Does early education influence Key Stage 1 attainment? Evidence for England from the millennium cohort study.

George, Anitha ; Stokes, Lucy ; Wilkinson, David 等

There is a body of evidence that shows that early education improves cognitive and social development for children while they are still attending, but the longer-term impacts depend on the quality of early education. Much of this evidence in England relates to a period when attendance rates at early education were around 60 per cent. Since then, early education has expanded through the guarantee of free provision for three- and four-year-olds, such that attendance at early education is now almost universal. This paper uses data from the Millennium Cohort Study to consider whether, in an era of near universal provision, early education is still associated with detectable improvements in outcomes for children. The analysis focuses on attainment in Key Stage 1 assessments when children were aged seven and finds that the overall impact of early education on Key Stage 1 attainment is modest, but that the impact is generally greater for those children who experienced poverty when they entered early education.

Keywords: Early education; child attainment, Key Stage 1, Millennium Cohort Study

JEL Classification: J13

Introduction

The longer-term impact of early education has attracted considerable attention. Notable studies include the Effective Pre-school, Primary and Secondary Education (EPPSE) project which showed early education in England to be a key determinant of cognitive and social/ behavioural development for children through to the end of Key Stage 1 (KS1) at age seven (Sylva et al., 2004). These impacts were sustained beyond age seven such that children in years tive and six of primary schooling, at ages ten and eleven, also had better outcomes (Sammons et al., 2008; Sylva et al., 2008). However, these longer-term impacts depend on the quality and effectiveness of the early years' provision. Quality here is measured using the Early Childhood Environment Rating Scales, ECERS-R (Harms et al., 1998) and ECERS-E (Sylva et al., 2006). At age fourteen, pre-school quality continued to predict children's academic attainment in maths and science as well as social/behavioural outcomes (Sylva et al., 2012).

In the US, the National Institute of Child Heahh and Human Development (NICHD) Study of Early Child Care also found a positive relationship between high quality childcare and cognitive and language skills (NICHD, 2000). Another major US study, Magnuson et al. (2007), also found a positive effect of prekindergarten on children's readiness for school, in terms of cognitive skills, but negative effects on children's behaviour. Attending prekindergarten located in the same public school as kindergarten however, did not result in negative consequences for behaviour, while the positive cognitive benefits were still observed. Correspondingly, Belsky et al. (2007), using the NICHD, found children who had experienced higher quality care obtained higher vocabulary scores at fifth grade and those spending longer hours in childcare had more problematic behaviours reported by their teachers.

Policy developments in England have acknowledged the benefits of early education with all four-year-olds entitled to a free early education place from 1998 onwards, with the free entitlement extended to all three-year-olds in 2004. By legislation, local authorities are currently required to make available sufficient free early education places offering 570 hours a year over no fewer than 38 weeks of the year. Every child in England is eligible for this free early education following their third birthday from either the start of the subsequent term, based on a three-term school year, or on specified dates (1) again depending on when their birthday occurs. This continues until they reach compulsory school age, the beginning of the terra following their fifth birthday.

Attendance is not compulsory, but take-up of the free entitlement is high. In January 2012, 98 per cent of the four-year-old population and 93 per cent of the three-year-old population benefited from some free early education (Department for Education, 2012). Most of these children accessed the full free entitlement such that when these figures are adjusted for part-time take-up, 92 per cent of the four-year-old free entitlement hours were used and 89 per cent of three-year-old hours. Early education is provided in a range of settings, including schools, as well as by private and voluntary sector organisations.

There is also a potential role for early education to limit attainment gaps for disadvantaged children. There are big differences in cognitive, and social and emotional development between children from rich and poor backgrounds at the age of three and this gap widens by the age of five (Dearden et al., 2010). However, there is a body of evidence that shows disadvantaged children are especially likely to benefit from attending better quality pre-school provision (e.g. Currie, 2001; Melhuish et al., 2008; Sammons et al., 2004; Stephen, 2006; Sylva et al., 2004, 2008).

Again this is reflected in policy such that from September 2012 a new targeted entitlement for two-year-olds to access free early education was introduced as part of the Government's fairness premium, to drive up social mobility and improve life chances. The primary focus is on disadvantaged children, who are currently less likely to access the benefits of early education.

