View in EDS HTML Full Text PDF Full Text

Student Intersectional Sociodemographic and School Variation in GCSE Final Grades in England Following COVID-19 Examination Cancellations

Saved in:
Bibliographic Details
Title: Student Intersectional Sociodemographic and School Variation in GCSE Final Grades in England Following COVID-19 Examination Cancellations
Language: English
Authors: Lucy Prior (ORCID 0000-0001-6557-9082), George Leckie (ORCID 0000-0003-1486-745X)
Source: Oxford Review of Education. 2025 51(5):763-784.
Availability: Routledge. Available from: Taylor & Francis, Ltd. 530 Walnut Street Suite 850, Philadelphia, PA 19106. Tel: 800-354-1420; Tel: 215-625-8900; Fax: 215-207-0050; Web site: http://www.tandf.co.uk/journals
Peer Reviewed: Y
Page Count: 22
Publication Date: 2025
Document Type: Journal Articles
Reports - Research
Education Level: Secondary Education
Higher Education
Postsecondary Education
Descriptors: Foreign Countries, COVID-19, Pandemics, Secondary Education, Exit Examinations, Colleges, Public Schools, Secondary Schools, Small Schools, State Schools, Selective Admission, Intersectionality, Scores, Student Characteristics, Grade Inflation, Test Bias, Prediction
Geographic Terms: United Kingdom (England)
DOI: 10.1080/03054985.2024.2407624
ISSN: 0305-4985
1465-3915
Abstract: In 2020, COVID-19 forced the cancellation of all student end-of-school examinations in England. Schools were asked to provide centre assessment grades (CAGs), offering their best estimates for what students would have achieved had they sat their examinations. Although initially replaced in favour of grades calculated via an algorithm, students were eventually awarded their CAGs following widespread public outcry over the calculated grades. Whether CAGs were unfairly awarded across different student groups and schools in 2020 compared to previous years is a key question. However, existing analyses of bias in CAGs are limited by a lack of attention to potential interactions between student characteristics and thus to hidden differential grade inflation across intersectional groups. We addressed this by examining student GCSE performance in 2018, 2019 and 2020 via a Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy (MAIHDA) analysis of intersectional sociodemographic variation which we cross-classified with schools given their role in generating CAGs. Overall, a picture of stability emerged where, despite substantial overall grade inflation in 2020, the use of CAGs did not appear to have generated new or divergent intersectional relationships in comparison to previous years, suggesting CAGs showed a similar susceptibility to bias as normal examinations.
Abstractor: As Provided
Entry Date: 2026
Accession Number: EJ1500707
Database: ERIC
Full text is not displayed to guests.
FullText Links:
  – Type: pdflink
    Url: https://content.ebscohost.com/cds/retrieve?content=AQICAHj0k_4E0hTGH8RJwT4gCJyBsGNe_WN95AvKlDbXJGqwxwGHb4xCVKZibyxqUxZw7To3AAAA4zCB4AYJKoZIhvcNAQcGoIHSMIHPAgEAMIHJBgkqhkiG9w0BBwEwHgYJYIZIAWUDBAEuMBEEDJh3VQYPM4Krdi1cmAIBEICBm5NRE152rJ9yELb2sS8uMMZWzAuQ6kj7JwHis8kTCEn4J8R_ud3ODRp3ptLNGfpkpP1SItHtPZAkBJsEstqXo87BJFRHBlGp7Fh_-YrYrY4ipnAY2AhjbHZupJso7xfDHap3qXFhaLPOAL2E88DVgNsxFgiLtaMjY2VSoEC7cjHu2XuYQV7G7xGPOsQRAEGrtWX9w4IfI8w-Jo2i
Text:
  Availability: 1
  Value: <anid>AN0188054556;oxr01oct.25;2025Sep22.06:14;v2.2.500</anid> <title id="AN0188054556-1">Student intersectional sociodemographic and school variation in GCSE final grades in England following Covid-19 examination cancellations </title> <p>In 2020, Covid-19 forced the cancellation of all student end-of-school examinations in England. Schools were asked to provide centre assessment grades (CAGs), offering their best estimates for what students would have achieved had they sat their examinations. Although initially replaced in favour of grades calculated via an algorithm, students were eventually awarded their CAGs following widespread public outcry over the calculated grades. Whether CAGs were unfairly awarded across different student groups and schools in 2020 compared to previous years is a key question. However, existing analyses of bias in CAGs are limited by a lack of attention to potential interactions between student