View in EDS HTML Full Text PDF Full Text

Initial Estimates of Teacher Value-Added in English Primary Schools

Saved in:
Bibliographic Details
Title: Initial Estimates of Teacher Value-Added in English Primary Schools
Language: English
Authors: John Jerrim (ORCID 0000-0001-5705-7954), Rebecca Allen, Maria Palma Carvajal, Raj Chande, Rob Coe, Calum Davey, Shaun Dillon (ORCID 0009-0005-8086-5139), Claire Maud, Sam Sims, Ourania Ventista
Source: British Educational Research Journal. 2025 51(6):2942-2963.
Availability: Wiley. Available from: John Wiley & Sons, Inc. 111 River Street, Hoboken, NJ 07030. Tel: 800-835-6770; e-mail: cs-journals@wiley.com; Web site: https://www.wiley.com/en-us
Peer Reviewed: Y
Page Count: 22
Publication Date: 2025
Document Type: Journal Articles
Reports - Research
Education Level: Elementary Education
Descriptors: Foreign Countries, Value Added Models, Teacher Effectiveness, Teacher Influence, Elementary School Teachers, Elementary School Students, Reading Achievement, Mathematics Achievement, Attendance, Barriers, Student Characteristics
Geographic Terms: United Kingdom (England)
DOI: 10.1002/berj.4207
ISSN: 0141-1926
1469-3518
Abstract: A sizeable literature investigating teacher test score value-added--the extent to which pupils make different rates of progress under different teachers--has emerged in the United States. While there is much interest in estimating teacher value-added in other countries such as England, progress has been limited by the lack of datasets linking teachers and pupils. We overcome this issue by drawing on internal assessment data from primary schools across two multi-academy trusts. Our results suggest that a substantial proportion of the progress primary pupils make in reading and mathematics occurs across (rather than within) the teachers to which they are assigned. There is, however, no clear evidence of teacher effects on attendance. Similar results are obtained using different model specifications and approaches. The paper concludes by clearly outlining some of the remaining challenges with estimating teacher value-added in England's primary schools, and the next steps that should be prioritised in this line of research.
Abstractor: As Provided
Entry Date: 2025
Accession Number: EJ1490548
Database: ERIC
Full text is not displayed to guests.
FullText Links:
  – Type: pdflink
    Url: https://content.ebscohost.com/cds/retrieve?content=AQICAHj0k_4E0hTGH8RJwT4gCJyBsGNe_WN95AvKlDbXJGqwxwHk2MHJOIA21lnht9D2XIvVAAAA4zCB4AYJKoZIhvcNAQcGoIHSMIHPAgEAMIHJBgkqhkiG9w0BBwEwHgYJYIZIAWUDBAEuMBEEDEkQ1xfyzxSuynAmFgIBEICBmyOJLiuSusgYy2jQlr-lybQS69LLRQuHZO00-MiCHGuVwT1mSa3BzCMBLGxoHO41kp_lp19szO2hn1GhgfYADbSokRCC9d29Z9B-25Tszqi7X2qbJLi9kzOHu36qgu86_8LCb42fzEUJkTcWDBfrmFqFblKLmafWFRemfxxhkEgq5QUpHmNGM2eV7GrBT-ZNEYWlz8MtDvPcuaDy
Text:
  Availability: 1
  Value: <anid>AN0189830383;bed01dec.25;2025Dec08.07:05;v2.2.500</anid> <title id="AN0189830383-1">Initial estimates of teacher value‐added in English primary schools </title> <p>A sizeable literature investigating teacher test score value‐added—the extent to which pupils make different rates of progress under different teachers—has emerged in the United States. While there is much interest in estimating teacher value‐added in other countries such as England, progress has been limited by the lack of datasets linking teachers and pupils. We overcome this issue by drawing on internal assessment data from primary schools across two multi‐academy trusts. Our results suggest that a substantial proportion of the progress primary pupils make in reading and mathematics occurs across (rather than within) the teachers to which they are assigned. There is, however, no clear evidence of teacher effects on attendance. Similar results are obtained using different model specifications and approaches. The paper concludes by clearly outlining some of the remaining challenges with estimating teacher value‐added in England's primary schools, and the next steps that should be prioritised in this line of research.</p> <p>Keywords: primary teachers; teacher value‐added; teacher effectiveness</p> <p>Key insights What is the main issue that the paper addresses?Using data from two multi‐academy trusts, the paper provides one of the first attempts to measure teacher value‐added amongst primary school teachers in England. What are the main insights that the paper provides?There are substantial differences in value‐added scores across teachers in reading and mathematics, but with little evidence of a teacher effect on attendance.</p> <hd id="AN0189830383-2">INTRODUCTION</hd> <p>Teachers are widely regarded as one of the most important resources available to schools (Burroughs et al., [<reflink idref="bib13" id="ref1">13</reflink>]). Evidence from the international literature—primarily the United States—suggests that being assigned an average rather than an ineffective teacher boosts pupils' test scores by around 0.1 to 0.15 standard deviations over the course of an academic year (Bacher‐Hicks & Koedel, [<reflink idref="bib9" id="ref2">9</reflink>]). Unfortunately, school systems across the world are struggling to recruit and retain enough high‐quality teaching staff (UNESCO, [<reflink idref="bib40" id="ref3">40</reflink>]), including England—the empirical setting of this paper. There is hence much interest in developing a better understanding of the impact, allocation and retention of England's teachers (Maisuria et al., [<reflink idref="bib30" id="ref4">30</reflink>]).</p> <p>Teachers also play a central role in theoretical models of school improvement. This is illustrated in Figure 1, which draws on the school improvement framework set out by Reezigt ([<reflink idref="bib33" id="ref5">33</reflink>]). Within this framework there are three broad pillars, which combine to drive school improvement. The first is the <emph>context</emph> in which schools and the education workforce operates. This includes centralised pressures to improve (e.g., market mechanisms, school inspections), goal setting and targets (e.g., for pupil outcomes) and school autonomy. These factors then influence schools' decision‐making and the actions of senior leaders. This includes the extent to which senior leaders put pressure on more junior staff to drive up standards, the goals staff are set (e.g., monitoring of pupil outcomes) and the culture within the school. The result is a process of cyclical improvement within many schools, including continuing self‐evaluation, planning to improve activities and evaluation.</p> <p> <img src="https://imageserver.ebscohost.com/img/embimages/rdk/BED/01dec25/berj4207-fig-0001.jpg?ephost1=dGJyMNXb4kSepq84yOvqOLCmsE6epq5Srqa4SK6WxWXS" alt="berj4207-fig-0001.jpg" title="1 Reezigt's model of school improvement. Adapted from Reezigt ([33])." /> </p> <p></p> <p>This then leads on to the actions of individual teachers in the classroom. The context in which a school operates, and the decisions made by its senior leaders, will determine the autonomy teachers have in the classroom and the decisions they take. The school culture will determine this willingness to learn, the extent to which they collaborate with colleagues and their commitment/motivation in their job. This will then impact on their effort in the school's cycle of school improvement. It is this that will lead to higher‐quality teaching, more positive learning environments and—ultimately—pupil outcomes.</p> <p>Measuring and understanding teacher value‐added (TVA) can play different roles in this process. For instance, in the United States, TVA has been used for accountability processes by certain states within individual schools (Amrein‐Beardsley & Holloway, [<reflink idref="bib7" id="ref6">7</reflink>]). This includes sometimes being linked to sanctions with regard to individual members of staff (Amrein‐Beardsley & Holloway, [<reflink idref="bib7" id="ref7">7</reflink>]), which then obviously feeds into the culture and environment in which teachers work (e.g., the pressure to improve). A long line of research has discussed the problems with using TVA for such purposes (McCaffrey et al., [<reflink idref="bib32" id="ref8">32</reflink>]), including the pressure it puts on teachers (Koedel et al., [<reflink idref="bib26" id="ref9">26</reflink>]), the perverse incentives it entails (Rothstein, [<reflink idref="bib35" id="ref10">35</reflink>]) and whether reliable enough inferences can be made at the individual level for such purposes. We thus firmly believe that measures of TVA shall not be used as part of school accountability. Rather, the use of TVA in this school improvement process should be at a broader level, helping inform the education sector more generally about factors linked to higher value‐added scores. This includes, for instance, different training pathways into teaching, professional development interventions, factors related to the retention of high value‐added teachers and how staff with different value‐added scores are allocated across different groups.</p> <p>Progress building evidence on the effects of teachers on pupils in England has, however, unfortunately been somewhat slow to develop. This is due, in part, to a lack of available data. Whereas researchers in other countries—most notably the United States—have had access to anonymised data linking pupils to their class teachers, such data have rarely been available in the United Kingdom. Moreover, with national assessments only conducted at sporadic points (primarily at ages 11 and 16), it has not typically been possible to produce estimates of TVA as has routinely been done in the United States.</p> <p>This, in turn, means that there are several key issues we currently know relatively little about. For instance, while some studies in England have tried to explore whether more 'effective' teachers tend to be assigned to certain groups of pupils (e.g., pupils eligible for Free School Meals), these have had to rely on proxy measures for TVA such as years of experience, teacher qualifications or specialisation (Allen & Sims, [<reflink idref="bib6" id="ref11">6</reflink>]). Clearly, with better measures, evidence regarding the allocation of teachers to pupils with particular characteristics could be further refined. Likewise, while we know that around a third of teachers in England are no longer working in state schools 5 years after qualifying (Maisuria et al., [<reflink idref="bib30" id="ref12">30</reflink>]), there is a dearth of evidence on the characteristics of those leaving their job. For instance, if the teachers leaving the profession are those that are failing to produce learning gains for their pupils, such high rates of attrition may be less of a concern.</p> <p>Fortunately, this situation is starting to change. The National Institute of Teaching (NIOT)—working with its four founding multi‐academy trusts (MATs)—is developing the Teacher Education Dataset (TED). This draws on MATs' and schools' internal management data to develop a resource that can be used for education research purposes. Critically, it includes information on pupil test scores (at least annually and often termly) and the allocation of staff to different classes. It thus creates the possibility to estimate TVA in research studies in England for the first time. In doing so, it has the potential to address several pressing academic and education policy issues we currently know little about.</p> <p>The only existing study using these data is Coe et al. ([<reflink idref="bib15" id="ref13">15</reflink>]). They focus on several technical aspects of estimating TVA using these data, with the sample primarily consisting of secondary school teachers and pupils. Indeed, their data comprised just 79 primary school teachers, compared to 1635 secondary school teachers. Likewise, the only other prior study we are aware of to measure TVA in England was focused on the latter stages of secondary school, based on data now around 25 years old (Slater et al., [<reflink idref="bib36" id="ref14">36</reflink>]).</p> <p>An important related body of work—much of which was conducted in the late 1990s and early 2000s—was produced by Christine Merrell and Peter Tymms within the Progress in Primary School (PIPS) project. This work also explored pupil progress in primary schools using models with some similarities to those applied in the TVA literature. For instance, Tymms et al. ([<reflink idref="bib38" id="ref15">38</reflink>]) used data from 1700 pupils to explore the progress they made during their first year at school. The reported correlation between start and end‐of‐year assessments was 0.75, with attendance at school found to be the most important predictor of progress made. Tymms et al. ([<reflink idref="bib39" id="ref16">39</reflink>]) built on this work, demonstrating how assessments of 4‐year‐olds correlate reasonably strongly (0.7) with the achievement in reading and mathematics at age 7. This work was later put into an international context by Tymms et al. ([<reflink idref="bib37" id="ref17">37</reflink>]), who compared pupil progress during the first year of school across England, Scotland, New Zealand and parts of Australia. However, perhaps the closest analogue to estimating TVA within this body of work is presented within Tymms et al. ([<reflink idref="bib39" id="ref18">39</reflink>]). This documents the intra‐cluster correlation (ICC) in pupil achievement that occurs between classes at the end of Reception (age 4) and the end of Key Stage 1 (age 7). They report ICCs of 0.24 and 0.22 in mathematics at the end of Reception and Key Stage 1, respectively.