Much of the evidence for England discussed above comes from the Effective Provision of Pre-school Education (EPPE) project, which was commissioned in the late 1990s, and its extensions through primary and secondary education, the EPPE 3-11 and EPPSE studies. (2) However, there have been considerable changes in early education since then, covering the expanded access discussed above, revisions to the early years curriculum, the role and training of early years professionals and graduate leaders as well as greater emphasis on supporting parents. Given these changes, it is important to update the evidence of the longer-term impact of early education on child attainment. Clearly it is not possible to establish the full impact of the current provision of early education in England, because time is needed to observe longer-term outcomes, but there is scope to consider the impact of early education on early primary school attainment for a later cohort of children than considered in the EPPE study.

Given the near universal take-up of early education for this later cohort of children, it may be problematic to detect an overall impact of early education. Our analysis tries to detect an overall impact, but also focuses on whether early education is associated with better outcomes for children with different characteristics, with our emphasis being on whether outcomes differ for children that experienced early poverty in comparison to those that did not.

This paper focuses on the longer-term impact of early education on attainment at the end of KS1, (3) when children are aged seven, for children in England in the Millennium Cohort Study (MCS), born between September 2000 and August 2001. It builds on earlier work (Hopkin et al., 2009) which looked at attainment for these children at the end of the Foundation Stage, measured by the Foundation Stage Profile (FSP), when children were aged five. That study found only a limited impact of early education on FSP scores. However, at the time, the FSP was still in the relatively early stages of implementation. In particular, it was not moderated, and this may have contributed to this finding.

In addition, the MCS data analysed did not include any measures of early education quality, which other studies had shown to be important predictors of later outcomes. However, another study, Hansen and Hawkes (2009), using data from the MCS, did find that early education, identified through formal group care at nine months, was positively associated with school readiness test scores at three, but found no association between formal group care and problem behaviour.

A smaller sub-study of the MCS (Mathers et al., 2007) of around 300 early education settings and roughly 600 children, does include the same ECERS quality indicators that were available in the EPPE study. Analysis of these data (Hopkin et al., 2010) found no association between attendance at a high quality early education setting and FSP scores, but did show an association between high quality early education and other cognitive outcomes, measured by the naming vocabulary assessment from the British Ability Scales. (4) Unfortunately, the MCS sub-study only has information on KS1 attainment for a little over 400 cohort members, so is not suitable for our analysis.

The purpose of this paper is to use the full MCS data to assess whether attendance at early education is related to child attainment at age seven and whether this varies by early childhood poverty, which we define as household income below 60 per cent of the median.

Although we do not have measures of the quality of early education in our data, we do include a quality measure for schools attended by children during KS1. This is taken from school inspection reports undertaken by the regulatory body Ofsted (Office for Standards in Education, Children's Services and Skills). This is the first time school inspection data has been linked to the MCS data and our results indicate that overall school effectiveness is strongly related to child attainment at KS1, although part of the Ofsted judgement relates to pupil achievement, so this finding is in line with expectations.

The remainder of the paper is structured as follows: firstly we outline the national curriculum assessment at Key Stage 1; we then introduce the data--the Millennium Cohort Study and data from school inspection reports; and finally present our results and conclusions.

National Curriculum assessments at Key Stage 1

Ali children in (state) maintained primary schools in England are required to be assessed by teachers against the National Curriculum in reading, writing, speaking and listening, mathematics and science when they reach the end of KS1.

The assessments provide a measurement of achievement against the precise attainment targets of the National Curriculum (Department for Education, 2008). They measure the extent to which pupils have the specified knowledge, skills and understanding which the National Curriculum expects pupils to have mastered by the end of KS1. The standards have been designed so that most pupils will progress by approximately one level every two years. This means that by the end of KS1, when most pupils are aged seven, they are expected to achieve Level 2.

If, based on judgements of levels for attainment targets in reading, writing and mathematics, teachers reach an overall judgement of level 2, they should then consider whether the performance is just into level 2 (2C), securely at level 2 (2B) or at the top of level 2 (2A). These refinements are known as sub-levels. Children judged not to have attained level 2 are further judged as either having 'achieved level 1' or to be 'working towards level 1'. At the other extreme some children are judged to have 'achieved at level 3 or above'.

Attainment at each of these levels is turned into a points score (5) for each pupil so that an overall average points score can be calculated. This average points score is based only on the scores for reading, writing, mathematics and science (the judgement on speaking and listening is excluded) and in our sample ranges from 3 to 22.5.

This gives us a number of different attainment measures to consider. For overall attainment at KS1 we consider the average points score across the four areas of learning discussed above. We also consider attainment across all five areas of learning. For all five areas we consider attainment at level 2, the level of attainment pupils are expected to achieve by the end of KS1, see table 1. However, for all five areas of learning there is more disaggregated information such that in these models we can use a more detailed measure of attainment. For science, and speaking and listening, we consider three different levels of attainment: achieving below level 2;6 achieving at level 2; and achieving at level 3 or above. (7) For reading, writing and mathematics we consider five different levels of attainment; making use of the three sub-levels of attainment for these areas of learning.

