Academies 2: the new batch the changing nature of academy schools in England

The English education system has undergone large-scale restructuring through the introduction of academy schools. The most salient feature of these schools is that, despite remaining part of the state sector, they operate with more autonomy than the predecessors they replaced. Two distinct time periods of academy school introduction have taken place, under the auspices of different governments. The first batch was initiated in the 2002/03 school year by the Labour government of the time, and was a school improvement programme directly aimed at turning around badly performing schools. The second batch involved a mass academisation process following the change of government in May 2010 and the Academies Act of that year, which resulted in increased heterogeneity of new academies. This paper compares the two batches of introduction with the aim of getting a better understanding of their similarities and differences, and their importance for education policy. To do so, we study what types of schools were more likely to change to academy status in the two programmes, and the impact of this change on the quality of new pupil enrolments into the new types of school. Whilst we do point out some similarities, these are the exception rather than the norm. For the most part, our analysis reveals a number of marked dissimilarities between the two programmes, in terms of both the characteristics of schools that become academies and the changes in pupil intakes that occurred post-conversion.


Introduction
A feature of the evolving nature of education systems in a number of countries has been the introduction of school reforms operating through the creation of new school types. The best known recent examples of these are free schools ("friskolor") in Sweden, charter schools in the United States and academy schools in England. These new school types have mostly been introduced from a premise that the pre-existing education system was not delivering high enough standards for children. Whilst there are notable differences between them (in particular whether they are brand new schools or conversions from existing schools), they all represent movements away from what can be thought of as the traditional, state controlled, local or community school. More specifically, these new schools are typically decentralised from local authority/municipality/school district control, and thus have more operational autonomy. This paper focuses on the academy schools of England. The introduction of academies, which began in the early 2000s, is turning out to be one of the most radical and encompassing programmes of school reform seen in the recent past amongst advanced countries. Unlike traditional community schools, academies are autonomous, state-funded schools which are managed and run outside of local authority control. In almost all cases, they are conversions of already existing predecessor schools that inherit pupils already enrolled in the school. In essence, academies are school 'takeovers' which, because of their nature, enable schools to operate with considerably more autonomy and strategic independence than they did in their The newly elected Conservative-Liberal Democrat Coalition quickly ushered in the 2010 Academies Act, a legislative change that widened the academies' remit. Prior to the Act, only secondary schools could become academies, and in order to convert they were required to sign up a sponsor. After the Academies Act, there was more heterogeneity as primaries were also 1 They are different from most US charter schools which are typically set up from scratch. A closer comparison to the typical charter school in England is the free schools, a recent addition to the education landscape. These are new schools often set up by parent or community groups. Conversely, the closer US comparison to English academies would be 'in-district' charters where an already existing public school is converted to a charter as a school takeover. Although these are less commonplace than US charters as a whole, there are places where conversions of public schools to charters have taken place (like Boston and New Orleanssee Abdulkadiroglu et al., 2016). 2 In England, secondary schooling takes place from ages 11-12 to 15-16 (years 7 to 11) and primary from ages 4-5 to 10-11 (years 1 to 6). allowed to become academies, and a sponsor was no longer required for conversion to take place.
In fact, the introduction of the Act dramatically changed the aim of the programme. While the Labour Government introduced academies as a remedial programme aimed at replacing failing schools, the Coalition Government expanded the programme as part of a project aimed at introducing greater autonomy and competition into the state school sector. While new Labourtype 'sponsored' academies have continued to open, most of the expansion of the academy sector has come from 'converter' academies: high-performing schools that change their institutional arrangements to acquire greater freedom in terms of budgeting, hiring of staff, pay and working condition negotiations, performance management, taught curriculum, and length of the school day. To put things in perspective, 'converters' represent more than 80% of the expansion of the academy sector in the past five years.
Since May 2010, the scale and scope of the academy programme has experienced further developments. In March 2016, the new Conservative Government announced that all schools would have to become, or be in the process of becoming, academies by 2022. However, following strong resistance from parents, teachers' associations and local councilsas well as some politiciansthe plan was officially abandoned in October 2016. Nevertheless, it is still the case that the government aspires to have as many schools as possible to voluntarily gain academy status, although the model is now 'encouragement' rather than 'compulsion'.
Notwithstanding these recent developments, the stark differences in the purposes of the academies programme before and after May 2010 have remained. This aim of this paper is to compare and contrast these two (pre-and post-May 2010) programmes. One reason why this investigation is needed is the continued extrapolation of findings from the first programme to the post-May 2010 conversions, by policymakers, media commentators andin some casesacademics. The extent to which this extrapolation is valid depends on the extent to which the two programmesand, more precisely, the schools becoming academies before/after May 20010are alike.
To uncover what similarities and differences characterise the two regimes, we undertake an empirical study of academy conversions that took place between the academic years 2002/2003 and 2012/2013, with two particular research questions in mind. Firstly, we study the characteristics of secondary schools that gain academy status. It is well documented that the pre-2010 conversions were under-achieving schools usually serving inner-city disadvantaged neighbourhoods. A general perception, as well as the nature of the Coalition Government's academy programme, suggests that this is not true for the new batch of academies. In this paper, we present rigorous empirical evidence on the validity or otherwise of this perception. 3 The second focus is on how academy conversion impacts on a school's pupil intake and to what extent this differs between the two regimes. Early evidence (Eyles and Machin, 2015;Machin and Vernoit, 2011;Machin and Wilson, 2008;Wilson, 2011) suggests that the Labour academies enrol pupils with higher end of primary school test scores and other advantageous background characteristics (e.g. fewer pupils eligible for free meals) post-conversion. We therefore consider how similar or different this trend has been for the pre-and post-May 2010 academy schools.
