizen & Platenga, 2015). There is
also evidence from Norway, where the present study was
conducted, that income inequality among young adults was
reduced by an increase in preschool coverage in the late
1970s, when these adults were between 3 and 7 years of age
(Havnes and Mogstad, 2011). In this article, the authors used
a difference-in-differences (DID) design based on variation
in enrollment rates pre– and post–preschool reforms to iden-
tify the causal effect.
ECEC for Infants and Toddlers
Compared to the evaluations of ECEC programs for pre-
school-aged children, there is less evidence about the effec-
tiveness of such programs for infants and toddlers. Yet the
potential benefits of ECEC for infants and toddlers is of
interest given the consequences of learning stimulation and
environmental experiences for brain growth during the first
2 to 3 years of life (Fox & Rutter, 2010; Mustard, 2006;
Nelson & Sheridan, 2011; Shonkoff & Phillips, 2000). And
observational studies do, in fact, show that ECEC in the ear-
liest years (0–3 years) can promote language and cognitive
development. For example, in one U.S. study, more time in
center care between 6 and 36 months of age predicted better
language and school readiness scores at 3 years, controlling
for background factors and child care quality (National
Institute of Child Health and Human Development [NICHD],
2000). Also, in nationally representative U.S. data, Loeb,
Bridges, Bassok, Fuller, and Rumberger (2007) found that
math and reading achievement (measured at the start of kin-
dergarten) was higher for children who entered center care
prior to kindergarten, with effect sizes largest (Cohen’s d
approximately .11) for those who entered between ages 2
and 3 compared to earlier or later.
Within this line of nonexperimental work, there is also
some evidence that high-quality child care is most strongly
and positively associated with language skills and school
readiness for low-income 3-year-olds (McCartney, Dearing,
Taylor, & Bub, 2007), thereby reducing income-related
achievement gaps in math and literacy through the elemen-
tary school years (Dearing et al., 2009). Yet larger ECEC
effects for low-income children are not found consistently.
Ruzek and colleagues (Ruzek, Burchinal, Farkas, &
Duncan, 2014), for example, found higher quality infant
and toddler ECEC was associated with better cognitive out-
comes at 2 years, but not more so for low-income children
than for others.
The potential in the early years for ECEC to narrow
achievement gaps has also received increasing attention out-
side of the United States. In nonexperimental studies from
Canada, ECEC attendance prior to 4 years predicted a nar-
rowing gap in early language skills and academic readiness
between children of low versus high socioeconomic status
(Geoffroy et al., 2007; Geoffroy et al., 2010); children of
mothers with low education who attended formal child care
scored between 36% and 87% of a standard deviation higher
on a range of achievement and cognitive tests after school
entry, compared to their counterparts not attending ECEC. In
a large U.K. sample, using a similar methodology, Cote,
Doyle, Petitclerc, and Timmins (2013) found that center-
based care (at 9 months) was associated with better cogni-
tive skills at age 3 and 5 (about 10% of a standard deviation
compared to attending other types of care) and was more
than twice as strongly so associated for children of mothers
Norway’s National Scale-Up
3
with low education. Yet this interaction was not evident for
low family income, and the association was inconsistent at
ages 5 and 7.
Notably, across most of these studies of ECEC between 0
and 3 years (in the United States and elsewhere), many low-
income children attended informal care settings, not ECEC
centers. Two exceptions, one from Canada and one from the
United States, addressed larger scale implementations of
center-based ECEC programs for children age 0 to 3. In
Quebec, for example, a quasi-experimental study of the roll-
out of universal early child care coverage found no effect on
language skills at age 4 and negative impacts on behavioral,
social, and motor outcomes, particularly for children enter-
ing as infants and toddlers and particularly for children of
less educated parents (Baker, Gruber, & Milligan, 2008;
Kottelenberg & Lehrer, 2014). It is notable, however, that
the Quebec program covered relatively few 0- to 2-year-
olds, involved relatively high adult to child ratios in care
settings, and resulted in most infants and toddlers being
cared for in accredited home care rather than in center care
(Japel, Tremblay, & Cote, 2005).
In the U.S. study, Duncan and Sojourner (2013) used a
large-scale randomized trial of a high-quality infant and tod-
dler ECEC intervention (the Infant Health and Development
Program) to extrapolate population-level effects of either a
universal program or one specifically targeted at children
from low-income families. The authors estimated that prior
to school entry, in both cases, income disparities in cognitive
abilities would be strongly reduced or eliminated, with treat-
ment gains being substantial even 2 years after the end of the
intervention (e.g., IQ score differences of 86% of a standard
deviation between low-income children in the treatment and
control groups).
ECEC Policy in Norway
In the present study, we build on the existing literature to
examine Norway’s national scale-up of universal ECEC for
infants and toddlers. Scale-up of ECEC in Norway has been
a gradual process across the past three decades, with the
country having first set an aim of publicly subsidizing uni-
versal access to high-quality ECEC (beginning at age 1) in
the 1980s. In turn, through the 1990s, Norwegian ECEC
policy was focused on establishing federal regulations for
quality of care (e.g., teacher educational requirements and
an educational content framework) for both center-based
care and family day care (Ministry of Education, 2014). And
in 2002, Norwegian municipalities were mandated to pro-
vide access with “a right to child care for all children” in
ECEC centers and/or family day care units, leading to an
increasing number of ECEC slots for 1- and 2-year-olds.
Since the 2002 mandate, progression toward universal
access has occurred incrementally as public spending has
increased, and correspondingly, family fees for care have
decreased. In 2004, a maximum fee for full-time care was
introduced (NOK 2,750 or about US$400 per month at the
2004 exchange rate) and lowered further in 2006 (NOK
2,250 or about US$360 per month at the 2006 exchange
rate), but sliding scales less than the maximum fee varied at
the discretion of municipalities. Progress toward universal
access also varied considerably across municipalities due to
idiosyncratic local circumstances. Obstacles to increased
ECEC slots included, for example, (a) lack of available
spaces for building new centers, high building cost, and/or
lack of available contractors (particularly for the major cit-
ies); (b) lack of qualified staff (particularly for the smaller
municipalities); (c) concerns in some municipalities about
overcoverage due to year-by-year variations in birth rates
and for some municipalities conservative predictions of
demand; and (d) local concerns about the availability of
long-term earmarked funding from the government (Aspland
Virak, 2006, 2009; Rindfuss, Guilkey, Morgan, Kravdal, &
Guzzo, 2007). Nonetheless, across the 5-year time period
investigated in the present study, the national coverage for
1- to 2-year-olds increased from 40% to 75% (Sæther, 2010;
Scheistrøen, 2012).
With regard to age of entry, Norwegian parents with new-
borns receive up to 1 year paid leave; as a result, few chil-
dren enter nonparental care prior to 9 months of age (Ministry
of Children and Equality, 2014). After parental leave, par-
ents have the choice of enrolling their children in publically
subsidized ECEC or receiving cash benefits (approximately
NOK 3,500 or US$560 per month at the 2006 exchange rate,
2006 being when many families in the present study were
eligible) for staying home with their children until age 3.
With regard to quality, across the years of interest in the
present study, national structural quality regulations in
Norway stipulated that at least 30% to 35% of the staff
should be ECEC teachers (with 3-year university college
degrees) and that there should be a leader with ECEC teacher
education in each center. Adult to child ratios of 1:3 for those
younger than 3 and 1:6 for older children were also recom-
mended but not enforced by law. In ECEC centers, most
children younger than 3 are in infant-toddler groups of about
nine children with three staff, one of them a trained teacher.
Yet variations in group size and age composition are allowed,
as long as the staff requirements are met.
The pedagogical content is guided by a “Framework” cur-
riculum plan that sets out guidelines regarding the values and
purpose of ECEC centers, their curricular objectives, and
educational approaches (Ministry of Education, 2006). With
regard to language development, which is of interest in the
present study, the curriculum plan specifies that there should
be stimulating verbal and nonverbal interactions in all every-
day situations, age-appropriate use of learning materials, and
a learning-rich environment including work with symbols,
books, and reading. Thus, Norwegian ECEC is considered
one national program, although the implementation of this
Dearing et al.
4
program varies between ECEC centers within the limits of
the legal requirements for structural quality as well as the
“Framework” plan.