</p> <p>Although the aforementioned studies have made a significant contribution to our understanding of pupil progress during primary school, it is clear that research attempting to estimate TVA in England remains limited. This highlights the need for further work in England focused on estimating TVA within the primary school sector. This paper starts to address this gap in the literature drawing on NIOT's TED, presenting evidence from 291 primary school teachers across 29 schools within two MATs. Given the ongoing development of the TED, we begin by providing an overview of the test score measures available. The quality of these tests is essential to producing robust estimates of TVA, so it is important to document their test–retest correlations, links with national examination scores and distributional characteristics. This further builds on the work of Coe et al. ([<reflink idref="bib15" id="ref19">15</reflink>]), where there was only a limited breakdown of investigations of assessment quality within the primary school sample. Research question 1 is therefore:</p> <p></p> <ulist> <item> RQ1 What are the statistical properties of the tests available to estimate TVA in the TED data?</item> </ulist> <p>Having investigated assessment quality, we turn to a second key issue within the TVA literature—the way in which pupils with certain characteristics are distributed across different teachers. This is important as teachers are unlikely to be randomly assigned to classes, which must be accounted for in the estimates of TVA via statistical controls. By documenting the way in which pupils with different characteristics (e.g., eligible for Pupil Premium funding, Special Educational Needs [SEN]) are distributed across different teachers, we hope to develop a better understanding of how strong this 'selection' (non‐random allocation) is, as well as this being an interesting and understudied issue in England in its own right. In doing so, we will decompose estimates of pupil segregation into two components—the part that is driven by school selection (between‐school differences) and the part that is driven by teacher–class allocation (within‐school differences). As different mechanisms are likely to be driving these different selection processes, this decomposition may provide useful insights for modelling TVA. Research question 2 is therefore:</p> <p></p> <ulist> <item> RQ2 What is the level of segregation of pupils across primary school teachers/classes? To what extent is it being driven by between‐school vs. within‐school selection processes?</item> </ulist> <p>Next, we get into the specifics of estimating TVA in England's primary schools. Studies from American elementary and middle schools have suggested that relatively simple models prove effective at measuring TVA, so long as they include rich controls for pupils' prior achievement (Bacher‐Hicks & Koedel, [<reflink idref="bib9" id="ref20">9</reflink>]). The same authors also suggest that, despite the theoretical advantages of including school fixed effects to control for pupil and teacher selection into different schools, there is little discernible benefit from doing so (Bacher‐Hicks & Koedel, [<reflink idref="bib9" id="ref21">9</reflink>]). Our third research question probes issues related to this matter for England. Specifically, we investigate the extent that estimates of TVA change when one simply controls for prior achievement, after a rich set of pupil and class covariates are added, and when school fixed effects are included as well. This reveals how much model specification matters for estimates of TVA. Research question 3 is therefore:</p> <p></p> <ulist> <item> RQ3 How does teacher value‐added in English primary schools vary across different model specifications?</item> </ulist> <p>Next, we present a summary of key results for TVA amongst primary teachers in England. This includes consideration of the ICC—the percentage of the conditional variation in test score outcomes across (as opposed to within) teachers—and the correlation of TVA estimates across different school outcomes. Estimates are presented for reading, mathematics and attendance, with findings discussed in the context of equivalent results reported from the United States. Our fourth and fifth research questions are thus:</p> <p></p> <ulist> <item> RQ4 What proportion of the variation in pupil progress over the course of an academic year occurs within vs. between teachers?</item> <p></p> <item> RQ5 How does TVA in English primary schools compare across reading, mathematics and attendance?</item> </ulist> <p>One alternative approach to estimating TVA that has gained some traction in the United States uses student percentile growth (Kurtz, [<reflink idref="bib27" id="ref22">27</reflink>]). This is based on a quantile regression model, with the progress made by each pupil referenced to that of their cohort peers and then aggregated to the level of the teacher. Our sixth research question implements such an approach for the first time in England, comparing the results to the 'standard' valued‐added model approach more commonly found in the literature:</p> <p></p> <ulist> <item> RQ6 How do the results from TVA models compare to those using a student growth percentile approach?</item> </ulist> <p>Finally, given the infancy of the TED, it is important we clearly document its current limitations regarding the estimation of TVA across primary schools. These are set out in answer to research question 7, noting how evidence on certain matters will likely become clearer over time, as the dataset grows:</p> <p></p> <ulist> <item> RQ7 What caveats are needed when interpreting estimates of TVA in English primary schools?</item> </ulist> <hd id="AN0189830383-4">INSTITUTIONAL CONTEXT</hd> <p>English primary schools educate students for 7 years, with a different teacher assigned to each year group by the headteacher. While teachers may express preferences and often specialise in teaching particular age groups, movement between year groups is common. A survey of teachers shows that each summer approximately two‐thirds of teachers will remain in the same year group, with the rest teaching a different year group (Allen, [<reflink idref="bib2" id="ref23">2</reflink>]). The allocation is carefully considered by headteachers: 94% of primary headteachers report factoring in the strengths and weaknesses of the previous year's teacher to ensure equitable teaching expertise across the school (Allen, [<reflink idref="bib2" id="ref24">2</reflink>]).</p> <p>All primary students receive some lessons without their class teacher due to their entitlement to 10% preparation, planning and assessment (PPA) time. In more than half of cases, these lessons are led by individuals who are not qualified teachers, such as teaching assistants (TAs) or higher‐level teaching assistants (HLTAs). Foreign languages, drama, music and physical education are the subjects most commonly delivered during PPA cover. In addition, it is increasingly common for students to be taught by two teachers who share a class. This arrangement often arises when teachers work part‐time or when one of the teachers has senior leadership responsibilities within the school.</p> <p>Primary classrooms frequently include a second adult—a TA—who works alongside the class teacher (Jerrim & Sims, [<reflink idref="bib25" id="ref25">25</reflink>]). TAs may provide general support or work specifically with SEN children. Their presence is more prevalent in younger year groups, reflecting the additional support requirements of younger students. For instance, 74% of Reception classes reportedly have a full‐time TA, compared to just 30% of Year 6 classes (Allen, [<reflink idref="bib2" id="ref26">2</reflink>]). This variation illustrates how the deployment of TAs is influenced by the differing needs of pupils as they progress through primary education.</p> <p>England does not have a system of nationally administered annual assessments. Instead, primary students are assessed in English and mathematics on entry at age 4 and exit at age 11. In addition, there is a phonics reading assessment at age 6, a multiplication check at age 9 and an optional mathematics and English assessment at age 7. However, primaries make extensive use of formal assessments administered across multiple schools in all year groups, with 70% of teachers reporting they do so for at least one subject in Reception, rising to 82% in Year 5 (Allen, [<reflink idref="bib3" id="ref27">3</reflink>]). In mathematics, 22% of teachers use assessments written by curriculum providers, 38% by assessment companies and 11% by MATs. In reading, 19% of teachers use assessment written by curriculum providers, 43% by assessment companies and 9% by MATs. In writing/grammar, 8% of teachers use assessments written by curriculum providers, 25% by assessment companies and 9% by MATs. Relatively little is known about the administration of these assessments in the classroom or about the statistical properties of these assessments, although in the case of those provided by commercial assessment companies, one study reported their predictive validity in relation to government assessments (Allen et al., [<reflink idref="bib5" id="ref28">5</reflink>]).</p> <p>The schools in this study all share a particular governance structure, as they belong to fairly large MATs. A MAT is an organisation that oversees and manages multiple academies, sharing resources, leadership and governance to improve educational outcomes across its schools. Primary schools within MATs often serve more disadvantaged communities, as academisation has historically been focused on improving schools in areas of higher socioeconomic need. While academies have some operational differences from local authority‐maintained schools—such as not being required to follow the National Curriculum—they are otherwise very similar in their day‐to‐day functioning.</p> <hd id="AN0189830383-5">DATA</hd> <p>The data we use are drawn from the TED covering the 2022/23 and 2023/24 academic years. This brings together data held in schools' management information systems (MISs) into an anonymised database for education research purposes. Within this paper we use data from primary schools within two participating MATs. MAT1 includes data from 7 primary schools, 34 teachers and 906 pupils. The analogous figures for MAT2 are 22 primary schools, 257 teachers and 7281 pupils. A small number (2%) of primary classes in MAT2 are recorded as having more than one teacher. In such cases, we assign the teacher that spent the largest share of their time with the class.[<reflink idref="bib1" id="ref29">1</reflink>] The first MAT includes pupils in school years 3 to 5 (ages 7/8 to 9/10), while the second MAT includes pupils in school years 1 to 6 (ages 5/6 to 10/11).</p> <p>An important feature of the TED is that it includes anonymised class and teacher identifiers. This makes it possible to link pupils to classes and subsequently to their teachers. It is thus possible to establish teacher–pupil–class allocations across academic years, thus creating opportunities to estimate TVA scores.</p> <p>Within both MATs, pupils sit standardised reading and mathematics assessments at least once—and up to three times—during each academic year. These are the Rising Star NTS mathematics and reading assessments (see https://<ulink href="http://www.hachettelearning.com/assessment">www.hachettelearning.com/assessment</ulink> for further details). The properties of these tests have been reviewed by the Education Endowment Foundation in their 'measures database' (Education Endowment Foundation, [<reflink idref="bib17" id="ref30">17</reflink>]), from which we reproduce key information in Table 1. All the tests have been standardised based on a large, representative UK sample. Depending on subject and year group, most take somewhere between 40 and 85 min to complete. Levels of internal consistency are reasonably high, with Cronbach's alpha typically sitting between 0.8 and 0.95. Less information is available, however, on the temporal stability of the tests (test–retest correlations) or how they correlate with Key Stage 2 test scores.