To preview our findings, we find that the programmes are significantly different in terms of the school characteristics that are associated with conversion; in particular, school average Key Stage 4 (end of secondary schooling) attainment, incidence of free-school meal eligible pupils (a proxy for low family income) and inspectorate (OFSTED) ratings. Schools converting under the two policy time periods are also different in terms of the quality of new enrolments to the school as measured by pupils' prior Key Stage 2 (end-of-primary) achievement and eligibility for free school meals. This clearly shows that the programmes are not prima facie comparable; and therefore that findings from the earlier conversions cannot be meaningfully extrapolated to later conversions. That said, we also highlight some commonalities suggesting that a more careful and nuanced comparison could in some circumstances be undertaken. We discuss this later in more detail, but one clear instance is that a comparison of Labour-type 'sponsored' academies in both regimes may be a relatively legitimate comparison to draw. Conversely, any extrapolation from 'sponsored' academies to Coalition 'converters' would be a distorted comparison likely to lead to biased conclusions.
Our evidence contributes to a small, but growing literature on the nature and effects of academy schools. The first round of studies focussed on a small number of early Labour academies and investigated the effect of conversion on pupil achievement at the end of secondary schooling (see Machin and Wilson, 2008;Machin and Vernoit, 2011;Price Waterhouse Coopers, 2008;and National Audit Office, 2010). More recently, Eyles and Machin (2015) conducted a comprehensive causal evaluation of the Labour academies programme studying both intake and end-of-secondary school performance effects, while  investigate a longer-term post-compulsory schooling impact of academies (mainly studying students' university enrolment). The authors found significant changes in the ability 3 Note that we focus on secondary schools as there were not primary academies during the Labour period. See Eyles, Machin and McNally (2016) for a study of primary school academisation post-2010. composition of pupils enrolling at academies after conversion. This result echoes the findings of Wilson (2011). However, no empirical investigation has been undertaken to quantitatively assess the (dis)similarities between the Labour and Coalition Government academies. Our study fills this gap.
Our work also contributes to a larger literature on the effect of different school types on pupil enrolment and performance. While research on the impact of attending different types of schools is extensive (Neal, 1997;Gibbons and Silva, 2011), there is much less evidence on what happens when the type of school attended by pupils changes. One study closely related to this paper, albeit from an earlier time period, is the analysis of grant-maintained (GM) schools by Clark (2009) in England in the late 1980s-early 1990s. 4 Clark documented significant test score improvements, but provided limited evidence on changes in pupil intake.
In the US the growing body of work on charter schools is clearly related to our work because charters have similarities to academies even though most charters are new schools while academies are not (see Abdulkadiroglu et al, 2011;Hoxby and Murarka, 2009;Angrist et al, 2010;and Dobbie and Fryer, 2011). A smaller number of studies have focused on the less frequent conversions of already existing schools to charters (Abdulkadiroglu et al., 2016), or the introduction of charter-like practices to US public schools (Fryer, 2014). Most of these studies analysed the causal effect of charter attendance on students' performance by exploiting the use of lotteries by oversubscribed schools; therefore, they cannot provide (by construction) evidence on the effect of charter conversion on pupil intake. Finally, the evidence on Swedish free schools is also related to our work (Ahlin, 2003, Björklund et al., 2005, and Sandström and Bergström, 2005; in particular, the analysis by Bohlmark and Lindahl (2007) who show that free schools tend to enrol pupils from better off families and affect the composition of pupils in public sector schools.
Although the background of the academy policy differs from both the US charters and the Swedish free school programme, we believe our work sheds light on related issues about school reform that are of general interest. Firstly, we provide evidence on how the characteristics of pupils and schools change as the impetus of the programme changes from addressing underperformance to providing more autonomy to all schools. Secondly, we provide such evidence in a setting where expansion of a more autonomous school sector has not happened at the margin, but has taken centre stage and become a reality for more than 60% of secondary schools over a six-year period. These features make the on-going English academy reform a truly exceptional experiment in terms of its scale and celerity. Given the growing trend towards more decentralised and autonomous education systems, our work is relevant for academics and policy makers alike.
The remainder of the paper is structured as follows. In the next section, we discuss the institutional context and the data we use in our empirical analysis. In Section 3 we spell out in more detail the methods that we use. Section 4 presents our results on the association between school characteristics and the probability of converting to academy, while Section 5 discusses our results on changes in pupil intake. Finally, Section 6 provides conclusions.

The English school system
Compulsory education in England is organised into Key Stages (KS) that stretch over the course of primary and secondary education. In terms of primary schooling, pupils enter school at age 4-5. This is known as the Foundation Stage. They then move on to KS1 (up until age 6-7), before finishing at KS2. The latter covers the last four years of primary schooling (completing at ages 10-11), and ends with the externally-marked KS2 standardized national tests (SATs) where proficiency in English, Mathematics and Science is assessed.
Secondary schooling covers ages 11-12 through to ages 15-16 and is divided into KS3 (the three years up to age 13-14) and KS4 (ages 14-15 and 15-16), which marks the end of compulsory education. At the end of KS4, pupils sit externally-marked academic (GCSEs) and/or vocational (NVQ/BTEC) tests in a range of subjects, although, English, Mathematics and Science are compulsory for every student at this stage. In our analysis, we focus on secondary schooling because this is where the majority of conversions to academies has taken place and because there were no primary school academies prior to May 2010.
For each of the Key Stages the central government sets learning targets (levels) and runs a national curriculum. Average attainment within a school at the various Key Stages as well as measures of school value added are published alongside other school characteristics (such as size and composition) in annual performance tables. These are highly salient in the education and policy debate and the media. They are also routinely used by parents when choosing a school for their children.
Despite the principle of 'free choice' guiding the process of school applications and admissions practices, parental freedom to choose is limited when schools are over-subscribed.
When this occurs, various admissions criteria are used to prioritise students, usually favouring those who live nearby, those with special educational needs or in care of the Local Authority (LA), and those with siblings in the school. Religious schools can prioritise students according to their faith, while a small proportion of secondary schools select on admission tests (Grammar schools).
The institutional arrangements of secondary schools determine more precisely whether the school, as opposed to the LA, holds responsibility for admissions and the margin of freedom the school enjoys over taught curriculum, personnel management and strategic decision making.
