• E-mail
• Print
• Comment
• Font Size
• Digg
• del.icio.us
• Discuss article

Early Education and Children's Outcomes: How Long Do the Impacts Last?*

Posted on: Tuesday, 3 January 2006, 03:03 CST

By Goodman, Alissa; Sianesi, Barbara

Abstract

We evaluate the effects of undergoing any early education (before the compulsory starting age of 5) and of pre-school on a cohort of British children born in 1958. In contrast to most available studies, we are able to assess whether any effects on cognition and socialisation are long-lasting, as well as to estimate their net impact on subsequent educational attainment and labour market performance. Controlling for a particularly rich set of child, parental, family and neighbourhood characteristics, we find some positive and long-lasting effects from early education. Specifically, pre-compulsory education (pre-school or school entry prior to age 5) was found to yield large improvements in cognitive tests at age 7, which, though diminished in size, remained significant throughout the schooling years, up to age 16. By contrast, attendance of pre-school (nursery or playgroup) was found to yield a positive but short-lived impact on test scores. The effects on socialisation appear to be more mixed: we found some positive, though short-lasting, effects of pre-compulsory education on teachers' reports of social adjustment (only at age 7); on the other hand, we found some adverse behavioural effects according to parental reports at age 7 which persisted up to age 11. In adulthood, pre-compulsory education was found to increase the probabilities of obtaining qualifications and of being employed at age 33. For both pre-compulsory education and pre-school per se, we found evidence of a marginally significant 3-4 per cent wage gain at age 33.

I. Introduction

Early childcare and pre-school policies have become an important focus of the government's strategy for improving the well-being of children, either through the enabling effect that childcare has by allowing parents to work or through other more direct effects of early education on children. The aim of this paper is to shed light on the question of how effective early pre-school and schooling are at improving the well-being of children, and whether any impacts are likely to be long-lasting. In achieving this aim, we add to a well- established literature both from the UK and from around the world (especially the US).

We add to this literature in two ways. First, by using data from the National Child Development Study (NCDS), a single cohort of people born in 1958, we can control for a particularly large set of parental, family, individual and neighbourhood characteristics in assessing the effects of early education on children. second, by considering a cohort that has now reached adulthood, rather than a contemporary cohort, we are able to assess whether any effects we find are long-lasting. Most studies to date, which have looked at early education amongst contemporary cohorts of children, are unable to be informative as to the longer-term effects. For example, the EPPE study (see section II) has only followed children up to age 7 so far, whereas the NCDS is currently available up to age 42.

In this paper, we consider the returns to two different early education 'treatments'. First, we consider the impact of a child obtaining any early education prior to the age of 5, whether this takes place in a school setting (through early entry to primary school) or in a pre-school setting (such as state-maintained or private nursery, or playgroup). This answers the question of whether any formal human capital intervention before the age of 5 is beneficial for child outcomes, a question that, to our knowledge, has not been explicitly looked at before. second, we restrict our treatment of interest to attendance at pre-school only, answering the question of whether attendance at nursery or playgroup before entering primary school has shortor long-term effects.

We consider a wide range of outcomes, from cognitive development, as measured by a series of tests taken by the children at ages 7, 11 and 16, to socialisation, derived from both parental and teacher assessments of social skills also at 7, 11 and 16, to educational attainment and labour market outcomes up to the age of 42.

Our aim is to consider the total policy effects of early education, without holding the value of other inputs constant. By doing this, we are aiming to capture the contributions of the education itself and of any other concurrent and subsequent family and school inputs that change as a result of early education participation. In estimating these policy effects, we use ordinary least squares (OLS), fully interacted linear regression modelling and propensity score matching.

To assess how 'economically significant' our estimated effects of early education are, throughout we compare their magnitude with the influences that other factors have on children's cognitive and social development - for example, the education and social class of the parents, and whether the child grew up in a family experiencing severe difficulties, such as alcoholism or mental illness. We also consider whether there are different effects for subgroups of the population - for example, whether children from lower social class backgrounds gain more or less from early education, and whether there are differential effects for boys and girls.

The paper proceeds as follows. section II reviews previous research in this area and places our work in its context. section UI describes the data we use, the early education 'treatments' we consider and the outcomes against which we assess their benefits. section IV discusses our methodology. section V describes patterns of participation in various forms of early education, looks at raw differences in performance, summarises the main determinants of early educational investments and presents all the results. section VI concludes.

II. A review of the literature and our contribution

A prolific body of research on early childhood programmes in the US focuses on the effects of programmes intended for disadvantaged children, in particular from ethnic-minority, low-educated, single- parent or low-income families. Such programmes include small-scale model, or 'exemplary', programmes such as the High/Scope Perry Pre- School Project, as well as ongoing, large-scale public programmes such as Head Start. An important strength of this research from an evaluation point of view is that often these programmes have been evaluated experimentally - that is, their services were randomly allocated to some children, with some other children being randomly assigned to a control group. Albeit not completely free from methodological problems,1 the evidence from these studies has consistently pointed to short-term cognitive improvements as well as long-term gains in terms of academic achievement, reduction in special education placement, employment, earnings and crime. Parents were positively affected as well, with benefits being reported in terms of maternal employment and increased parental involvement in the child's school. see, for example, the reviews by Currie (2001), Waldfogel (1999), Karoly et al. (1998) and Barnett (1995).

Although there is thus ample evidence from the US robustly documenting the benefits from high-quality, intensive early education interventions for disadvantaged children, this evidence tells us nothing about the effectiveness of ordinary community services (a different 'treatment') for children from all backgrounds (a different target population). There is, in fact, much less consensus about the effectiveness of typical pre-school programmes for children generally, on which the evidence, though growing, is more limited as well as more controversial. This is because, even in the US, studies of larger and more representative programmes have had to rely on a non-experimental design and have typically been able to include only a few regressors to control for selection issues. A small number of studies have been able to address this matter by using rich data, exploiting either the detailed information available in the British cohort studies (NCDS and BCS70) or data specifically collected on pre-school-age children (such as EPPE in the UK or NICHD in the US).