</p> <p>1 TABLE Summary of previously published information on the assessment measures used within this paper.</p> <p> <ephtml> <table><thead valign="bottom"><tr><th align="left" /><th align="left">NTS maths</th><th align="left">NTS reading</th></tr></thead><tbody valign="top"><tr><td align="left">UK standardised</td><td align="left">Yes</td><td align="left">Yes</td></tr><tr><td align="left">Length</td><td align="left">Approx. 40 min KS1;85 min KS2</td><td align="left">Approx. 40–60 min</td></tr><tr><td align="left">Correlation with Key Stage 2 scores</td><td align="left">Not reported</td><td align="left">Not reported</td></tr><tr><td align="left">Cronbach's alpha</td><td align="left">0.82–0.96</td><td align="left">0.83–0.91</td></tr><tr><td align="left">Standardisation sample</td><td align="left">Large representative sample</td><td align="left">Large representative sample</td></tr><tr><td align="left">Test–retest correlation</td><td align="left">Not reported</td><td align="left">Not reported</td></tr></tbody></table> </ephtml> </p> <p>1 <emph>Note</emph>: Information has been drawn from the Education Endowment Foundation's measures database: https://educationendowmentfoundation.org.uk/projects‐and‐evaluation/evaluation/eef‐outcome‐measures‐and‐databases/attainment‐measures‐database/am‐database?keyStage=33442.</p> <p>As the discussion above illustrates, only partial details are available on the psychometric properties of these tests. For instance, an anonymous reviewer of this paper noted how the algorithm used to convert raw test data into standardised scores has—on some assessments—been suspected of introducing measurement error into the data. Moreover, in general, understanding of these tests (including their correlations over time and with other independent measures) are not widely understood. Such issues have partly motivated RQ1, where we provide further investigation and discussion of the assessment measures available within the data analysed. Nevertheless, as the same tests are used amongst pupils in the same year group within each MAT, assessment scores should be comparable across schools and teachers. We standardise all scores to mean zero and standard deviation one within school year group for each school term. Within our analysis, we calculate the average score across all the assessments a pupil completed in a subject within a given academic year.</p> <p>The data also include several other measures we will use within our analysis. This includes annual attendance (percentage of school sessions attended) for the 2022/23 and 2023/24 academic years. Given their skew, we transform these variables by first inverting them (so that they measure the percentage of sessions absent) and then take the natural logarithm. The distribution of the original and transformed versions of these variables can be found in Appendices S1 and S2.</p> <p>In addition, the following variables are used:</p> <p></p> <ulist> <item> <emph>Pupil Premium eligible</emph>. A binary variable indicating whether the pupil was eligible for Pupil Premium funding—an indicator of low family income.</item> <p></p> <item> <emph>Special Educational Needs</emph>. A binary variable indicating whether the pupil has SEN.</item> <p></p> <item> <emph>English as an Additional Language</emph>. A binary variable indicating whether the pupil's first language is English or whether they speak English as an additional language.</item> <p></p> <item> <emph>Male</emph>. A binary variable indicating whether the pupil is a boy or a girl.</item> </ulist> <p>Within some of our TVA models we also include class‐average pupil characteristics as controls. This includes class averages of prior achievement and prior levels of attendance.</p> <hd id="AN0189830383-6">METHODOLOGY</hd> <p>As our analysis for RQ1 primarily relies on simple descriptive statistics (correlations, histograms, scatterplots), we quickly turn to our methodological approach for RQ2—the way in which pupils with different background characteristics are distributed across different teachers.</p> <hd id="AN0189830383-7">RQ2</hd> <p>Our approach to RQ2 focuses on the use of two segregation indices—the dissimilarity index (<emph>D</emph>) and Theil's <emph>H</emph>. Both capture the extent that different individuals (pupils) are distributed across groups (primary teachers/classes). Such indices have been widely used to study the segregation of pupils with different characteristics across schools (Gutiérrez et al., [<reflink idref="bib21" id="ref31">21</reflink>]; Jenkins et al., [<reflink idref="bib24" id="ref32">24</reflink>]), but less frequently to study the segregation of pupils across different classes/teachers. Indeed, we believe this is the first study to present such evidence within the English setting.</p> <p>The segregation index <emph>D</emph> is widely used due to its simple, intuitive interpretation. It is calculated as follows:1 <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0001" display="block" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mi>D</mi><mo linebreak="goodbreak">=</mo><mfrac><mn>1</mn><mn>2</mn></mfrac><mspace width="0.25em" /><munderover><mo movablelimits="false">∑</mo><mrow><mi>i</mi><mo>=</mo><mn>1</mn></mrow><mi>I</mi></munderover><mfrac><msub><mi>n</mi><mi mathvariant="italic">ij</mi></msub><msub><mi>N</mi><mi>j</mi></msub></mfrac><mo linebreak="goodbreak">−</mo><mfrac><msub><mi>n</mi><mi mathvariant="italic">ik</mi></msub><msub><mi>N</mi><mi>k</mi></msub></mfrac></mrow></semantics></math> </ephtml> where <emph>I</emph> = total number of primary school classes/teachers in the trust; <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0002" display="inline" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><msub><mi>n</mi><mi mathvariant="italic">ij</mi></msub></mrow></semantics></math> </ephtml> = number of pupils in primary class <emph>i</emph> for group <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0003" display="inline" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><msub><mi>N</mi><mi>j</mi></msub></mrow></semantics></math> </ephtml> (e.g., number of boys in class <emph>i</emph>); <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0004" display="inline" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><msub><mi>N</mi><mi>j</mi></msub></mrow></semantics></math> </ephtml> = total number of pupils in group <emph>j</emph> (e.g., total number of boys in the MAT); <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0005" display="inline" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><msub><mi>n</mi><mi mathvariant="italic">ik</mi></msub></mrow></semantics></math> </ephtml> = number of pupils in primary class <emph>i</emph> for group <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0006" display="inline" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><msub><mi>N</mi><mi>k</mi></msub></mrow></semantics></math> </ephtml> (e.g., number of girls in class <emph>i</emph>); <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0007" display="inline" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><msub><mi>N</mi><mi>k</mi></msub></mrow></semantics></math> </ephtml> = total number of pupils in group <emph>k</emph> (e.g., total number of girls in the MAT); <emph>j</emph> and <emph>k</emph> are two groups (e.g., boys and girls).</p> <p>Values of this index range from zero (no segregation across groups) to one (complete segregation across groups). The figures can be interpreted as the percentage of pupils with a certain characteristic (e.g., Pupil Premium eligible) that would need to change group (e.g., teacher/class) so that they would be evenly distributed across groups. A value of 0.3 would, for instance, suggest that 30% of Pupil Premium children would need to change class/teacher to ensure an even distribution of Pupil Premium children across classes. However, this index also suffers two limitations. First, it really captures unevenness rather than segregation—meaning that it partly 'reflects randomness in the allocation of individuals to units' (Allen et al., [<reflink idref="bib4" id="ref33">4</reflink>], p. 40). Second, it is not possible to decompose values into 'between' and 'within' school components. This means the index is not well suited to studying the extent to which any apparent segregation of pupils is being driven by between‐school (e.g., Pupil Premium children attending different schools) or within‐school (e.g., Pupil Premium children allocated to different classes/teachers within their chosen school) selection.</p> <p>We thus also use Theil's <emph>H</emph> as an alternative measure of segregation, with details of its calculation presented in Massey and Duncan ([<reflink idref="bib31" id="ref34">31</reflink>]) and implemented within the Stata DSEG package (Guinea‐Martin & Mora, [<reflink idref="bib20" id="ref35">20</reflink>]). Values of <emph>H</emph> can range between zero (no segregation) and infinity, with higher values representing greater levels of segregation. While this index has been criticised for absolute values of <emph>H</emph> being difficult to interpret (Conceição & Ferreira, [<reflink idref="bib16" id="ref36">16</reflink>]), it remains useful for making relative comparisons (e.g., higher or lower levels of segregation across different characteristics) and can be decomposed into within vs. between sources of segregation. This makes it attractive for our purposes, to establish the extent to which segregation across classes is occurring through school choice/selection compared to teacher/class allocations.</p> <p>The first characteristic we consider is gender. We do so because segregation of boys and girls across primary teachers/classes is likely to be minimal. This hence provides a useful minimum benchmark against which the segregation indices for other characteristics can be judged (e.g., Pupil Premium eligibility, SEN, English as an Additional Language [EAL], low/high levels of prior achievement).</p> <hd id="AN0189830383-8">RQ3, RQ4 and RQ5</hd> <p>In RQ3, RQ4 and RQ5 we turn to estimation of TVA models. We follow the extensive US literature in specifying our models for TVA. This is based on the 'two‐step' approach set out by Bacher‐Hicks and Koedel ([<reflink idref="bib9" id="ref37">9</reflink>], p. 100).</p> <p>With respect to mathematics test scores,[<reflink idref="bib2" id="ref38">2</reflink>] the most extensive version of the first step model is specified as follows:2 <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0008" display="block" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mi>M</mi><mo>%5f</mo><msub><mn>24</mn><mi mathvariant="italic">ijk</mi></msub><mo linebreak="goodbreak">=</mo><mi>α</mi><mo linebreak="goodbreak">+</mo><mi>β</mi><mo>.</mo><mi>T</mi><mo>_</mo><msub><mn>23</mn><mi mathvariant="italic">ijk</mi></msub><mo linebreak="goodbreak">+</mo><mi>γ</mi><mo>.</mo><mi mathvariant="italic">Ab</mi><mo>_</mo><msub><mn>23</mn><mi mathvariant="italic">ijk</mi></msub><mo linebreak="goodbreak">+</mo><mi>δ</mi><mo>.</mo><msub><mi>X</mi><mi mathvariant="italic">ijk</mi></msub><mo linebreak="goodbreak">+</mo><mi>π</mi><mo>.</mo><msub><mi>C</mi><mi mathvariant="italic">jk</mi></msub><mo linebreak="goodbreak">+</mo><msub><mi>u</mi><mi>k</mi></msub><mo linebreak="goodbreak">+</mo><msub><mi>ε</mi><mi mathvariant="italic">ij</mi></msub></mrow></semantics></math> </ephtml> where <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0009" display="inline" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mi>M</mi><mo>%5f</mo><msub><mn>24</mn><mi mathvariant="italic">ijk</mi></msub></mrow></semantics></math> </ephtml> = standardised scores across the mathematics tests pupil <emph>i</emph> sat during the 2023/24 academic year; <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0010" display="inline" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mi>T</mi><mo>%5f</mo><msub><mn>23</mn><mi mathvariant="italic">ijk</mi></msub></mrow></semantics></math> </ephtml> = a vector of prior test scores (reading and mathematics) sat during the 2022/23 academic year; <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0011" display="inline" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mi mathvariant="italic">Ab</mi><mo>_</mo><msub><mn>23</mn><mi mathvariant="italic">ijk</mi></msub></mrow></semantics></math> </ephtml> = level of pupil absences during the 2022/23 academic year; <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0012" display="inline" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><msub><mi>X</mi><mi mathvariant="italic">ijk</mi></msub></mrow></semantics></math> </ephtml> = a vector of background demographic controls; <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0013" display="inline" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><msub><mi>C</mi><mi mathvariant="italic">jk</mi></msub></mrow></semantics></math> </ephtml> = a vector of class characteristics, including class average prior achievement and class average prior attendance; <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0014" display="inline" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><msub><mi>u</mi><mi>k</mi></msub></mrow></semantics></math> </ephtml> = school fixed effects; <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0015" display="inline" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><msub><mi>ε</mi><mi mathvariant="italic">ij</mi></msub></mrow></semantics></math> </ephtml> = error term, assumed to have mean zero; <emph>i</emph> = pupil; <emph>j</emph> = teacher/class; <emph>k</emph> = school.