Grammar schools in England: a new analysis of social segregation and academic outcomes

Abstract

The UK government is planning to increase the number of pupils attending state-funded selective grammar schools, claiming that this will assist overall standards, reduce the poverty attainment gap and so aid social mobility. Using the full 2015 cohort of pupils in England, this article shows how the pupils attending grammar schools are stratified in terms of chronic poverty, ethnicity, language, special educational needs and even precise age within their year group. This kind of clustering of relative advantage is potentially dangerous for society. The article derives measures of chronic poverty and local socio-economic status segregation between schools, and uses these to show that the results from grammar schools are no better than expected, once these differences are accounted for. There is no evidence base for a policy of increasing selection, and so there are implications for early selection policies worldwide. The UK government should consider phasing the existing selective schools out.

Keywords:

• Segregation
• selection
• educational disadvantage
• school effects
• grammar schools

Introduction

This is an article about the social composition of the pupil intakes to selective ‘grammar’ schools in England, and their progress in attainment in comparison to their peers not at grammar schools. The focus is on England, but there are implications for all such selective systems that seek to improve attainment for the most able students at the risk of reducing social cohesion in schools and beyond. Are grammar schools a good example of a meritocratic system in operation (Moore 1996)?

The authors are looking at different ways of estimating disadvantage in schools in England based on existing data-sets, creating new variables to encompass individual ‘trajectories’ of disadvantage, and applying these to analyses of school intakes and outcomes. For example, we take a variable such as whether a pupil is eligible for free school meals (FSM) in any year (or whether data are missing), and collate this for every year the pupil was in compulsory schooling. FSM is a measure of family poverty. The results can be used to create new variables, such as how many years a child has been FSM-eligible, both for the individual and for those they go to school with. We do the same thing with other background variables such as living or going to school in a deprived area, having a special educational need, having English as an additional language and ethnic classification. We have already shown that the variables used in our new approach can explain the apparent difference in school performance between the more deprived North East and less deprived South East of England. The claim that schools in the North East were doing worse with equivalent pupils was shown to be false because, by relying on flag variables such as whether a pupil was deemed FSM-eligible or not, their evidence did not take account of the duration of poverty concerned (Gorard 2016).

Policy

This article applies the same kind of approach to an analysis of the 163 selective state-funded grammar schools in England. Historically grammar schools were widespread in the United Kingdom, set up as part of a planned tri-partite system after 1944 which in practice became a two-tier system of grammar schools and secondary modern schools (the schools that took the majority of pupils not in grammar schools, with a slightly different curriculum). Children took a series of tests at age 10/11 (the 11+), and those with high scores were selected to attend grammar schools, with the remainder going to secondary modern schools.

The system has largely been abolished in Scotland and Wales, but retained in Northern Ireland, where it is made more segregated by having sectarian grammar schools for Catholic and Protestant families (Gallagher and Smith 2000). Early work in school effectiveness has suggested a consistent grammar school effect on attainment (Daly 1991). However, secondary-age educational outcomes in Northern Ireland are very variable in terms of academic performance, and pupils from deprived backgrounds or with special educational needs are under-represented in the grammar schools there (Borooah and Knox 2015). Internationally, many countries such as China and Singapore have highly selective schools within a national system, and some such as Germany and Austria retain an entirely selective tracked school system. This is linked adversely to the social mix of pupils in their schools (as discussed later), and so perhaps to social cohesion and inclusion.

In England, the number of grammar schools peaked at 1298 in 1964, and then dropped as low as 150 in 1989, before returning to the current level by 2004 (House of Commons Library 2017). The 163 remaining schools are disproportionately Academies and/or single-sex, and have a sixth form (for post-16 academic study). Their pupil intake has increased since 1980 (Bolton 2016). These schools are over-subscribed and popular with many local parents (Lloyds Bank 2016). At time of writing this is a topical issue, because the UK government campaigned in the 2017 General Election with a manifesto promise to change the law to allow more selective schools to be created, and to permit existing schools to become grammar schools under certain conditions.