The 1970 British Cohort Study (BCS70) has been extensively explored in the book by Osborn and Milbank (1987), whose study has been viewed as the 'most statistically convincing' and described as 'the first major evaluation of British preschool education' (Cleveland and Krashinsky, 1998). Based on a sample of 8,500 children, with additional information merged in from a census of all pre-school institutions, they assess separately the effects on cognitive and behavioural development at ages 5 and 10 of several types of ordinary pre-school programmes, finding similar types of effects for nearly all. Based on 'analyses of variance' controlling for a number of important socio-economic and family factors, they find that pre-school generally boosts cognitive attainment at ages 5 and 10. In terms of problem behaviour, pre-school attendance was found to have no effect at age 5 but to increase some types of behavioural problems at 10, in particular conduct disorder, although the latter associations were relatively weak. The study also found weak evidence for the benefits of nursery education being slight\ly greater for socially disadvantaged children, although this difference was small compared with the general benefit of pre- school for all children.

The BCS70 has also been looked at by Feinstein, Robertson and Symons (1998), together with the National Child Development Study (NCDS). They consider the effects of hours spent in (any type of) childcare relative to being neither in pre-school nor with the mother (i.e. relative to being cared for by a relative or childminder) on cognitive ability tests and on measures of social adjustment (at 7, 11 and 16 for the NCDS; at 5 and 10 for the BCS70). For the 1958 NCDS cohort, pre-school was found to have no effects on social adjustment and positive effects on cognitive tests up to age 11 (particularly on maths skills), which then fade away by age 16. For the 1970 BCS70 children, by contrast, the evidence is of marginally worse social adjustment and reduced vocabulary at 5, worse reading skills at 11 and no effects on maths skills. The authors thus conclude that 'over about a decade (1962-1973), the pre- school experience appears to have ceased to improve test scores in children as they enter secondary school'.

Methodologically, the Feinstein et al. study is very ambitious, in that the authors try to estimate the cognitive and non-cognitive development production functions, allowing the choice of pre-school hours and the allocation of maternal time as well as children's intermediate performance to be endogenous. It is not likely, however, that the results are very robust, given how hard it is to identify structural parameters without enough exogenous variations for all endogenous (initial and intermediate) inputs. By contrast, in this paper, we aim to identify policy parameters - that is, the total effect of pre-school on achievement (i.e. not holding other inputs and intermediate outcomes constant).

The second strand of non-experimental research is based on recent 'ad hoc' studies, which specifically sample pre-school-age children, collect relevant background information and follow both their subsequent preschool experience and their subsequent cognitive and behavioural development.

In the US, the National Institute of Child Health and Human Development initiated the NICHD Study of Early Child Care, a longitudinal study to determine the relationship between children's childcare experiences and their developmental outcomes from infancy to school age. In 1991, over 1,300 newborn children across 10 sites throughout the country were sampled; they were then followed up and are currently in their seventh year in school. Children in higher- quality childcare arrangements were found to perform better on tests of cognitive skills and language ability than children in lower- quality care. However, children who spent more time in childcare were found to display more behavioural problems,2 particularly aggression, at 4 and later in kindergarten than children who spent less time in childcare.3 The findings from the NICHD study thus caution against overarching conclusions about whether childcare harms or benefits children; rather, childcare effects crucially depend on the characteristics of that care (its quality, continuity and intensity), as well as on the characteristics of the child and family (see Waldfogel (1999)).

Another 'ad hoc' US study is the newly available Early Childhood Longitudinal Study - Kindergarten Class of 1998-99 (ECLS-K). This is a large and nationally representative sample of children who entered kindergarten in the autumn of 1998, and is currently available through to the spring of 2000, when most children are completing first grade. Magnuson, Ruhm and Waldfogel (2004) use these data to address a similar question to that in the NICHD. However, they are also able to test the robustness of their findings to potential selection on unobserved variables by instrumenting pre-kindergarten attendance with state expenditures on these programmes. They find that pre-kindergarten increases maths and reading skills at kindergarten entry but is also associated with an increase in behavioural problems. Furthermore, while the cognitive gains largely dissipate by the spring of first grade, the negative effects on classroom behaviour do not. The largest and most lasting academic gains were found for disadvantaged children.

The British counterpart to NICHD, the Effective Provision of Pre- School Education (EPPE), is a longitudinal study begun in 1996 and funded by the Department for Education and Skills. Over 3,000 children were sampled from a range of pre-school providers, and then followed from the start of pre-school (age 3 years plus) through to primary-school entry and across the infant period of primary education (age 7 years plus). An additional small sample of 314 'home' children was recruited at entry to primary school to act as comparison for the pre-school sample. The study found a positive impact of pre-school, on both cognitive and overall social development, at entry to primary school (age rising 5 years) and at the end of Year 1 in primary school (age 6). At the end of Key Stage 1 (age 7), the cognitive gains had been reduced in size but were still present, while the impact on social behaviour had faded out. The exception to the latter was that very long durations in pre- school were associated with higher levels of anti-social behaviour at entry to primary school which were still visible at age 7.

The great advantage of studies of the EPPE, NICHD and ECLS-K kind is that they look at the effects of relatively recent pre-school provision, which should closely reflect current provision; naturally, the price they have to pay is that they are only able to assess extremely short-term outcomes (up to age 7), leaving them to speculate about how the early effects they uncovered would later develop.4

This is where our paper fits in: as we have briefly reviewed, while results for the impact of early education on cognitive development have been consistently positive, existing studies have produced mixed results for behavioural development, in particular that early education may have adverse effects on children's behavioural adjustment. Although cognitive outcomes have traditionally received most attention in the economics literature, the importance of non-cognitive (or social) skills as determinants of educational attainment and labour market success has recently been increasingly appreciated. In such a context, without access to data on longterm outcomes, it remains unclear whether the uncovered increases in relatively low levels of problem behaviour will be consequential for subsequent education and labour market outcomes (an issue also raised by Magnuson et al. (2004)).