Monitoring of ECEC quality in Norway is a collaborative
responsibility of the ECEC centers and their municipalities,
with a focus primarily on inspection of structural quality
standards, safety and hygiene, and evaluation of the educa-
tional content (the latter varying in formality across munici-
palities). At the time of the present study, structural quality
standards were mostly in accordance with regulations and
recommended standards (Winsvold & Guldbrandsen, 2009)
while not entirely met in all centers across the country
(Brenna et al., 2010; OECD, 2015; UNICEF Innocenti
Research Center, 2008). Observed quality has just recently
been studied in Norway (Bjørnestad & Os, in press), show-
ing that most classrooms meet only minimal quality stan-
dards as assessed with the Infant/Toddler Environment
Rating Scale–Revised ITERS-R (Harms, Cryer, & Clifford,
2006), in particular with regard to hygiene, safety, and access
to play materials.
The Present Study
We examined whether Norway’s scale-up to national
ECEC coverage beginning at age 1 (a) had consequences for
children’s early language skills and, if so, (b) differentially
affected the language of children from low-, middle-, and
high-income families and communities. Our focus on early
language skills was driven by evidence that differences in
early language help explain a considerable portion of the dif-
ferences in school performance between lower and higher
income children during elementary school (Durham, Farkas,
Hammer, Tomblin, & Catts, 2007) and that early caregiving
environments are critical to developing early language skills
(Kuhl, 2004; Mustard, 2006; Pruden, Hirsh-Pasek, Golinkoff,
& Hennon, 2006; Shonkoff & Phillips, 2000; Snow, Burns,
& Griffin, 1998).
For our analyses, we used population-based data (n =
63,350) on children living in more than 400 municipalities in
Norway. Within these data, we focused our analyses on five
birth cohorts of children who were sampled after the 2002
mandate for universal access. In these birth cohorts, we
examined whether the ECEC policy scale-up (i.e., increas-
ing availability of public ECEC over time) led to better
mother-reported child language skills at age 3. More specifi-
cally, we estimated three types of statistical models that
addressed complementary questions about the consequences
of the scale-up.
First, using two-stage least square regression models, we
attempted to isolate the causal effects of attending ECEC on
individual children’s language. Specifically, we used ECEC
attendance rates within each child’s municipality, and for her
or his same birth cohort, as an instrument for that child’s
own ECEC attendance. The logic guiding these models was
that attendance rates provided an indicator of the availability
of ECEC over time within municipalities. In turn, we
expected that availability affected children’s use of ECEC,
for reasons unrelated to family selection, thereby allowing
us to examine the effects of exogenous variance in ECEC
use. Because we were also interested in whether ECEC use
had differential impacts on children from low-, middle-, and
high-income families, we estimated our models separately
for these groups.
Second, we estimated DID models that exploited varia-
tions in ECEC use across birth cohorts and municipalities to
estimate the causal effects of the policy scale-up at the level
of municipalities. Specifically, we estimated the effects of
changes in ECEC availability across cohorts, with the expec-
tation that increasing availability should predict improved
language skills within municipalities. Here again, we used
ECEC attendance as an indicator of the likely availability of
ECEC within municipalities for each birth cohort. And here
again, we estimated models separately for low-, middle-,
and high-income groups, although in this case with a focus
on community-level income (i.e., within-municipality aver-
age family income for each birth cohort).
Third, we used fixed-effects regression to extend our
study of the differential impacts of ECEC by income group.
Specifically, we investigated whether the scale-up of uni-
versal ECEC in Norway predicted a narrowing of achieve-
ment gaps between low- and high-income children (in this
case, language skill gaps). Our expectation was that munici-
palities that experienced a narrowing of ECEC-use gaps
between low- and high-income children (i.e., municipalities
in which increasing use of ECEC rose faster for low-income
children than for high-income children) would also evi-
dence the greatest narrowing of language skill gaps between
these groups.
Method
Participants
Data are from the Norwegian Mother and Child Cohort
Study (MoBa; for complete details, see Magnus et al., 2006,
and www.fhi.no/morogbarn). Beginning in 1999, pregnant
women in Norway who received routine exams at birth units
delivering more than 100 births per year were invited to par-
ticipate during their 17th-gestational-week visit. As of
October 2010, 90,725 mothers of 108,639 children had
enrolled and completed baseline assessments, which repre-
sented 42.1% of all eligible mothers in Norway. Written
informed consent was obtained, and the study was approved
by the Regional Committee for Medical Research Ethics and
the Norwegian Data Inspectorate.
Questionnaires covering demographics, health, life-
style, and child development were administered during the
17th, 22nd, and 30th weeks of gestation and at ages 0.5,
1.5, and 3.0 years (questionnaires are available online:
Norway’s National Scale-Up
5
www.fhi.no/moba-en). Retention rates at 1.5 and 3.0 years
were 72.4% and 59.3%, respectively. The present study uses
data collected during pregnancy (demographics), at 1.5 years
(child care use), and at 3.0 years (language skills). Linkage to
the National Income and the Medical Birth Registries pro-
vided data on family income and infant health, respectively.
For the purposes of this study, we restricted the sample to
2002 through 2006 birth cohorts (n = 63,471 children)
because the 2002 cohort was the first full cohort given the
age 3 questionnaire and the 2006 cohort was the last for
which registry data on family income were available by the
child’s age of 1.5 years. Moreover, by restricting analyses to
these cohorts, we targeted a time of exceptional population
increase in ECEC use during infancy due to policy changes.
In Appendix A in the online supplemental material, we detail
participation rates for the cohorts relative to population
births in eligible birthing units.
Measures
Child language. Two separate language screening measures
were used as indicators of broad levels of language ability,
from language difficulties and delays to typical language
ability. Mother reports were elicited to measure children’s
grammatical complexity at 3 years of age with the Norwe-
gian version of a six-point scale previously used in a large-
scale community study in the United Kingdom (Dale, Price,
Bishop, & Plomin, 2003). The initial development of the
measure was informed by the MacArthur Communicative
Development Inventory: U.K. Short Form (Dionne, Dale,
Boivin, & Plomin, 2003). A mother chose one of six state-
ments that best described her child’s ability, ranging from
“Not yet talking” (one point) to “Talking in long and com-
plex sentences, such as ‘when I went to the park, I went on
the swings’ or ‘I saw a man standing in the corner’” (six
points). For general communication skills including recep-
tive and expressive language, mothers answered six items
from the communication domain of a normed Norwegian
translation of the Ages and Stages Questionnaire (ASQ; Jan-
son & Squires, 2004; Squires, Bricker, & Potter, 1999). The
items included four original 36-month ASQ items, and one
item each from the 18- and 48-month ASQ. Mothers
responded on a three-point scale (yes, a few times, not yet) to
communicative behavior descriptions (e.g., “When looking
at a picture book, does your child tell you what is happening
or what action is taking place in the picture?”).
We combined these two measures (correlated by r = .62)
by computing the mean of the standardized scores. We then
transformed the final composite by log10 to correct for skew-
ness, inverted values such that higher scores would reflect
better language skills (in standard deviation units), and set
the lowest score to equal zero. For more details about the
distribution of items, psychometric properties, see Appendix
B in the online supplemental material.
ECEC arrangements. Mothers reported the type of child
care used at age 1.5 years, representing the child’s primary
care arrangement. Choices included “at home with mother
or father,” “at home with unqualified child minder,” “fam-
ily day care,” and “center care.” From these reports, we
computed a dummy variable indicating whether children
were in center-based ECEC versus any of the other arrange-
ments that were not regulated to include educational con-
tent. We also analyzed the data by comparing children in
center care exclusively with those in the “home with
mother or father” group, but these models produced results
that were statistically indistinguishable from the center-
based care versus other arrangement analysis, and the pol-
icy relevance of our findings was maximized by comparing
children in regulated center care with all other children in
care arrangements that did not have a federally regulated
educational curriculum.
Household income. We made use of annual tax records for
each participating mother and for the fathers who had agreed
to participate (77.6%). In cases wherein father’s income was
missing, we imputed this income using an expectation maxi-
mization algorithm, basing estimates on historical tax
records on mother’s income, assets, and debt dating back to
1993 as well as demographic information including total
family income that was self-reported during pregnancy. We
calculated a ratio of family income-to-needs, dividing total
income by the OECD poverty line for each particular year
(50% of the median income, adjusted for family size; OECD,
2011). The distribution of household income in the sample
relative to the population distribution is provided in Appen-
dix C in the online supplemental material.
For analyses, we divided families into three income
groups: low income (< 25th percentile), middle income, and
high income (>75th percentile). This approach allowed our
adjustments for selection and our estimates of the exogenous
components of ECEC use to vary across the three groups,
with the assumption that both selection forces and rate of
change in ECEC access due to policy change and municipal-
ity-specific factors likely differed for low- versus high-
income families.