In September 2016, the UK Prime Minister announced that the law banning state-funded schools in England (other than the 164 grammars in place in 1997) from using academic selection to allocate their pupil places will be removed (Foster, Long, and Roberts 2016). Instead, new schools such as free schools and academies would be able to become, in effect, new grammar schools. This was described by the Prime Minister and the Department for Education as a way to provide more good school places through ‘schools that work for everyone’.¹ They see it as a responsibility of the state to provide especially for pupils who are gifted and talented, and they feel that a comprehensive school environment cannot fully nurture the potential of such students. Since the 2017 General Election, although the unit at the Department for Education dedicated to planning the increase of grammar schools has been maintained, it is more likely that any immediate increase in selection will be sought by allowing the existing grammar schools to expand further by opening satellite schools (Allen-Kinross 2017). This policy does not require direct legislation, and £50 million has already been earmarked to fund it. The claims used to try and justify this change of policy to increase selection were that:

•	Pupils generally perform better at grammar schools than they do at non-selective schools.
•	The poorest children attending grammar schools do even better so that such schools actually reduce the poverty attainment gap and promote social mobility.
•	There is little or no harmful consequence for the other pupils in the rest of the schools.

This article considers these three claims using a national data-set for England, and a number of newly derived variables developed by the authors specifically for this kind of analysis. The article starts by looking at some of the prior evidence on selective schools, their effectiveness and implications for other types of school. The article then summarises the methods and data used, before presenting the most up-to-date evidence on each of the three claims, and ends by considering the implications for policy and practice in England and elsewhere.

Attainment at school and subsequent outcomes

Grammar schools use examinations to select children aged 10 or 11 who are predicted to do well in subsequent examinations at age 16. They select well, as evidenced by the high raw-score outcomes of these pupils five years later. This seems to confuse some commentators, members of the public and even policy-makers who assume that these good results are largely due to what happens in the school rather than the nature of the children selected. This is not a correct interpretation, as has long been pointed out in the sociology of education (Heyns 1974). A counterfactual is needed to tell us what would have happened to these children if they had attended a different school. This is often attempted by looking not at raw-score outcomes, but at the amount of progress made by each pupil while at the school (the ‘value-added’ model).

Some value-added studies comparing grammar schools with comprehensive or other types of school have suggested that the former have better pupil outcomes even once prior attainment is accounted for (Atkinson, Gregg, and McConnell 2006; Levaçić and Marsh 2007; Prais 2001). However, these studies also suggest that the subsequent lower attainment of the much larger number of pupils in the associated secondary modern schools at least outweighs any such gains. The system of selection is zero-sum at best. Intriguingly, these same value-added results appear if the model is still based on pupils attending grammar schools but uses their Key Stage 1 (KS1) results for prior attainment in standardised tests when aged seven and their Key Stage 2 (KS2) results when aged 11 as the outcome. This cannot be due to attending a grammar school because pupils only move to grammar schools after KS2, and so this odd result suggests that the purported grammar school effect is in fact a form of unmeasured pre-selection (Manning and Pischke 2006). This issue also exemplifies how hard it is to assess whether one school or type of school is genuinely more effective than another, even once the prior attainment and other characteristics of school intakes have been accounted for. Even if a difference is found in results, it may still be due to unknown or unconsidered prior differences in intakes. If the intakes to grammar schools really are already on a path to success based on their KS1 results, then that subsequent success at Key stage 4 (KS4) aged 16 must not be mistakenly attributed to having attended a grammar school in the meantime.

Anyway, such school effectiveness analyses are error-ridden (Gorard 2010), and very sensitive to assumptions about errors (Televantou et al. 2015). Their ensuing school ‘effects’ are small, volatile across years, inconsistent across different kinds of achievement (Marks 2015) and heavily dependent upon the precise model used (Darling-Hammond 2015). Therefore, published school performance measures based on value-added scores are likely to be profoundly misleading, particularly for those such as parents and policy-makers unfamiliar with the high level of uncertainty in the estimates for individual schools (Perry 2016).

What is really needed is a series of robust randomised evaluations (Gorard, See, and Siddiqui 2017). But allocating pupils to selective schools or not, at random, is not really feasible for quite good practical and ethical reasons. Instead, it should be possible to gain access to the 11+ scores and use a kind of regression discontinuity design, but these scores are not currently being made available to us.