Our paper addresses this directly, by assessing both the persistence of any given effect (i.e. whether 'skills beget skills' or whether the 'home' children rapidly catch up once in school) and the long-term, net benefits of pre-school education (i.e. assessing how gains and losses in different dimensions are traded off in the labour market), giving us a much fuller understanding of the longer- run returns to investments in early education. Given the obvious trade-off between how recent the early education experience of the sample is and how long into the lives of the children the effects can be evaluated, the extent to which the childcare and home care of the NCDS children that we study accurately reflect today's large- scale public and private provision remains unknown.

III. Data, treatments and outcomes of interest

1. Data

Among the uniquely rich data from the British cohort studies, the NCDS keeps detailed longitudinal records on all children born in a single week in March 1958. We make extensive use of each sweep of the study, starting with the baseline 1958 Perinatal Mortality Survey (the birth survey), in which information was obtained from the mother and from medical records on characteristics of the mother (age, weight, marital status, education, paid occupation, previous abortions and other complications, smoking behaviour, social class of mother's father), on the father's social class and age, and on characteristics of the child (gender, birth weight, any serious illness). The 1965 follow-up contains information from the parent, head teachers and class teachers, the schools health service and the child, and allows us to identify the type of pre-school the child attended, as well as cognitive and behavioural outcomes at age 7, the latter assessed both by the teacher and by the parent. Subsequent sweeps are used to gather information on outcomes: indicators of cognitive and behavioural development at ages 11 and 16, educational achievement, interests and attitudes, and labour market outcomes (wages and employment status) in adulthood (at ages 23, 33 and 42).

We further code in information about state-maintained provision of nursery places at the local authority (LA) level, taken from Blackstone (1971, appendix 4; originally from the Department for Education and Science), as well as other information about LA characteristics from the 1961 Census and from the Institute of Municipal Treasurers and Accountants' Education Statistics (1958 and 1965).

In terms of sample selection, we exclude anyone not present in the birth survey5 as well as Scottish children, the latter both because their school system differs in a number of respects from the English one and because for them we would not have all the information at the LA level that we have for children in England and Wales. Finally, in this analysis, we do not consider day nurseries or children attending them.6

We are left with a sample of 12,172 children with non-missing information on at least one early education question. This sample size is greater than those of experimental studies in the US or other 'ad hoc' studies discussed in section II.

2. Treatments

Our research addresses itself to two questions:

1. Howlong do the effects of early education on a given outcome - say, test scores - persist?

2. What is the net effect of early education on other subsequent outcomes (such as educational attainment or wages)?

In order to answer these questions, we need to define carefully what the early education 'treatment' of interest is. One can think of early education encompassing both pre-school education and early school entry:

* Pre-school education is defined as taking place in establishments regularly attended by the child outside his or her own home in order to participate in educational activities, before starting school. These include formal centre-based care in the form of nurseries (both local authority and independent) and other more informal settings such as playgroups, which also give children the opportunity to interact with peers and typically expose children to learning experiences.

* Early school entry (or infant school) refers to starting school before reaching the compulsory school age of 5. In contrast to the more informal pre-school education, children who start school early share the building with grown school children, are looked after by teachers and are taught a more academically oriented and structured curriculum. Almost all NCDS children (94 per cent) who started school early did so aged between 4 and less than 5 years, with another 4 per cent of them aged in the half-year bracket below.

Note that, having being born in the same calendar week, the NCDS children will have received different amounts of schooling by the time their outcomes are measured (e.g. tests at 7) depending on when they started formal schooling.7

In the light of this, we consider the effects of the following two broad types of early education:

* Pre-compulsory education: We define 'pre-compulsory education' to encompass any form of education prior to the statutory school- starting age of 5. This includes both pre-school education and early school entry. We thus look at the effect of any type (or combination) of precompulsory education compared with starting school at the statutory age of 5 without having had any form of pre- school.

* Pre-school education: We also consider the effects of just pre- school education - either private or LA nursery, or playgroup - irrespective of whether the child subsequently went to school early. The comparison state includes alternatives such as staying at home, as well as starting school early. We stress that both the treatment and the comparison states are allowed to include early entry.8 We thus capture the full effect of preschool attendance, in that pre- school is realistically allowed to affect early school start.

This second definition of treatment is comparable to that used in other studies, in particular Osborn and Milbank (1987) and the more recent EPPE evaluation (see also the definition in Blackstone (1971, p. 92)). The former study explicitly excludes infant classes from its pre-school classification.9 The EPPE study also restricts itself to an analysis of strictly pre-school settings and does not consider the effects on children of entering school before the age of 5. More specifically, the children entered the EPPE study at different ages, but were tested after having received the same amount of formal schooling (i.e. at entry to primary school, at the end of Year 1 and at the end of Year 2).10

In the following, when there is no need to distinguish between the two types of treatment, we will use the encompassing term 'early education'.

3. Outcomes

The NCDS allows us to analyse the effects of early educational investments on an exceptional variety of outcomes, measured from early childhood to mature adulthood. Being able to look at the effects on such a range of diverse dimensions puts us in the rare position of attaining a very full picture of what types of skills and behaviours are affected by early education.

Cognitive development

We look at cognitive functioning when aged 7, 11 and 16. We consider a summary measure of overall cognitive achievement and separate measures of mathematical skills, language/reading skills and, for the younger ages, verbal and non-verbal general ability and motor-perceptual ability (a copying design test, i.e. a non-verbal test of cognitive ability based on spatial awareness and eye-hand coordination). Having separate tests for different aspects of cognitive development allows us to evaluate which aspect is receptive to early education.

