EDUCATIONFORUM

THE EARLY YEARS

Preschool Influences on The advantages of home learning environment

and preschool are apparent years later in
Mathematics Achievement children’s math achievement.

Edward C. Melhuish,1* Kathy Sylva,2 Pam Sammons,3 Iram Siraj-Blatchford,4 Brenda Taggart,4
Mai B. Phan,5 Antero Malin1

U
niversal preschool is being consid- Preschool centers (141) were randomly school). Multilevel models are a standard
ered as a policy option in many parts chosen in six areas demographically similar form of regression analysis, particularly
of the world, but the most influential to England overall. The preschool stage of suited to data exhibiting a hierarchical struc-
evidence relates to disadvantaged groups. the study involved children from nursery ture (12), and they provide a method of ana-
Preschool improves disadvantaged child- classes, playgroups, private day nurseries, lyzing mathematics achievement at age 5,
ren’s school readiness, educational with 30 child, family, area, and

Downloaded from www.sciencemag.org on November 13, 2008

achievements, and social adjust- 0.5 preschool variables as covariates
ment (1). It is not the only influ- (11). Residual effects associated
ence—parental support also bene- with individual preschools after
fits children’s development, par- 0.4 these variables were accounted
ticularly if combined with center- for provided a measure of a pre-
based programs (2). Additionally, school’s effectiveness in promot-
the longer-term effects of pre- ing numeracy. Preschools where
school for disadvantaged children 0.3 children performed better than
Effect size

are mediated by the schools sub- expected on the basis of prior

sequently attended (3, 4). attainment and background were
Studies with disadvantaged 0.2 deemed more effective; pre-
children may have little rele- schools where children perfor-
vance for the general popula- med worse than expected were
tion. Nonetheless, such evi- 0.1 deemed less effective (11).
dence has fueled an increasing Children in English state pri-
interest in the universal provi- mary schools take national assess-
sion of preschool education (pre- ments at 7 and 11 years. Analyses
0.0
kindergarten) as a means of of data from 540,000 pupils at-
ht

ion

ion

ss
t
en
om le

tu

es
om

ne
ig

advancing children’s school tending 15,000 schools produced

on ma

sta

en
at

at

m
we

ve
nc
uc

uc

on

tiv
fe

ic

cti
yi

readiness and later attainment school effectiveness measures

rth

ed

ed

vir

c
cio vs.

ffe

ffe
il
en
Bi

m
’s

’s
So le (

le

le

(5). Some argue that preschool standardized for all English state
er

er

Fa

g
ec

oo

oo
th

th
Ma

in
Mo

Fa

experience is critical for chil- primary schools. We controlled

rn

ch

ch
ea

es

ys
Pr
el

dren’s future competence, cop- for prior ability, eligibility for free
ar
m

im
Ho

ing skills, health, and later school meals (poverty marker),

Pr

employment (6). Furthermore, Highest effects for predictors of mathematics attainment at age 10. The effect gender, age, ethnicity, English as
it is argued that the benefits out- sizes are in standard deviation units to facilitate comparison between predictors. second language, school composi-
weigh the costs (7). tion, and area characteristics.
England has high levels of preschool use centers run by a local authority, nursery School-level residuals from multilevel models
from age 3 onward, which produces benefits schools, and integrated children’s centers of age 11 mathematics attainment provided
over no preschool in the early school years and, thus, included all types of preschool measures of school effectiveness in promot-
(8). Our study concerns longer-term effects in centers in England at the time of the study. ing mathematics. Schools’ effectiveness var-
the general population in England. This study Children’s cognitive ability at ages 3 to 4 and ied by child ability (fig. S1). Therefore, school
considered the influence of home environ- 5 and mathematical attainment at age 10 effectiveness was derived separately for chil-
ment on children’s development [often a were assessed and family data obtained by dren of below-average, average, and above-
stronger factor than socio-demographic char- interview (11). The parental interview when average ability. Effectiveness measures for
acteristics (9)], and preschool and school children were age 3 to 4 covered learning three successive years were averaged and then
effectiveness (10). activities enabling the creation of a home matched to children by ability in this longitu-
learning environment (HLE) index (11). The dinal study (11).
1Institute for the Study of Children, Families and Social typical child attended preschool for 18 Children (n = 2558) were clustered by
Issues, Birkbeck, University of London, 7 Bedford Square, months part-time, and primary school for preschool (n = 141) and school (n = 960).
2
London, WC1B 3RA, UK. University of Oxford, Oxford OX2
6PY, UK. 3University of Nottingham, Nottingham, NG8 more than 5 years full-time by age 10. Statistically significant variables were kept
1BB, UK. 4Institute of Education, University of London, Children’s numeracy at the start of pri- in the model. Then, measures of the home
WC1H 0AL, London, UK. 5University of Kent, Canterbury, mary school (age 5) was analyzed, with con- learning environment, preschool effective-
Kent, CT2 7NZ, UK. trols for background influences and prior ness, and primary school effectiveness
*Author for correspondence. E-mail: [email protected] attainment at age 3 to 4 years (start of pre- were added.