The Millennium Cohort Study

Our analysis is based on children in the Millennium Cohort Study (MCS). (8) The MCS follows the lives of around 19,000 babies, born between September 2000 and August 2001 in England and Wales, and between November 2000 and January 2002 in Scotland and Northern Ireland. KS1 assessments only apply to children in schools in England, so our focus is on these children. We use data from the first four surveys. The survey for the first sweep took place when the children were aged about nine months. Data were collected from parents (or a very few guardians) about themselves and their babies. Some detail about childcare provision is included. The second survey was carried out when the children were around three years old. The third survey was conducted in 2006 when the children were around five years old and were entering primary school; and the fourth survey in 2008 when the children were around seven years old. For further details about the survey see Hansen (2010).

The MCS collects detailed information on a wide range of topics. These have included child's health and development, early education, schooling and childcare, parenting activities, grandparents' and friends' involvement with the children, details of siblings, parental health, employment,

education and earnings, housing, local area and household demographics and family context. Some parts of the interview are completed via self-completion modules; these have included questions about the child's behaviour, the respondent's relationship with their partner, mental health and attitudes to parenting.

The MCS sample is clustered geographically and disproportionately stratified to over-represent areas with higher proportions of ethnic minorities in England, areas of high child poverty, and the three smaller countries of the UK. All our analysis takes into account the survey design, including non-response weighting.

The survey includes cognitive and behavioural assessments for children at age three, five and seven, whilst KS1 attainment data has been merged into the survey data for children in England. This was possible for 95 per cent of children in maintained schools. We only include children where complete assessment data are available. This leaves a sample of 6,762 children.

Attendance at early education

Information on attendance at early education was collected at the first three sweeps of the MCS. Here we focus on the information from the third survey, where respondents were asked whether their child had ever attended any of the following types of provider: nursery school or class, playgroup, preschool, childminder or day nursery.

Ideally we would have liked to use data from earlier sweeps of the survey to give us a full early education history, but inconsistencies in questions asked and in reporting of spells of provision mean that we are limited to sweep 3 data. The MCS questions for sweep 3 were designed partly with the intention of gathering data on previous attendance, as information collected at the first two sweeps was felt not to have sufficiently captured this (Jones, 2008).

Overall 93 per cent of MCS respondents attended early education as defined above. This figure is broadly in line with the take-up rates of free entitlement discussed earlier. Almost all of these children (87 per cent of the total) attended either a nursery school or class, playgroup or preschool, which typically have a clear education component to the early years experience. For the remaining group of children (6 per cent of the cohort) the extent to which they received some early education may be questionable, particularly if they only remained with the day nursery or childminder until their first birthday, below which age little formal education takes place.

Roughly 5 per cent of the cohort attended only a day nursery up to the age of five. Just under one-quarter of these children ended their time at the day nursery before their first birthday with over one-half continuing to attend a day nursery at least until their third birthday. The remaining 1 per cent either attended only a childminder or attended a childminder and a day nursery (but not a nursery school or class, playgroup or preschool). The majority of these children, 80 per cent, continued to attend a childminder or a day nursery until at least their third birthday.

Children who attended a day nursery or childminder at age three would be eligible for the free entitlement to early education. We therefore argue that the majority of children who attended only a childminder, or a day nursery, or both, are likely to have experienced some early education albeit in a less formal environment than a nursery school or class, playgroup or preschool.

School inspections in England

Ofsted is an independent national body that undertakes regular inspections of schools. The purpose of inspection is to provide an independent external evaluation of a school's effectiveness and a diagnosis of what it should do to improve. This is based on a range of evidence available to inspectors that is evaluated against a national framework, which sets out the statutory basis for inspections conducted under section 5 of the Education Act 2005. This is regularly updated with a new framework operational from September 2012 (Ofsted, 2012).

The framework highlights three essential functions of school inspections: to provide parents with an expert and independent assessment of how well a school is performing, and help inform those who are choosing a school for their child; to provide information to the Secretary of State for Education and to Parliament about the work of schools and the extent to which an acceptable standard of education is being provided; and to promote the improvement of individual schools and the education system as a whole.