Social/Behavioural development

Although cognitive skills have traditionally received most attention in the economics literature, the importance of behavioural and attitudinal skills as determinants of attainment and labour market outcomes has been increasingly recognised. In parallel with cognitive development, we thus look at a number of measures of social adjustment at 7, 11 and 16. A unique feature of our data is that we can consider social behaviour assessed by three different people close to the child: the primary school teacher, the secondary school teacher and the mother.11

At ages 7 and 11, the teacher assessed the child's behavioural deviance and maladjustment using the Bristol Social Adjustment Guides (BSAG), a test for measuring the extent of disturbance in children's social adjustment and behaviour. It is designed to be as free as possible from personal judgement and tests a variety of personal and interpersonal dimensions, resulting in an overall assessment of maladjustment.

At 16, the teacher was asked to report on the child's behaviour by answering a set of specific questions related to: aggressive behaviour (for example, whether the child fights frequently, destroys things, is irritable, or is resentful/aggressive when corrected), peer relationships (whether not much liked by other children, bullied by other children or bullies other children), other types of antisocial behaviour (is often disobedient, tells lies or steals things) and socio-emotional development (whether the child often worries, is miserable, cannot settle for more than a few moments, has difficulties concentrating, is upset by a new situation or is unresponsive and apathetic). Very similar questions were asked of the mother at all ages; in addition to an overall parental rating, for age 7 we have grouped the answers into those relating more to 'interpersonal skills' and those relating to 'selfcontrol skills'.

We finally consider the potential effects on Objective' measures of social maladjustment in terms of pre-delinquent or delinquent behaviour at 11 and offending behaviour at 16 (trouble with police, having been to court), reported by the teacher and the parent respectively.

Educational attainment

School outcomes are measured by special education, the attainment of any qualification beyond basic level (i.e. above Level 1) and the achievement of higher qualifications (at Level 4 or 5). Any effect on academic functioning in terms of educational attainment includes the effects on cognitive development and socialisation.

TABLE 1

Selected outcomes

Labour market success

We aim to establish whether there is evidence of any continuing early education influence in adulthood in terms of economic success in the labour market, as measured by employment status at 33 and 42 (irrespective of whether as an employee or self-employed) and wages at 33 and 42.12

It should be noted that section V presents our results on just a selected subset of these outcomes, as summarised in Table 1; however, results for all outcomes are available on request from the authors.

IV. Methodology

1. The evaluation problem

We are interested in the causal effect of early education - in the form of either pre-compulsory education or pre-school education - on the cognitive and behavioural development of those children attending early education (the so-called average effect of treatment on the treated - ATT).13 Note also that we are concerned with the total effect of early education on achievement - that is, not holding other family and school inputs constant.14 Such an effect thus encompasses the indirect contribution of any other concurrent and subsequent family and school inputs that change as a result of participation in early education and that in turn affect cognitive development.15

In order to estimate the impact of early education on an outcome such as test scores for the children who received early education, one would ideally need to compare the average test score of these children and the average test score fthat these same children would have achieved had they not received early education. However, since a given child either receives early education or does not, the average test score that early educated children would have achieved in the absence of early education remains an unobserved counterfactual. The evaluation problem consists in providing unbiased estimates of this average counterfactual through the use of appropriate methods and usually untestable assumptions.

Trying to estimate the counterfactual no-education outcomes for early educated children using the observed mean outcomes of 'home' children will yield unbiased estimates of the impact of early education only if children participate in early education based on characteristics that are unrelated to outcomes ('random assignment'). By contrast, one might in general expect families to involve their children in early education on the basis of characteristics that also influence children's subsequent outcomes. The so-called 'selection bias' problem is that some of the difference in outcomes between early educated children and 'home' children is attributable to these pre-education differences, not to the early education itself.

2. Selection on observables

If we had information detailed enough to capture all the outcome- relevant differences between children receiving early education and children not doing so, we could adjust the raw comparison for those pre-education diff\erences, thus obtaining unbiased estimates of the causal impact of early education. This is the 'selection-on- observables' assumption.

Before describing in some detail the extensive individual, family and local neighbourhood information we have assembled in order to control credibly for selection into early education, we briefly sketch the estimation methods we use that rely on the selection-on- observables assumption.

A standard ordinary least squares (OLS) regression controlling linearly for the set of observable characteristics and participation in early education may suffer from two potential sources of bias from observables alone. First, if the true model were non-linear in terms of the characteristics, the OLS estimate of the effect of early education would, in general, be biased. second, this regression constrains the impact to be homogeneous, i.e. the same for all children; if, by contrast, the effect varies according to some of the child's characteristics, OLS will not, in general, recover the ATT. Both these biases are exacerbated if some children fall outside the so-called common support of the observables - that is, if there are children participating in early education for whom there are no comparable children in the non-participating sample. In this case, performing OLS might hide the fact that the researcher is actually comparing incomparable children by using the (linear) extrapolation.

In contrast to standard parametric methods such as OLS, matching is robust in the sense that it does not at all restrict the way in which the effect of early education may vary according to individual characteristics. In addition, the focus of matching methods is on the careful choice of an appropriate comparison group.

More specifically, if there are early educated children who are not comparable to anyone in the home group, they are dropped from the analysis. Matching then involves explicitly selecting and pairing to each early educated child a child who did not receive early education but who has the same characteristics as the early educated child. More generally, the matching process attaches appropriate weights to the observations in the non-participation group, so as to realign their distribution of characteristics to the one in the group of early educated children. It is important to note that one can easily check how well matching has balanced the available observables between the two groups. If balancing cannot be achieved, the researcher needs to accept the fact that the two groups being compared are simply too different in terms of the observables and that there is not enough information in the available data to achieve sufficiently close - and thus reliable - matches.