</p> <p>Note that the residuals from this model ( <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0016" display="inline" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><msub><mi>ε</mi><mi mathvariant="italic">ij</mi></msub></mrow></semantics></math> </ephtml> ) in effect capture how much better or worse a pupil did across their 2023/24 mathematics tests than would be predicted based on their prior achievement and other variables included in the model.[<reflink idref="bib3" id="ref39">3</reflink>] After this model is estimated, a second‐stage regression is estimated with this residual entering as the dependent variable. This second‐stage regression is of the form3 <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0017" display="block" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><msub><mi>ε</mi><mi mathvariant="italic">ij</mi></msub><mo linebreak="goodbreak">=</mo><mi>τ</mi><mo linebreak="goodbreak">×</mo><msub><mtext>Teacher</mtext><mi>j</mi></msub><mo linebreak="goodbreak">+</mo><msub><mi>μ</mi><mi mathvariant="italic">ij</mi></msub></mrow></semantics></math> </ephtml> where <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0018" display="inline" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><msub><mtext>Teacher</mtext><mi>j</mi></msub></mrow></semantics></math> </ephtml> = a set of dummy variables for each teacher; <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0019" display="inline" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><msub><mi>μ</mi><mi mathvariant="italic">ij</mi></msub></mrow></semantics></math> </ephtml> = the residual error term.</p> <p>The vector of <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0020" display="inline" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mi>τ</mi></mrow></semantics></math> </ephtml> parameters from Equation (<reflink idref="bib3" id="ref40">3</reflink>) then represents our measures of TVA. These are, in effect, the mean of the residuals calculated from Equation (<reflink idref="bib2" id="ref41">2</reflink>) for each teacher. Higher averages thus correspond to higher TVA scores—teachers whose pupils tend to make greater progress over the course of an academic year than predicted. The ICCs from these models are also considered in RQ4, along with the consistency of TVA estimates across mathematics, reading and attendance in RQ5.</p> <p>Separate sets of models are used for our three outcomes (mathematics, reading and attendance) with multiple imputation (<emph>m</emph> = 10) by chained equations used to account for missing covariate data. All models are estimated separately by MAT. Three versions of this model are estimated, with the TVA estimates from each compared under RQ3. The first specification includes just lagged test scores (the vector <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0021" display="inline" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mi>T</mi><mo>%5f</mo><msub><mn>23</mn><mi mathvariant="italic">ijk</mi></msub></mrow></semantics></math> </ephtml> ) with no other controls.[<reflink idref="bib4" id="ref42">4</reflink>] The second specification additionally includes controls for prior absences ( <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0022" display="inline" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mi mathvariant="italic">Ab</mi><mo>_</mo><msub><mn>23</mn><mi mathvariant="italic">ijk</mi></msub></mrow></semantics></math> </ephtml> ), pupil demographics ( <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0023" display="inline" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><msub><mi>X</mi><mi mathvariant="italic">ijk</mi></msub></mrow></semantics></math> </ephtml> ) and class characteristics ( <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0024" display="inline" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><msub><mi>C</mi><mi mathvariant="italic">jk</mi></msub></mrow></semantics></math> </ephtml> ). School fixed effects are then also added in the final specification. Our primary interest is the extent to which including additional controls over just prior test scores changes our estimates of TVA.</p> <hd id="AN0189830383-9">RQ6</hd> <p>Student growth percentile (SGP) models are an alternative to the value‐added model approach set out in Equation (<reflink idref="bib2" id="ref43">2</reflink>). We follow the procedure of Betebenner ([<reflink idref="bib10" id="ref44">10</reflink>], [<reflink idref="bib11" id="ref45">11</reflink>]) to estimate SGP models using the TED data for primary pupils. This can be summarised as follows.</p> <p>First, we estimate 99 separate quantile regression models, from the first percentile up to the 99th percentile. The specification of this model is similar to that presented in Equation (<reflink idref="bib2" id="ref46">2</reflink>), though only including the prior achievement vector ( <ephtml> <math altimg="urn:x-wiley:01411926:media:berj4207:berj4207-math-0025" display="inline" overflow="scroll" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mi>T</mi><mo>%5f</mo><msub><mn>23</mn><mi mathvariant="italic">ijk</mi></msub></mrow></semantics></math> </ephtml> ) as covariates. From this model, one can produce predicted scores at each percentile, based on the distribution of all students with the same prior attainment. One then finds which of these predicted percentile scores falls closest to the pupil's actual score on the test. Pupils are then assigned this value, which will fall between 1 and 100. Thus, in the SGP model, TVA is calculated as the mean of the rank of their pupils' test score growth. This contrasts with our modelling approach for RQ3–RQ5 above, where TVA is estimated as the mean of the magnitude of their pupils' test score growth.[<reflink idref="bib5" id="ref47">5</reflink>] An attraction of this approach is the comparatively straightforward interpretation of the results (e.g., a student assigned an SGP of 65 can be said to have made greater progress over the course of the year than 65% of their academic peers—i.e., other children in their cohort with similar levels of prior achievement).</p> <p>As under the value‐added model approach, we calculate the average of these SGP scores for each teacher within the dataset. A teacher with an SGP score of 50 will be an 'average' value‐added teacher; their pupils made on average greater progress than 50% of their academic peers. Our primary interest in RQ6 is the extent to which TVA estimates based on this approach are similar to those based on the standard value‐added model presented in Equation (<reflink idref="bib2" id="ref48">2</reflink>). When making these comparisons, we ensure that the value‐added and SGP models are based on analogous model specifications (i.e., both approaches only include prior achievement controls).</p> <hd id="AN0189830383-10">RESULTS</hd> <p></p> <hd id="AN0189830383-11">RQ1: Basic properties of commercial primary school tests</hd> <p>Appendices S1 (MAT1) and S2 (MAT2) provide a detailed set of descriptive statistics investigating the distribution and inter‐test correlations across primary schools' internal assessments. A summary of these results is provided in Table 2.</p> <p>2 TABLE Summary of the correlations across measures.</p> <p> <ephtml> <table><thead valign="bottom"><tr><th align="left">(a) Mathematics</th></tr><tr><th align="left" /><th align="left">MAT1</th><th align="left">MAT2</th></tr></thead><tbody valign="top"><tr><td align="left">Start/end‐of‐year correlations</td></tr><tr><td align="left">Year 1</td><td align="left">—</td><td align="char" char=".">0.76</td></tr><tr><td align="left">Year 2 (autumn/spring)</td><td align="left">—</td><td align="char" char=".">0.86</td></tr><tr><td align="left">Year 3</td><td align="left">0.78</td><td align="char" char=".">0.85</td></tr><tr><td align="left">Year 4</td><td align="left">0.87</td><td align="char" char=".">0.85</td></tr><tr><td align="left">Year 5</td><td align="left">0.86</td><td align="char" char=".">0.86</td></tr><tr><td align="left">Year 6 (autumn/spring)</td><td align="left">—</td><td align="char" char=".">0.92</td></tr><tr><td align="left">Average</td><td align="left">0.83</td><td align="char" char=".">0.84</td></tr><tr><td align="left">Annual average scores</td></tr><tr><td align="left">Correlation 22/23 to 23/24</td><td align="left">0.87</td><td align="char" char=".">0.85</td></tr><tr><td align="left">Correlation with KS2 scores</td></tr><tr><td align="left">Year 6 term 1</td><td align="left">—</td><td align="char" char=".">0.90</td></tr><tr><td align="left">Year 6 term 2</td><td align="left">—</td><td align="char" char=".">0.92</td></tr><tr><td align="left">Year 6 average score</td><td align="left">—</td><td align="char" char=".">0.93</td></tr></tbody></table> </ephtml> </p> <p></p> <p> <ephtml> <table><thead valign="bottom"><tr><th align="left">(b) Reading</th></tr><tr><th align="left" /><th align="left">MAT1</th><th align="left">MAT2</th></tr></thead><tbody valign="top"><tr><td align="left">Start/end‐of‐year correlations</td></tr><tr><td align="left">Year 1</td><td align="left">—</td><td align="char" char=".">0.74</td></tr><tr><td align="left">Year 2 (autumn/spring)</td><td align="left">—</td><td align="char" char=".">0.85</td></tr><tr><td align="left">Year 3</td><td align="left">0.68</td><td align="char" char=".">0.79</td></tr><tr><td align="left">Year 4</td><td align="left">0.77</td><td align="char" char=".">0.79</td></tr><tr><td align="left">Year 5</td><td align="left">0.75</td><td align="char" char=".">0.79</td></tr><tr><td align="left">Year 6 (autumn/spring)</td><td align="left">—</td><td align="char" char=".">0.83</td></tr><tr><td align="left">Average</td><td align="left">0.73</td><td align="char" char=".">0.78</td></tr><tr><td align="left">Annual average scores</td></tr><tr><td align="left">Correlation 22/23 to 23/24</td><td align="left">0.80</td><td align="char" char=".">0.83</td></tr><tr><td align="left">Correlation with KS2 scores</td></tr><tr><td align="left">Year 6 term 1</td><td align="left">—</td><td align="char" char=".">0.81</td></tr><tr><td align="left">Year 6 term 2</td><td align="left">—</td><td align="char" char=".">0.81</td></tr><tr><td align="left">Year 6 average score</td><td align="left">—</td><td align="char" char=".">0.85</td></tr></tbody></table> </ephtml> </p> <p>2 <emph>Note</emph>: Figures refer to Pearson correlations. See Appendices S1 (MAT1) and S2 (MAT2) for further details.</p> <p>The table illustrates the correlation between the tests pupils sit at the start and end of the academic year. This can be interpreted as an approximate 8‐month test–retest correlation. Mostly, such correlation typically sits around 0.8, though it is generally slightly higher for mathematics than reading. Test–retest values of this magnitude over an 8‐month period are relatively high, pointing towards consistent measurement of mathematics/reading skills over time.</p> <p>The table also (under 'annual average scores') reports the correlation between pupils' average scores achieved across the assessments they sat in 2022/23 and compared to the average across all assessments they sat in 2023/24. This is important as the former is the key covariate—and later the outcome—in our TVA models. Such correlation falls around 0.85 for mathematics and 0.75 for reading. These strong associations illustrate how baseline scores explain most of the variation in the outcome, with the remaining residual variation being limited. This is important, as it will help limit the scope for unobserved confounding to bias estimates of TVA.</p> <p>Finally, the table illustrates how well scores on the autumn and spring Year 6 tests correlate with children's Key Stage 2 SATs scores (tests taken at the end of Year 6). These correlations are again high, standing at around 0.85 in reading and 0.9 in mathematics. This illustrates both the predictive validity of the internal assessments that schools use and how they are very closely related to their scores on another test that teachers will have not seen before.