Inspectors are required to report on the quality of education provided in the school and must cover the achievements of pupils, the quality of teaching, the behaviour and safety of pupils, and the quality of leadership in, and management of, the school. The measure we use in the analysis captures the overall effectiveness of the school. This identifies 'how effective and efficient are the provision and related services in meeting the full range of learners' needs?'. As an overall school effectiveness measure it encapsulates, to some degree, all the dimensions of the inspection, and hence is related to pupil achievement. However, most of the children in any one school are not in the MCS sample, (9) so the inspection judgement provides an indicator of school quality whilst our focus is typically on one or two children in each school.

In line with the current inspection framework, the framework that applied when MCS cohort members were in KS1 (Ofsted, 2005) uses a four-point scale to provide a judgement of the overall effectiveness of the school. The possible judgements are grade 1: outstanding; grade 2: good; grade 3: satisfactory; and grade 4: inadequate. Inspections were short and focused, taking no more than two days in a school and concentrated on close interaction with senior managers, taking self-evaluation as the starting point. Schools were generally inspected once every three years, but inspections occur more frequently for schools causing concern. All inspection reports are made publicly available.

We merged inspection data from 2005-8 with the MCS data using a unique school identifier. Details of the inspection with date closest to the interview date for the cohort member is merged onto the information for each cohort member. For 7 per cent of the MCS sample in maintained schools it was not possible to merge in data from an inspection report.

Descriptive statistics

Children in the MCS sample typically completed KS1 in the academic year 2007/8. (10) National Statistics for this year indicate a KS1 assessment average points score of 15.3 (see table 1), whilst the average points score for the MCS sample is 15.8. On average, our sample includes children with slightly higher attainment levels, differences which were statistically significant at conventional levels. Significant differences are also evident across all the five KS1 scales with a slightly higher percentage of MCS children achieving level 2 than nationally.

Table 1 also shows the average points score and the percentage of children achieving level 2 by whether the child attended early education. The average points score for children who attended early education is slightly higher at 15.9, compared to 15.5 for the minority that did not attend any early education. This difference is statistically significant at the 10 per cent level. Similarly, across the five areas of learning, there was a slightly higher percentage of children who attended early education achieving at least a level 2 attainment compared to those that did not attend. These differences ranged from 2 to 4 percentage points with differences in attainment in reading, writing, and speaking and listening significant at the 10 per cent level, whilst differences in attainment in maths and science were not significant even at the 10 per cent level.

The association between attending early education and KS1 attainment, although positive, is not significant at conventional levels. However, in order to assess the impact of early education, we need to consider how much difference early education makes to child development over time. Ideally we would want a measure of development upon starting early education that can be compared with later indicators of attainment, but children start early education at different times, so in the context of the survey such a measure does not exist. The survey does include development measures at age three, which we can compare with our attainment measure at age seven. The age three indicator we use is the naming vocabulary assessment from the British Ability Scales discussed earlier. Although the naming vocabulary assessment captures a limited aspect of child development at age three, whilst the KS1 assessment is a broader measure at age seven, we believe the comparison is valid. Hopkin et al. (2009 and 2010) found that the naming vocabulary indicator was a strong predictor of a broad measure of child development at age five as measured by the FSP.

In order to compare the naming vocabulary and KS1 measures, we use weighted data to convert them both into z scores with mean 0 and standard deviation of 1. Table 2 shows the standardised scores at age three and age seven by whether the child attended early education or not and by whether they lived in a household with income below the poverty level. This poverty indicator relates to when the child was aged three; i.e. at the same time as the naming vocabulary assessment.

Living in poverty is clearly associated with lower attainment both at age three and age seven. The standardised scores for children not living in poverty were all positive, indicating that attainment in the assessments for these children was above the mean. In contrast, standardised scores for children living in poverty were all negative, indicating attainment below the mean.

The difference in standardised scores indicates that children who attended early education had higher assessment scores. At age three, this was true both for children living and not living in poverty. This is important because it suggests that higher ability children were more likely to have attended early education and this may bias the estimated impact of early education. We experimented with instrumental variable models to control for these differences in early education, using variables that measured whether the mother of the child was recorded as working when the child was nine months old and three years old, but we found these variables to be very weak instruments. Being unable to properly control for the decision to attend early education may have bearing on our findings, but given the high levels of attendance at early education we feel the impact on the findings is likely to be small.

By age seven, for children who were not living in poverty at age three, the attainment gap for those that did and did not attend early education reduced from 0.14 to 0.05, a difference that was not statistically significant. However, for children who were living in poverty at age three, the attainment gap for those that did and did not attend early education increased from 0.11 to 0.25 and remained statistically significant. This suggests that, for children living in poverty, attending early education may help to reduce the poverty attainment gap with non-poor children.