Covariates. Medical Birth Registry information was
retrieved for child gender, birth weight (dichotomized: <
2,500 and ≥2,500 grams), Apgar score 5 min after birth,
multiple birth (e.g., twins), and congenital syndromes
(including Down syndrome, cleft lip and palate, and limb
malformations). Parental education, partner status (single
vs. partnered), non-Norwegian background, and number of
siblings were reported by mothers at the 17th gestational
week. Mothers reported on their anxious/depressive symp-
toms (Tambs & Moum, 1993) and partner/spouse relation-
ship satisfaction (Rosand, Slinning, Eberhard-Gran,
Roysamb, & Tambs, 2011) at 0.5, 1.5, and 3.0 years.
Dearing et al.
6
Statistical Analyses
We employed an array of analytic techniques to probe the
causal effects of ECEC scale-up in Norway. Here, as our pri-
mary models, we focus the article on results from two-stage
least squares (TSLS), generalized DID, and fixed-effect
regression analyses, techniques that can provide quasi-experi-
mental tests of causal hypotheses when correctly specified
(e.g., Angrist & Pischke, 2009; Murnane & Willett, 2011).
For both our TSLS and DID models, we exploited varia-
tions in ECEC use across cohorts and municipalities as a
proxy for variations in ECEC availability. Although we did
not have access to administrative data on actual ECEC avail-
ability in Norway, we were able to estimate availability of
ECEC within cohort by municipality clusters (i.e., each
child was nested within a cohort of children based on birth
year and municipality). Specifically, we used the proportion
of children attending ECEC within each cohort by munici-
pality cluster as an estimate of availability. To do so, we took
a jackknife approach—iteratively excluding children one by
one to compute the proportion—so that the cohort by munic-
ipality proportion in ECEC for child i excluded child i when
calculating that child’s corresponding proportion. In our
TSLS models, proportion of children in ECEC served as the
instrument; in our DID models, proportion of children in
ECEC served as the treatment. Note that this approach led to
an analytical sample of 63,350 children (from 63,471), drop-
ping 121 children who were the only children represented
within their cohorts by municipality cluster.
The general logic guiding these models was as follows.
The rollout of universal ECEC in Norway created an empiri-
cal opportunity because children born at different times in dif-
ferent municipalities had varying access to ECEC for reasons
beyond family selection. Within a short historical window,
there is little reason to expect the language in the child popula-
tion to improve across birth cohorts unless influenced by far-
reaching environmental changes in Norway, such as influential
policy shifts. Moreover, the interaction of (a) when children
were born by (b) where families lived (i.e., birth cohort by
municipality variations in ECEC scale-up) likely rendered the
progression toward universal provision of ECEC difficult for
families to forecast, and local idiosyncratic causes of ECEC
scale-up rates within each municipality helped rule out the
influences of other national policy changes or population
trends. Below, we provide details of our TSLS and DID model
specifications and key assumptions.
TSLS (instrumental variable) analyses. The first and sec-
ond stages of our TSLS models were estimated as Equation
1 and Equation 2, respectively. In the first stage (Equation
1), ECEC use was regressed on our instrument (i.e., the pro-
portion attending ECEC in the corresponding cohort by
municipality cluster for child i) and our covariate set, indi-
cated with Ws. In the second stage (Equation 2), children’s
language scores were regressed on predicted values of ECEC
from the first stage, plus covariates.
ECEC ProportionECEC
WW
ii
ir
ri i
=+
++…+ +
+
ππ
ππυ
01 1
21 1. Equation 1
Language ECEC W
W
ii
i
rrii
=+ ′+
+…++
+
ββ β
βµ
01 2
1. Equation 2
To remove omitted between-municipalities heterogene-
ity, we estimated our TSLS models with municipality fixed
effects. Our covariate set included all child and family vari-
ables in Table 1 as well as cohort by municipality averages
for these variables. To examine variations by family income,
we estimated the TSLS model separately for children in
low-, middle-, and high-income families.
For these TSLS models, we examined three key assump-
tions, one of which is related to instrument strength and two
of which are related to instrument validity: (a) the instrument
should be strongly associated with the treatment (i.e., F-test
statistic of 10 or greater), (b) the instrument(s) should be
independent of factors (other than treatment) that influence
the outcome variable, and (c) the instrument should influence
the outcome variable only through the treatment (i.e., the
exclusion restriction). Satisfying the first assumption, in all
cases, F values from our first-stage models exceeded the cri-
terion of 10 (see online supplemental material Table S4).
We also found that the study covariates were balanced
across levels of our instrument. In two-way (cohort by
municipality) fixed-effects regression models, nearly all
associations between cluster-specific levels of the covariates
and cluster-specific levels of ECEC participation were very
small and null (see online supplemental material Table S5),
with only two exceptions being evident: for children in high-
income families, ECEC participation was negatively related
to parent education and preterm births. Moreover, consistent
with the exclusion restriction, there was no evidence that our
instrument was associated with language scores once con-
trolling for ECEC use and none of our covariates provided
alternative pathways through which the instrument affected
child language (see online supplemental material Table S6).
DID analyses. We estimated generalized DID models in the
present study to examine whether rate of expansion of ECEC
availability was, in turn, predictive of rate of improvement
in children’s language skills (for similar empirical approaches
to the study of ECEC policy expansion, see Bassok, Fitzpat-
rick, & Loeb, 2014; Cascio & Schanzenbach, 2013). DID
designs include two orders of differencing—a pre-post treat-
ment difference and a comparison of the size of pre-post
treatment differences across groups with varying levels of
exposure to the treatment—and may be generalized to quasi-
experiments in which there are more than two groups (e.g.,
7
in the present study, the cohort by municipality clusters that
differ in proportion of ECEC availability) and continuous
treatment variables (e.g., in the present study, the proportion
of children in ECEC within a cluster); in essence, general-
ized DID models are two-way fixed-effects models (e.g., in
the present study, cohort by municipality). To examine vari-
ations by community income, we estimated the DID models
separately for the poorest 25% of the clusters (low income),
the middle 50% of the clusters (middle income), and the
richest 25% of the clusters (high income). Specifically, for
each income group, our DID equation took the following
form, where Language for cohort by municipality cluster cm
was regressed on proportion of children in ECEC for cluster
cm, and γc and δm were vectors of cohort and municipality
fixed effects:
Language ProportionECEC
cm cm
cmcm
=+
++ +
αβ
γδ ε
1
. Equation 3
Within-municipality regression estimates of language gap
changes. In addition to our TSLS and DID models, we esti-
mated a within-municipality fixed-effects regression that was
also specified to estimate DID. In this case, we were inter-
ested in examining associations between income-related dis-
parities in ECEC use and, in turn, income-related disparities
in early language skills. And more specifically, we examined
whether within-municipality changes in ECEC-use dispari-
ties predicted changes in language skill disparities.
Prior to estimating the fixed-effects regression model, the
first-level differencing involved computing within-munici-
pality difference scores for the 2002 and 2006 cohorts. For
both rate of ECEC use and language skills, we subtracted the
average for low-income children from the average for high-
income children. For example, for differences in average lan-
guage skills, the difference score for cohort c and municipality
m was computed as LanguageGapHighIncomeLang
cm cm
=
−LowIncomeLangcm . Next, to create a set of DID indicators,
we subtracted the language gap for the 2002 cohort from the
language gap for the 2006 cohort. That is, for municipality
m we computed ∆LanguageGapm = LanguageGap m2006
−LanguageGap m2002 . And in a similar fashion, we created
DID scores for ECEC use as well as the study covariates.
In turn, to examine whether narrowing of ECEC-use gaps
over time (i.e., smaller difference scores for the 2006 cohort
compared with the 2002 cohort) predicted narrowing lan-
guage gaps, we estimated a (within-municipality) fixed-
effects regression model as presented in Equation 4;
∆LanguageGapm was the change score indicating the extent
to which language score differences between low- and high-
income children narrowed or widened across cohorts for
municipality m, ∆ECECGapm1 was the change score
TABLE 1
Descriptive Statistics for Child Language, Early Childhood Education and Care Use, and Study Covariates (n = 63,350)
M (SD) or % Range Percentage missing
Early language skills 7.86 (1.04) 0.00–8.58 40.38
Early childhood education and care variables
Use by 18 months 45.73% 27.25
Child covariatesa
Boy 51.00% 0.00
Apgar score at 5 min 9.40 (0.78) 0.00–10.00 0.00
Preterm/low birth weight 4.30% <1.00
Multiple birth (e.g., twins) 3.51% 0.00
Congenital syndromes 5.04% 0.00
Parent and family covariates
Maternal age 29.59 (4.58) 14.00–47.00 3.59
Income-to-needs 2.16 (0.78) 0.00–10.00 0.00
Years of parent educationb15.02 (2.44) 8.00–18.00 4.45
Number of siblings 0.79 (0.90) 0.00–15.00 0.00
Single parent 4.61% 2.54
Non-Western background 10.11% 5.52
Maternal anxiety/depression 1.25 (0.30) 1.00–4.00 32.43
Relationship satisfaction 1.79 (0.72) 1.00–6.00 14.48
aInformation about all parent variables was taken from the questionnaire at the 17th gestational week for the Norwegian Mother and Child Cohort Study,
and from the 6-month interview for the Behavior Outlook Norwegian Developmental Study. bYears of parent education is for the most educated parent in
the household.