[<reflink idref="bib6" id="ref49">6</reflink>]</p> <p>Appendices S1 and S2 provide further details, including demonstrating how the tests do not suffer from ceiling or floor effects and contain a reasonable amount of variation.</p> <hd id="AN0189830383-12">RQ2: Uneven distribution of pupils across primary school classes</hd> <p>Table 3 presents results from our analysis of primary school pupil segregation across classes. Figures are presented for the dissimilarity index (<emph>D</emph>) and Theil's <emph>H</emph>, with the latter also decomposed into between‐school and within‐school components.</p> <p>3 TABLE Estimates of primary school pupil segregation across different classes.</p> <p> <ephtml> <table><thead valign="bottom"><tr><th align="left">(a) Trust 1</th></tr><tr><th align="left" /><th align="left">Dissimilarity index</th><th align="left">Theil index</th><th align="left">Theil between</th><th align="left">Theil within</th></tr></thead><tbody valign="top"><tr><td align="left">Gender</td><td align="char" char=".">0.117</td><td align="char" char=".">0.016</td><td align="char" char=".">0.004</td><td align="char" char=".">0.012</td></tr><tr><td align="left">Pupil Premium</td><td align="char" char=".">0.216</td><td align="char" char=".">0.062</td><td align="char" char=".">0.021</td><td align="char" char=".">0.041</td></tr><tr><td align="left">SEN</td><td align="char" char=".">0.251</td><td align="char" char=".">0.064</td><td align="char" char=".">0.030</td><td align="char" char=".">0.034</td></tr><tr><td align="left">EAL</td><td align="char" char=".">0.353</td><td align="char" char=".">0.146</td><td align="char" char=".">0.127</td><td align="char" char=".">0.019</td></tr><tr><td align="left">Low achievers</td><td align="char" char=".">0.464</td><td align="char" char=".">0.188</td><td align="char" char=".">0.099</td><td align="char" char=".">0.089</td></tr><tr><td align="left">High achievers</td><td align="char" char=".">0.386</td><td align="char" char=".">0.149</td><td align="char" char=".">0.078</td><td align="char" char=".">0.070</td></tr></tbody></table> </ephtml> </p> <p></p> <p> <ephtml> <table><thead valign="bottom"><tr><th align="left">(b) Trust 2</th></tr><tr><th align="left" /><th align="left">Dissimilarity index</th><th align="left">Theil index</th><th align="left">Theil between</th><th align="left">Theil within</th></tr></thead><tbody valign="top"><tr><td align="left">Gender</td><td align="char" char=".">0.134</td><td align="char" char=".">0.022</td><td align="char" char=".">0.004</td><td align="char" char=".">0.018</td></tr><tr><td align="left">Pupil Premium</td><td align="char" char=".">0.291</td><td align="char" char=".">0.090</td><td align="char" char=".">0.056</td><td align="char" char=".">0.035</td></tr><tr><td align="left">SEN</td><td align="char" char=".">0.244</td><td align="char" char=".">0.059</td><td align="char" char=".">0.009</td><td align="char" char=".">0.050</td></tr><tr><td align="left">EAL</td><td align="char" char=".">0.286</td><td align="char" char=".">0.095</td><td align="char" char=".">0.063</td><td align="char" char=".">0.032</td></tr><tr><td align="left">Low achievers</td><td align="char" char=".">0.297</td><td align="char" char=".">0.092</td><td align="char" char=".">0.026</td><td align="char" char=".">0.066</td></tr><tr><td align="left">High achievers</td><td align="char" char=".">0.262</td><td align="char" char=".">0.078</td><td align="char" char=".">0.025</td><td align="char" char=".">0.053</td></tr></tbody></table> </ephtml> </p> <p>3 <emph>Note</emph>: The dissimilarity index illustrates the percentage of pupils with the characteristic that would need to change teacher/class in order for the distribution to be even across classes. Lower values of this index therefore indicate a greater degree of evenness in the distribution of pupils across classes. Lower (higher) values of the Theil index indicate lower (higher) segregation of pupils with the characteristic across different primary classes (teachers). The 'Theil between' column indicates the extent to which segregation occurs across schools within the MAT (i.e., sorting of pupils with that particular characteristic into different schools). The 'Theil within' column indicates the extent to which segregation occurs within schools (i.e., sorting of pupils with the characteristic into different classes).</p> <p>As expected, the estimates point towards non‐random allocation of pupils across teachers/classes. Take MAT2, for example. With respect to gender—a characteristic for which we would anticipate the extent of non‐random allocation across primary school classrooms to be low—the dissimilarity index equals 0.13. This means that 13% of boys and girls would need to swap classes for the gender balance to be completely evenly spread. The analogous figures for other pupil characteristics are notably higher: 0.29 for Pupil Premium status; 0.24 for SEN; 0.29 for EAL; 0.26 for low prior achievement; 0.3 for high prior achievement. Together, this illustrates how—within these MATs—there is some degree of segregation of pupils with certain characteristics into different classes.</p> <p>At the same time, we note how the overall degree of segregation is relatively mild; while there is evidence of non‐random allocation of pupils across classes, the extent of the selection does not stand out as overly strong. For instance, within the broader segregation literature, values of <emph>D</emph> below 0.3 have been classified as low, and between 0.3 and 0.6 as moderate (Iceland & Sharp, [<reflink idref="bib23" id="ref50">23</reflink>]). Each of the six characteristics considered for MAT2 thus fall into the 'low' category, while in MAT1 they are all either moderate (EAL, low/high prior achievement) or low (Pupil Premium, SEN). Moreover, while values of both <emph>D</emph> and <emph>H</emph> for each characteristic are higher than for gender, the differences are relatively modest in size.</p> <p>Table 3 is also useful for helping to build our understanding of the extent to which the overall level of segregation occurs within versus between schools. The generally higher degree of segregation for EAL and prior achievement in MAT1 seems largely driven by greater between‐school differences. This may possibly be due to this MAT comprising a more diverse range of primary schools. More generally, the within‐school <emph>H</emph>‐index values for Pupil Premium, SEN and EAL are all relatively close to the value for gender. This indicates that segregation by these characteristics is, to a certain extent, being driven by differences in school selection or residential separation; the part attributable to differences in class assignments within school appears relatively weak. The within‐school <emph>H</emph>‐index values for low/high levels of achievement do, however, continue to suggest that academically weaker/stronger pupils are—at least to some extent—unevenly allocated across classes within schools. Appendix S8 provides further evidence on this issue, presenting estimates of the dissimilarity index by school (capturing unevenness in the distribution of pupils with given characteristics across classes).</p> <p>Thus, in summary, there is indeed evidence of an uneven distribution of pupils with certain characteristics across primary school teachers/classes. However, the extent of this uneven allocation appears to be moderate, and to some extent driven by differences in pupil composition across schools. This, in turn, has some important implications for our estimates of TVA. Whether such estimates truly capture the effect of teachers partly depends on the extent to which the selection of pupils into different teachers/classes can be controlled for (in terms of factors that are also likely to be correlated with future achievement). The moderate levels of segregation by key observable characteristics in Table 3 suggest that these selection effects may not be overly strong.</p> <hd id="AN0189830383-13">RQ3: Consistency of TVA estimates across model specifications</hd> <p>In RQ3, we turn to the consistency of TVA estimates across different model specifications. Appendix S3 provides a full set of scatterplots illustrating the strength of the association of our TVA estimates when using three different sets of controls (lagged test scores only, inclusion of pupil and class characteristics, addition of school fixed effects). These are then summarised in Table 4 via reporting of the Pearson correlations.</p> <p>4 TABLE Comparison of TVA estimates across model specifications.</p> <p> <ephtml> <table><thead valign="bottom"><tr><th align="left">(a) MAT1</th></tr><tr><th align="left" /><th align="left">Correlation</th></tr></thead><tbody valign="top"><tr><td align="left">Maths</td></tr><tr><td align="left">Lagged scores vs. full controls</td><td align="char" char=".">0.88</td></tr><tr><td align="left">Full controls vs. full controls + school FE</td><td align="char" char=".">0.87</td></tr><tr><td align="left">Reading</td></tr><tr><td align="left">Lagged scores vs. full controls</td><td align="char" char=".">0.88</td></tr><tr><td align="left">Full controls vs. full controls + school FE</td><td align="char" char=".">0.89</td></tr><tr><td align="left">Attendance</td></tr><tr><td align="left">Lagged attendance vs. full controls</td><td align="char" char=".">0.98</td></tr><tr><td align="left">Full controls vs. full controls + school FE</td><td align="char" char=".">0.69</td></tr></tbody></table> </ephtml> </p> <p></p> <p> <ephtml> <table><thead valign="bottom"><tr><th align="left">(b) MAT2</th></tr><tr><th align="left" /><th align="left">Correlation</th></tr></thead><tbody valign="top"><tr><td align="left">Maths</td></tr><tr><td align="left">Lagged scores vs. full controls</td><td align="char" char=".">0.88</td></tr><tr><td align="left">Full controls vs. full controls + school FE</td><td align="char" char=".">0.87</td></tr><tr><td align="left">Reading</td></tr><tr><td align="left">Lagged scores vs. full controls</td><td align="char" char=".">0.91</td></tr><tr><td align="left">Full controls vs. full controls + school FE</td><td align="char" char=".">0.91</td></tr><tr><td align="left">Attendance</td></tr><tr><td align="left">Lagged attendance vs. full controls</td><td align="char" char=".">0.99</td></tr><tr><td align="left">Full controls vs. full controls + school FE</td><td align="char" char=".">0.91</td></tr></tbody></table> </ephtml> </p> <p>4 <emph>Note</emph>: Figures refer to the Pearson correlation between the TVA estimates across model specifications.</p> <p>Overall, the correlations are high, generally sitting close to 0.9 or above. The one slight exception is the addition of school fixed effects in the attendance model for MAT1, where the correlation is somewhat weaker (0.69). These results show that once we account for prior achievement, we also take care of many other factors that could confound the results, including most differences between schools. Thus, while these results cannot rule out unobserved confounders biasing estimates of TVA using these data, the limited change to our estimates after controlling for the key additional observed confounders is reassuring. It is also consistent with studies from elementary schools in the United States, which have found 'simply conditioning on lagged achievement yields approximately unbiased VAMs' (Gershenson, [<reflink idref="bib18" id="ref51">18</reflink>], p. 1).</p> <hd id="AN0189830383-14">RQ4: Estimates of the ICC</hd> <p>Table 5 presents estimates of the ICC from our preferred model specification, with controls included for prior achievement, pupil and class characteristics. These are based on two‐level random effects (multi‐level) models, with pupils (level 1) clustered within teachers/classes (level 2). Figures refer to the percentage of the variation in the 2023/24 outcomes that occurs within primary school classes/teachers. Appendix S6 provides additional analyses considering how the ICCs change when using different model specifications, and without using multiple imputation.</p> <p>5 TABLE Estimates of the conditional ICCs.</p> <p> <ephtml> <table><thead valign="bottom"><tr><th align="left" /><th align="left">MAT 1</th><th align="left">MAT 2</th></tr></thead><tbody valign="top"><tr><td align="left">Maths</td><td align="char" char=".">33.7%</td><td align="char" char=".">22.0%</td></tr><tr><td align="left">Reading</td><td align="char" char=".">27.8%</td><td align="char" char=".">11.6%</td></tr><tr><td align="left">Absences</td><td align="char" char=".">1.1%</td><td align="char" char=".">0.9%</td></tr></tbody></table> </ephtml> </p> <p>5 <emph>Note</emph>: Figures refer to the (conditional) ICC capturing the (conditional) variation in 2023/24 outcomes that occurs across teachers.