Table 3 shows average points scores according to the inspection rating of the school attended. This is the first paper to incorporate data on school inspections in analysis of the MCS. Here we find that children who attended 'outstanding' schools had higher average KS1 points scores than children who attended schools rated less highly. Again here it is possible that higher ability children went to better schools and so we would need to control for school selection. The MCS includes a set of questions identifying parents' reasons for choosing the school their child attended. These include choosing a school for academic reasons, because it was near, or for family reasons. We included these variables in our models and found that, once we had included our other control variables, these school choice variables were neither related to school characteristics in terms of the Ofsted rating nor to pupil attainment at the KS1 assessment.

Statistical analysis

To further explore the relationship between early education, poverty, school inspection and KS1 attainment, we estimate attainment models for the average points score at KS1 and the probability of achieving different levels of attainment in the five areas of KS1 learning; we include early education, poverty and inspection ratings as control variables together with a raft of other control variables that are related to child development. The results are summarised in tables 4 and 5 with the full model specifications shown in Appendix tables A1 and A2.

As well as the early education, poverty and inspection variables discussed above, the models include variables that can be grouped into four categories:

* child characteristics;

* family characteristics;

* school characteristics; and

* the home learning environment.

The child characteristics include gender, ethnicity, age, whether the child received help at school due to disability/ behavioural problems, whether the school has told the parent/guardian that their child has special needs, parental reports of: whether the child enjoyed school usually or always; whether the child talked about school; whether the child was reluctant to go to school; whether the child was bullied at school more than once or twice.

We also include measures of development collected at the second sweep of the MCS when the children were aged three. The measures are the naming vocabulary element of the British Ability Scales discussed above and the Bracken School Readiness assessment.

The family characteristics include mother's qualifications, whether the school provides the child with free meals, (11) whether a language other than English was mostly or only spoken at home and parent/guardian satisfaction with education at current school.

The school characteristics include whether the school attended is a faith school, whether the child has changed schools between sweeps 3 and 4 of the MCS (i.e. between the ages of five and seven), and indicators for whether the child is in a different school year than would be expected for children born between September 2000 and August 2001. (12)

The survey asks a number of questions about the home learning environment for children when they are aged three, five and seven. We include in our models one indicator at age five and one indicator at age seven which our analysis showed were related to KS1 attainment. These were whether the parent/guardian ever draws or paints with their child (asked at sweep 3 when the children were aged five); and whether the parent/ guardian ever plays indoor games with their child (asked at sweep 4 when the children were aged seven).

Table 4 presents results of Ordinary Least Squares estimates for the KS1 average points score. The first column shows the association between attending early education and the average points score. The estimated coefficient of 0.39 is the same as the difference in average points shown in table 1.

When we add an interaction term for attendance at early education and experience of early poverty as well as separate poverty indicators (column 2), we find that attending early education matters to those children who experienced early poverty, although this was only significant at the 10 per cent level. The data allow us to identify the age the child started in early education as well as whether they lived in a household in poverty at age nine months and three years. We include in our models separate poverty indicators at each point in time, but when we interact poverty with early education we use a poverty indicator that relates to the age the child started in early education. For children who started at the setting that provided them with early education before they were one year old we use the poverty indicator from the first MCS survey, when they were nine months old. For children who started on or after their first birthday we use the poverty indicator from when they were three years old. We experimented with different indicators of poverty with no significant change in our results.

Experience of poverty itself was associated with lower attainment at KS1, whilst the difference in the early education coefficients by whether the child had experienced early poverty were not statistically significant.

Including our full set of control variables (see column 3 of table 4 and Appendix table A1 for the full model estimates) has little impact on the results. In these models we include measures of child development at age three so that the estimates now relate to improvements in child development between ages three and seven. Early education for children who had not experienced early poverty was not associated with KS1 attainment, but for children who had experienced early poverty, early education was associated with higher KS1 attainment, which was again only significant at the 10 per cent level.

In this model, the difference between the coefficient on early education and the coefficient on early education for those children who experienced early poverty was now also statistically significant, again at the 10 per cent level. The difference in these coefficients (-0.12 and 0.29) indicates a higher average points score of 0.41 for those children that attended early education and lived in poverty compared to children who attended early education and had not lived in poverty.

In this model, experience of early poverty (at age nine months or three years) is again associated with lower attainment. However, the size of the poverty coefficients is similar in magnitude to the 0.41 difference in the early education coefficients. This means that early education can largely offset the lower attainment associated with transitory poverty--the experience of poverty at one of these points in time--but not for the children who experienced poverty both when aged nine months and three years.