Secondary schools can take one of the following alternative structures: community schools, voluntary-controlled (VC) schools, foundation schools, voluntary-aided (VA) schools and, since their introduction in the early 2000s, academy schools. 5 Although academies are the focus of our work, we briefly discuss the structures of other school types for comparison. 6 Community and VC schools are mainly managed by the LA and by a governing body predominantly composed of members of staff and LA representatives. Since the majority of VC schools are religiously denominated, their governing body also includes members of the foundation supporting the school. Responsibility for recruiting, human resources decisions and admissions is in the hands of the LEA. As a result, these schools are characterised by very little autonomy.
VA and foundation schools are similarly structured even though voluntary aided schools are mainly religiously denominated while foundation schools are not. These two types of school are run as partnerships between the state and voluntary sector, and their associated foundations have a significant representation on their governing bodies. These institutional arrangements grant VA and foundation schools more autonomy than community and VC institutions.
Lastly, academy schools enjoy a much larger degree of autonomy than any of the other school types in the state system. Academies were gradually introduced by the Labour Government in September 2002 with the aim of replacing failing schools. The programme was designed as a targeted intervention aimed at addressing underperformance by changing organizational structure, providing schools with more autonomy and linking schools to Government-approved sponsors enlisted to drive educational improvements. Indeed, sponsorsusually educational charities or business groupswere seen as crucial in bringing to schools a more entrepreneurial 'ethos'.
The election of the Conservative-Liberal Democrat Coalition Government in May 2010 dramatically changed the aims of the programme. The Academies Act of June 2010 allowed as 5 Any of these school types can be a selective 'Grammar' and admit pupils on the basis of entry tests. 6 One final school type was the Community Technology College (CTC), which enjoyed relative autonomy from the LA. In particular, the majority of their governing body was filled with representatives of the sponsorusually a business or voluntary groupand the school had control over staffing and admission decisions. However, CTCs followed the national teaching curriculum and were characterised by a strong emphasis on technological, scientific and practical subjects. CTCs represented a very small share of the secondary education sector and nearly all have converted to academies, so we do not consider this group in our analysis. many schools as possibleincluding primaries -to 'convert' to academies and removed the requirement that would-be academies had to find a sponsor. Although a number of new Labourtype 'sponsored' academies have opened since, more than 80% of the expansion of the academy sector has happened through Coalition-type 'converters'. Importantly, in the new regime, greater emphasis is put on the voluntary process of conversion to academy status. While schools classed as 'eligible for intervention' can still have sponsored academy status imposed upon them, underperforming (rather than failing) schools can now voluntarily enter in discussions with sponsorsmainly large well performing chains (e.g. the Ark or Harris networks)and become academies under their aegis in order to address low standards. On the other hand, all that is required from would-be 'converters' is that they discuss the possibility of converting with 'such persons as they think appropriate'; typically, this includes parents, members of staff and pupils, and the foundation or religious body backing the school (if any). They must then register their intention to apply with the Department for Education (DfE) and send in information about school attainment, pupil progress and school finances for the past three years, before finally providing the most recent school report prepared by the school inspectorate (OFSTED).
The latter is a key determinant of approval for conversion. OFSTED visits schools every three to five years (although this frequency depends on school performance, with poor performers visited more regularly) and inspections result in publicly available reports rating schools from 'Outstanding' to 'Inadequate' on their overall quality as well as on specific aspects such as teaching, management and pupil behaviour. 7 Originally, schools with 'Outstanding' overall rating were fast-tracked and pre-approved for conversion by the DfE. However, as the Coalition and then Conservative Governments continued their push to expand the programme, schools with 'Good, with Outstanding Features' rating as well as schools displaying upward trajectories in attainments and quality of management/teaching practices were subsequently allowed to convert. Broadly speaking, the process of (voluntary) conversion can take between three and five months, though the time lapse can expand and the exact timing depends on specific aspects of the proposed conversion and the volume of new applications. Despite these differences, both 'sponsored' and 'converter' academies enjoy similarly wide margins of autonomy. Although they remain part of the state sector, they fall outside the control of the LA in terms of key strategic decisions and day-to-day management. This is administered by the head-teacher and a self-appointed board of governors with a limited number of representatives from the LA. This body has responsibility for hiring the staff, negotiating pay 7 More details on OFSTED, its inspections and the possible outcomes are provided in Eyles and Machin (2015). and working conditions, and deciding on matters such as career development, discipline and performance management. Furthermore, academies enjoy more autonomy in terms of the majority of the taught curriculum (except for English, Maths, Science and IT where they have to follow the national curriculum), as well as of the structure and length of the school day.
According to DfE (2014) survey of academies' freedoms, 63% of secondary academies have changed the performance management system for their teachers following conversion; 60% have changed the offered curriculum; and 51% have changed school leadership. On the other hand, only 10% and 6% have increased the length of the school day or school term respectivelymainly because of opposition from teachers. Interestingly, the majority of schools report that they converted to academy in order to 'gain autonomy to use funding as they see fit' and 'raise educational standards', and approximately 70% claim that the changes introduced following conversion either moderately or substantially improved attainments.

Data construction and sample selection
The main data source used in our analysis is the National Pupil Database (NPD). The NPD is obtained by linking several centrally-gathered datasets that collect information on the characteristics of all pupils in state-maintained primary and secondary schools, their attainment at the various Key Stages, and the characteristics of the schools they attend. These databases are used by the DfE to assess schools' performance and funding needs and, as such, are highly reliable and contain very few missing observations. A core element of the NPD is the Pupil Level Annual Census data (PLASC) which contains information on the characteristics of pupils in each year of compulsory schooling. These data comprise both time varying characteristicssuch as free school meal eligibility status (a proxy for low parental income), special education needs status, and the identity of the school attended in each yearalongside time invariant factors such as ethnicity. The data also contain the school attended by students in every year of their schooling. The school attended in year 7 is a crucial piece of information since it allows us to identify the school where pupils start their secondary education and evaluate any effect of academy conversion on pupil intake.