</p> <p>There are four key points to note. First, the ICCs are sizable in reading and mathematics across both MATs. For instance, in MAT2, 22% of the conditional variation in 2023/24 mathematics scores occurs within primary classes/teachers. Second, the ICCs are larger in mathematics than reading (34% vs. 28% of the variation occurs within teachers/classes in MAT1 and 22% vs. 12% in MAT2). This is consistent with evidence from the international literature, where elementary teachers have been found to have greater impact on pupils' mathematics than reading test scores (Hanushek & Rivkin, [<reflink idref="bib22" id="ref52">22</reflink>]). Third, the ICCs are consistently larger in MAT1 than in MAT2, potentially suggesting greater variation in the impact of primary school teachers on children's test scores across these trusts. Finally, the ICCs for attendance are very close to zero, indicating little evidence of variation across primary teachers/classes. This suggests that primary teachers have very little impact on pupil attendance in England, which contrasts with findings from the United States (Gershenson, [<reflink idref="bib18" id="ref53">18</reflink>], [<reflink idref="bib19" id="ref54">19</reflink>]; Liu & Loeb, [<reflink idref="bib28" id="ref55">28</reflink>]).</p> <hd id="AN0189830383-15">RQ5: Comparisons of TVA across reading, mathematics and attendance</hd> <p>Figure 2 provides scatterplots comparing the TVA estimates across reading and mathematics (panel a) and for each subject compared to attendance (panels b and c). The results presented are for MAT2, with the analogous estimates for MAT1 provided in Appendix S4.</p> <p> <img src="https://imageserver.ebscohost.com/img/embimages/rdk/BED/01dec25/berj4207-fig-0002.jpg?ephost1=dGJyMNXb4kSepq84yOvqOLCmsE6epq5Srqa4SK6WxWXS" alt="berj4207-fig-0002.jpg" title="2 Comparison of TVA estimates across mathematics, reading and absences. Correlation = 0.54 (left) and − 0.04 (right). Correlation = −0.10. See Appendix S4 for analogous results for MAT1." /> </p> <p></p> <p>The correlation between TVA estimates in reading and mathematics is reasonably strong, standing at 0.79 in MAT1 and 0.54 in MAT2. These are in line with the 0.66 reported by Coe et al. ([<reflink idref="bib15" id="ref56">15</reflink>]), based on an analysis of data from MAT1 comparing autumn and summer‐term test scores in 2022/23.</p> <p>The correlation between TVA in reading/mathematics and TVA in absences is clearly much weaker. Across the two MATs, the Pearson correlation sits between −0.04 and −0.12. This is similar to analogous analyses conducted in the United States, such as the correlation of −0.12 for TVA in mathematics and absences reported by Liu and Loeb ([<reflink idref="bib28" id="ref57">28</reflink>]), and values very close to zero found by Gershenson ([<reflink idref="bib18" id="ref58">18</reflink>], [<reflink idref="bib19" id="ref59">19</reflink>]). The interpretation US academics have placed on these near‐zero correlations is that they show teachers who are effective at boosting children's test scores are not the same teachers who are effective at boosting children's attendance. However, given the low ICCs for attendance reported under RQ4, an alternative interpretation is that this finding reflects the fact that primary teachers have very little impact on pupil attendance. In other words, these near‐zero correlations reflect the fact that variation in TVA on attendance is simply capturing spurious noise.</p> <hd id="AN0189830383-17">RQ6: Comparison of student growth percentile approaches to estimating TVA</hd> <p>RQ6 compares estimates of TVA from a standard value‐added model to those based on an SGP approach. These results are presented in Figure 3 for MAT2, with those for MAT1 provided in Appendix S5. Figures along the horizontal axis provide the TVA estimates using a version of the value‐added model described in Equation (<reflink idref="bib2" id="ref60">2</reflink>), with those from the analogous SGP approach presented on the vertical axis.</p> <p> <img src="https://imageserver.ebscohost.com/img/embimages/rdk/BED/01dec25/berj4207-fig-0003.jpg?ephost1=dGJyMNXb4kSepq84yOvqOLCmsE6epq5Srqa4SK6WxWXS" alt="berj4207-fig-0003.jpg" title="3 Comparison of TVA estimates across value‐added model (VAM) and student growth percentile (SGP) approaches. Correlation = 0.97 (left) and 0.98 (right). Results are for MAT2. Analogous results for MAT1 are presented in Appendix S5. Estimates from the VAM are presented along the horizontal axis, with those from the SGP on the vertical axis." /> </p> <p></p> <p>The two sets of estimates are very strongly correlated. For both MATs, the correlations across the standard value‐added and SGP approaches stand at 0.97 or above in reading and mathematics. It is thus evident that the use of either model leads to very similar substantive results, replicating similar findings from the United States (Kurtz, [<reflink idref="bib27" id="ref61">27</reflink>]). This is an important finding, as non‐statistical audiences often find results from SGPs conceptually simpler and more straightforward to interpret than estimates from value‐added models (Lockwood & Castellano, [<reflink idref="bib29" id="ref62">29</reflink>]). The use of SGPs may thus prove a more effective way to present results of analyses investigating TVA to audiences that do not work in this field.</p> <hd id="AN0189830383-19">RQ7: Caveats with estimates of primary TVA scores</hd> <p>A series of caveats need to be kept in mind when interpreting the above results. Whereas prior US studies have access to data spanning several years, the TED at the time of writing contains information for just 2022/23 and 2023/24. This means we are currently unable to examine cross‐year stability in TVA estimates, verifying that high value‐added teachers in one year are also high value‐added teachers in the next. It also means our estimates of individual TVA will be relatively imprecise, given the limited number of pupils taught by each primary teacher (<emph>n</emph> ~ 30). Indeed, studies from the United States suggest that TVA estimates from more than one timepoint may be needed before they become reasonably stable (Bitler et al., [<reflink idref="bib12" id="ref63">12</reflink>]). It also means we are unable to perform some of the common validation checks often carried out in US studies, such as whether future teacher allocations predict pupils' prior test‐score growth (Rothstein, [<reflink idref="bib34" id="ref64">34</reflink>]) and those that exploit teacher mobility across years (Chetty et al., [<reflink idref="bib14" id="ref65">14</reflink>]). These current challenges will, however, hopefully be resolvable once the TED matures.</p> <p>As set out in Allen et al. ([<reflink idref="bib1" id="ref66">1</reflink>]), extra data are also required to conduct additional validation checks to provide further reassurance that our estimates of TVA are capturing the effect of teachers. This includes replicating previous US findings—such as TVA estimates increasing with teacher experience—that one would anticipate to hold in England as well. Likewise, further data are required to cross‐validate the TVA estimates against other external measures, such as student surveys and classroom observations, as has been done in the United States (Bacher‐Hicks et al., [<reflink idref="bib8" id="ref67">8</reflink>]). While not currently possible, such analyses should be a priority when additional data become available.</p> <p>There are some issues that are unlikely to be resolved, and where appropriate caveats will always be needed. One is that TVA estimates in primary schools—using schools' internal assessment data—will almost certainly be restricted in focus to the core subjects of reading and mathematics. It is thus not possible to rule out that high value‐added teachers are those who prioritise these subjects, potentially to the detriment of others (which go unmeasured). Likewise, we do not know how high/low value‐added teachers compare in terms of developing pupils' broader outcomes, such as their wellbeing, meta‐cognition and self‐confidence.</p> <p>Another such example is missing data. In the literature from the United States, state‐wide assessments are often used that likely have comparatively low levels of missing data. When using schools' internal assessment data, it is likely that levels of missing data will be greater. This is particularly true for certain year groups—depending on how and when schools issue such assessments—and may impact pupils with some characteristics more than others. We discuss this issue of missing data in Appendix S9 in further detail and illustrate how our substantive estimates of TVA appear broadly similar whether missing covariate data are imputed or not. Imputation is not, however, a silver bullet to solve issues surrounding missing data, essentially involving estimating what the missing values would likely be from the data observed. We consequently believe that this issue of missing data is particularly worthy of further investigation as the TED database matures and develops.</p> <p>Finally, one selection problem that is difficult to measure (and thus overcome) is how TAs are allocated across classes. For instance, most primary teachers in England will be paired with a TA to work with, and such allocations are unlikely to be random. There may also be a risk of collinearity, with the same teacher–TA matches occurring over time. This may consequently mean it will always prove difficult to tease apart the relative contribution of different members of staff.</p> <p>Readers should bear these caveats in mind when interpreting estimates of TVA for primary school teachers in England. Whether they truly capture approximately unbiased estimates of teacher effects on pupil learning will no doubt continue to be debated. The results presented above, however, do lead us to be cautiously optimistic in their potential to inform key issues in education policy and research.</p> <hd id="AN0189830383-20">CONCLUSIONS</hd> <p>Teachers are the lifeblood of the education system. They are the individuals who take responsibility for educating the next generation of young people, and in whom parents place their trust. An extensive academic literature has thus been devoted to studying this profession, including various indicators of teacher effectiveness. A valuable subset of this literature that has emerged—primarily from the United States—has focused on estimating teacher value‐added (TVA) on pupils' test scores. In other words, how much academic progress do pupils make when they are taught by a given member of staff relative to others? Unfortunately, analogous evidence outside of the United States has been comparatively limited due to a lack of available data.</p> <p>This paper has sought to start addressing this issue in a country (England) where previous attempts to estimate TVA—particularly amongst primary school teachers—have been very limited. It thus represents one of the first such studies within this empirical setting to date. Our results have shown promising signs regarding the quality of newly available data to estimate TVA, including reassuring psychometric properties of the tests conducted and relatively mild degrees of selection of primary pupils into different classes. We find evidence of substantial teacher effects on test scores, with between 20% and 30% of the conditional variation in mathematics progress over the course of an academic year occurring amongst pupils taught within the same class (and by the same teacher). Yet no analogous effect has been found for pupil attendance. Reassuringly, broadly similar findings emerge across subjects (reading and mathematics), model specification and estimation approach.</p> <p>Many of these findings resonate closely with equivalent results from the United States. For instance, the standard deviation of TVA we have estimated for England (~0.2) is similar to the values widely reported in the United States (Bacher‐Hicks & Koedel, [<reflink idref="bib9" id="ref68">9</reflink>]). Likewise, our finding that estimates of TVA are broadly stable across different model specifications and statistical approaches (ordinary least squares in value‐added models vs. quantile regression in student percentile growth models) is consistent with findings from US studies such as Chetty et al. ([<reflink idref="bib14" id="ref69">14</reflink>]) and Kurtz ([<reflink idref="bib27" id="ref70">27</reflink>]). However, while we replicate US findings that TVA on test scores is very weakly correlated with TVA on attendance (as reported by Gershenson, [<reflink idref="bib18" id="ref71">18</reflink>], [<reflink idref="bib19" id="ref72">19</reflink>] and Liu & Loeb, [<reflink idref="bib28" id="ref73">28</reflink>]), our substantive interpretation of this null effect differs. Rather than indicating that those teachers who are effective at raising test scores are not the same teachers who are effective at raising attendance (Gershenson, [<reflink idref="bib18" id="ref74">18</reflink>], [<reflink idref="bib19" id="ref75">19</reflink>]), we believe this near‐zero correlation (accompanied by the near‐zero ICC) reflects that primary school teachers have little impact on whether their pupils turn up for school or not.