A close alternative occurs when over-subscribed schools allocate places by lottery. Using such an approach, it has been shown that the lauded Charter schools in the United States are no better than other schools, and perhaps slightly worse (Clark et al. 2015). In fact, detailed analyses using as much data as possible have tended to show little or no substantive difference between the effectiveness of any types of schools within a national school system. Schools differ largely in terms of who attends them (Gorard and See 2013), and this seems to include grammar schools. Grammar schools in England did not confer a real advantage in the past and in their prime (Halsey and Gardner 1953; Halsey, Heath, and Ridge 1980), and they do not do so now (Coe et al. 2008; Sullivan et al. 2014). They do not increase social mobility in comparison to comprehensive schools, and do not assist working-class pupils with social class mobility, although the figures for income mobility are not so clear (Boliver and Swift 2011).

Segregation

The foregoing work seems to contradict, for the most part, the first argument (see earlier) for retaining and expanding grammar schools. What about the third argument, that there are no harmful consequences for others? Internationally, it is quite clear that the extent to which pupils are clustered together with others like them socially and ethnically as well as in terms of ability is much higher in countries with selective systems (Jenkins, Micklewright, and Schnepf 2008; OECD 2014). Such ‘segregation’ tends to be low in developed countries with little or no diversity of schooling such as those in Scandinavia, linked to low achievement gaps, higher average attainment and also a high percentage of very skilled students (Alegre and Ferrer 2010). Segregation tends to be high in countries with tracking or selection at a young age such as Germany, Austria, Belgium and Hungary.

In England, the still largely comprehensive system means that social, racial and economic segregation between schools is lower than in the latter group of countries (Gorard 2015a). But there is considerable variation between areas such as local authorities. The few authorities that have retained selection and grammar schools have the highest level of socio-economic status segregation in England. The correlation between FSM segregation and the number of grammar schools in any area is 0.62 (Gorard, Hordosy, and See 2013). In areas that have grammar schools, those living in the most disadvantaged parts are less likely to attend a grammar even where they have high prior attainment scores (Cribb, Sibieta, and Vignoles 2013). Of course, grammar schools are not the only kinds of schools in England (or elsewhere) that have heavily stratified intakes. Faith-based schools, fee-paying schools and even the most popular comprehensive schools can create as big a problem of skewed pupil intakes (Coe et al. 2008; Gorard 2015b).

There is also a clear link between relative attainment at school and the age of a pupil within their year group (Gorard 2015c). Younger children in each age cohort have lower test scores, worse non-cognitive skills on average, are rated weaker by their teachers and are less happy and more often bullied at school (Crawford, Dearden, and Greaves 2011). This is all worsened by selecting pupils by ability, as is done in grammar schools at the young age of 10 or 11 when their age in year matters more (Campbell 2014).

This segregation of pupils of different types between schools is not merely a question of who goes to school with whom. Segregation, whether racially or by religion or social class, may have alarming and dangerous consequences for the school system and for society more widely in the longer term.

Selective schools can make pre-existing inequalities worse by providing differential opportunities to learn (Schmidt et al. 2015), poorer instruction at school, less qualified teachers, substandard resources for the lower tracks (Harris and Williams 2012; Kalogrides and Loeb 2013) and altering teachers’ responses to children (Strand and Winston 2008). The kind of stratification created by grammar schools can widen the gap between privileged and not so privileged in terms of civic knowledge (Collado, Lomos, and Nicaise 2015), emotional and behavioural problems (Muller and Hofmann 2016), and even achievement in many studies (Condron 2013; Danhier and Martin 2014; Goldsmith 2011; Mendolia, Paloyo, and Wlaker 2016; Yeung and Phuong Nguyen-Hoang 2016). They can increase the direct impact of socio-economic status and low expectations (Parker et al. 2016), and affect relationships between pupils and teachers in the remaining non-selective schools (Vieluf et al. 2015), and between pupil peers, leading to poorer social skills (Gottfried 2015).

This all contradicts the third argument about no damage caused by having selection (see earlier). What is needed as a priority is more evidence about the second argument. Do grammar schools provide a specific advantage for poorer children?

Methods

The new research presented here is based on the National Pupil Database for England – specifically the 2015 KS4 cohort, with attainment, school and background information for every year that these pupils have been in compulsory schooling. There are 549,203 pupils with relatively complete records, of whom 75,787 (14%) are listed as eligible to receive FSM. There are 171,397 in local authority areas that contain at least one grammar school (defined here as selective areas), of which 22,402 attended a grammar school at KS4. The same analyses (in the following) have also been conducted with the 2014 and 2016 KS4 cohorts with the same substantive results.