Since its inception in 2002, PLASC has been collected each January; in addition, there have been collections in May and October since 2006. We use the January collection because this contains all relevant pupil level information and is available in all the years that we study.
For our analysis, we also gather data on pupils' KS2 and KS4 attainment. For KS2, we focus on test scores across the three compulsory subjects assessed at this stage: English, Math and Science. Test scores in Science span the 0-80 range (coming from the sum of two separate papers each marked out of 40), while the English and Math tests are marked out of 100 (both composed of the sum of two separate test scores, each marked out of 50). We use KS2 test scores averaged across the three subjects. 8 At KS4, we consider attainment in the more academic GCSE (General Certificate of Secondary Education) and the vocationally-oriented GNVQs and BTECs. GCSEs are graded A*-G and the points score calculations give an A* a score of 8 and a G a score of 1 with grades in between going up in increments of 1.
GNVQ/BTEC are converted into GCSE equivalents before being assigned a point score. The point measure used in the analysis refers to the grades obtained in the best eight qualifications.
Capping qualifications at the best eight allows us to deal with the possibility that pupils entering many examsbut performing poorly on all of themwould still appear to do well. However, this restriction is inconsequential for our results.
Finally, the NPD also collects information at the school levelin particular data on school types and institutional arrangementsthrough the Annual School Census and Edubase. We use these data in conjunction with files provided by the DfE, which give information on all academies that have opened (or are in the process of doing so), to identify schools that become academies, their time and mode of conversion ('sponsored' or 'converter' route), and the predecessor school that underwent transformation.
We use this information to construct two datasets. The first one is a school-level panel of secondary schools covering the period spanning the school years 2002/2003 to 2012/2013. This dataset includes all schools meeting our selection criteria (explained below) and comprises of an indicator for whether the school is an academy in a given academic year alongside a rich set of school level characteristics measured in the previous year. We use this dataset to study which school characteristics are associated with conversion to academy status and investigate whether these associations change from the Labour Government years to the Coalition Government period. For our analysis, we only retain secondary schools that cover the whole of the Key Stage Table 1 reports the cumulative number of open academies in our sample of secondary schools over the years covered by our investigation. 10 We define a school being as open as an academy in a given year if it was an academy in or before September of the previous year. For instance, a school that converted in September 2011 would be classed as being an academy from the 2011/12 academic year whereas a school opening in October would be classed as opening in 2012/13. A schools' first year of treatment is thus the first full academic year that it begins with academy status. We studied the extent to which our results are robust to this definition.
Overall, academies went from representing just a few exceptional cases in the early 2000s to covering more than half of secondary schools by 2016 (Column 1). There are, of course, no The second dataset that we construct is a pupil-level database that assigns pupils to the school they attend at the beginning of secondary school (when children are aged 11-12 and right after the KS2 exams). We use this dataset to study whether schools converting to an academy change their pupil intake post-conversion. For most of our analysis, we only consider schools that convert to academies during the Coalition Government yearsthat is from the academic year 2010/2011 to present. This is because Wilson (2011) and Eyles and Machin (2015) have already presented evidence on the impact of Labour's academies on pupil intake, reporting 11 or age 15. We only consider schools that had age 11-15 pupils in all years of our analysis. This cuts the sample further because of school closures and missing data in some years (this occur if, for instance, some institutions became middle schools mid-way through the sample). In some robustness checks, we re-estimated all our results using an unbalanced panel that included schools with missing years (but still dropped middle/upper schools). Our findings were completely unchanged. 10 Note that these numbers do not precisely match the official statistics published by the DfE because of the restrictions we impose on our data. However, our data reproduces the main trends in terms of sector expansion and the relative importance of 'sponsored' and 'converter' academies. positive effects. However, in some cases, we provide complementary evidence for the Labour years. We apply the same restrictions as in the school-level dataset. Furthermore, we only consider the following pupils: prior to conversion. These restrictions are applied so that our analysis of the effect of academy conversion on pupil intake composition compares the changes in the composition of actual academies (pre-and post-conversion) to changes in the composition of future academies. We return to this point in the next section, where we discuss our empirical methods.

Research design I -School characteristics and academy conversion
In the first part of our analysis, we investigate which characteristics predict academy conversion and whether the association between these characteristics and academy status changes between the Labour and the Coalition Government periods.
In order to do so, we estimate the following linear probability model for school s in year t: located (as well as the academy -there are no cross-LA relocations) and are school-year dummies. Finally, is an error term which we assume to be uncorrelated with all other observable characteristics in the model. However, we allow for a degree of autocorrelation in the residuals within schools over time, and cluster standard errors at the school level.
The parameters of interest are and which capture the associations between school characteristics and the probability of conversion in the 'Labour Years' and 'Coalition Years', respectively. The differences between these coefficients (and their statistical significance) are key to this paper since they allow us to test the notion that the nature of academies has changed since the Academies Act of 2010.
The way in which we have specified the dependent variable makes our analysis similar to a transition model where schools are 'at risk' of becoming academies. This allows us to isolate the effect of school characteristics at time t-1 on the probability of conversion at time twithout confounding this estimate with an effect of these characteristics on the probability of being an academy from time t onwards (i.e. survival). Since academies do not convert back to nonacademy status, the observations after conversion has occurred do not add any useful element of variation to our analysis and are essentially disregarded (by coding our dependent variable as missing).
In our analysis, we are also interested in analysing whether the association between school characteristics and academy status changes for different types of academies. For example, during the Coalition Government years we observer schools becoming 'converter' (Coalitiontype) as well as 'sponsored' (Labour-type) academies. Moreover, both type of academies can be 'stand-alone' or be part of a 'chain'i.e. belong to a group working together under a shared structure and jointly administered by an Academy Trust.