</p> <p>Our findings should, of course, have a caveat given the limitations of our study and the data currently available. First, we appreciate ongoing debates regarding whether estimates from such models purely capture the effect of teachers, as previously discussed in response to RQ7 (and thus not repeated here). Second, data are only currently available from two MATs, including a moderate sample of 257 primary teachers. Future studies should thus look to replicate our findings using bigger samples from a more diverse set of schools, including those under local authority (rather than academy) control. Third, our analysis on the allocation of higher/lower value‐added teachers is limited by the fact that data are only currently available for two academic years. Further research is needed to understand how higher and lower value‐added teachers are allocated and moved across classes, as well as linking this to information about their promotion and retention. Fourth, our study focuses on primary school teachers and not those working in the secondary sector, where the estimation of TVA may well prove more complex (see Allen et al., [<reflink idref="bib1" id="ref76">1</reflink>] for a discussion of this issue). Finally, future studies should attempt to estimate teacher effects across a broader array of pupil outcomes, including the development of their socioemotional skills (e.g., self‐confidence, wellbeing, school engagement).</p> <p>Our findings nevertheless continue to have some important implications for education policy and research. They suggest that producing estimates of TVA at scale is likely possible amongst primary teachers in England that prove robust enough for research purposes and for inferences to be made at an aggregate level. This, in the future, will open up opportunities for important new pieces of evidence to emerge, such as how schools choose to allocate their highest value‐added teachers and the extent to which these individuals are retained by their school (and within the teaching profession). They may also prove useful in understanding the efficacy of different routes into teaching, and professional development opportunities designed to help teachers boost pupils' reading and mathematics skills. While the results of this paper represent just the first step along this pathway, we nevertheless believe the findings presented serve as a firm foundation upon which progress towards these goals can be developed.</p> <hd id="AN0189830383-21">FUNDING INFORMATION</hd> <p>This research was supported by the Nuffield Foundation. The Nuffield Foundation is an independent charitable trust with a mission to advance social wellbeing. It funds research that informs social policy, primarily in education, welfare and justice. It also funds student programmes that provide opportunities for young people to develop skills in quantitative and scientific methods. The Nuffield Foundation is the founder and co‐funder of the Nuffield Council on Bioethics and the Ada Lovelace Institute. The Foundation has funded this project, but the views expressed are those of the authors and not necessarily the Foundation. Visit https://<ulink href="http://www.nuffieldfoundation.org/">www.nuffieldfoundation.org/</ulink>.</p> <hd id="AN0189830383-22">CONFLICT OF INTEREST STATEMENT</hd> <p>The authors declare no conflicts of interest.</p> <hd id="AN0189830383-23">DATA AVAILABILITY STATEMENT</hd> <p>The data that support the findings of this study are due to be made available within a secure setting by the National Institute of Teaching.</p> <hd id="AN0189830383-24">ETHICS STATEMENT</hd> <p>Ethical approval for secondary analysis was granted by the UCL Institute of Education's ethics research committee.</p> <p>GRAPH: Data S1.</p> <ref id="AN0189830383-25"> <title> Footnotes </title> <blist> <bibl id="bib1" idref="ref29" type="bt">1</bibl> <bibtext> For instance, a class may have two teachers, one working with the class for 60% of the week and the other 40% of the week. Within our analysis, we would focus on the teacher working 60% of their time with the class. Where there are two teachers with a 50%/50% time split, one of the teachers is selected at random.</bibtext> </blist> <blist> <bibl id="bib2" idref="ref23" type="bt">2</bibl> <bibtext> The models for reading test scores and absences include the same set of covariates, with just the outcome variable changed.</bibtext> </blist> <blist> <bibl id="bib3" idref="ref27" type="bt">3</bibl> <bibtext> We calculate 10 sets of residuals for each—one for each imputation—and then take the average.</bibtext> </blist> <blist> <bibl id="bib4" idref="ref33" type="bt">4</bibl> <bibtext> For school absences, the only control included in this specification is lagged school absences.</bibtext> </blist> <blist> <bibl id="bib5" idref="ref28" type="bt">5</bibl> <bibtext> When covariates other than prior achievement are included in the model, unexplained by these other factors.</bibtext> </blist> <blist> <bibl id="bib6" idref="ref11" type="bt">6</bibl> <bibtext> With the same tests being used each year, it could be the case that teachers know the content and exact questions used in the internal assessments, and coach pupils so that they know the answers. The same is not possible with KS2 tests, which change each year, and thus the questions will not be known by teachers in advance.</bibtext> </blist> <blist> <bibl id="bib7" idref="ref6" type="bt">7</bibl> <bibtext> [Corrections added on 10 July 2025, after first online publication: The 6th author's first name has been corrected in this version.]</bibtext> </blist> </ref> <ref id="AN0189830383-26"> <title> REFERENCES </title> <blist> <bibtext> Allen, B., Jerrim, J., & Sims, S. (2025). Distinctive challenges with estimating teacher value added in English secondary schools.</bibtext> </blist> <blist> <bibtext> Allen, R. (2024). What is primary teaching like in England today? https://teachertapp.com/uk/articles/what‐is‐primary‐teaching‐like‐in‐england‐today/</bibtext> </blist> <blist> <bibtext> Allen, R. (2025). Assessments in schools today. https://teachertapp.com/uk/articles/assessments‐in‐schools‐today/</bibtext> </blist> <blist> <bibtext> Allen, R., Burgess, S., Davidson, R., & Windmeijer, F. (2015). More reliable inference for the dissimilarity index of segregation. The Econometrics Journal, 18, 40 – 66. https://doi.org/10.1111/ectj.12039</bibtext> </blist> <blist> <bibtext> Allen, R., Jerrim, J., Parameshwaran, M., & Thomson, D. (2018). Properties of commercial tests in the EEF database. EEF Research Paper 001 (February 2018).</bibtext> </blist> <blist> <bibtext> Allen, R., & Sims, S. (2018). Do pupils from low‐income families get low‐quality teachers? Indirect evidence from English schools. Oxford Review of Education, 44 (4), 441 – 458. https://doi.org/10.1080/03054985.2017.1421152</bibtext> </blist> <blist> <bibtext> Amrein‐Beardsley, A., & Holloway, J. (2017). Value‐added models for teacher evaluation and accountability: Commonsense assumptions. Educational Policy, 33 (3), 516 – 542. https://doi.org/10.1177/0895904817719519</bibtext> </blist> <blist> <bibl id="bib8" idref="ref67" type="bt">8</bibl> <bibtext> Bacher‐Hicks, A., Chin, M. J., Kane, T. J., & Staiger, D. O. (2019). An experimental evaluation of three teacher quality measures: Value‐added, classroom observations, and student surveys. Economics of Education Review, 73, 101919. https://doi.org/10.1016/j.econedurev.2019.101919</bibtext> </blist> <blist> <bibl id="bib9" idref="ref2" type="bt">9</bibl> <bibtext> Bacher‐Hicks, A., & Koedel, C. (2023). Estimation and interpretation of teacher value‐added in research applications. Handbook of the Economics of Education, 6, 93 – 134.</bibtext> </blist> <blist> <bibtext> Betebenner, D. W. (2009). Norm‐ and criterion‐referenced student growth. Educational Measurement: Issues and Practice, 28, 42 – 51. https://doi.org/10.1111/j.1745‐3992.2009.00161.x</bibtext> </blist> <blist> <bibtext> Betebenner, D. W. (2011). A technical overview of the student growth percentile methodology: Student growth percentiles and percentile growth trajectories/projections. The National Center for the Improvement of Educational Assessment. <ulink href="http://www.nj.gov/education/njsmart/performance/SGP%5fTechnical%5fOverview.pdf">http://www.nj.gov/education/njsmart/performance/SGP%5fTechnical%5fOverview.pdf</ulink></bibtext> </blist> <blist> <bibtext> Bitler, M., Corcoran, S., Domina, T., & Penner, E. (2021). Teacher effects on student achievement and height: A cautionary tale. Journal of Research on Educational Effectiveness, 14 (4), 900 – 924. https://doi.org/10.1080/19345747.2021.1917025</bibtext> </blist> <blist> <bibtext> Burroughs, N., Gardner, J., Lee, Y., Guo, S., Touitou, I., Jansen, K., & Schmidt, W. (2019). A review of the literature on teacher effectiveness and student outcomes. In Teaching for excellence and equity. IEA Research for Education, vol. (Vol. 6, pp. 7 – 17). Springer. https://doi.org/10.1007/978‐3‐030‐16151‐4_2</bibtext> </blist> <blist> <bibtext> Chetty, R., Friedman, J. N., & Rockoff, J. E. (2014). Measuring the impacts of teachers I: Evaluating bias in teacher value‐added estimates. American Economic Review, 104 (9), 2593 – 2632.</bibtext> </blist> <blist> <bibtext> Coe, R., Ventista, O., Chande, R., Maud, C., Kime, S., & Dillon, S. (2024). Estimating teacher value‐added from schools' internal assessments in England. https://doi.org/10.31219/osf.io/tqnjk</bibtext> </blist> <blist> <bibtext> Conceição, P., & Ferreira, P. (2000). The young person's guide to the Theil index: Suggesting intuitive interpretations and exploring analytical applications. UTIP Working Paper No. 14. LBJ School of Public Affairs, The University of Texas at Austin; Internet and Telecoms Convergence Consortium, Massachusetts Institute of Technology. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=228703</bibtext> </blist> <blist> <bibtext> Education Endowment Foundation. (2025). Attainment measures database. https://educationendowmentfoundation.org.uk/projects‐and‐evaluation/evaluation/eef‐outcome‐measures‐and‐databases/attainment‐measures‐database/am‐database?keyStage=33442</bibtext> </blist> <blist> <bibtext> Gershenson, S. (2016a). Should value‐added models control for student absences? IZA Discussion Paper No. 9978. https://docs.iza.org/dp9978.pdf</bibtext> </blist> <blist> <bibtext> Gershenson, S. (2016b). Linking teacher quality, student attendance, and student achievement. Education Finance and Policy 2016, 11 (2), 125 – 149. https://doi.org/10.1162/EDFP_a_00180</bibtext> </blist> <blist> <bibtext> Guinea‐Martin, D., & Mora, M. (2022). Computing decomposable multigroup indices of segregation. The Stata Journal, 22 (3), 521 – 556. https://doi.org/10.1177/1536867X2211244</bibtext> </blist> <blist> <bibtext> Gutiérrez, G., Jerrim, J., & Torres, R. (2020). School segregation across the world: Has any progress been made in reducing the separation of the rich from the poor? The Journal of Economic Inequality, 18, 157 – 179. https://doi.org/10.1007/s10888‐019‐09437‐3</bibtext> </blist> <blist> <bibtext> Hanushek, E. A., & Rivkin, S. G. (2010). Generalizations about using value‐added measures of teacher quality. American Economic Review, 100 (2), 267 – 271.</bibtext> </blist> <blist> <bibtext> Iceland, J., & Sharp, G. (2013). White residential segregation in U.S. metropolitan areas: Conceptual issues, patterns, and trends from the US census, 1980 to 2010. Population Research and Policy Review, 32 (5), 663 – 686. https://doi.org/10.1007/s11113‐013‐9277‐6</bibtext> </blist> <blist> <bibtext> Jenkins, S., Micklewright, J., & Schnepf, S. (2008). Social segregation in secondary schools: How does England compare with other countries? Oxford Review of Education, 34 (1), 21 – 37.