The original pupil-level variables involved in the headline analyses are as follows:

•	School attended.
•	Local authority area.
•	Birth month and year – used to compute age in years.
•	Sex – girls tend to have better results than boys.
•	Ethnic origin or group.
•	English as an additional or second language.
•	Special needs with a statement.
•	Special needs without a statement.
•	Whether the pupil moved to the school in the last two years.
•	FSM-eligibility at KS4 – a flag variable showing whether a pupil is from a home officially classified as below the poverty line.
•	EverFSM6 – whether a pupil has been eligible for FSM in any of the past six years.
•	Index of Deprivation as a Child Index (IDACI) score – a measure of average deprivation for the area where the pupil lives or goes to school.
•	KS1 points score – attainment at age seven.
•	KS2 points score – attainment at age 11.
•	KS4 capped points score – attainment at age 16.

Pupil-level variables derived from the data are as follows:

•	The month of birth in the school year – to distinguish between summer and winter-born pupils.
•	The number of years in total a pupil was eligible for FSM up to KS4 – a more accurate measure of enduring poverty.
•	Whether a pupil goes to school in an area with grammar schools.
•	Whether a pupil goes to a grammar school.

A further school-level-derived variable is the segregation residual for FSM-eligibility (Gorard, Taylor, and Fitz 2003). This residual is the amount by which each school’s intake deviates from the national average. In this case, it is the difference between the number of FSM pupils in each school divided by the number of FSM pupils in England (or any area) and the number of all pupils in each school divided by the number of all pupils in England (or any area).

The data are considered for two geographical areas – all pupils in England, and those going to school in an area with grammar schools (a ‘selective area’). Some pupils will cross local authority boundaries in order to attend grammar schools, but to also define the areas they come from as ‘selective’ would make the situation less clear, without altering the substantive results. We will investigate local case studies of selective areas in a further article.

The real number variables (see earlier) are averaged for each area and for pupils attending grammar schools or not. The differences are converted into simple ‘effect’ sizes² by dividing them by the overall standard deviation for each variable. The categorical variables are used to create odds ratios – the number in each category in a grammar school times the number not in that category and not in a grammar school, divided by the number not in each category but in a grammar school times the number in each category not in a grammar school. This kind of analysis demonstrates the differences between the pupil intakes for grammar schools and the rest. These differences must be taken into account when considering the relative ‘effectiveness’ of grammar schools.

The effectiveness of grammar schools and grammar school areas is assessed via four regression models based on four different groups of pupils – all pupils in England, FSM-eligible pupils in England, all pupils in selective areas, and FSM-eligible pupils in selective areas. Each model has the same basic structure. The outcome variable to be explained is the KS4 attainment score for each pupil. The predictors for all models at the first stage are all of the other variables listed, except for the last two. These include prior attainment and the background characteristics of each pupil.

The first two models then add a second step with one more potential explanatory variable – whether the pupil goes to school in a selective area. The last two models do not use this variable, as they only concern pupils in selective areas. The last step for all four models also has one potential explanatory variable – whether the pupil attends a grammar school. In this way, the amount of variation explained at each stage, and the coefficients for the explanatory variables, provide an overall estimate of the possible impact of attending a grammar school or not, shorn of the known differences in the intakes to each type of school.

The data represent all pupils in state-maintained schools in England (for whom there is an official record). Therefore, issues such as statistical generalisation, clustered standard errors and significance testing are not relevant to any part of this article.

Results

The intakes to grammar schools

It is immediately apparent from Tables 1 and 2 that the characteristics of pupils attending grammar schools are markedly different in many ways from those attending other state-funded schools in England. Those attending grammar schools, on average, live in less deprived areas and are older in their year group, and even where they are FSM-eligible they will have been so for fewer years. The latter point, that even the duration of poverty of pupils officially defined as living in poverty is different, is particularly important and will be discussed further in terms of the attainment of grammar school pupils.

Table 1. Numeric characteristics of grammar school pupils compared to all pupils, England 2015.