Dearing et al.
8
indicating the extent to which ECEC-use differences
between low- and high-income children narrowed or wid-
ened across cohorts for municipality m, and W1i to Wri were
municipality-level averages for the child and family covari-
ates presented in Table 1.
∆∆
LanguageGapECECGap
WW
mm
ir
ri i
=+ +
+…
++
+
ππ
ππυ
01 1
21 1. Equation 4
For this analysis, we included only the 173 municipalities
that had a reasonable representation of lower income and
higher income households (we chose 10 households in each
of these categories as the threshold, although the results
were similar if we included municipalities with fewer low-
and high-income households).
Missing data. Despite complete data on some key indicators
such as family income, there were considerable missing data
due to attrition. Most notably, 27.25% of children were miss-
ing ECEC data at 1.5 years, and 40.37% of children did not
have complete language data at 3.0 years. Although likeli-
hood of having missing values on the study covariates or lan-
guage outcome was by and large unrelated to the “treatment”
of interest, likelihood of attrition was higher for more socially
disadvantaged families and children with congenital syn-
dromes. To account for this attrition, our statistical models
were estimated using multiple imputation for missing values,
combining estimates and standard errors according to Rubin’s
Rules (Rubin, 1987). Given the sample size and complexity
of our models, we were limited to using five imputations (see
online Appendix E for further details about missing data).
Results
Descriptive Statistics
In Table 1, we provide descriptive statistics for the study
variables. In addition, in Figure 1, we plot the proportion of
children in ECEC for each birth cohort. In the figure, it is
evident that ECEC use increased across cohorts by nearly 30
percentage points or more for children from low-, middle-,
and high-income households. Correspondingly, for low-
income children, use of parental care had the most rapid
declines across cohorts, while for middle- and high-income
children, declining use was fairly similar for parental care
and family day care.
TSLS Models
In Table 2, we summarize results from our TSLS models
estimating the effects of ECEC use at 18 months on lan-
guage at 3 years. For low-income children, the estimated
effect of ECEC use was statistically significant and more
than twice as large as the estimates for middle- and high-
income children. Based on the TSLS estimate, low-income
children who were in ECEC at 18 months had language
scores at 3 years that were, on average, 89% of a standard
deviation higher than those for low-income children who did
not attend ECEC. For middle- and high-income children, on
the other hand, ECEC use was associated with approxi-
mately 31% and 29% of a standard deviation difference in
FIGURE 1. Trends in center-based early childhood education
and care and other arrangements across birth cohorts.
9
language scores, respectively; the estimate was null for
high-income children and only approached significance for
middle-income children. It is worth noting, however, that
even for the low-income children, the 95% confidence inter-
val was quite wide, including values ranging from 28% to
149% of a standard deviation; given similarly large confi-
dence intervals for the middle- and high-income groups, the
estimates for the three income groups were not statistically
distinguishable from one another.
DID Models
In Table 3, we summarize results from our DID models.
While these estimates did not significantly differ from zero
for children in middle- and high-income communities (i.e.,
children in cohort by municipality clusters that were, on aver-
age, middle and high income), increasing ECEC availability
significantly predicted improved language for children in
low-income communities. Regarding effect size, it is critical
to note that the coefficients in these models represent the esti-
mated changes in language scores (in percentages of a stan-
dard deviation) given a change of 0% to 100% “availability”
(i.e., no children within a cluster versus all children in a clus-
ter attending ECEC). Thus, given a one standard deviation
change in our ECEC availability indicator (i.e., a 15.4 per-
centage point increase), we would expect a 5.39% standard
deviation increase in the average language scores of children
within clusters. It is also important to note that confidence
intervals for the three income groups overlapped, indicating
they were not statistically distinguishable from one another.
Within-Municipality Regression Estimates of Language
Gap Changes
As a final analytical step, we estimated ordinary least
squares regression models examining whether municipali-
ties that evidenced narrowing gaps in ECEC use between
higher and lower income children (from 2002 to 2006) also
evidenced narrowing gaps in language scores (see Table 4).
We estimated unadjusted models and, in turn, models that
controlled for (a) within-municipality mean levels of the
covariates in Table 1 and (b) changes across cohorts in the
covariates. In total, 58.29% of municipalities demonstrated
narrowing gaps between the proportions of lower versus
higher income children who were in ECEC at 18 months,
and these narrowing gaps in ECEC use were positively pre-
dictive of narrowing gaps in language scores. In the two
models adjusting for covariate levels or changes, every 10
TABLE 2
Predicting Language Scores at 36 Months from Early Childhood Education and Care Use at 18 Months: Instrumental Variable Models
Low income
(n = 17,395)
Middle income
(n = 31,073)
High income
(n = 14,882)
b
[95% CI]
b
[95% CI]
b
[95% CI]
Early childhood education and
care use at 18 months
.89***
[0.28–1.49]
.31*
[–0.05–0.67]
.29
[–0.12–0.70]
Note. In the two-stage least squares models, low, middle, and high income refer to family income levels. Low-income families were at the 25th percentile or
lower, and high-income families were at the 75th percentile or higher.
*p < .10. ***p < .01.
TABLE 3
Predicting Language Scores at 36 Months from Early Childhood Education and Care Coverage at 18 Months: Difference-in-Difference
Models
Low income
(n = 495)
Middle income
(n = 1,006)
High income
(n = 489)
b
[95% CI]
b
[95% CI]
b
[95% CI]
Early childhood education and care
coverage at 18 months
.35***
[0.10–0.60]
.21
[–0.11–0.52]
−.04
[–0.44–0.36]
Note. Sample sizes for the difference-in-difference models indicate the number of cohort-by-municipality clusters of children. In these models, low, middle,
and high income refer to cluster-level averages of family income. Low-income communities were those at the 25th percentile or lower, and high-income
communities were those at the 75th percentile or higher.
***p < .01.
10
percentage point narrowing of the gap in ECEC use between
lower and higher income households was associated with a
1.56 percentage point narrowing in the language skill gap.
Sensitivity and Robustness Checks
In addition to our primary models reported here, we con-
ducted a number of analyses to examine the sensitivity of
our results to respecification and, more generally, to exam-
ine the robustness of our main findings. In Appendix F in the
online supplemental material, we provide a brief overview
of results from these sensitivity and robustness checks.
Discussion
To date, most evidence from the United States on ECEC
at scale comes from targeted programs for disadvantaged
preschool children (3- to 5-year-olds); international evi-
dence on universal scale-up beginning at younger ages pro-
vides a useful addition to the cumulative knowledge. In the
present study, we investigated the consequences of Norway’s
national scale-up of universal ECEC, beginning at age 1, for
children’s early language skills. In doing so, we gave special
attention to differential consequences for children from
low-, middle-, and high-income families. In a population-
based sample, we found that scale-up of Norway’s universal
ECEC led to improvements in children’s early language
skills, with low-income children’s evidencing this most
robustly. Our results were, by and large, consistent with the
hypothesis that attending large-scale public ECEC is improv-
ing low-income children’s language skills in Norway and
thereby may be narrowing early achievement gaps between
low- and high-income children. More specifically, our
results provided three complementary pieces of evidence on
the effects of universal ECEC scale-up in Norway.
First, in TSLS regression models, we found that attending
ECEC at 18 months was predictive of better language skills
at age 3, primarily for low-income children. On average,
low-income children attending ECEC were estimated to
have language skills approximately 90% of a standard devia-
tion higher than low-income children not in ECEC. A word
of caution is required when interpreting the size of this
effect, however, given that our 95% confidence interval for
low-income children ranged from 29% to 149% of a stan-
dard deviation. Despite this wide range, the estimates
reached statistical significance, and even the lower bound of
nearly one third of a standard deviation would be of consid-
erable practical importance given the long-term risks associ-
ated with limited early language skills (Durham et al., 2007).