</bibtext> </blist> <blist> <bibtext> Jerrim, J., & Sims, S. (2019). Teachers in primary and secondary schools: TALIS 2018. Department for Education research report. https://<ulink href="http://www.gov.uk/government/publications/teachers‐in‐primary‐and‐secondary‐schools‐talis‐2018">www.gov.uk/government/publications/teachers‐in‐primary‐and‐secondary‐schools‐talis‐2018</ulink></bibtext> </blist> <blist> <bibtext> Koedel, C., Mihaly, K., & Rockoff, J. E. (2015). Value‐added modeling: A review. Economics of Education Review, 47, 180 – 195. https://doi.org/10.1016/j.econedurev.2015.01.006</bibtext> </blist> <blist> <bibtext> Kurtz, M. D. (2018). Value‐added and student growth percentile models: What drives differences in estimated classroom effects? Statistics and Public Policy, 5 (1), 1 – 8. https://doi.org/10.1080/2330443X.2018.1438938</bibtext> </blist> <blist> <bibtext> Liu, J., & Loeb, S. (2019). Engaging teachers: Measuring the impact of teachers on student attendance in secondary school. The Journal of Human Resources, 56 (2), 1216‐8430R3. https://doi.org/10.3368/jhr.56.2.1216‐8430R3</bibtext> </blist> <blist> <bibtext> Lockwood, J. R., & Castellano, K. E. (2015). Alternative statistical frameworks for student growth percentile estimation. Statistics and Public Policy, 2 (1), 1 – 9. https://doi.org/10.1080/2330443X.2014.962718</bibtext> </blist> <blist> <bibtext> Maisuria, A., Roberts, N., Long, R., & Danechi, S. (2023). Teacher recruitment and retention in England. House of Commons Library No. 07222. https://researchbriefings.files.parliament.uk/documents/CBP‐7222/CBP‐7222.pdf</bibtext> </blist> <blist> <bibtext> Massey, D., & Duncan, N. (1988). The dimensions of residential segregation. Social Forces, 67 (2), 281 – 315.</bibtext> </blist> <blist> <bibtext> McCaffrey, D. F., Lockwood, J. R., Koretz, D. M., & Hamilton, L. S. (2004). Evaluating value‐added models for teacher accountability. https://<ulink href="http://www.rand.org/content/dam/rand/pubs/monographs/2004/RAND%5fMG158.sum.pdf">www.rand.org/content/dam/rand/pubs/monographs/2004/RAND%5fMG158.sum.pdf</ulink></bibtext> </blist> <blist> <bibtext> Reezigt, G. (2001). A framework for effective school improvement. https://cordis.europa.eu/docs/projects/files/SOE/SOE2972027/70595591‐6_en.pdf</bibtext> </blist> <blist> <bibtext> Rothstein, J. (2010). Teacher quality in educational production: Tracking, decay, and student achievement. Quarterly Journal of Economics, 125 (1), 175 – 214.</bibtext> </blist> <blist> <bibtext> Rothstein, J. (2013). Effects of value‐added policies. Focus, 29 (2), 23 – 24.</bibtext> </blist> <blist> <bibtext> Slater, H., Davies, N. M., & Burgess, S. (2012). Do teachers matter? Measuring the variation in teacher effectiveness in England. Oxford Bulletin of Economics and Statistics, 74 (5), 629 – 645.</bibtext> </blist> <blist> <bibtext> Tymms, P., Merrell, C., Hawker, D., & Nicholson, F. (2014). Performance indicators in primary schools: A comparison of performance on entry to school and the progress made in the first year in England and four other jurisdictions: Research report. Department for Education.</bibtext> </blist> <blist> <bibtext> Tymms, P., Merrell, C., & Henderson, B. (1997). The first year at school: A quantitative investigation of the attainment and progress of pupils. Educational Research and Evaluation, 3 (2), 101 – 118. https://doi.org/10.1080/1380361970030201</bibtext> </blist> <blist> <bibtext> Tymms, P., Merrell, C., & Henderson, B. (2000). Baseline assessment and progress during the first three years at school. Educational Research and Evaluation, 6 (2), 105 – 129. https://doi.org/10.1076/1380‐3611(200006)6:2;1‐E;F105</bibtext> </blist> <blist> <bibtext> UNESCO. (2024). Global report on teachers: Addressing teacher shortages and transforming the profession (978‐92‐3‐100655‐5). (CC BY‐SA 3.0 IGO).</bibtext> </blist> </ref> <aug> <p>By John Jerrim; Rebecca Allen; Maria Palma Carvajal; Raj Chande; Rob Coe; Calum Davey; Shaun Dillon; Claire Maud; Sam Sims and Ourania Ventista</p> <p>Reported by Author; Author; Author; Author; Author; Author; Author; Author; Author; Author</p> </aug> <nolink nlid="nl1" bibid="bib13" firstref="ref1"></nolink> <nolink nlid="nl2" bibid="bib40" firstref="ref3"></nolink> <nolink nlid="nl3" bibid="bib30" firstref="ref4"></nolink> <nolink nlid="nl4" bibid="bib33" firstref="ref5"></nolink> <nolink nlid="nl5" bibid="bib32" firstref="ref8"></nolink> <nolink nlid="nl6" bibid="bib26" firstref="ref9"></nolink> <nolink nlid="nl7" bibid="bib35" firstref="ref10"></nolink> <nolink nlid="nl8" bibid="bib15" firstref="ref13"></nolink> <nolink nlid="nl9" bibid="bib36" firstref="ref14"></nolink> <nolink nlid="nl10" bibid="bib38" firstref="ref15"></nolink> <nolink nlid="nl11" bibid="bib39" firstref="ref16"></nolink> <nolink nlid="nl12" bibid="bib37" firstref="ref17"></nolink> <nolink nlid="nl13" bibid="bib27" firstref="ref22"></nolink> <nolink nlid="nl14" bibid="bib25" firstref="ref25"></nolink> <nolink nlid="nl15" bibid="bib17" firstref="ref30"></nolink> <nolink nlid="nl16" bibid="bib21" firstref="ref31"></nolink> <nolink nlid="nl17" bibid="bib24" firstref="ref32"></nolink> <nolink nlid="nl18" bibid="bib31" firstref="ref34"></nolink> <nolink nlid="nl19" bibid="bib20" firstref="ref35"></nolink> <nolink nlid="nl20" bibid="bib16" firstref="ref36"></nolink> <nolink nlid="nl21" bibid="bib10" firstref="ref44"></nolink> <nolink nlid="nl22" bibid="bib11" firstref="ref45"></nolink> <nolink nlid="nl23" bibid="bib23" firstref="ref50"></nolink> <nolink nlid="nl24" bibid="bib18" firstref="ref51"></nolink> <nolink nlid="nl25" bibid="bib22" firstref="ref52"></nolink> <nolink nlid="nl26" bibid="bib19" firstref="ref54"></nolink> <nolink nlid="nl27" bibid="bib28" firstref="ref55"></nolink> <nolink nlid="nl28" bibid="bib29" firstref="ref62"></nolink> <nolink nlid="nl29" bibid="bib12" firstref="ref63"></nolink> <nolink nlid="nl30" bibid="bib34" firstref="ref64"></nolink> <nolink nlid="nl31" bibid="bib14" firstref="ref65"></nolink>
Header DbId: eric
DbLabel: ERIC
An: EJ1490548
AccessLevel: 3
PubType: Academic Journal
PubTypeId: academicJournal
PreciseRelevancyScore: 0
IllustrationInfo
Items – Name: Title
  Label: Title
  Group: Ti
  Data: Initial Estimates of Teacher Value-Added in English Primary Schools
– Name: Language
  Label: Language
  Group: Lang
  Data: English
– Name: Author
  Label: Authors
  Group: Au
  Data: <searchLink fieldCode="AR" term="%22John+Jerrim%22">John Jerrim</searchLink> (ORCID <externalLink term="https://orcid.org/0000-0001-5705-7954">0000-0001-5705-7954</externalLink>)<br /><searchLink fieldCode="AR" term="%22Rebecca+Allen%22">Rebecca Allen</searchLink><br /><searchLink fieldCode="AR" term="%22Maria+Palma+Carvajal%22">Maria Palma Carvajal</searchLink><br /><searchLink fieldCode="AR" term="%22Raj+Chande%22">Raj Chande</searchLink><br /><searchLink fieldCode="AR" term="%22Rob+Coe%22">Rob Coe</searchLink><br /><searchLink fieldCode="AR" term="%22Calum+Davey%22">Calum Davey</searchLink><br /><searchLink fieldCode="AR" term="%22Shaun+Dillon%22">Shaun Dillon</searchLink> (ORCID <externalLink term="https://orcid.org/0009-0005-8086-5139">0009-0005-8086-5139</externalLink>)<br /><searchLink fieldCode="AR" term="%22Claire+Maud%22">Claire Maud</searchLink><br /><searchLink fieldCode="AR" term="%22Sam+Sims%22">Sam Sims</searchLink><br /><searchLink fieldCode="AR" term="%22Ourania+Ventista%22">Ourania Ventista</searchLink>
– Name: TitleSource
  Label: Source
  Group: Src
  Data: <searchLink fieldCode="SO" term="%22British+Educational+Research+Journal%22"><i>British Educational Research Journal</i></searchLink>. 2025 51(6):2942-2963.
– Name: Avail
  Label: Availability
  Group: Avail
  Data: Wiley. Available from: John Wiley & Sons, Inc. 111 River Street, Hoboken, NJ 07030. Tel: 800-835-6770; e-mail: cs-journals@wiley.com; Web site: https://www.wiley.com/en-us
– Name: PeerReviewed
  Label: Peer Reviewed
  Group: SrcInfo
  Data: Y
– Name: Pages
  Label: Page Count
  Group: Src
  Data: 22
– Name: DatePubCY
  Label: Publication Date
  Group: Date
  Data: 2025
– Name: TypeDocument
  Label: Document Type
  Group: TypDoc
  Data: Journal Articles<br />Reports - Research
– Name: Audience
  Label: Education Level
  Group: Audnce
  Data: <searchLink fieldCode="EL" term="%22Elementary+Education%22">Elementary Education</searchLink>
– Name: Subject
  Label: Descriptors
  Group: Su
  Data: <searchLink fieldCode="DE" term="%22Foreign+Countries%22">Foreign Countries</searchLink><br /><searchLink fieldCode="DE" term="%22Value+Added+Models%22">Value Added Models</searchLink><br /><searchLink fieldCode="DE" term="%22Teacher+Effectiveness%22">Teacher Effectiveness</searchLink><br /><searchLink fieldCode="DE" term="%22Teacher+Influence%22">Teacher Influence</searchLink><br /><searchLink fieldCode="DE" term="%22Elementary+School+Teachers%22">Elementary School Teachers</searchLink><br /><searchLink fieldCode="DE" term="%22Elementary+School+Students%22">Elementary School Students</searchLink><br /><searchLink fieldCode="DE" term="%22Reading+Achievement%22">Reading Achievement</searchLink><br /><searchLink fieldCode="DE" term="%22Mathematics+Achievement%22">Mathematics Achievement</searchLink><br /><searchLink fieldCode="DE" term="%22Attendance%22">Attendance</searchLink><br /><searchLink fieldCode="DE" term="%22Barriers%22">Barriers</searchLink><br /><searchLink fieldCode="DE" term="%22Student+Characteristics%22">Student Characteristics</searchLink>
– Name: Subject
  Label: Geographic Terms
  Group: Su
  Data: <searchLink fieldCode="DE" term="%22United+Kingdom+%28England%29%22">United Kingdom (England)</searchLink>
– Name: DOI
  Label: DOI
  Group: ID
  Data: 10.1002/berj.4207
– Name: ISSN
  Label: ISSN
  Group: ISSN
  Data: 0141-1926<br />1469-3518
– Name: Abstract
  Label: Abstract
  Group: Ab
  Data: A sizeable literature investigating teacher test score value-added--the extent to which pupils make different rates of progress under different teachers--has emerged in the United States. While there is much interest in estimating teacher value-added in other countries such as England, progress has been limited by the lack of datasets linking teachers and pupils. We overcome this issue by drawing on internal assessment data from primary schools across two multi-academy trusts. Our results suggest that a substantial proportion of the progress primary pupils make in reading and mathematics occurs across (rather than within) the teachers to which they are assigned. There is, however, no clear evidence of teacher effects on attendance. Similar results are obtained using different model specifications and approaches. The paper concludes by clearly outlining some of the remaining challenges with estimating teacher value-added in England's primary schools, and the next steps that should be prioritised in this line of research.
– Name: AbstractInfo
  Label: Abstractor
  Group: Ab
  Data: As Provided
– Name: DateEntry
  Label: Entry Date
  Group: Date
  Data: 2025
– Name: AN
  Label: Accession Number
  Group: ID
  Data: EJ1490548
PLink https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=eric&AN=EJ1490548
RecordInfo BibRecord:
  BibEntity:
    Identifiers:
      – Type: doi
        Value: 10.1002/berj.4207
    Languages:
      – Text: English
    PhysicalDescription:
      Pagination:
        PageCount: 22
        StartPage: 2942
    Subjects:
      – SubjectFull: Foreign Countries
        Type: general
      – SubjectFull: Value Added Models
        Type: general
      – SubjectFull: Teacher Effectiveness
        Type: general
      – SubjectFull: Teacher Influence
        Type: general
      – SubjectFull: Elementary School Teachers
        Type: general
      – SubjectFull: Elementary School Students
        Type: general
      – SubjectFull: Reading Achievement
        Type: general
      – SubjectFull: Mathematics Achievement
        Type: general
      – SubjectFull: Attendance
        Type: general
      – SubjectFull: Barriers
        Type: general
      – SubjectFull: Student Characteristics
        Type: general
      – SubjectFull: United Kingdom (England)
        Type: general
    Titles:
      – TitleFull: Initial Estimates of Teacher Value-Added in English Primary Schools
        Type: main
  BibRelationships:
    HasContributorRelationships:
      – PersonEntity:
          Name:
            NameFull: John Jerrim
      – PersonEntity:
          Name:
            NameFull: Rebecca Allen
      – PersonEntity:
          Name:
            NameFull: Maria Palma Carvajal
      – PersonEntity:
          Name:
            NameFull: Raj Chande
      – PersonEntity:
          Name:
            NameFull: Rob Coe
      – PersonEntity:
          Name:
            NameFull: Calum Davey
      – PersonEntity:
          Name:
            NameFull: Shaun Dillon
      – PersonEntity:
          Name:
            NameFull: Claire Maud
      – PersonEntity:
          Name:
            NameFull: Sam Sims
      – PersonEntity:
          Name:
            NameFull: Ourania Ventista
    IsPartOfRelationships:
      – BibEntity:
          Dates:
            – D: 01
              M: 12
              Type: published
              Y: 2025
          Identifiers:
            – Type: issn-print
              Value: 0141-1926
            – Type: issn-electronic
              Value: 1469-3518
          Numbering:
            – Type: volume
              Value: 51
            – Type: issue
              Value: 6
          Titles:
            – TitleFull: British Educational Research Journal
              Type: main
ResultId 1