	Grammar	Not grammar	Standard deviation	‘Effect’ size
Mean IDACI score	0.12	0.17	0.17	0.31
Mean birth month in the year	6.23	6.52	3.49	0.08
FSM years by KS2, all pupils	0.15	0.98	1.88	0.44
FSM years by KS4, all pupils	0.26	1.16	2.91	0.31
FSM years by KS2, FSM pupils	2.12	3.49	2.39	0.57
FSM years by KS4, FSM pupils	5.12	6.86	2.99	0.58

Table 2. Categorical characteristics of grammar school pupils compared to all pupils, England 2015.

	Grammar (%)	Not grammar (%)	Odds ratio
Non-White UK	25.1	19.3	1.40
White any	74.5	80.5	–
Black any	3.0	5.0	–
Pakistani/Bangladeshi	14.7	8.8	–
Chinese	1.5	0.3	–
Other/mixed	6.2	5.5	–
English as an additional language (EAL) students	11.3	14.0	0.78
Special educational needs (SEN) students with statement	0.3	3.9	0.07
SEN students no statement	5.3	12.7	0.38
EverFSM11	7.3	31.5	0.17
EverFSM6	5.3	25.6	0.16
FSM2015	2.4	14.3	0.15

Those attending grammar schools, on average, are less likely to be White UK or Black in ethnic origin, less likely to have English as an additional language, much less likely to report any special educational need, especially statemented ones, and are substantially less likely to be FSM-eligible at age 15.

Of course, some of these differences could be due to the kind of pupil populations in areas where the minority of 163 grammar schools remains, which could differ from the rest of the country. To assess this, Tables 3 and 4 compare the characteristics of pupils attending grammar schools with only those pupils in areas with grammar schools. These make it clear that the differences in Tables 1 and 2 are not produced by the geography of where grammar schools still exist. In fact, pupils in grammar schools are even less representative of their local areas than they are of pupils in England as a whole (the ‘effect’ sizes are the same or larger, especially for the IDACI scores, and the odds ratios are lower, especially for FSM-eligibility). In particular, even the few FSM-eligible pupils in grammar schools have been eligible for noticeably fewer years than in the rest of the school system (also see the following).

Table 3. Numeric characteristics of grammar school pupils compared to pupils in selective areas, 2015.

	Grammar	Not grammar	Standard deviation	‘Effect’ size
Mean IDACI score	0.12	0.22	0.16	0.61
Mean month in the year	6.23	6.54	3.48	0.09
FSM years by KS2, all pupils	0.15	0.97	1.81	0.45
FSM years by KS4, all pupils	0.26	1.60	2.81	0.48
FSM years by KS2, FSM pupils	2.12	3.45	2.40	0.55
FSM years by KS4, FSM pupils	5.12	6.81	3.00	0.56

Table 4. Categorical characteristics of grammar school pupils compared to pupils in selective areas, 2015.

	Grammar (%)	Not grammar (%)	Odds ratio
Non-White UK	25.1	18.1	1.52
White any	74.5	81.7	–
Black any	3.0	4.0	–
Pakistani/Bangladeshi	14.7	8.8	–
Chinese	1.5	0.3	–
Other/mixed	6.2	5.2	–
EAL students	11.3	12.8	0.87
SEN students with statement	0.3	4.4	0.07
SEN students no statement	5.3	13.0	0.37
EverFSM11	7.3	31.5	0.17
EverFSM6	5.3	25.7	0.16
FSM2015	2.4	14.4	0.15

Children aged 10 or 11 are put forward by their families and then tested for entry into grammar schools on the basis of their ability, prior attainment and motivation. However, this selection process indirectly also selects for a wide range of other social characteristics most of which should not be relevant. It is understandable that pupils with serious learning challenges will be less likely to pass an 11+ test of ability or attainment, even assuming it is a fair test in that respect. It is also understandable that children born later in the school year and those for whom English is not their first language might tend to do worse – although the test can and should make allowance for this. But it is harder to see why the family income, ethnic origin and precise area of residence should be so stratified. For the present, the key point is that those who go to grammar schools differ from the rest of the schools in England by far more than their talent as tested by the 11+. Therefore, grammar schools cannot yet be said to be obtaining better results with equivalent pupils even once that prior ability and attainment, as demonstrated in the 11+ test, is taken into account.

This also means that grammar schools and other schools in the same areas are much more segregated by any of these indicators of possible disadvantage than the rest of England is. For example, the segregation residual in terms of FSM-eligibility in 2015 was 0.00002 averaged across all schools, whereas it was double that (0.00004) in areas with grammar schools and over 10 times that (0.00022) in grammar schools themselves. However good grammar schools are (or not), this must be set against the real dangers from such a deliberate policy of socio-economic segregation between schools (as described earlier).