Second, in DID models, we found that the increasing
availability of ECEC in Norway led to improvements in the
average language skills within low-income municipalities.
The size of these improvements were, however, consider-
ably smaller than those estimated for the effects of ECEC
use on individual children’s language scores—a 15 percent-
age point increase in the availability of ECEC predicted
slightly more than a 5% of standard deviation increase in the
average language scores within low-income municipalities.
One reason for this seemingly small effect was likely the fact
that not all children in low-income municipalities were,
themselves, living in low-income families; based on our
TSLS models, we would expect less robust effects of ECEC
on the language skills of children in middle- and high-
income families compared with those children in low-
income families. Moreover, our initial DID models did not
address for whom ECEC availability was increasing most
rapidly (e.g., was availability increasing at similar or faster
rates for low- vs. high-income children as a function of cost
structures?).
Third, therefore, we estimated regression models focused
on the question of for whom ECEC attendance increased
within municipalities. More specifically, we compared
municipalities according to disparities in ECEC attendance
between low- versus high-income children and the implica-
tions of changes over time in these disparities for language
outcomes. Municipalities in which use of ECEC increased
more rapidly over time for low-income children than for
high-income children also evidenced the greatest narrowing
of language skill gaps between these groups of children. In
2006, in the population-based sample we examined, there
remained an ECEC-use disparity of approximately 20 per-
centage points between low- and high-income children (see
Figure 1); our regression model estimates indicated that
closing this gap would narrow income gaps in early lan-
guage skills by a little more than 3 percentage points. While
these effects are modest in absolute terms, they should be
evaluated with attention to the fact that strong early language
skills are excellent predictors of long-term achievement and
well-being outcomes (e.g., Durham et al., 2007; Farkas &
TABLE 4
Municipality-Level Association Between Changes in Early
Childhood Education and Care–Use Gap and Changes in
Language Skill Gap Between Lower and Higher Income
Households
Change in Early Childhood
Education and Care–Use Gap
Change in Child
Language Gap
b (SE)
Unadjusted OLS estimates .14** (.06)
OLS adjusted for average
within-municipality
covariate levels
.16*** (.07)
OLS adjusted for within-
municipality covariate rates
of change
.16*** (.06)
Note. OLS = ordinary least squares.
**p < .05. ***p < .01.
Norway’s National Scale-Up
11
Beron, 2004). Thus, rather small changes in early language
due to universal ECEC scale-up might still prove to have
considerable implications for these children’s life chances
and for hopes of reducing social disparities in Norway.
Optimism about the potential benefits of Norway’s ECEC
program is justified by the fact that our findings are consis-
tent with other Norwegian studies. There is evidence that the
scale-up of ECEC for preschoolers in Norway in the late
1970s improved life chances into early adulthood (Havnes &
Mogstad, 2011); in addition, a recent study from Norway’s
capital, Oslo, showed that children who, due to a lottery,
entered ECEC on average 6 months earlier than their coun-
terparts (i.e., about 15 months of age rather than 19 months)
scored 12% of a standard deviation higher on math and read-
ing tests at age 6 (Drange & Havnes, 2015). Yet it is critical
to recognize that any hope of reducing social disparity via
ECEC relies on strong rates of participation in public ECEC
among disadvantaged families. While participation among
low-income children grew rapidly during the time period we
studied, only a little more than half of the municipalities in
Norway actually narrowed rates of ECEC use between low-
and high-income families. With national rates of ECEC use
for 1- to 2-year-olds now near 80%, disparities may be
decreasing but remain a concern as Norway has recently
increased its ECEC policy focus on participation rates
among socially disadvantaged families (OECD, 2015).
Extending the Cumulative Knowledge About ECEC at
Scale: Relevance in the United States and Internationally
Beyond Norway, our findings of improved early language
skills for low-income children during ECEC scale-up are
consistent with those demonstrated in randomized trials of
infant and toddler ECEC (Duncan & Sojourner, 2013) and
nonexperimental studies of high-quality infant and toddler
care, including those studies that have demonstrated larger
effects of high-quality care for children in low-income fami-
lies (Geoffroy et al., 2007; McCartney et al., 2007). Our
findings are also consistent with evaluations of universal
preschool programs at scale in the United States for which
positive achievement gains have been most pronounced
among low-income children but also evident among middle-
income children (Gormley et al., 2005; Gormley et al.,
2008). Our findings extend these existing lines of work
because we have examined such effects within a large-scale,
national implementation of quality-regulated ECEC, begin-
ning at age 1, when brain maturation, language learning, and
cognitive development are rapid.
Compared with effect sizes from previous studies, our
estimated effects were smaller, however. Some studies, in
fact, report considerably larger effect sizes. One case in
point are the projections based on analyses of low birth
weight children in the United States (Duncan & Sojourner,
2013); at age 3, estimated population gaps in IQ would be
closed by 87% assuming a 2-year universal ECEC program.
Somewhat smaller effect sizes were reported in Duncan and
Magnuson’s (2013) meta-analysis of preschool interventions
(i.e., 21% of a standard deviation), particularly for studies
dating after 1980 (i.e., 16% of a standard deviation). Also
relevant are effect sizes for nonexperimental studies of
infant and toddler care, ranging from 9% to 16% of a stan-
dard deviation for school readiness scores for each addi-
tional year children attend center care in the NICHD Study
of Early Child Care and Youth Development, for example
(NICHD, 2000).
Given that Norway’s ECEC program lasts until school
entry (the year children turn 6), it is possible that our esti-
mates at age 3 are smaller than they would be at school entry.
However, when comparing our estimated effects of early
ECEC in Norway to preschool interventions in the United
States, the supportive sociopolitical context of Norway must
be considered (Dearing & Zachrisson, 2017). Beyond paid
parental leave, Norway offers free health care (including
well-being clinics for parents and children) and a more fam-
ily-friendly labor market than does the United States (e.g., 5
weeks paid holiday). In addition, there are lower levels of
material deprivation associated with child poverty in Norway
compared with the United States (UNICEF Innocenti
Research Center, 2012). It is reasonable to assume that the
counterfactual condition to regulated ECEC for low-income
children (e.g., being cared for by parents or unqualified child
minders), is to a lesser extent than in the United States, asso-
ciated with disadvantaged developmental contexts. Thus,
the baseline for children’s language development may well
be higher; indeed, in the population-based data analyzed in
the present study, language score differences between low-
and high-income children are only about 15% of a standard
deviation, on average. In Canada, a national context some-
what more comparable to Norway than is the United States,
evidence on ECEC appears mixed: ECEC attendance for
disadvantaged children prior to age 4 predicted 36% to 87%
of a standard deviation higher scores on school achievement
in Canada (Geoffroy et al., 2007; Geoffroy et al., 2010), but
this finding is in stark contrast to evaluations finding null (or
negative) effects on behavior of ECEC policies in Quebec
(Baker et al., 2008; Kottelenberg & Lehrer, 2014).
Our results should also be interpreted in light of the
national curriculum (“Framework” plan). While structural
quality appears high, this curriculum provides very general
guidelines for pedagogical practice, which likely varies con-
siderably across centers. Relatively speaking in international
comparisons, Norwegian ECEC teachers may often empha-
size free play, minimize staff-child interactions, and avoid
formalized direct instructions (OECD, 2015), potentially
leading to low instructional quality. Considering the empha-
sis given to high-quality curriculum as a core “active ingre-
dient” in ECEC promoting cognitive development (Duncan
& Magnusen, 2013), it is reasonable to expect that a more
Dearing et al.
12
academically focused and structured curriculum could lead
to larger effect sizes for language development than we
detected.
As a final note about the relevance of our findings in the
United States and internationally, it is worth considering that
while there is some evidence of negative behavioral conse-
quences of early, extensive, and continuous ECEC (e.g.,
Belsky, 2001; Huston, Bobbit, & Bentley, 2015), this does
not appear to be the case Norway (Dearing, Zachrisson, &
Nærde, 2015; Zachrisson, Dearing, Lekhal, & Toppelberg,
2013). One reason may be that paid parental leave policy
makes 1 year the most common age of entry into ECEC in
Norway, whereas many children in the United States enter
some form of nonparental care by 9 months (Halle et al.,
2009). More generally, however, the internal validity of many
studies demonstrating negative consequences of ECEC has
been called into question (Dearing & Zachrisson, 2017).