The boundary between OLS and matching is a thin one; nothing prevents the researcher from bringing the OLS estimate closer and closer to the matching one by imposing common support before running the regression as well as by specifying an increasingly flexible OLS model, hi particular, one can allow the effect of early education to vary according to each observable characteristic and thus estimate the ATT via OLS regression by implementing a fully interacted linear regression model in which the early education indicator is interacted with each one of the observable variables. An additional advantage of this specification is that we can actually test for the presence of heterogeneous effects.

If no heterogeneity in impacts is found, the estimate of the early education effect from simple OLS will basically coincide with the one from the fully interacted model. If, in addition, the two groups of interest can be selected so as to be comparable (i.e. there is no serious common support problem and matching can achieve a good balancing of characteristics), both sets of estimates will be very close to the one from the matching estimator. As we will see in section V, this often turns out to be the case in our application.

But how credible is the selection-on-observables assumption in our application? Given that we need to capture all those factors that jointly determine early education attendance and children's cognitive and social development, the nature, extent and accuracy of the available observables are crucial for the credibility of our estimates. The NCDS birth cohort provides a uniquely rich source of information in this regard.

TABLE 2

Summary of control variables

Inputs into a child's early process of cognitive and non- cognitive development comprise the child's hereditary endowment of mental capacity and physical capability, the local characteristics and amenities of the environment the child is growing up in, and parental and family inputs. These last inputs include parental resources such as time and income, their allocation among the children within a family, the more general type of family and learning background offered to the child and, most importantly to our purposes, parental choices as to early child education. Table 2 summarises the types of information we can control for in the NCDS combined with LA-level data. Such exhaustive controls are direct or proxy variables for the following four groups of factors.16

The child

In addition to gender, we control for the child's health status at birth through relative birth weight and any serious illness noted at birth;17 breastfeeding is also included because it is generally viewed as important for early child development. Belonging to a non- white ethnic group and English not being the mother's usual language with the child should reflect variations in language acquisition and language spoken at home. To proxy the child's genetic endowment, in addition to these factors we control for key measures summarising the child's early development: the presence of disability, whether the child has attended a welfare clinic under 1 year of age, and whether he or she is not walking alone by 1 years, is not speaking by 2 years or is incontinent by day after 3 years.

The parents

In addition to summary indicators of the pregnancy history (interval between marriage and first birth and an indicator of how 'difficult' the child was to obtain by combining past miscarriages, neonatal deaths and complications in pregnancy), this set of variables includes various proxies for 'parental quality' and hence for the quality of parental inputs to the child's development.

As far as parental health and demographic characteristics are concerned, we control for father's and mother's ages (older mothers may have different attitudes towards children's needs), mother's smoking behaviour (both prior to and during pregnancy), mother's obesity and her region of birth. To measure variation in the quality of time parents devote to children and to proxy material resources, we also use social class of the father and of both the maternal and paternal grandfathers. Parental human capital should capture the value parents are likely to attach to the education of their child, as well as the more general quality of parental time. In addition to the typical measure in terms of mother's and father's years of completed education, we have created indicators for whether the mother reads a newspaper most days as well as books most weeks, and similarly for the father.

Several indicators of early maternal employment are crucial to control not only for family financial resources but also for the mother's work propensity (or necessity). In particular, we control for her labour supply and social class in terms of her type of job at the start of pregnancy, the intensity of the work commitment and how far into the pregnancy she worked.

The home/learning environment

The home and learning environment during the formative first years of a child's life is likely to be critically related to early child education as well as to cognitive development and behaviours.

To capture the nature of family life in its essence, we use a selected range of variables from those that were recorded by the health visitor.18 These are whether in the family there was any: physical illness or disability; mental illness, neurosis or mental subnormality; difficulty relating to divorce, separation or desertion; domestic tension; in-law conflict; alcoholism; or other serious difficulty affecting the child's development.

In addition to these important factors reflecting the emotional, psychological, financial and more general overall climate of the family in which the child is growing up, we use proxies for the available quantity of parental time: whether the child has a twin, the child's birth order, the presence of any older brother, any older sister or a close sibling (i.e. the child was born less than 2 years after the sibling) and family size. The received evidence is that large family size is associated with an increased risk of problem behaviour and poorer cognitive development (Osborn and Milbank, 1987).

Together with family structure and size, parental marital status is an important variable to reflect how parental and family resources (of both time and income) are allocated among children within the same family. Furthermore, a mother on her own may experience social isolation and higher levels of stress, which in turn negatively impact on her child's development. We therefore control for the mother's marital status at birth and whether the child has ever been in care by age 7.

Local characteristics

The social, demographic and economic profile of a region or local environment can exert a strong influence on the way a child grows and develops, as well as affect the pattern of availability and use of pre-school and pre-compulsory education services.

Coarse measures of the local environment in which the child is growing up are the region within Britain and the type of administrative district. Additionally, we use information at the finer level of the local authority, capturing the demographic or relevant age composition of theLA (number of primary/secondary students per 1,000 residents), the local income and employment structure (number of economically active females as a percentage of number of economically active males, share of males in semiskilled and unskilled occupations among economically active males) and the weight given by the LA to nursery as opposed to other levels of education (share of nursery pupils in total pupils, number of nursery pupils per teacher as a percentage of total number of pupils per teacher). Finally, a number of indicators are used to capture the quality of primary and secondary education in the LA: primary teachers' salary-cost per pupil, secondary teachers' salary-cost per pupil, all primary-school costs per pupil and all secondary-school costs per pupil. In addition to the quality of schooling, cost indicators summarise several important features of the area, such as the pupil-teacher ratios and the age of school buildings.

To conclude, the set of observables we can control for is extensive; we feel that a case can indeed be made that we have data on the relevant child characteristics, family background influences, parental inputs and local characteristics upon which parents are likely to take their early education decisions.