The outcomes from grammar schools

The KS4 attainment of pupils in selective areas and the rest of England is very similar. The mere existence of grammar schools in an area does not seem to drive up standards or reduce the gap between FSM-eligible pupils and the rest (Table 5). The ‘effect’ sizes listed are a summary of the extent to which FSM-eligible pupils, in each row, obtain lower results than none-FSM pupils. As would be expected, because they are selected by ability, pupils in grammar schools have higher than average attainment at KS4, showing that the 11+ test is reasonably good at selecting those who will do well five years later.

Table 5. Comparison of KS4 attainment (total capped points) by FSM, England 2015.

	Not FSM-eligible	EverFSM6	Standard deviation	‘Effect’ size
England	328	260	98	−0.69
Selective areas	334	259	99	−0.76
Grammar schools	413	395	44	−0.41

However, Table 5 also shows that FSM-eligible pupils, using the official Department for Education measure of EverFSM6, have better KS4 scores in grammar schools to such an extent that the ‘effect’ size or poverty attainment gap is noticeably lower there. This is the basis of the second argument for grammar schools – that they aid social mobility by encouraging poorer children to levels of attainment more similar to their better-off peers. Table 5 also shows that pupils in selective areas not attending grammar schools have a higher poverty gradient than in other parts of England, suggesting, as earlier, that whatever good grammar schools might do for those who attend them, this is at least negated by the harm done to those who do not attend.

Another way of looking at this can be seen in Table 6. The link between prior attainment (KS1) and results aged 16 (KS4) is considerably lower in grammar schools than in schools more generally. This is presumably at least partly because they have a much smaller range of prior attainment (being selective). This will be addressed in the regression models that follow. But the link between KS4 results and FSM-eligibility in grammar schools is also lower. Again, this could partly be because very few pupils in grammar schools have been FSM-eligible for their entire school career. For example, in 2015 only 86 pupils in all of the grammar schools in England (0.4%) had been FSM-eligible for nine of the previous years. The figure for selective areas is 6249 pupils (3.6%), and for England is 22,143 pupils (4%). Grammar schools not only take very few FSM pupils (see earlier), they also only take the less chronically poor even among those few (here by less than a tenth of their fair share). One consequence is that the correlations between poverty, prior attainment and KS4 results are higher in selective areas than they are in England on average, despite the lower correlations in grammar schools. The differences may appear small but it must be recalled that the number of pupils in selective areas not in grammar schools will be around 5–10 times as many as the number in grammar schools. If there is any advantage from grammar schools for disadvantaged pupils in the models that follow it would tend to be zero-sum.

Table 6. Correlation between KS4 attainment, KS1 and number of years FSM-eligible, 2015.

KS4 attainment points	Years FSM-eligible	KS1 attainment points
England all pupils	−0.31	0.68
England FSM pupils	−0.15	0.51
Selective areas all pupils	−0.33	0.70
Selective areas FSM pupils	−0.16	0.63
Grammar schools all pupils	−0.08	0.34
Grammar schools FSM pupils	−0.01	0.23

The relevance of how many years a pupil has been FSM-eligible (i.e. living in poverty) is illustrated well in Figure 1. Pupils who have never been known to be eligible for FSM (on the far left) have a much higher level of KS4 attainment than any pupil with even one year of FSM-eligibility. This is well known. Less well known is the fact that, on average, KS4 attainment declines with every year of FSM-eligibility. This matters because grammar schools are not only taking just a fraction of their fair share of FSM pupils; the few FSM pupils they do take are disproportionately those towards the left of Figure 1 who are likely to do better anyway whatever school they attend. This means that the other schools in selective areas are not only taking more than their fair share of FSM pupils, but are also disproportionately dealing with the more chronically poor in their areas.

Figure 1. Crossplot of the number of years any pupil is FSM-eligible and their average KS4 point score, all schools in England 2015.