Limitations and Strengths of the Present Study
One of the more serious limitations to the present study
was the high rates of attrition in this population-based sam-
ple, with evidence that more disadvantaged and higher risk
children were less likely to be retained through age 3. As
described here, and in our supplementary materials, our
results proved robust across approaches to handling this lim-
itation (i.e., multiple imputation and listwise deletion).
While our primary approach (multiple imputation) is recom-
mended for bias reduction with high levels of attrition that is
correlated with study covariates (e.g., Graham, 2009), it is
possible that our estimates of the effects of ECEC scale-up
do not, in fact, apply to the most seriously disadvantaged or
developmentally at-risk children in Norway.
It is also important to note that in the population-based
sample, our language measure was based on maternal
reports. There is, however, excellent evidence of the validity
of parent reports of child language at age 3 via concurrent
and predictive correlations with direct assessment tools
(e.g., Feldman et al., 2005; Thal, O’Hanlon, Clemmons, &
Fralin, 1999). Even so, the two maternal report measures
employed in the population-based study are screening tools
for language problems/delay.
Finally, it is critical to note that our instrument in the TSLS
models and treatment variable in the DID models was, in
essence, a proxy indicator. Ideally, we would have exploited
public administrative data on publicly funded ECEC avail-
ability within municipalities; however, for purposes of pro-
tecting participants’ anonymity, municipalities were not
identifiable in MoBa, and therefore we could not link public
ECEC data to MoBa. However, MoBa’s being a population-
based sample boosts our confidence in our estimates of ECEC
availability. Moreover, an important strength of the present
study was our ability to employ a variety of methods aimed at
probing the causal hypothesis and determining whether the
results of our TSLS and DID models were robust to alterna-
tive specifications.
While randomized experiments are the safest method for
ensuring internal validity, methodologists increasingly
encourage cause probing in nonexperimental work. Best
practice recommendations highlight the importance of (a)
using statistical methods and designs that help rule out omit-
ted variables bias (Duncan, Magnusson, & Ludwig, 2004;
Foster, 2010; McCartney, Bub, & Burchinal, 2006; Shadish,
Cook, & Campbell, 2002) and (b) examining robustness and
sensitivity across multiple methods that rely on somewhat
different assumptions regarding plausible alternative
hypotheses (Jo & Vinokur, 2011; Morgan & Winship, 2015;
Murnane & Willett, 2011; Shadish et al., 2002). In the pres-
ent study, we examined alternative specifications of our
instrumental variable, DID, and within-municipality regres-
sion models, as summarized in the online appendices. Our
results, by and large, proved robust across this array of alter-
native methods.
Conclusion
Even as preschool programs at scale become increasingly
common in the United States and elsewhere, debate should
continue about the relative costs and benefits of universal
versus targeted ECEC policies (e.g., do targeted public pro-
grams produce similar (or larger) gains among socially dis-
advantaged children than do universal programs?). Yet our
findings in Norway juxtaposed with the few universal pre-
school program evaluations in the United States should help
push that debate away from targeted approaches that exclude
middle-class and higher income children. Furthermore,
whereas most policy discussions in the United States are
centered on preschool children, we believe these findings
should provoke more conversation about the value of ECEC
at scale for the younger children. Although our findings are
limited to shorter term outcomes, with fade-out a legitimate
concern (Bailey, Duncan, Odgers, & Yu, 2017), the present
study increases evidence that nations can implement publi-
cally subsidized and regulated ECEC programs for very
young children at scale with a potential benefit of narrowing
achievement gaps.
Acknowledgments
This research was supported by a grant from the Research Council
of Norway. Henrik D. Zachrisson was at the Norwegian Institute of
Public Health when this work was initiated.
References
Angrist, J. D., & Pischke, J.-S. (2009). Mostly harmless econo-
metrics: An empiricist’s companion. Princeton, NJ: Princeton
University Press.
Aspland Virak. (2006). Analyse av barnehagetall pr 20.09.2006
[Analyses of ECEC numbers by Sept 20th 2006]. Retrieved
Norway’s National Scale-Up
13
from http://www.regjeringen.no/en/dep/kd/dok/rapporter_
planer/rapporter/2007/analyse-av-barnehagetall-pr-20092006.
html?id=456310
Aspland Virak. (2009). Analyse av barnehagestatistikk—status
for utbygging og ventelister pr. 20. september 2009 [Analysis
of ECEC statistics—status for expansion and waiting lists by
Sept 20th 2009]. Retrieved from http://www.regjeringen.no/
upload/KD/Vedlegg/Barnehager/Rapporter%20og%20planer/
Sluttrapport_analyse%20av%20barnehagetall%20per%20
20.09.09.pdf
Bailey, D., Duncan, G. J., Odgers, C. L., & Yu, W. (2017).
Persistence and fadeout in the impacts of child and adolescent
interventions. Journal of Research on Educational Effectiveness,
10(1), 7–39. doi: 10.1080/19345747.2016.1232459
Baker, M., Gruber, J., & Milligan, J. (2008). Universal child
care, maternal labor supply, and family well-being. Journal of
Political Economy, 116(4), 709–745. doi: 10.1086/591908
Barnett, W. S. (2010). Universal and targeted approaches to
preschool education in the United States. International
Journal of Child Care and Education Policy, 4(1), 1–12. doi:
10.1007/2288-6729-4-1-1
Barnett, W. S. (2011). Effectiveness of early educational interven-
tion. Science, 333, 975–978. doi: 10.1126/science.1204534
Bassok, D., Fitzpatrick, M., & Loeb, S. (2014). Does state preschool
crowd-out private provision? The impact of universal preschool
on the childcare sector in Oklahoma and Georgia. Journal of
Urban Economics, 83, 18–33. doi: 10.1016/j.jue.2014.07.001
Belsky, J. (2001). Emanuel Miller Lecture—Developmental
risks (still) associated with early child care. Journal of Child
Psychology and Psychiatry and Allied Disciplines, 42, 845–859.
Bjørnestad, E., & Os, E. (in press). Quality in Norwegian child-
care for toddlers using ITERS-R. European Early Childhood
Education Research Journal.
Brenna, L. R., Bjerkestrand, M., Broström, S., Fagerli, B., Hernes,
I., Hornslien, Ø., … Tørresdal, B. (2010). Med forskertrang
og lekelyst. Systematisk pedagogisk tilbud til alle førskole-
barn (NOU 2010:8) [With urge for research and playful-
ness. Systematical educational programs for all preschool
children (Research report 2010:8)]. Retrieved from http://
www.regjeringen.no/nb/dep/kd/dok/nouer/2010/nou-2010-8.
html?id=616123
Camilli, G., Vargas, S., Ryan, S., & Barnett, W. S. (2010). Meta-
analysis of the effects of early education interventions on cogni-
tive and social development. Teachers College Record, 112(3),
579–620.
Cascio, E. U., & Schanzenbach, D. W. (2013). The impacts of
expanding access to high-quality preschool education (No.
w19735). Cambridge, MA: National Bureau of Economic
Research.
Ceci, S. J., & Papierno, P. B. (2005). The rhetoric and reality of gap
closing: When the “have-nots” gain but the “haves” gain even
more. American Psychologist, 60, 149–160. doi: 10.1037/0003-
066X.60.2.149
Cote, S. M., Doyle, O., Petitclerc, A., & Timmins, L. (2013). Child
care in infancy and cognitive performance until middle child-
hood in the Millennium Cohort Study. Child Development,
84(4), 1191–1208. doi: 10.1111/cdev.12049
Dale, P. S., Price, T. S., Bishop, D. V., & Plomin, R. (2003).
Outcomes of early language delay: I. Predicting persistent
and transient language difficulties at 3 and 4 years. Journal of
Speech, Language, and Hearing Research, 46, 544–560. doi:
10.1044/1092-4388(2003/044)
Dearing, E. (2014). The state of research on children and fami-
lies in poverty: Past, present, and future empirical avenues of
promise. In K. McCartney, H. Yoshikawa, & L. Forcier (Eds.),
Improving the odds for America’s children (pp. 203–216).
Cambridge, MA: Harvard Education Press.