To test for these differences, we estimate the following model: where (2) redefines the dependent variable to take value one only for academy transitions into a specific subgroup -for example only 'chain academies'and taking values zero or missing as previously discussed. In order to test whether the effect of school characteristics on the probability of becoming an academy are significantly different across subgroups, we set the problem up as a seemingly unrelated equation system so that our tests for the differences in the parameters take into account the joint variance/covariance structure of the error terms across equations.

Research design II -Academy conversion and school intake
In the second part of our analysis, we investigate whether schools converting to academy change their pupil intake following conversion. In order to do so, we use pupil-level data (described in Section 2.2) and estimate the following model: where denotes the background of pupil i (aged 11-12) starting secondary school s in school year t; and denote school and time fixed effects respectively; denotes whether the school becomes an academy at some point during the observation window; and ( ≥ = ) is an indicator variable taking value one for all school years in which the school is operating as an academy. Finally, is an error term which we assume to be uncorrelated with all other observable characteristics in (3). We allow for a degree of autocorrelation in the residuals of pupils in the same school and over years, and cluster standard errors at the school level.
The main coefficient of interest is which captures the effect of academy conversion on intake. One concern with our specification is that it only controls for time-fixed school unobservables; however, schools that become/do not become academies could differ along other unobservable dimensions that might be time varying and/or time trending. In order to mitigate this problem, we only estimate the model in equation (3)  The empirical model in equation (3) imposes an average effect of academy conversion on intake quality that is constant across all post-conversion years. To relax this assumption, we estimate the following flexible event study specification where we allow for separate 'academy effects' in the different school years postand pre-conversion: where the expression ( = ) denotes a set of indicators that separately identify the school years prior to conversion year (from − 5 to − 1; with − 6 being the excluded baseline year) and at/after conversion (from to + 2); and capture the effect of academy status on intake in that specific year. Note that estimates of for the years prior to conversion allow us to determine whether actual converters (the 'treated' schools) differ from future converters (the 'control' schools) in the years leading up to the 'switch' to academy. This helps us assess the validity of our assumption that treated and control schools would have had parallel trends in the outcome variable in the absence of treatment .
As for the school-level analysis of the determinants of conversion, we are interested in whether the effects on intake are heterogeneous for different types of academies. In particular, we are interested in heterogeneity along the margins of 'Sponsored' vs. 'Converter' and 'Standalone' vs. 'Chain' academies. In order to do so assess these differences, we interact the variable in equation (3) with an indicator denoting whether the academy belong to ones of these groups.

Key descriptive statistics: pupil-level data
In Table 2b we present descriptive statistics on the characteristics of pupils in our sample. These are the variables we use to characterise pupils' background (i.e., the variables in equations 3 and 4). KS2 attainment of pupils when they enter secondary school (

Preliminary graphical evidence
We present our first set of results graphically in Figure 1.

Main regression analysis
In this section, we provide more formal evidence on the changes in the nature of academies by estimating the model laid out in equation (1) to study the effect of school and local characteristics measured at time t-1 on the probability of conversion at time t. Our estimates are presented in Table 3. The table has the following structure. Columns (1) to (3) present simple models that only consider KS4 attainment, the incidence of FSM eligible pupils and the percentages of male and White British pupils. Columns (4) to (6) add further controls for OFSTED ratings and predecessor school types as well as LA dummies. Lastly, Columns (7) to (9) include a control for the percentage of secondary schools that are academies in the LA where the school is located as well as dummies indicating whether the local council is (majority) Labour held or (majority) Conservative-held. 13 Further, Columns (1), (4) and (7) focus on the Labour Years, while Columns (2), (5) and (8) present the estimates for the Coalition period, and Columns (3), (6) and (9) the difference in the effects between the two periods. 13 The omitted group includes councils held by the Liberal Democrats or with no clear majority.
Starting with the first three columns, our results show that during the Labour years lower KS4 attainment and a higher share of FSM pupils were positively and significantly associated with probability of conversion. This pattern is completely reversed during the Coalition years: higher KS4 attainment and a lower percentage of FSM pupils are strongly associated with a higher probability of becoming an academy. As shown in Column (3) the differences in these associations are not only strikingthe coefficients push in clearly opposing directionsthey are also highly significant. As for the percentage of male and White British pupils in the school, we find that these are not strongly associated with academisation during either period, and their effects do not change significantly. 14 In Columns (4) to (6), we include LA dummies as well as OFSTED ratings and predecessor school types. The omitted categories are respectively 'Grade 1' (Outstanding) and Community schools. Note that since we are not able to collect consistent OFSTED data for the late 1990s and early 2000s, we lose some schools for which we are unable to gather inspection results preconversion. The results show that during the Labour years schools with the lowest inspection results -Grade 4 schoolswere more likely to become academies. This finding is completely reversed during the Coalition years: the coefficient on this dummy in Column (5) is negative and significant, and Column (6) shows that this difference is highly significant. Similarly, we find that 'satisfactory' schools are significantly less likely to become academies after 2010. Lastly, we find that 'good' schools are less likely to become academies than 'outstanding' schools in both periods. Although this is somewhat puzzling for the pre-2010 'sponsored' academies, the stronger and significantly more negative effect after 2010 is in line with expectations. In terms of predecessor school types, we find evidence that Foundation schools are more likely to convert after 2010 and Voluntary aided more likely to become academies before 2010. Finally, we find that controlling for OFSTED inspections does not substantially affect the change in the association between the percentage of FSM eligible pupils and academy conversion. However, the changes in the association between academy and KS4 test scores are reducedalthough they remain sizeable and significant. This is not surprising since OFSTED inspections incorporate a broader assessment about the quality of teaching which should translate into higher KS4 attainment (at least in as much as these do not solely capture family background and pupils' own ability, as opposed to school effectiveness). 15 14 Note that we re-estimated these regressions using non-linear probit models to account more properly for the small probability of academisationespecially in the Labour Years. We found very similar results, which are available on request. 15 In Appendix Table 1, we present additional results where we replace KS4 attainments of year 11 pupilsas a measure of school 'output'with the KS2 of year 7 students -as a proxy for 'intake' quality. We find consistent results. This is because school average KS2 and KS4 achievements are highly correlated.