In order to take all of this into account, we ran four multivariate regression models and the headline results are presented in Table 7. The R value using all of the pupil background and prior attainment variables listed earlier in the article, for all pupils in England, was 0.81 (66% of total variation in KS4 outcomes explained). This was about the same for only those pupils in selective areas (0.82 or 68%) and slightly lower when considering the FSM-eligible subset of pupils in either group (0.72 or 59%). This is similar to the usual educational effectiveness models. Additional background variables could improve this somewhat, but there will always be an error term due to missing data, measurement errors and so on (Gorard 2010). The first model in each category is slightly better than it would be traditionally due to knowledge of exactly how many years each pupil has been FSM-eligible rather than using the traditional binary classification (and this approach will be refined further and applied to other explanatory variables in future articles). If the model is run with ‘current’ FSM-eligibility (for 2015) at KS4 (as is standard) rather than the number of years eligible, the R value for the first model drops from 0.81 to 0.77, and all other models have proportionately lower R values. Although this difference is small it does suggest that the new measure of chronic poverty is picking up variation that neither current FSM nor EverFSM6 status does.

Table 7. R values from multi-stage regression models predicting total capped KS4 points, England 2015.

	All pupils	FSM pupils	Pupils in selective area	FSM pupils in selective area
Background predictors	0.81	0.77	0.82	0.77
Whether in selective area	0.81	0.77	–	–
Whether in grammar school	0.81	0.77	0.82	0.77

None of the models improves at all when knowledge is added of whether a pupil goes to school in a selective area or not. This means that if grammar schools are at all differentially effective, their effect is indeed zero-sum and wiped out by exactly equivalent harm done to the rest of the nearby school system. However, adding knowledge of whether a pupil goes to a grammar school also improves none of the four models at all. With only 163 grammar schools, it could be argued that they would not be expected to add much to the full model for all pupils in England. But they do not add anything to the smaller model restricted only to areas with grammar schools either. On this basis, grammar schools appear to be no more or less effective than other schools, once their clear difference in intake has been taken into account. And this is true both for FSM-eligible and non-FSM-eligible pupils.

Table 8 makes much the same point, by presenting the standardised coefficients (equivalent to ‘effect’ sizes) for each variable in the four models. Prior attainment at KS2 is by far the best predictor of KS4 outcomes, followed by prior attainment at KS1, the number of years eligible for FSM and whether a pupil has any kind of special educational need. The least important variables, with almost negligible ‘effect’ sizes, are the date of birth in the school year, the level of deprivation in the area of residence, whether the school is in a selective area and whether it is a grammar school. However, the level of deprivation would increase in importance if FSM-eligibility was not available because the two are correlated. Similarly, the month of birth in the school year appears far less important than in reality because the prior attainment scores are acting as proxies to a great extent (younger pupils do less well at KS1, KS2 and KS4).

Table 8. Standardised coefficients from four multi-stage regression models predicting total capped KS4 points, England 2015.

	All pupils	FSM pupils	Pupils in selective area	FSM pupils in selective area
KS2 average points (prior attainment)	0.57	0.50	0.55	0.49
KS1 average points (prior attainment)	0.10	0.10	0.11	0.11
Sex of pupil	0.08	0.07	0.07	0.08
Month in the year (summer born)	0.04	0.03	0.04	0.03
Number of years known to be FSM-eligible	−0.10	−0.06	−0.10	−0.06
IDACI scores (deprivation)	−0.05	−0.02	−0.05	−0.03
Special need (SEN)	−0.10	−0.14	−0.10	−0.14
Mean number of years FSM-eligible, school	−0.05	−0.06	−0.07	−0.08
Joined school in last two years (mobility)	−0.08	−0.10	−0.07	−0.09
English as an additional language	0.08	0.11	0.06	0.09
Non-White UK (ethnic minority)	0.06	0.09	0.05	0.08
Schooled in a selective area	−0.01	−0.01	–	–
Attends grammar school	0.04	0.03	0.07	0.06

Conclusions

As stated at the outset, ideally a randomised control would allocate pupils to selective schools or not in order to assess whether grammar schools confer any advantage for attainment. An alternative would be to use regression discontinuity based on 11+ scores. Such powerful designs are based on two fair groups for comparison, and this makes missing variables irrelevant because they can be assumed to be unbiased between the two groups. All weaker designs, such as here or where pupils in grammar schools are matched with those in non-selective areas, face the problem that those who attend grammar schools may differ by more than the surface variables that are taken into account in the analysis. For example, those applying to grammar school may be more motivated to succeed already or have parents who are more engaged in their education. Such missing values could create an illusory ‘grammar school effect’.