Dearing, E., McCartney, K., & Taylor, B. A. (2009). Does higher
quality early child care promote low-income children’s math and
reading achievement in middle childhood? Child Development,
80(5), 1329–1349. doi: 10.1111/j.1467-8624.2009.01336.x
Dearing, E., & Zachrisson, H. D. (2017). Concern over internal,
external, and incidence validity in studies of child-care quan-
tity and externalizing behavior problems. Child Development
Perspectives, online first. doi: 10.1111/cdep.12224
Dearing, E., Zachrisson, H. D., & Nærde, A. (2015). Age of
entry into early childhood education and care as a predictor of
aggression faint and fading associations for young Norwegian
children. Psychological Science, 26(10), 1595–1607. doi:
10.1177/0956797615595011
Dionne, G., Dale, P. S., Boivin, M., & Plomin, R. (2003). Genetic
evidence for bidirectional effects of early lexical and gram-
matical development. Child Development, 74, 394–412. doi:
10.1111/1467-8624.7402005
Drange, N., & Havnes, T. (2015). Child care before age two and
the development of language and numeracy: Evidence from a
lottery. IZA Discussion Paper No. 8904. Bonn, Germany: IZA
Institute of Labor Economics.
Duncan, G. J., & Magnuson, K. (2013). Investing in preschool pro-
grams. Journal of Economic Perspectives, 27(2), 109–132. doi:
10.1257/jep.27.2.109
Duncan, G. J., Magnuson, K. A., & Ludwig, J. (2004). The endo-
geneity problem in developmental studies. Research in Human
Development, 1, 59–80. doi: 10.1207/s15427617rhd0101&2_5
Duncan, G. J., & Sojourner, A. J. (2013). Can intensive early child-
hood intervention programs eliminate income-based cognitive
and achievement gaps? Journal of Human Resources, 48(4),
945–968. doi: 10.1353/jhr.2013.0025
Durham, R. E., Farkas, G., Hammer, C. S., Tomblin, J. B., & Catts,
H. W. (2007). Kindergarten oral language skill: A key variable
in the intergenerational transmission of socioeconomic status.
Research in Social Stratification and Mobility, 28, 294–305.
doi: 10.1016/j.rssm.2007.03.001
European Commission. (2011). Communication from the commis-
sion: Early childhood education and care: Providing all our
children with the best start for the world tomorrow. Retrieved
from http://eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri=
COM:2011:0066:FIN:EN:PDF
Farkas, G., & Beron, K. (2004). The detailed age trajectory of
oral vocabulary knowledge: Differences by class and race.
Social Science Research, 33(3), 464–497. doi: 10.1016/j.ssre-
search.2003.08.001
Farran, D. C. (2016). We need more evidence in order to cre-
ate effective pre-K programs. Evidence Speaks Reports, 1.
Washington, DC: Brookings Institution.
Feldman, H. M., Dale, P. S., Campbell, T. F., Colborn, D. K., Kurs-
Lasky, M., Rockette, H. E., & Paradise, J. L. (2005). Concurrent
and predictive validity of parent reports of child language at
Dearing et al.
14
ages 2 and 3 years. Child Development, 76(4), 856–868. doi:
10.1111/j.1467-8624.2005.00882.x
Foster, E. M. (2010). Causal inference and developmental psychol-
ogy. Developmental Psychology, 46(6), 1454–1480.
Fox, N. A., & Rutter, M. (2010). Introduction to the special section on
the effects of early experience on development. Child Development,
81(1), 23–27. doi: 10.1111/j.1467-8624.2009.01379.x
Geoffroy, M. C., Cote, S. M., Borge, A. I., Larouche, F., Seguin, J.
R., & Rutter, M. (2007). Association between nonmaternal care
in the first year of life and children’s receptive language skills
prior to school entry: The moderating role of socioeconomic
status. Journal of Child Psychology and Psychiatry, 48(5), 490–
497. doi: 10.1111/j.1469-7610.2006.01704.x
Geoffroy, M. C., Cote, S. M., Giguere, C. E., Dionne, G., Zelazo,
P. D., Tremblay, R. E., … Seguin, J. R. (2010). Closing the
gap in academic readiness and achievement: The role of early
childcare. Journal of Child Psychology & Psychiatry, 51(12),
1359–1367. doi: 10.1111/j.1469-7610.2010.02316.x
Gormley, W. T., Phillips, D., & Gayer, T. (2008). Preschool pro-
grams can boost school readiness. Science, 320, 1723–1724.
doi: 10.1126/science.1156019
Gormley, W. T., Jr., Gayer, T., Phillips, D., & Dawson, B. (2005).
The effects of universal pre-K on cognitive development.
Developmental Psychology, 41, 872–884. doi: 10.1037/0012-
1649.41.6.872
Graham, J. W. (2009). Missing data analysis: Making it work in
the real world. Annual Review of Psychology, 60, 549–576. doi:
10.1146/annurev.psych.58.110405.085530
Hackman, D. A., Farah, M. J., & Meaney, M. J. (2010).
Socioeconomic status and the brain: Mechanistic insights from
human and animal research. Nature Reviews Neuroscience, 11,
651–659. doi: 10.1038/nrn2897
Halle, T., Hair, M. N., Weinstein, D., Vick, J., Forry, N., &
Kinukawa, A. (2009). Primary child care arrangements of US
infants: Patterns of utilization by poverty status, family struc-
ture, maternal work status, maternal work schedule, and child
care assistance. Washington, DC: Child Trends and Office of
Planning Research and Evaluation.
Harms, T., Cryer, D., & Clifford, R. M. (2006). Infant/Toddler
Environment Rating Scales (rev. ed.). New York, NY: Teachers
College Press.
Havnes, T., & Mogstad, M. (2011). No Child Left Behind:
Subsidized child care and children’s long-run outcomes.
American Economic Journal: Economic Policy, 3(2), 97–129.
doi: 10.1257/pol.3.2.97
Hoff, E. (2006). How social contexts support and shape language
development. Developmental Review, 26(1), 55–88. doi:
10.1016/j.dr.2005.11.002
Huston, A. C., Bobbit, K. C., & Bentley, A. (2015). Time spent
in child care: How and why does it affect social develop-
ment? Developmental Psychology, 51, 621–634. doi: 10.1037/
a0038951
Janson, H., & Squires, J. (2004). Parent-completed developmen-
tal screening in a Norwegian population sample: A comparison
with US normative data. Acta Paediatrica, 93, 1525–1529. doi:
10.1111/j.1651-2227.2004.tb02641.x
Japel, C., Tremblay, R. E., & Cote, S. (2005). “Quality counts:
Assessing the quality of daycare services based on the Quebec
Longitudinal Study of Child Development.” Choices, 11,
1–42.
Jo, B., & Vinokur, A. D. (2011). Sensitivity analysis and bound-
ing of causal effects with alternative identifying assumptions.
Journal of Educational and Behavioral Statistics, 36, 415–440.
doi: 10.3102/1076998610383985
Kottelenberg, M. J., & Lehrer, S. F. (2014). Do the perils of uni-
versal childcare depend on the child’s age? CESifo Economic
Studies, 60, 338–365. doi: 10.1093/cesifo/ifu006
Kuhl, P. K. (2004). Early language acquisition: Cracking the
speech code. Nature Reviews Neuroscience, 5, 831–843. doi:
10.1038/nrn1533
Loeb, S., Bridges, M., Bassok, D., Fuller, B., & Rumberger, R.
W. (2007). How much is too much? The influence of pre-
school centers on children’s social and cognitive development.
Economics of Education Review, 26, 52–66. doi: 10.1016/j.
econedurev.2005.11.005
Magnus, P., Irgens, L. M., Haug, K., Nystad, W., Skjaerven, R., &
Stoltenberg, C. (2006). Cohort profile: The Norwegian Mother
and Child Cohort Study (MoBa). International Journal of
Epidemiology, 35, 1146–1150. doi: 10.1093/ije/dyl170
Magnuson, K., & Shager, H. (2010). Early education: Progress and
promise for children from low-income families. Children and
Youth Services Review, 32, 1186–1198. doi: 10.1016/j.childy-
outh.2010.03.006
McCartney, K., Bub, K. L., & Burchinal, M. (2006). Selection,
reflection, detection. Monographs of the Society for Research
in Child Development, 71(Serial No. 285), 105–126. doi:
10.1111/j.1540-5834.2006.00407.x
McCartney, K., Dearing, E., Taylor, B. A., & Bub, K. L.
(2007). Quality child care supports the achievement of low-
income children: Direct and indirect pathways through
caregiving and the home environment. Journal of Applied
Developmental Psychology, 28(5/6), 411–426. doi: 10.1016/j.
appdev.2007.06.010
Ministry of Children and Equality. (2014). Family benefits.