characteristics and thus to hidden differential grade inflation across intersectional groups. We addressed this by examining student GCSE performance in 2018, 2019 and 2020 via a Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy (MAIHDA) analysis of intersectional sociodemographic variation which we cross-classified with schools given their role in generating CAGs. Overall, a picture of stability emerged where, despite substantial overall grade inflation in 2020, the use of CAGs did not appear to have generated new or divergent intersectional relationships in comparison to previous years, suggesting CAGs showed a similar susceptibility to bias as normal examinations.</p> <p>Keywords: Intersectionality; examinations; schools; bias; GCSE; MAIHDA</p> <hd id="AN0188054556-2">Introduction</hd> <p></p> <hd id="AN0188054556-3">Context</hd> <p>In 2020, Covid-19 caused widespread disruption to the educational system in England, and the decision was taken to cancel GCSE (age 15/16) and A-level (age 17/18) national examinations (GOV.UK, [<reflink idref="bib13" id="ref1">13</reflink>]). The Department for Education (DfE) and the exams regulator (Ofqual) were tasked with providing an alternative solution for student grades, enabling students to progress to the next stage of education or other destinations. Schools were asked to provide their best estimate of the grade that students would have gained had they sat their examination (centre assessment grade; CAG) (Ofqual, [<reflink idref="bib38" id="ref2">38</reflink>]). However, CAGs were overly optimistic and, charged with preventing grade inflation and maintaining grading consistency across years, Ofqual initially replaced CAGs with 'calculated' grades derived using their Direct Centre-level Performance (DCP) algorithm (Kelly, [<reflink idref="bib22" id="ref3">22</reflink>]; Ofqual, [<reflink idref="bib36" id="ref4">36</reflink>]). This led to 40% of results being downgraded by one or more grades, and a public furore with media reports of bias and that students were 'robbed' of their deserved grades (BBC News, [<reflink idref="bib1" id="ref5">1</reflink>]; The Guardian, [<reflink idref="bib16" id="ref6">16</reflink>], [<reflink idref="bib17" id="ref7">17</reflink>]; Kelly, [<reflink idref="bib22" id="ref8">22</reflink>]). As a result of this outcry, the government reverted to using CAGs (or the calculated grade where this was higher; Ofqual, [<reflink idref="bib37" id="ref9">37</reflink>]).</p> <hd id="AN0188054556-4">Why investigate potential biases?</hd> <p>Like examinations in normal years, CAGs will directly impact the destinations of students, whether this be further study at A-level, apprenticeships, university, or entry into the labour market. Any distortions in grades resulting from potential biases may therefore have affected the relative (dis)advantage of student groups in terms of their destinations. For instance, the grade inflation of CAGs presented universities with an oversubscription problem in 2020 with many more disadvantaged students meeting the criteria for entry than in usual years (Kelly, [<reflink idref="bib22" id="ref10">22</reflink>]). Given the longer-term impacts of student grades on employment opportunities, income, and associated social status (Boliver, [<reflink idref="bib3" id="ref11">3</reflink>]; Murphy & Wyness, [<reflink idref="bib34" id="ref12">34</reflink>]), understanding potential biases arising from the use of CAGs will provide insight into future social mobility patterns and wider inequalities. Additionally, evaluating the 2020 grades could offer lessons for any future scenarios that may disrupt examinations (e.g. future pandemics, teacher strikes, exam boycotts, centre malpractice, etc.) and lead to the use of teacher judgements in the future.