The new variable created for this analysis – the number of years each pupil has been eligible for FSM by KS4 – explains at least part of the difference between grammar schools results and those of other schools, over and above the obvious difference in prior attainment. Not only do grammar schools in England take only a tiny proportion of pupils who are or have ever been eligible for FSM, but those they do take have been eligible for fewer years. This, and the use of segregation figures, age in the school year and other derived variables, makes the presented re-analysis of the grammar school ‘effect’ novel. It may also help explain the substantive difference in findings between this study and earlier work using the less complete data available at that time, and without considerations of levels of enduring poverty (for example, Schagen and Schagen 2005).

Every grammar school creates a much larger number of schools around it that cannot be comprehensive in intake, of necessity, because they are denied a supply of so many of the most high-attaining children. It does not matter whether these schools are officially designated as ‘secondary modern’ schools, or whether they offer different curricula or not. Also, the two sets of co-existing schools will and must differ in terms of a whole set of other pupil characteristics that appear to be related to selection at the 11+ but are not directly selected for. These include special educational needs, first language and, less obviously, ethnicity and poverty. The secondary modern schools also receive less funding per pupil within their local authorities than do grammar schools (Levaçić and Marsh 2007). Policy-makers and unthinking advocates always focus on grammar schools, whereas an equivalent claim to ‘we must have more grammar schools’ would be ‘we must have a lot more schools that are deprived of the most talented 15–20% of pupils in their catchment areas’. In areas with selective schools, the system is a clear driver towards increased social and economic segregation between schools, and all of the dangers that this entails – such as lower self-esteem and aspiration, poorer role models, poorer relationships and a distorted sense of justice (Gorard and See 2013).

On the basis of the prior evidence and the new analyses presented here, the policy of selective schools has little to recommend it. Dividing children into the most able and the rest from an early age does not appear to lead to better results for either group, even for the most disadvantaged. This means that the kind of social segregation experienced by children and young people in selective areas of the United Kingdom, and in selective schools and countries around the world, is for no clear gain. The United Kingdom, among others, has spent 20 years or more including pupils with special educational needs and disabilities into mainstream schooling. The argument was that whatever specific provision was needed could be provided in most mainstream schools, they could be taught separately for some tasks, and that it was better to allow them to socialise and gain educational experiences with their non-SEN peers. It is hard to see why exactly the same argument does not apply to the most able students in each area.

Put together, the findings mean that grammar schools in England endanger social cohesion for no clear improvement in overall results. The policy is a bad one and, far from increasing selection, the evidence-informed way forwards would be to phase out the existing 163 grammar schools in England. This is not to decry the schools that are currently grammars, or the work of their staff. But, overall, they are simply no better or worse than the other schools in England once their selected and privileged intake is accounted for. There is no reason for them to exist. Nor do the findings of this article and others mean that schools should not use any form of internal ‘setting’ by ability in any phase. In the absence of grammars it is still currently possible for high-ability pupils to be taught together for at least some of the week (and whether that is an effective approach would be another article). But the richer and more able pupils currently in grammar schools would then mix with a wider range of peers, especially those with learning challenges, in all classes and years that did not use setting, all vertical structures such as ‘houses’, and in sports, play and extra-curricular activities. The importance of inclusion is not just about those with learning difficulties or disabilities; it can also be about including the most able and advantaged at age 11 in the general school mix.

There are other factors leading to stratified entries to schools. Some of these, such as changes in the local economy, transport in rural areas and residential segregation, are beyond a quick policy or educational fix (Gorard 2018). But some, such as the continued existence (and planned expansion) of schools that are selective by religion, could be abolished at a stroke. Any improvement in one is likely to have a beneficial impact on the others (via the ‘Belfast model’; Gorard, Taylor, and Fitz 2003), but none of these further problems is an argument for making the stratification worse by retaining or expanding grammar schools.

Disclosure statement

No potential conflict of interest was reported by the authors.

Funding

The work reported here was funded by Economic and Social Research Council (ESRC) [grant ES/N012046/1].

Notes

1. See Accessed April 2017. https://engage.number10.gov.uk/good-school-places/.

2. The term ‘effect’ size is in inverted commas throughout because although this is the standard term for this index of average difference it should not be read to imply any sort of causal relationship.