Retrieved from http://www.regjeringen.no/en/dep/bld/topics/
family-policies/family-policy.html?regj_oss=1&id=670514
Ministry of Education. (2006). Rammeplan for barnehagens
innhold og oppgaver [Framework plan for content and duties
in kindergarten]. Retrieved from http://www.regjeringen.no/
nb/dep/kd/dok/lover_regler/reglement/2006/rammeplan-for-
barnehagens-innhold-og-opp.html?id=278626
Ministry of Education (2014). Early childhood education and care
policy. Retrieved from http://www.regjeringen.no/en/dep/kd/
Selected-topics/kindergarden/early-childhood-education-and-
care-polic.html?id=491283
Morgan, S. L., & Winship, C. (2015). Counterfactuals and causal
inference: Methods and principles for social research (2nd ed.).
New York, NY: Cambridge University Press.
Murnane, R. J., & Willett, J. B. (2011). Methods matter: Improving
causal inference in educational and social science research.
New York, NY: Oxford University Press.
Mustard, J. (2006). Experience-based brain development: Scientific
underpinnings of the importance of early child development in a
global world. Paediatrics and Child Health, 11, 571–572.
Nelson, C. A., & Sheridan, M. A. (2011). Lessons from neurosci-
ence research for understanding causal links between family
Norway’s National Scale-Up
15
and neighborhood characteristics and educational outcomes. In
G. J. Duncan, & R. J. Willett (Eds.), Whither opportunity (pp.
27–46). New York, NY: Russell Sage.
National Institute of Child Health and Human Development
Early Child Care Research Network. (2000). The relation
of child care to cognitive and language development. Child
Development, 71, 960-980. Retrieved from http://www.jstor.
org/stable/1132337
Obama, B. (2014). President Barack Obama’s state of the union
address 2014. Retrieved from http://www.whitehouse.gov/the-
press-office/2014/01/28/president-barack-obamas-state-union-
address
Organisation for Economic Co-operation and Development.
(2006). Starting strong II: Early childhood education and care
policy. Paris: Author.
Organisation for Economic Co-operation and Development.
(2011). Poverty. Society at a glance 2011: OECD social indica-
tors. Paris: Author.
Organisation for Economic Co-operation and Development.
(2015). Norway—Early childhood education and care policy
review. Retrieved from http://www.oecd.org/norway/Early-
Childhood-Education-and-Care-Policy-Review-Norway.pdf
Pruden, S. M., Hirsh-Pasek, K., Golinkoff, R. M., & Hennon, E.
A. (2006). The birth of words: Ten-month-olds learn words
through perceptual salience. Child Development, 77(2), 266–
280. doi: 10.1111/j.1467-8624.2006.00869.x
Rindfuss, R. R., Guilkey, D., Morgan, S. P., Kravdal, O., &
Guzzo, K. B. (2007). Child care availability and first-birth
timing in Norway. Demography, 44, 345–372. doi: 10.1353/
dem.2007.0017
Rosand, G. M., Slinning, K., Eberhard-Gran, M., Roysamb, E., &
Tambs, K. (2011). Partner relationship satisfaction and mater-
nal emotional distress in early pregnancy. BMC Public Health,
11, 161. doi: 1471-2458-11-161
Rubin, D. B. (1987). Multiple imputation for nonresponse in sur-
veys. New York, NY: Wiley.
Ruzek, E., Burchinal, M., Farkas, G., & Duncan, G. J. (2014). The
quality of toddler child care and cognitive skills at 24 months:
Propensity score analysis results from the ECLS-B. Early
Childhood Research Quarterly, 28, 12–21. doi: 10.1016/j.
ecresq.2013.09.002
Sæther, J. P. (2010). Barn i barnehage—foreldrebakgrunn og
utvikling de seneste årene [Children in center care—Parental
background and recent developments]. Oslo: Statistics
Norway.
Sameroff, A., & Chandler, M. (1975). Reproductive risk and the
continuum of caretaking casualty. In F. D. Horowitz (Ed.),
Review of child development research (Vol. 4). Chicago, IL:
University of Chicago Press.
Scheistrøen, J. (2012). Undersøking om foreldrebetaling i barne-
hagar, januar 2012 [Investigation of parent fees in child care
centers, January 2012]. Retrieved from http://www.ssb.no/
utdanning/artikler-og-publikasjoner/undersoking-om-foreldre-
betaling-i-barnehagar-januar-2012
Shadish, W. R., Cook, T. D., & Campbell, D. T. (2002).
Experimental and quasi-experimental designs for generalized
causal inference. Boston, MA: Houghton, Mifflin.
Shonkoff, J. P., & Phillips, D. A. (Eds.). (2000). From neurons
to neighborhoods. The science of early childhood development.
Washington, DC: National Academy Press.
Snow, C. E., Burns, M. S., & Griffin, P. (Eds.). (1998). Preventing
reading difficulties in young children. Washington, DC:
National Academy Press.
Squires, J., Bricker, D. D., & Potter, L. (1999). The ASQ user’s
guide (2nd ed.). Baltimore, MD: Brookes.
Tambs, K., & Moum, T. (1993). How well can a few question-
naire items indicate anxiety and depression? Acta Psychiatrica
Scandinavia, 87(5), 364–367.
Thal, D. J., O’Hanlon, L., Clemmons, M., & Fralin, L. (1999).
Validity of a parent report measure of vocabulary and syntax
for preschool children with language impairment. Journal of
Speech, Language, and Hearing Research, 42(2), 482–496.
UNICEF Innocenti Research Center. (2008). The child care tran-
sition: A league table of early childhood education and care
in economically advanced countries. Innocenti Report Card 8.
Florence, Italy: Author.
UNICEF Innocenti Research Center. (2012). Measuring child pov-
erty: New league tables of child poverty in the world’s richest
countries (Report Card 10). Florence, Italy: Author.
van Huizen, T., & Platenga, J. (2015). Universal child care and
children’s outcomes: A meta-analysis of evidence from natu-
ral experiments. Discussion Paper Series 15-13. Utrecht,
the Netherlands: Utrecht School of Economics, Tjalling C.
Koopmans Research Institute.
Winsvold, A., & Guldbrandsen, L. (2009). Kvalitet og kvantitet—
kvalitet i en barnehagesektor i sterk vekst [Quality and quan-
tity—Quality in a child-care sector in rapid growth]. Oslo,
Norway: NOVA.
Yoshikawa, H., Aber, J. L., Bergman, L. R., & Beardslee, W. R.
(2012). The effects of poverty on the mental, emotional, and
behavioral health of children and youth. American Psychologist,
67(4), 272–284. doi: 10.1037/a0028015
Yoshikawa, H., Weiland, C., Brooks-Gunn, J., Burchinal, M. R.,
Espinosa, L. M., Gormley, W. T., … Zaslow, M. J. (2013). Investing
in our future: The evidence base on preschool education. Ann
Arbor, MI: Society for Research in Child Development. Retrieved
from http://fcd-us.org/resources/evidence-base-preschool
Zachrisson, H. D., Dearing, E., Lekhal, R., & Toppelberg, C. O.
(2013). Little evidence that time in child care causes exter-
nalizing problems during early childhood in Norway. Child
Development, 84(4), 1152–1170. doi: 10.1111/cdev.12040
Authors
ERIC DEARING is a professor of applied developmental psychol-
ogy in the Lynch School of Education at Boston College and a senior
researcher at the Norwegian Center for Child Behavioral
Development. His research is focused on the role of children’s lives
outside of school for their success in school, with a special interest in
the ways family, early education and care, and neighborhood condi-
tions affect children’s achievement and psychological well-being.
HENRIK DAAE ZACHRISSON is senior researcher at the
Norwegian Center for Child Behavioral Development and a profes-
sor at the Center for Educational Measurement at the University of
Dearing et al.
16
Oslo. His research is on consequences of child care and social
inequality for children’s development.
ARNSTEIN MYKLETUN is a professor in the Department of
Community Medicine at the University of Tromsø, a senior
researcher in the Department of Mental Health and Suicide at the
Norwegian Institute of Public Health, and a researcher at the Centre
for Research and Education in Forensic Psychiatry and Psychology
at the Haukeland University Hospital, Bergen, Norway. He is also
head of research at the Center for Work and Mental Health, Nordland
Hospital Trust, Bodø, Norway. His research is in the areas of epide-
miology, public health, psychiatry, and work medicine.
CLAUDIO O. TOPPELBERG is a research scientist at Judge
Baker Children’s Center and an assistant professor at Harvard
Medical School. Dr. Toppelberg’s research in child/adolescent
psychopathology has two foci: (a) the relations of language, neuro-
cognitive, and emotional/behavioral development and (b) the
development of immigrant and dual-language children and national
childhood policies that affect both.