3. Subgroup analysis

As well as estimating the effects of early education, according to our two definitions, on the population at large, we can assess whether some children benefit more from early education than others. We explore such potential heterogeneity in the effects of early education by performing a variety of subgroup analyses: by gender, birth order, father's social class, parental education, home environment and mother's work status.19 Table 3 summarises the subgroups we look at and the criteria we chose to define them.

TABLE 3

Subgroup analysis

V. Results

1. Differences in performance and determinants of participation in early education

To obtain a general idea of the differences in outcomes between children attending early education and non-participants, we perform some simple exploratory analysis for selected outcome measures.

Since there are no natural units for test results, for this descriptive analysis we follow Krueger (1999) and scale test scores into percentile ranks. Specifically, the non-participating children are assigned percentile ranks based on their scores, from 0 (lowest) to 100 (highest). We then determine where in the distribution of the non-participating children every early educated child would fall, based on the participating child's own test score, and assign him or her the corresponding percentile score.

FIGURE 1

Kernel density estimates of the distribution of percentile ranks for treated and non-treated children

The upper two panels of Figure 1 show the distribution of percentile ranks in maths and general ability tests at age 11 for participating and nonparticipating children, where participation refers either to pre-compulsory education or to pre-school education. Since by construction the distribution for non- participating children is uniform, these plots reveal at a glance how differently the participating children perform. In particular, children with early education are visibly concentrated among the top ranks, which shows how much stronger on these tests these children are. In fact, further on in these children's lives, those with some early education still clearly outperform those who were cared for only at home in childhood in terms of markedly higher wages at age 33, as shown by the bottom panel of Figure 1.

As to the pattern of participation in early education for children born in 1958, Table 4 shows that as many as 60 per cent of them had some form of education prior to the statutory school- starting age of 5. For the large majority (three-quarters), this early education was only in terms of an early start to infant school. A considerable number, however, attended a pre-school placement, with roughly equal shares (over 11 per cent) being in the maintained and private sectors, hi the latter, almost equal proportions (5-6 per cent) started their early education in a nursery and in the less formal setting of a playgroup.

If we by contrast focus strictly on pre-school education (either private or LEA nursery, or playgroup) without taking into account whether the child subsequently went to school early, the picture changes markedly (Table 5). Around 15 per cent of the children received pre-school education; this percentage is considerably smaller than the 60 per cent who received any early education, as most of these were early school entrants. In fact, although as many as 85 per cent of the children had had no pre-school education at all, over half of them started infant school before age 5. The type of provision for pre-school attenders closely reflects the pattern above: an equal split between maintained and independent institutions, and, within the latter, between nurseries and playgroups.

TABLE 4

Sample split by type of pre-compulsory education

TABLE 5

Sample split by type of pre-school education

Although the overall incidence of pre-school education is considerably lower in our sample than amongst present-day children, our sample size of 1,684 children receiving any, or some combination of, pre-schooling is comparable to the EPPE sample, with the added advantage that we are able to follow them up well into adulthood.

As to the determinants of early education investment, we find that the average personal, parental, home environment and regional characteristics of the children attending pre-school or receiving pre-compulsory education differed in important ways from those of children with no pre-school or pre-compulsory education experience.

On the one hand, we found a higher incidence of some child characteristics that we might also think of as contributing to poorer educational outcomes among early educated children (incontinence, disability, welfare clinic attendance, having a twin, not being spoken to in English). The differences are, however, small. By contrast, large disparities working in the opposite direction were found in terms of family factors (mother's human capital, social class of the father, mother and even maternal grandfather). A child who is underprivileged for social or economic reasons (proxied by social class) was shown to be far less likely to have had any pre-school or pre-compulsory education experience than a relatively more advantaged child. These disparities are likely to be further compounded by the large differences we found in the availability and experience of pre-school and pre-compulsory education depending on the social, demographic and economic profiles of the regions and LAs in which the children lived. In particular, living in urban areas or in LAs with lower-than-median share of unskilled male workers, higher average social class (and presumably also income) or higher share of working females were all found to translate into higher use of early education services.20

Since the most influential differences are those that have traditionally been associated with an increased risk of problem behaviour and poorer cognitive development, the raw performance comparisons in Figure 1 are very likely to overestimate the benefits of early education. The next two subsections investigate how robust the apparent benefits from early education actually are.

2. The effects of pre-compulsory education

We first consider the effects of any pre-compulsory education, defined as any formal education before age 5. This treatment is made up of early school entry, nursery and playgroup (or any combination of these); as shown in Table 4, this treatment group largely consists of those who started school early but did not have any prior nursery or playgroup experience. The comparison group is those who did not have any formal education before age 5.

Effects of pre-compulsory education on cognitive test scores at 7, 11 and 16

Of particular interest is the effect of pre-compulsory education on cognitive development. The first column of Table 6 shows our OLS estimates of the effects of pre-compulsory education on the standardised scores of a range of tests taken between ages 7 and 16. The results confirm that pre-compulsory education leads to consistently better test scores at age 7, both on average and separately in maths and reading. Importantly, we also find that these gains persist, though diminished in size, through to ages 11 and 16.

For example, our OLS results suggest that obtaining education before the age of 5 is associated with an increase of 9 per cent of a standard deviation in average test scores at age 7; by age 11, the gain is around 7 per cent; by age 16, this has declined to just over half its size at age 7, at around 5 per cent, but it remains highly statistically significant. The gains in terms of standard deviations of test scores are even bigger when we consider separate maths and reading tests. For example, our OLS results suggest an average gain in maths tests of 14 per cent of a standard deviation by age 7, reduced to 12 per cent by age 11 and again roughly halved from its original (age 7) level to around 7 per cent by age 16. The gains to reading test scores are of a similar magnitude and follow a roughly similar profile over time.