The last three columns add controls for the percentage of academies in the LA and the party controlling the local council. We find that a higher percentage of academies in the year prior to conversion has a positive and significant effect on academisation during the Coalition years, but not during the Labour years. Furthermore, we find that schools in Labour-held councils were slightly more likely to convert during the Labour years (with a marginally significant positive coefficient), but much less likely to become academies after 2010. The difference in the association with this variable between the two periods is highly significant. Conversely, we do not detect any significant pattern or change in Conservative-held councils. As for the variables already included in the previous specifications, these mainly retain their size and significance.
The only exception is the effect of the incidence of FSM eligible students at school, which is now found to switch much less between the two periods. This change is explained by the inclusion of the Labour-held council dummy which is strongly correlated with local (and school) levels of deprivation.
How sizeable are the effects we have documented? To answer this question we first use the estimates of the first three columns of  (4) to (6), the impact of a one standard deviation improvement in KS4 would be a 40% reduction for the Labour years, and a 7% increase for the Coalition years. Although these numbers are clearly smaller, it should be noted that they stem from variation in KS4 achievement conditional on OFSTED records and are likely to provide a lower bound to the overall effect of academic quality on academisation in the two regimes. Lastly, note that the estimates capturing the change in the effect of attainment on the probability of becoming an academy range from 4 percentage points (Column 3) to 1.4 percentage points (Column 6). This means that a one standard deviation change in attainment would have pushed schools in opposing directions by around 1.4 to 4 percentage points before and after the Academies Act of 2010. Scaled against an average probability of conversion over the whole period of 4.5%, this is a very sizeable change.
In order to assess the magnitude of the effect of the percentage of pupils eligible for FSM at the school, we consider a 10 percentage point change from the mean of 15.4%. This corresponds to approximately 75% of a standard deviation change (i.e., 10 percentage points over 12.9 percent). Using the first three columns of Table 3, we find that, during the Labour years, an increase in the percentage of FSM eligible pupils of this magnitude would have approximately corresponded to a 60% increase in the probability of becoming an academy. Conversely, during the Coalition years this same increase would have basically reduced the probability of conversion to zero. The change in the association between the percentage of FSM students and the probability of becoming and academy pre/post 2010 is similar in Columns (3) and (6)

Further results
In Table 4, we present additional results on the changing nature of academies. All regressions enter school characteristics simultaneously and control for both LA dummies and predecessor school type. 16 In Panel A of the table we investigate whether the changes in the associations between the Labour years and the Coalition years reflect the differences between 'sponsored' and 'converter' academies. We find that the patterns in the coefficients closely resemble our previous findings: the association between KS4 attainment is negative and significant for 'sponsored' academies, but becomes positive and significant for 'converters'. On the other hand, the association between the share of FSM eligible pupils and 'sponsored' academies is positive and significant, but becomes negative (though insignificant) for 'converters'. Once again, we find that for both variables the change in these associations is significant. Note also that rescaled against the probability of converting through a 'sponsored' or a 'converter' route (respectively at 1.1% and 3.4%) the magnitudes of the associations presented in Panel A of Table 4 are similar to those displayed in Tables 3. Consistent with the evidence presented in Figure 1, this suggests that most of the Labour years vs. Coalition years dichotomy can be explained by the differences in the relative number of schools becoming academies via the 'converter' route in the latter period.
In panel B, we investigate whether the association between school characteristics and conversion differs for academies that belong to chains and those that are stand-alone. This distinction is important given the prevalence of Multi-Academy Trusts (MATs) post 2010. Ladd and Fiske (2016) argue that the recent White Paper on academies emphasises the 'scaling up of what works' aspect of the programme much more than the potential for an academised system to 16 We do not tabulate the effects of predecessor schools types and the shares of White British and male pupils because these show no relevant changes or patterns compared to the estimates in Table 3. foster competition. We find that stand-alone academies have significantly higher KS4 test scores in the year prior to conversion than academies in chains. The size of the difference in this association is similar to what we found before. On other hand, we find that chain academies are not significantly more likely to have a higher fraction of FSM eligible pupils prior to conversion than stand-alone academies. Although the point estimate on the percentage of pupils eligible for FSM is higher for school that convert to chain academies (at 0.063) than for stand-alone ones (at 0.029), the two coefficients are not statistically different. Finally, we find similar patterns to those documented before with respect to the OFSTED ratings although the patterns are less stark and significant. Note that whereas approximately 70% of the 'sponsored' academies in our analysis are in chains, only 30% of the 'converters' are. This suggests that the chain/stand-alone margin substantially overlaps with the 'sponsored'/'converter' one, reinforcing our conclusion that differences along this latter margin explain the majority of the pre-/post-2010 differences  (3) and (4) (1) show that becoming an academy during the Coalition period leads to a small increase in the KS2 of enrolled pupils.

Main regression analysis
Recall that KS2 tests are taken at the end of primary school (i.e., they are a measure of prior ability) and that we are considering the school where pupils start their secondary education. Our findings therefore suggest that Coalition academies slightly improve their composition in terms of students' ability by approximately 1% of a standard deviation. Column (2) shows that after conversion the Coalition-year academies also attract pupils who are less likely to be eligible for FSM. This effect is precisely estimated and corresponds to approximately a 4% reduction in the probability that an FSM eligible pupil is enrolled at an academy. Lastly, Columns (3) and (4) show that academies do not change their composition in terms of gender, and only marginally their ethnic background after conversion. The negative conversion effect on the probability that a White British student is enrolled in the school is borderline significant and only represents a 0.5% reduction.
In Panel B, we shed some light on the time profile of these effects. To do so, we estimate the model laid out in equation (4) in Section 3.2. Column (1) shows a positive and significant effect on pupils' KS2 attainment two years after conversion (at 7.7% of a standard deviation).