The coefficients on Ofsted ratings are also reported in column 3 of table 4. Here we find a strong relationship between the Ofsted rating of the school and child attainment. Children who attended an 'outstanding' school had a KS1 average points score 0.74 higher than children who attended a 'satisfactory' school.

Although the coefficient was negative, children who attended a failing school (one rated 'inadequate') did not have significantly lower attainment levels than children who attended 'satisfactory' schools. Allen and Burgess (2012) find that schools which were only just failing their inspection see an improvement in scores in the two to three years following inspection compared with schools that just achieve a 'satisfactory' rating. Hence it is possible that the failing schools in our sample showed some improvement following inspection that brings them more into line with 'satisfactory' schools.

The coefficient on schools where the Ofsted data is missing is also not statistically significant, suggesting that attainment for children in these schools is broadly in line with that for children attending 'satisfactory' schools.

As noted above, these indicators have not been previously included in analysis of MCS data. It is important therefore to note that, despite these variables being strongly associated with child attainment, when we exclude them from our models the relationship between early education, poverty and child attainment is largely unchanged.

Finally we turn to models of the level of attainment for each of the five areas of learning. The available data are slightly different, such that for reading, writing and maths there are five categories, whilst for science and speaking and listening, level 2 achievement is not split into sub-levels, so we have just three categories. The indicators are all clearly ordered, so we estimate ordered probit models for each area of learning. The models reported in table 5, with the full specification reported in Appendix table A2, include the same control variables as the models in table 4. Here we report only the results for the full specification models, equivalent to column 3 of table 4.

The results vary by area of learning. Early education was associated with significantly higher attainment in maths for children who experienced early poverty, but not for others. Furthermore, for children who attended early education, attainment was also significantly higher for those who experienced early poverty than for children who did not experience early poverty (the difference between the two early education coefficients was statistically significant).

For writing, science and speaking and listening, early education was found to have no association with attainment at KS1. This was the case irrespective of whether children experienced early poverty. Furthermore, within the group that attended early education, differences in impact by experience of early poverty were also not evident.

For reading, whilst early education had no overall impact on attainment, there were differences between children who attended early education by whether they experienced early poverty. Here, attainment was higher for children who experienced early poverty than for those that did not.

Discussion and conclusions

The evidence presented in this paper suggests that any detectable impact of early years education on KS1 attainment is limited to children who experienced early poverty. Overall, the difference in attainment levels between children who did and did not attend early education were small and disappear when we control for family, child and school characteristics. However, some positive association remains for children who experienced early poverty in the overall KS1 average points score and for attainment in maths, although the former association is statistically weak.

This contrasts with much of the evidence of the impact of early education in England based on EPPE data, but is not dissimilar to other results using the MCS.

The results may be limited by a number of important factors. As noted, the past fifteen years have seen a large increase in the percentage of children in England attending early education through the introduction and expansion of a free entitlement to some hours of early education for three- and four-year-olds. The high take up of the free entitlement means that the group of children who do not receive any early education is now quite small, and the identification of an overall impact of early education relies on a comparison against this small group. This group of children who do not attend early education includes children with a variety of circumstances. Overall, children from disadvantaged backgrounds are over-represented in the group of non-participants, but there is also a significant group of children from non-disadvantaged backgrounds who perform well at the KS1 assessment.

Furthermore, the representative nature of the MCS sample means that in each early education setting there are typically only one or two children, meaning that it is not practical to collect data on setting quality for a large sample of settings. Other studies have highlighted the importance of early education quality on identifying the longer-term benefits of early education, such that failure to identify setting quality in this analysis leads us to identify average relationships between early education and subsequent attainment which are diluted and therefore not detectable.

A further concern is that because schools administering the KS1 assessment are primarily assessed on the value added between KS1 and KS2, (13) there is an incentive to mark down the KS1 attainment score in order to inflate the value added at KS2. The extent that this happens in practice is not possible to identify. However, the fact that such an incentive is built into the value judgements of schools and can potentially be relatively easily manipulated because KS1 scores are teacher-assessed, means that, in combination with the other factors discussed above, the lack of a strong association between attendance at early education and KS1 attainment may not be surprising.

Our analysis also has some limitations. We note in the paper the possibilities of selection bias in terms of children with higher ability being more likely to attend early education and attend higher quality schools. +Our attempts to identify variables that allow us to control for such potential biases were unsuccessful. Finding more appropriate variables to identify these possible biases would provide more confidence in the findings.

The limited positive findings for children who experienced early poverty may be a reflection of the policy over the period under consideration of targeting high quality early education for children from disadvantaged backgrounds.

A tentative conclusion from this research is that universal provision, with a focus on high quality services for disadvantaged children, can improve attainment levels for these children.