www.sciencemag.org SCIENCE VOL 321 29 AUGUST 2008 1161

Published by AAAS
 EDUCATIONFORUM

Findings socioeconomic status or education (r = 0.28 to societies and is increasingly sought by others.
Variables in the data set typically had less than 0.32) and showed independent effects slightly The HLE before school exerted a powerful
5% missing data. Multiple imputation was less than mother’s education but greater than effect. Although any preschool has benefits
used to produce estimates for missing data father’s education and family income. This (18), effective or higher-quality preschools
and to avoid any possible bias. Results for indicates that what parents do is as important have the greatest effect. Preschool learning
complete and imputed data were equivalent, as who parents are. environments can be improved through pro-
and imputed data results are reported. The Previous work with this sample had shown grams that target cognitive functioning (23)
HLE, preschool effectiveness, and primary that the effect of 1 year of part-time preschool and staff training (24).
school effectiveness all showed significant was equivalent to increasing family income by
effects on children’s mathematics achievement more than £10,000 (U.S. $19,000) a year (8). References and Notes
1. L. A. Karoly, M. R. Kilburn, J. S. Cannon, Early Childhood
at age 10 (P < 0.001). Total variance accounted We show that the effect of primary school was Interventions: Proven Results, Future Promises (RAND,
for was 22%. Effect sizes (ESs) are from the even more important than preschool (0.39 Santa Monica, CA, 2005).
final model, after we allowed for all other vari- versus 0.26 SD), but both were sufficiently 2. J. M. Love et al., Dev. Psychol. 41, 885 (2005).
ables (supporting online material text). large to be important for any government 3. J. Currie, D. Thomas, J. Hum. Resour. 35, 755 (2000).
4. A. J. Reynolds, S. Ou, J. D. Topitzes, Child Dev. 75, 1299
Low birth weight; girls (versus boys); and wishing to maximize educational achieve- (2004).
lower parental occupational, educational, or ment. They are greater than the effect for 5. E. Zigler, W. Gilliam, S. Jones, A Vision for Universal