However, we find almost no effect at the time of conversion (0.006) and a small but insignificant effect one year after (at 0.019). It is interesting to note that the coefficients on the years leading up to the time of conversion are small and insignificant (with the exception of c-5) suggesting that actual academiesthe treated group -are balanced in terms of their pupils' ability relative to future academiesthe control schoolsbefore they switch. This lends some support to our identifying strategy, which assumes that the timing of conversion is 'as good as random' within the subset of schools we have chosen for our analysis (see Section 3.2 for a discussion). Column (2) reveals that the average effect of conversion on FSM eligible pupils found in Panels A comes from the estimate of the impact in the year after the first in which the academy has opened (c+1). Conversely, the impact at time t is still negative, but small and less precisely estimated (-0.010), while the impact at time c+2 is essentially zero. Once again, we find no evidence of differential pre-trends between treated (actual academies) and controls (future academies), lending support to our approach. Finally, Columns (3) and (4) confirm that academies do not significantly change the gender and ethnic composition of their students.
How robust are these findings? As discussed, our sample includes only pupils in schools that convert between 2011 and 2013 in a time window going from six years prior to conversion to up to two years after (treated schools), and pupils in schools that will convert in 2014, 2015 and 2016 from six years to one year prior to conversion (control schools). This approach was taken to reduce potentially unobservable time-varying differences between treated and controls, and to guarantee we could test for pre-treatment effects using an equal number of (school-year) observations for the two groups. However, this approach is not the only one we have experimented with. 17 To begin with, we redefine the control group to include all the observations in future significantly change their intake after conversion (reconfirming the earlier findings of Wilson, 2011 andMachin, 2015). The effect on KS2 is precisely estimated and corresponds to a 7% of a standard deviation change, on average, over the three periods post-conversion. This compares to a 1% estimated effect for the academies opened during the Coalition period.
Further, we find that Labour academies significantly reduced their intake of pupils eligible for FSM. The estimated effectat around 7% of the baseline probabilityis almost twice as sizeable as the one we found for the Coalition years. Finally, Labour academies also attract fewer pupils of White British origin, but the effect is ten times larger than the one we found for the Coalition years and much more significant. We return to these comparisons below where we investigate the differences between 'sponsored' and 'converter' academies, as well as chain and stand-alone schools, during the Coalition Years.

Heterogeneous effects on intake
In Table 6 we investigate whether the effects of conversion during the Coalition years on intake are heterogeneous and depend on whether academies are 'sponsored' as opposed to 'converters', or part of a chain instead of stand-alone institutions. We focus only on KS2 test scores and pupils' eligibility for FSM since these are the variables where we detected significant effects previously.
The first three columns of Panel A reveal some evidence of heterogeneity. Whereas 'converter' academies do not experience any change in the ability of their intake, 'sponsored' academies that open during the Coalition years attract students with significantly higher KS2 test scores. The estimated effect corresponds to 5.3% of standard deviation, with the difference between the impact of 'coalition' and 'sponsored' academies being significant. This pattern suggests that the small effect documented in Table 5 is driven by 'converters,' and that 'sponsored' academies opening after the Academies Act of 2010 more closely resemble the 'sponsored' of the Labour period when it comes to changes in their pupil intake. The next three columns of Panel A focus on pupils' eligibility for FSM and reveal yet another dimension of heterogeneity. We now find that 'converters' are associated with a reduction in the share of FSM eligible pupils and drive the overall negative effect presented in Table 5. Conversely, 'sponsored' academies attract pupils who are more likely to be eligible for FSM after conversion with an estimated effect of approximately 9%. As shown in the last column, the difference between these two coefficients is significant. This suggests that along this margin the Labour and the Coalition years 'sponsored' academies are not very comparable.
In Panel B, we provide evidence of the effect of chain and stand-alone academies on intake.
The first three columns show that only academies in chains significantly alter their intakes along the dimension of KS2 scores. However, this effect is smaller than the one documented in Panel A for 'sponsored' academies. Similarly, the difference in the effect of conversion between chain and stand-alone academies is less pronounced than the difference between 'sponsored' and 'converters'. The last three columns show that stand-alone academies tend to attract pupils that are less likely to be eligible for FSM. The estimated effect is approximately 5% of the baseline probability and the difference between chain and stand-alone academies is significant. Note that chain academies also tend to attract fewer FSM eligible pupils, although the effect is not statistically different from zero and is significantly smaller than the impact of stand-alone conversions.
Given this heterogeneity, we also investigate whether there are interesting patterns when we consider conversion type and chain interactions (see Appendix Table 2). Although the results become somewhat noisier, we still find that 'sponsored' academies in the Coalition years experienced significant improvements in the average KS2 of their pupil irrespective of whether they are in a chain or stand alone. This is not true for 'converters' of any type. We also confirm the pattern detected in the top panel of Table 6 showing that 'sponsored' academies marginally increase their intake of FSM eligible pupils. Furthermore, the negative effects we found for 'converters' overall is similar when we consider stand-alone and chain academies separately.
All in all, these results confirms the conclusions we drew from the analysis of school level data: the most interesting dichotomy is the one between 'converters' and 'sponsored' academies.
The patterns of heterogeneity of our findings lend support to the idea that 'sponsored' academies in the Coalition years are partially comparable to the Labour-period 'sponsored' academies.
However, the pattern of heterogeneity on the percentage of FSM eligible pupils enrolled after conversion suggests that 'sponsored' academies of the Coalition period might still be different from the 'sponsored' academies of the Labour years along important dimensions.

Further results
We conclude our analysis by studying whether the intake effects documented in Table 6 vary depending on the political party in control of the local council and on the percentage of secondary academies in the LA. Our results are presented in Table 7, which mirrors the analysis of Table 3, Columns (7) to (9).