The paper focuses only on the assessment undertaken by children in maintained schools at the end of KS1.

However, there may be benefits to children in other dimensions. The MCS is a rich data source including other measures of cognitive ability at age seven as well as teacher and parent assessments of behavioural difficulties, all of which are worthy of analysis. Early education can also have wider benefits, for example, allowing parents able to find work to do so whilst their child is at an early education setting. Early education may also lead to longer-term benefits, so a full assessment of the impact of early education needs to consider a much wider array of outcomes than considered here.

Overall, the evidence from the EPPSE study provided a strong case for the expansion of early education, but more than a decade later with near universal take-up of early education the evidence base needs to be refreshed. One of the key messages from EPPSE is that longer-term outcomes from early education only occur when children attended high quality early education settings. Analysis of data, more recent than EPPSE, from the MCS, does not allow quality of early education to be identified. Improved data are required and the current proposal from the Department for Education, for a new study to evaluate the impact of early education in England, will provide a valuable opportunity to bring the evidence up to date.

REFERENCES

Allen, R. and Burgess, S. (2012), 'How should we treat under-performing schools? A regression discontinuity analysis of school inspection in England', The Centre for Market and Public Organisation Working Paper No. 12/287.

Belsky, J., Lowe Vandell, D., Burchinal, M., Clarke-Stewart, K.A., McCartney, K. and Tresch Owen, M. (2007), 'Are there long-term effects of early child care?', Child Development, 78, 2, pp. 681-701.

Currie, J. (2001), 'Early childhood intervention programs: what do we know?', Journal of Economic Perspectives, 15, 2, pp. 213-38.

Dearden, L., Sibieta, L. and Sylva, K. (2010), 'From birth to age five: evidence from the Millennium Cohort Study', in Goodman, A. and Gregg, P. (eds), Poorer children's educational attainment: how important are attitudes and behaviour?, Joseph Rowntree Report, March.

Department for Education (2008), National Curriculum Assessments at Key Stage I in England, Statistical First Release 21/2008.

--(2012), Provision for Children under five years of age in England: January 2012, Statistical First Release 13/2012.

Hansen, K. and Hawkes, D. (2009), 'Early childcare and child development', Journal of Social Policy, 38, 2, pp. 211-39.

Hansen. K. (ed.) (2010), Millennium Cohort Study First, Second, Third and Fourth Surveys: A Guide to the Datasets (Fifth edition), Centre for Longitudinal Studies, Institute of Education, University of London.

Harms, T., Clifford, M. and Cryer, D. (1998), Early Childhood Environment Rating Scale, Revised Edition (ECERS-R), Vermont, Teachers College Press.

Hopkin, R., Stokes, L. and Wilkinson, D. (2009), 'Using foundation stage proflle assessments to assess outcomes from early years education', National Institute Economic Review, 207, January.

--(2010), Quality, Outcomes and Costs in Early Years Education, Report to Office for National Statistics.

Jones, E. (2008), 'Childcare', Chapter 5 in Hansen, K. and Joshi, H. (eds), Millennium Cohort Study Third Survey: A User's Guide to Initial Findings, Centre for Longitudinal Studies, Institute of Education, University of London.

Magnuson, K., Ruhm, C. and Waldfogel, J. (2007), 'Does prekindergarten improve school preparation and performance?', Economics of Education Review, 26, pp. 33-51.

Mathers, S., Sylva, K. and Joshi, H. (2007), Quality of Childcare Settings in the Millennium Cohort Study, Research Report SSU/2007/ FR/025, SureStart/Department for Education and Skills.

Melhuish, E., Phan, M., Sylva, K., Sammons, P., Siraj-Blatchford, I. and Taggart, B. (2008), 'Effects of the home learning environment and preschool center experience upon literacy and numeracy development in early primary school, Journal of Social Issues, 64, 1, pp 95-114.

National Institute of Child Health and Human Development (NICHD) Early Childcare Research Network (2000), 'The relation of childcare to cognitive and language development', Child Development, 71,4, pp. 958-78.

Ofsted (2005), Every Child Matters: Framework for the Inspection of Schools in England from September 2005, July 2005, Reference No. 120100.

--(2012), The Framework for School Inspection, September 2012, Reference No. 120100.

Sammons, P., Elliot, K., Sylva, K., Melhuish, E., Siraj-Blatchford, I. and Taggart, B. (2004), 'The impact of pre-school on young children's cognitive attainments at entry to reception', British Educational Research Journal, 30, 5, pp. 691-712.