</p> <p>Studying CAGs can also inform broader debates around teacher assessments and teacher biases. Teacher assessments are widely used in educational systems for summative assessment, in England and internationally (Harlen, [<reflink idref="bib18" id="ref13">18</reflink>]), for example, the predicted grades schools provide as part of university applications (UCAS, [<reflink idref="bib51" id="ref14">51</reflink>]) or the teacher assessments that form part of Key Stage tests in English schools (GOV.UK, [<reflink idref="bib14" id="ref15">14</reflink>]). In all these settings, it is important to understand any unfairness arising from the use of teacher judgements, particularly that which is related to student characteristics, to ensure that educational assessment systems are not exacerbating social inequalities. A point that holds equally for all forms of assessment, including examinations. Evaluation of the 2020 CAGs offers a beneficial situation in which to examine these issues, as we can explore them for an entire cohort of students at the end of secondary schooling.</p> <hd id="AN0188054556-5">Conceptual background for teacher biases</hd> <p>A variety of different mechanisms could produce teacher biases. Explicit, conscious prejudice or discrimination against groups of students should hopefully be minimised through the protections arising from policies such as the Equality Act (2010) and we would not expect this to be a dominant source of bias, though it cannot be ruled out entirely. Stereotyping – the characterising of a group based on particular attributes that shapes how people interact with that group – is a commonly featured theoretical basis for biases (Magowan, [<reflink idref="bib29" id="ref16">29</reflink>]; Urhahne & Wijnia, [<reflink idref="bib52" id="ref17">52</reflink>]). Stereotyping is typically considered to be an unconscious process allowing efficient, if not necessarily accurate, judgements of student groups which may operate in positive or negative directions (Campbell, [<reflink idref="bib6" id="ref18">6</reflink>]; Ready & Wright, [<reflink idref="bib44" id="ref19">44</reflink>]). Stereotyping can feed through to teacher expectations and beliefs for pupils and thus possibly affect their assessments of students. For instance, discrimination along gender lines or by socioeconomic status could emerge through beliefs around girls outperforming boys or the expected performance of low achievers (Gibbons & Chevalier, [<reflink idref="bib12" id="ref20">12</reflink>]; Lindhal, [<reflink idref="bib28" id="ref21">28</reflink>]). Teachers may also boost grades to attempt to compensate for perceived disadvantage, as a form of encouragement, or to give students a tactical advantage, depending on the situation (Snell et al., [<reflink idref="bib46" id="ref22">46</reflink>]). Additionally, where bias is assessed in relationship to test scores, divergences could emerge because the test scores themselves may not represent the 'true' ability of students (Marcenaro-Gutierrez & Vignoles, [<reflink idref="bib30" id="ref23">30</reflink>]). Likewise, it is possible that teacher assessments and tests are capturing different aspects of student performance. For example, examinations represent performance at one point in time and include aspects of examination technique, whilst teachers' judgements may encompass a more holistic appraisal of a student. Often it is not possible to differentiate or confirm the action of these processes; rather researchers rely on the identification of systematic divergences along student characteristics as indicators of potential biases.</p> <hd id="AN0188054556-6">Previous research</hd> <p>Previous research has indicated systematic differences between teacher assessments and student test scores according to student sociodemographic characteristics, though the evidence for the strength and direction of these relationships is varied (Lee & Newton, [<reflink idref="bib25" id="ref24">25</reflink>]; Lee & Walter, [<reflink idref="bib27" id="ref25">27</reflink>]; Urhahne & Wijnia, [<reflink idref="bib52" id="ref26">52</reflink>]). For example, teachers are commonly found to overestimate girls' scores in comparison to boys (Marcenaro-Gutierrez & Vignoles, [<reflink idref="bib30" id="ref27">30</reflink>]; Plewis, [<reflink idref="bib40" id="ref28">40</reflink>]; Ready & Wright, [<reflink idref="bib44" id="ref29">44</reflink>]; Timmermans et al., [<reflink idref="bib50" id="ref30">50</reflink>]). A more mixed picture has emerged in relation to ethnicity: some have indicated potential biases against minority ethnic students (Plewis, [<reflink idref="bib40" id="ref31">40</reflink>]; Ready & Wright, [<reflink