TABLE 6

Effects of pre-compulsory education on cognitive test scores, and coefficients on other aspects of the home environment (OLS estimates)

In section FV.2, we discussed the use of other methodologies besides OLS for identifying the treatment effects of interest. The series of F-tests on heterogeneous effects based on the fully interacted linear model (FILM) suggest that there are no heterogeneous effects; in fact, such estimates basically coincide with the ones from standard OLS. Furthermore, our matching analysis suggests that the two groups of interest can be safely regarded as highly comparable,21 so the restrictions of t\he simple OLS regression are in effect not binding. This is indeed confirmed by the matching estimates, which coincide with those from FILM and from OLS. For simplicity, we thus present OLS estimates only.

How economically significant are these effects? One way of answering this is to compare the magnitude of the estimated treatment effects and the estimated coefficients on some other family and environmental factors we control for in our OLS specification. Although these other coefficients do not straightforwardly measure analogous 'treatment' effects (since the observable factors we control for are not chosen to identify the effects of other treatments besides the early education treatments of central interest to this paper), they give us some indication of how the effects of precompulsory education might compare with other important factors affecting children's development. Table 6 shows that whilst the positive effects of pre-compulsory education on test scores at age 7 are larger than the effect of having a father with a high social class, the advantage conferred by social class is magnified by age 11 and then stabilises,22 whilst the effect of precompulsory education diminishes. By ages 11 and 16, the positive education advantages are considerably smaller in magnitude than the positive advantages associated with social class. Similarly, whilst the positive effect of pre-compulsory education is of approximately the same magnitude as (but opposite direction to) the detrimental effect associated with living in a family with severe difficulties at age 7, the effects of family difficulties persist basically unchanged over time, whilst the effects of pre-compulsory education are greatly diminished. The other comparison we provide is with additional years of mother's education. At age 7, pre-compulsory education is worth the equivalent of between 3 and 5 years of extra maternal education. By age 16, the effect is reduced to the equivalent of around 1 to 2 years of maternal education.

Effects of pre-compulsory education on measures of socialisation at 7, 11 and 16

The picture from Table 7 of how pre-compulsory education affects social skills is more mixed than the picture described above for cognitive skills. In general, we find education before age 5 leads to better teacher-reported social skills at age 7. Attending some kind of pre-compulsory education leads to an improvement of 5 per cent of a standard deviation in the BSAG behavioural disturbance score at age 7. However, some particular types of social skills appear to be negatively affected at age 7: for example, pre- compulsory education leads to more parental reports of poor self- control at age 7 (such as irritability and difficulty concentrating). The effects appear to be neutral on parental reports of interpersonal skills, and the child's happiness or otherwise to attend school (the latter not shown).

The positive effects on teacher-reported social skills we found at age 7 do not appear to persist by age 11 or age 16 (the latter not shown), though the improvement does persist for some subgroups of the population (as we discuss later). Nor do there appear to be long-lasting effects for parental reports of social skills. This contrasts with the markedly large, negative effects on social skills of a difficult home environment across all ages, and a few instances of marginally positive effects of high social class and additional years of mother's education.

TABLE 7

Effects of pre-compulsory education on measures of socialisation, and coefficients on other aspects of the home environment (OLS estimates)

Effects of pre-compulsory education on education and labour market outcomes

Pre-compulsory education is found to have positive effects on both early schooling outcomes, such as the probability of having special education needs at age 7, and on later educational outcomes, such as the probability of obtaining qualifications at Level 2 or higher (Table 8). However, we cannot detect any impact of pre- compulsory education on the probability of obtaining higher education qualifications. We also detect some fairly weak evidence of positive labour market effects, with pre-compulsory education exerting marginally significant positive effects on the probability of being in employment at age 33 and on wages at age 33.23 By age 42, however, any effect has disappeared.

TABLE 8

Effects of pre-compulsory education on labour market and educational outcomes (OLS estimates)

Subgroup effects of pre-compulsory education

As far as gender differences are concerned, although the effect of precompulsory education on average cognitive test scores at age 7 is significantly larger for boys than for girls (driven by a larger improvement in reading scores for boys than for girls at this age), we do not find any evidence that these differences in effects persist beyond age 7.

Turning next to differences by social class: contrary to much research from the US on the effectiveness of early years' education (see section II), we find little evidence that children from more disadvantaged backgrounds (proxied by having a father in social class IV or V) gained any more, in absolute terms, from pre- compulsory education than those from high social class backgrounds (I or II). However, we do find that pre-compulsory education reduced the probability that a child from a low social class background would need special education at age 7, whilst it did not have any effect on this probability for children from higher social class backgrounds. Further work will assess whether parental social class has an effect on relative outcomes: in other words, if we take into account that children from low social class backgrounds tend to start from a lower base of test score results, we may see bigger relative improvements in their scores compared with children from high social class backgrounds as a result of early educational experiences.

We find considerable evidence that the benefits of pre- compulsory education accrue in more instances to second-born and later children, rather than to first-borns (or only children). What negative effects there may be for example, on parental reports of social skills - appear often to affect first-borns rather than second-borns. For example, we find that the improvements in reading scores at age 7, and in the average, maths and reading scores at ages 11 and 16, are all significantly larger for later-borns than for first-borns. Indeed, for all but test scores at age 7, we cannot find significant effects of pre-compulsory education for first- borns at all, whilst we always find positive and significant effects for second- and later-born children in their test score results. By contrast, the effects of pre-compulsory education on parental reports of poor self-control skills at age 7 (see Table 7) are driven by the effects on first-born children; for second-borns and others, pre-compulsory education does not appear to have negative effects on self-control. At age 11, pre-compulsory education has a beneficial effect on teachers' reports of social adjustment for second-boras but not for first-boms (resulting in no significant effects across the population as a whole).

These differences in the effects of pre-compulsory education for firstborns and later-borns persist well into adulthood. We find positive effects on wages both at 33 and at 42 for second- and later- born children, but no effects at all for first-boms. These differences between first- and later-borns are found to be statistically significant at the 5 per cent level.