Downloaded from www.sciencemag.org on November 13, 2008

income status were significantly and inde- father’s education and similar to that for fam- Preschool Education (Cambridge Univ. Press, New York,
2006).
pendently linked with lower mathematics ily income but less than that for mother’s edu- 6. M. McCain, J. F. Mustard, Early Years Study: Reversing the
scores (table S2). Ethnic group differences cation (see figure, page 1161). Analyses for Real Brain Drain (Publications Ontario, Toronto, 1999).
were regarded as unreliable owing to small low and higher income groups reveal that the 7. J. J. Heckman, Science 312, 1900 (2006).
8. K. Sylva, E. Melhuish, P. Sammons, I. Siraj-Blatchford, B.
group size. Cognitive ability of the average effects for the HLE and preschool and school
Taggart, “The Effective Provision of Pre-School Education
preschool child and the percent of children in effectiveness are remarkably similar for both (EPPE) Project: The final report” [Tech. Paper 12,
the primary school with special educational income groups, which indicates their impor- Department for Education and Skills (DfES), London,
needs (SEN) had weak, significant relations tance across the income spectrum. These 2004].
9. E. C. Melhuish et al., J. Soc. Issues 64, 95 (2008).
with mathematics achievement at age 10. effects are predictive, but we cannot assume 10. P. Sammons, School Effectiveness: Coming of Age in the
Other factors were not statistically significant causality. Observational studies, such as this 21st Century (Swets & Zeitlinger, Lisse, Netherlands,
after allowing for the above. After controlling study, do not have random assignment, so it is 1999).
11. Materials and methods and additional discussion are
for other child, parental, preschool, and school always possible that results may reflect selec- available as supporting material on Science Online.
variables, the HLE, preschool effectiveness, tion bias and/or the operation of unmeasured 12. H. Goldstein, Multilevel Statistical Models (Arnold,
and primary school effectiveness all showed variables (11). London, ed.3, 2003).
separate significant effects on mathematics Countries vary in preschool provision. 13. L. Feinstein, Economica 70, 73 (2003).
14. A. Sacker, I. Schoon, M. Bartley, Soc. Sci. Med. 55, 863
achievement at age 10 (P < 0.001). Some deliver preschool services universally (2002)
We examined the effects on mathematical (United Kingdom, Scandinavia, and France), 15. E. C. Melhuish, K. Petrogiannis, Eds., Early Childhood
attainment at age 10 of having high (1 SD or whereas other countries provide services to Care and Education: International Perspectives on Policy
and Research (Routledge, London, 2006).
more above mean), low (1 SD or more below some children only (United States) and some 16. Learning for Tomorrow’s World: First Results from PISA
mean), and medium (within 1 SD of the mean) are moving rapidly to increase provision 2003 (Organisation for Economic Cooperation and
scores in HLE and preschool and school (China) (15). However, there is international Development, Paris, 2004).
17. S. Loeb, M. Bridges, D. Bassok, B. Fuller, R. W.
effectiveness. The HLE had significant, posi- support for our findings. The PISA project
Rumberger, Econ. Educ. Rev. 26, 52 (2007).
tive effects at both high and medium levels, indicates that enhanced mathematics achieve- 18. P. Sammons et al., “Measuring the impact of pre-school
compared with low (ES = 0.40 and 0.21, ment is associated with preschool experience on children’s cognitive progress over the pre-school
respectively). Preschool effectiveness was internationally (16). In the United States, period” (Tech. paper 8a, Institute of Education,
University of London, London, 2002).
significant only for high compared with low prekindergarten improved mathematics and 19. F. E. Aboud, Early Child. Res. Q. 21, 46 (2006).
(ES = 0.26), whereas primary school effec- reading at kindergarten (17), with greatest 20. J. E. Montie, Z. Xiang, L. J. Schweinhart, Early Child. Res.
tiveness had significant effects for both high gains if preschool started between 2 and 3 Q. 21, 313 (2006).
21. S. Berlinski, S. Galiani, M. Manacorda, Giving Children a
and medium levels (ES = 0.33, and 0.39, years as found in England (18). Preschool Better Start: Pre-School Attendance and School Age
respectively) compared with low levels. boosted primary school achievement in Profiles (Institute for Fiscal Studies, London, 2007);
The sample was divided into families with Bangladesh (19), with similar results re- www.ifs.org.uk/wps/wp0618.pdf.
low annual incomes [<£17,000 (or ~U.S. ported for 10 countries (20). During pre- 22. S. Berlinski, S. Galiani, P. Gertler, The Effect of Pre-
Primary Education on Primary School Performance
$32,000), 52.5%] or higher incomes (>£17,000, school expansion in Uruguay, comparisons of (Institute for Fiscal Studies, London, 2006);
47.5%). The final multilevel model was run (i) siblings with and without preschool and www.ifs.org.uk/wps/wp0604.pdf.
separately for each income group. Results (ii) regional variations revealed clear pre- 23. A. Diamond, W. S. Barnett, J. Thomas, S. Munro, Science
318, 1387 (2007).
were similar for the two groups, which indi- school benefits in secondary school (21). 24. K. Sylva et al., Int. J. Early Years Educ. 15, 49 (2007).
cated that the effects apply across the income Similar Argentine data revealed that 1 year 25. We thank A. Leyland for technical advice. Supported
spectrum with minor differences (table S2). of preschool was associated with primary by the U.K. Department for Children, Schools and
Families. Our data will be publicly available as an
school attainment increases of 0.23 SD (22),
anonymized data set subject to government approval.
Discussion analogous to the effect of high versus low Further information is available from the authors
The effects observed for background vari- effective preschools reported here. on request.
ables were similar to other studies (13, 14). Our study demonstrates the relative mag- Supporting Online Material
However, HLE effects were substantial and nitude of home, preschool, and school effects www.sciencemag.org/cgi/content/full/321/5893/1161/DC1
occurred across the whole population. The likely to occur with universal preschool edu-
HLE had low correlations with parents’ cation, which is common in many advanced 10.1126/science.1158808

1162 29 AUGUST 2008 VOL 321 SCIENCE www.sciencemag.org

Published by AAAS