In Columns (1) to (3) we focus on changes in the KS2 attainment of pupils. Irrespective of the party controlling the local council or the share of academies in the LA, we still find no evidence that 'converter' academies significantly change the ability of their intake. On the other hand we find that post-2010 'sponsored' academies increase their students' ability, but this effect is declining in the percentage of secondary schools in the same area. In LAs with a zero percentage of academies, 'sponsored' academies increase the KS2 of their intake by almost 9% of a standard deviation. This effects is smallerat 6.4%in LAs with the average percentage of academies within our sample (12.3%), and negligibleat less than 1%in areas with the highest concentration of academies (40% or the 90 th percentile of the distribution). Although this result should be interpreted with caution, the pattern suggests that in areas with more competition from other academies 'sponsored' might find it hard to cherry pick their students or to entice parents of better achieving pupils to apply to join the school. Interestingly, we find no heterogeneity in intake effects when we consider differences between schools in areas with different political affiliations.
Columns (4) to (6) consider whether there is heterogeneity in the effect of conversion of admitting FSM eligible pupils. In this case, we find no evidence that 'sponsored' academies markedly change their intake. Both the main effect, and an interaction with the percentage of academies in the LA, are insignificant and small. All in all, the patterns are consistent with those presented in the top panel of

Conclusions
The last ten to fifteen years have seen radical developments in the English education landscape.
The most salient change in terms of school structures has been the introduction of academies: autonomous schools that remain part of the state sector, but operate outside the control of the The aim of this paper was to study whether the nature of academies has changed as a result of this shift. In order to do so, we have investigated and contrasted the characteristics associated with conversion during the Labour Government years and the Coalition Government period. We have also studied how the composition of schools that become academies after 2010 changes as a result of conversion, and whether the association between academisation and changes in intake differs across different types of academies and in the Labour and Coalition years.
Our evidence leaves no doubts that the second batch of academies differs markedly from the first. Schools that converted during the Labour years had low attainment and a high share of disadvantaged students before conversion. The opposite is true for schools that become academies after May 2010. We also find that the new academies tend to experience no changes in terms of the ability of the students they enrol, and some limited changes in the percentage of children eligible for free school meals. This is in contrast to evidence for the Labour academies which shows significant improvements in the ability of enrolled students and more substantial declines in the percentage of disadvantaged pupils after conversion.
Altogether this suggests that simple extrapolation from the evidence on the effects of the first batch of conversions to the second batch is not warranted and potentially very misleading.
This finding is important because extrapolation has been, and continues to be, commonplace 18 In some additional extensions, we studied whether the evidence presented above is heterogeneous for academies whose predecessors was either a Community or a Voluntary Controlled school (the least autonomous school types) as opposed a Foundation or Voluntary Aided school. We found no evidence of strong and significant heterogeneous patterns. Furthermore, this additional dimension did not help shed light on the differences/similarities between the Labour and Coalition Years, or the 'sponsored' and 'converter' academies. As a consequence, the results are not tabulated (but are available from the authors). among policymakers, media commentators andin some casesacademics. Nevertheless, our results also point to some commonalities between the two programmes. 'Sponsored' academies originally introduced by Labour but also proliferating during the Coalition periodhave fairly similar pre-conversion characteristics during both regimes. Moreover, they tend to change the ability of their pupil intake in similar ways across all years, though the change in the percentage of disadvantaged pupils they attract is somewhat different for the two batches. This suggests that a comparison of 'sponsored' academies in both regimes may be relatively legitimate and that an extrapolation of the lessons learned for the Labour academies might apply to a subset of Coalition academies. Despite this, it is important to stress that the academy sector has predominantly expanded through the 'converter' route. Our analysis reveals that there is too little overlap between the nature of these academies and the Labour batch to warrant any meaningful extrapolation. Indeed, the fact that these schools were typically designated as good or outstanding before conversion means that they may not see the need to innovate postconversion in the same way that the poorly performing 'sponsors' might (Academies Commission, 2013).
More broadly, our paper holds important lessons for the debate about the consequences of autonomy in state education. The revamped academy programme launched in May 2010 is an experiment of unprecedented scale in which highly performingas well as struggling schoolsare allowed more independence with the hope that this will bring improvements in standards across the board. In the span of five years, these autonomous institutions have gone from being a minor part of the education system to representing more than 60% of secondary and more than 20% of primary schools.
One of the main concerns with such a vast expansion of autonomy and differentiation in education has been the risk of increased stratification with students of different background segregated in different schools, either because of differential parental preferences for different types of schools, or because of school admission practices. Our analysis reveals that these concerns are partly warranted. The intake composition of Labour academies in terms of pupils' academic ability clearly improved more than for the Coalition academies. Nevertheless, we also found that 'converter' academies enrol fewer pupils with free school meal status. Conversely, the intake of Coalition 'sponsored' academies has become somewhat more skewed towards disadvantaged children. This suggests that, at least along this dimension, schools are becoming more stratified.
Although we are not yet able to tell whether this stems from parental free choice or school 'back door' (hidden) selection practices, understanding the mechanisms that lie behind this finding is an essential future research venture of relevance for designing policies that can help reap the potential benefits of more autonomous schooling, while at the same time mitigating any negative effects.       Discrepancies in the number of observations are due to missing OFSTED inspection grades. KS4 point score is calculated across best eight subjects and includes GCSE equivalent. KS4 point scores have been standardized. Omitted OFSTED group: Grade 1 (Outstanding). Omitted school type: community school. *: 10% significant; **: 5% significant; ***: 1% significant or better.       Table 5. Labour and Conservative held council identify whether the council where the school is located was held by a majority of the Labour or Conservative party in the academic year prior to conversion. Excluded groups include: Liberal Democrats or Other party majority; no clear majority. Number of sponsored/converter academies open within observation window by party majority as follows: Sponsored/Labour held: 35; Sponsored/Conservative: 86; Converter/Labour: 169; and Converter/Conservative: 666. Percentage (%) of academies refer to secondary schools open as academies in the academic year prior to the school's conversion. Sample mean of percentage of academies: 12.3%. *: 10% significant; **: 5% significant; ***: 1% significant or better.