Sammons, P., Sylva, K., Melhuish, E., Siraj-Blatchford, I., Taggart, B., Barreau, S. and Grabbe, Y. (2008), The Influence of School anal Teaching Quality on Children's Progress in Primary School, DCSF, Research Report No. DCSF-RR028.

Stephen, C. (2006), 'Early years education: perspectives from a review of the international literature', Scottish Executive, Edinburgh.

Sylva, K., Melhuish, E., Sammons, P., Siraj-Blatchford, I. and Taggart, B. (2004), The Effective Provision of Pre-School Education (EPPE) Project. Technical Paper 12 - The final report: effective pre-school education, DfES/Institute of Education, London.

--(2008), The Effective Provision of Pre-School Education 3-11 (EPPE 3-11): Final Report for the Primary Phase: Pre-school, School and Family Influences on Children's Development during Key Stage 2 (Age 7-11), Department for Children, Schools and Families Research report No. DCSF-RR061.

--(2012), Effective Pre-school, Primary and Secondary Education 3-14 Project (EPPSE 3-14)-Final Report from the Key Stage 3 Phase: Influences on Students' Development from age 11-14, London, Department for Education.

Sylva, K., Siraj-Blatchford, I., Taggart, B., Sammons, P., Melhuish, E., Elliot, K. and Totsika, V. (2006), 'Capturing quality in early childhood through environmental rating scales', Early Childhood Research Quarterly, 21(1), pp. 76-92.

Anitha George, Lucy Stokes and David Wilkinson *

* National Institute of Economic and Social Research. E-mail: d.wilkinson@niesr.ac.uk. This research was undertaken as part of the Indicators for Evaluating International Performance in Service Sectors (INDICSER) project, funded by the European Commission, Research Directorate General, as part of the 7th Framework Programme, Theme 8: Socio-Economic Sciences and Humanities, Grant Agreement no. 244709. The authors are grateful to the referees of this paper for their helpful comments.

NOTES

(1) The specified date for children born in the period I January to 31 March is I April following the child's third birthday. For children born in the period I April to 31 August: I September following the child's third birthday; and for children born in the period I September to 31 December: I January following the child's third birthday.

(2) The original Effective Provision of Pre-school Education (EPPE) study followed children to the end of Key Stage I. The extension Effective Provision of Pre-school and Primary Education (EPPE 3-11) project followed the same children to the end of primary school, while the Effective Provision of Pre-school, Primary and Secondary Education (EPPSE) covers the period to the end of Key Stage 3 (at age 14).

(3) KSI comprises the two years of schooling in maintained schools in England, when pupils are aged between five and seven (Years 1 and 2).

(4) The Naming Vocabulary assessment from the British Ability Scales is part of a set of cognitive assessments designed to gauge children's expressive language skills.

(5) Working towards level I has a points score equivalent of 3 points; achieving level I has a points score equivalent of 9 points; just into level 2-13 points; securely at level 2-15 points; at the top end of level 2-17 points; achieving level 3-21 points; and achieving level 4-27 points.

(6) The categories 'working towards level I' and 'achieving level I' are combined, for the purposes of the ordered probit, because very few children are recorded as 'working towards level I'.

(7) Similarly, the categories 'achieving at level 3' and 'achieving level 4' are combined because very few children are recorded as 'achieving level 4'.

(8) University of London, Institute of Education, Centre for Longitudinal Studies, Millennium Cohort Study: First to Fourth Surveys, 2001-2003, 2003-2005, 2006 and 2008, Linked Education Administrative Dataset: Secure Data Service Access [computer file]. Colchester, Essex: UK Data Archive [distributor], November 2011. SN: 6862.

(9) The data available to us include a unique school identifier for the schools of two-thirds of the children in our analysis sample. On average, for these children, there were two MCS children in any one school, and 65 per cent of schools in the analysis sample include just one MCS cohort member.

(10) Just 3 pupils completed KSI in academic year 2008/9.

(11) In England, children are eligible to receive free school meals if their parents are in receipt of any of the following benefits: Income Support, Income-based Job Seekers' Allowance, Income-related Employment and Support Allowance, Support under Part VI of the Immigration and Asylum Act 1999, the Guaranteed element of State Pension Credit, Child Tax Credit, provided they are not also entitled to Working Tax Credit and have an annual gross income of no more than 16,190 [pounds sterling], as assessed by Her Majesty's Revenue and Customs.

(12) Most children are in school year 2 at the time of the MCS sweep 4 interview, but a small minority (0.3 per cent) were in year 1 and a further 5 per cent were in year 3.

(13) KS2 assessment is at the end of primary school, age 11.