It is interesting to speculate as to why the effects are better for later-born children. It may be that first-borns - of whom 22 per cent are in fact only children - receive better-quality attention at home in the very early years of their lives, whilst later-borns have to share parental inputs with other siblings from the moment they are born. If this is the case, education before age 5 may play the role of substituting for parental care for later-borns, compensating them for an early deficit in parental inputs compared with first- borns. On the other hand, it may be that later-born children are able to gain more from pre-compulsory education because their older siblings spend time helping them and reinforcing what they learn - for example, practising reading. If this is the case, the home environment plays a more complementary role in early education for later-born children.

Age the child starts school

We have also considered whether there are any differences in the effects of pre-compulsory education depending on the age at which the child began nursery or early schooling.24 Table 9 shows the distribution of starting ages and the corresponding effects on test scores and wages compared with starting education at the statutory age of 5. Although, as discussed above, attending infant school or pre-school before the age of 5 has generally beneficial effects on cognitive and some labour market outcomes, we could not find much evidence for differences in these effects according to the age when the child started this pre-compulsory education. Most of the estimated effects - and all of those shown in the table - are not statistically different from one another, in the few cases where they are, these do not show a clear picture or story of an Optimal' age at which to start education in a formal setting. Specifically, we do not find any particular gain or penalty from starting at a very young age (although the albeit small wage gain from pre- compulsory education seems to be mostly driven by the gain for those who started two years or more before the compulsory age).

TABLE 9

Effects of starting age of pre-compulsory education (OLS estimates)

3. The effects of pre-school education

The next effects we consider are those of pre-school education, summarised in Table 10. The treatment group is made up of all children who had any nurseryor playgroup education before starting formal schooling. The control group consists of those who did not go to nursery or playgroup before starting school. Both treatment and control groups therefore contain children who started school early.

TABLE 10

Effects of pre-school education (OLS estimates)

Pre-school attendance leads to higher average test scores at age 7, driven in particular by a better performance in maths. Specifically, we find that at age 7, pre-school leads to an increase of 5 per cent of a standard deviation in average test scores (equivalent to the effects of an additional two years of maternal education) and to an increment of 8 per cent of a standard deviation in maths scores (OLS estimates).25 This evidence of positive effects on early cognitive performance is consistent with the findings from the EPPE study, which considers the effects of pre-school education on cognitive tests up to age 7 for a cohort of today's young children (see Section II).

However, unlike EPPE, we use a much older cohort, and are thus able to look at whether these effects persist beyond age 7. In contrast to precompulsory education more generally (i.e. also including early attendance at infant school as part of the treatment and only taking as comparison children those not starting infant school before 5), we find only weak, and somewhat mixed, evidence of any long-lasting effects on cognitive scores, having controlled for detailed child characteristics and background influences.

By age 11, we still detect an estimated increase of 4 per cent of a standard deviation in the average and reading test scores, with these effects significant at the 5 per cent and 10 per cent levels respectively. However, by age 16, there is only weak evidence of continued effects, with average and reading test results unaffected and an increase of 5 per cent of a standard deviation in maths scores, which is significant only at the 10 per cent level.

As to social skills, we do not find any evidence of an improvement in teacher-reported social skills for those attending pre-school. This contrasts with our results looking at pre- compulsory education more generally, where teachers' reports were favourable at age 7 for the treatment group (Table 7). Parental reports of the incidence of poor skills, particularly those related to aspects of the child's self-control, are, however, significantly higher at age 7 if the child attended pre-school. This apparently negative effect persists up to age 11, though at age 16 there no longer appear to be any significant effects.

There do not seem to be any long-lasting effects of attending pre- school on education and labour market outcomes (though we do detect a marginally significant positive effect on wages at age 33). This is consistent with the lack of persistence of effects on cognition and other skills that are known to contribute to academic achievement and labour market success.

It is perhaps not surprising that the estimated effects of attending preschool are smaller than those outlined in section V.2 looking at precompulsory education more generally. This is because members of both the treatment and the comparison groups may start their main schooling before the age of 5 here, whereas in the previous section the comparison group contained only those who did not attend any education before age 5. If entering formal education before age 5 is of any value, then the effects of pre-school education are likely to be smaller when children whose only early education was starting school before age 5 are included in both the treatment and comparison groups than when they are only included in the treatment group.

Subgroup effects of pre-school education

Unlike for pre-compulsory education more generally, we do not find differential effects on test scores for boys and girls from attending preschool. However, we do find some differential effects on social skills and longer-term labour market outcomes. In particular, we find that pre-school attendance may be more damaging to boys than to girls in terms of some measures of social skills: specifically, we find a higher probability of predelinquent/ delinquent behaviour at age 11, and a larger number of poor social skills reported by teachers at 16, for boys than for girls as a result of attending pre-school. Indeed, although pre-school appears to have negative effects on these outcomes for boys, it does not have any detectable effects for girls. By contrast, attending pre- school appears to have positive longer-term benefits for girls that we do not find for boys. For example, we find a significant effect of pre-school attendance on the probability of a girl obtaining a degree or other higher-education qualification as well as on wages at age 33 for women in employment;26 we do not find a similar effect for males.27

We also find some weak evidence that the benefits of pre-school education accrue in more instances to second-born and later children than to first-borns (or only children) and that what negative effects there may be on social skills affect first-borns more than second- and later-borns. However, this evidence is less strong than when we considered pre-compulsory education more generally. For example, we find that there is an improvement in maths test scores at 16 for later-borns attending pre-school, but not for first- borns. Pre-school is found to lead to increased parental reports of poor self-control at 7 for first-borns but not for later-borns.

We find little evidence of differential effectiveness of pre- school according to social class of the father. If anything, children from high social class backgrounds see greater benefits in terms of some test score results at some ages than children from low social class backgrounds. The EPPE study, too, did not uncover any consistent advantage from pre-scho

Source: Fiscal Studies

More News in this Category