idref="bib44" id="ref32">44</reflink>]; Tenenbaum & Ruck, [<reflink idref="bib48" id="ref33">48</reflink>]); others found gaps in favour of minority students (Burgess & Greaves, [<reflink idref="bib5" id="ref34">5</reflink>]; Gibbons & Chevalier, [<reflink idref="bib12" id="ref35">12</reflink>]) or no significant differences (Lindhal, [<reflink idref="bib28" id="ref36">28</reflink>]; Marcenaro-Gutierrez & Vignoles, [<reflink idref="bib30" id="ref37">30</reflink>]). Students with Special Educational Needs (SEN) are generally marked lower by teachers than on tests, leading to greater differences between the two assessment methods for these students (Burgess & Greaves, [<reflink idref="bib5" id="ref38">5</reflink>]; Campbell, [<reflink idref="bib6" id="ref39">6</reflink>]; Reeves et al., [<reflink idref="bib45" id="ref40">45</reflink>]; Thomas et al., [<reflink idref="bib49" id="ref41">49</reflink>]). However, the overall conclusion by Thomas et al. ([<reflink idref="bib49" id="ref42">49</reflink>]) was that different methods of assessment were largely the same in terms of group differences.</p> <p>Bias is identified against students of lower socioeconomic status across various teacher judgements (Boone & Van Houtte, [<reflink idref="bib4" id="ref43">4</reflink>]; Burgess & Greaves, [<reflink idref="bib5" id="ref44">5</reflink>]; Ready & Wright, [<reflink idref="bib44" id="ref45">44</reflink>]; Timmermans et al., [<reflink idref="bib50" id="ref46">50</reflink>]). However, in relation to the difference between predicted grades (as provided to universities for applications) and achieved grades in examinations, studies have identified that disadvantaged students are generally overpredicted (Wyness, [<reflink idref="bib54" id="ref47">54</reflink>]). Given the 2020 CAGs were meant to be the teachers' best prediction of how well a student would have done had they sat their examinations, rather than a contemporaneous estimate of student achievement, we may be more likely to see a positive bias towards disadvantaged students in the following analysis. Interviews with teachers around their CAG grading decisions suggest teachers were likely to be optimistic, representing student potential on a 'good' day (Holmes et al., [<reflink idref="bib19" id="ref48">19</reflink>]).</p> <p>Research should also consider how potential biases relate to students' prior attainment. Gibbons and Chevalier ([<reflink idref="bib12" id="ref49">12</reflink>]) found the strongest differences between teacher and test assessments were observed by prior attainment rather than sociodemographic characteristics, with low attaining students tending to receive more favourable ratings from teachers relative to their test scores. Regarding predicted examination grades, research highlights the potential ceiling and floor effects that may be present at the extreme ends of the achievement spectrum (Dhillon, [<reflink idref="bib8" id="ref50">8</reflink>]). Possible interaction effects whereby high achieving disadvantaged students receive lower predicted A-level grades than their more socially advantaged counterparts have also been noted (Murphy & Wyness, [<reflink idref="bib34" id="ref51">34</reflink>]; Wyness, [<reflink idref="bib54" id="ref52">54</reflink>]).</p> <p>Interaction effects between student characteristics have received less attention in this research area (Urhahne & Wijnia, [<reflink idref="bib52" id="ref53">52</reflink>]). Ready and Wright ([<reflink idref="bib44" id="ref54">44</reflink>]) provided some evidence that the association of teacher perceptions and a student's language status differed by a student's ethnic background, whilst others have explored interactions between student gender and behaviour in how this impacts teacher ratings (Urhahne & Wijnia, [<reflink idref="bib52" id="ref55">52</reflink>]). Previous research has recognised the importance of considering the wider institutional context of student experience, whether this be as moderators of teacher bias or as an important factor in itself. For instance, Martínez et al. ([<reflink idref="bib31" id="ref56">31</reflink>]) explored how classroom assessment practices could moderate the relationship between standardised tests and teacher scores. Timmermans et al. ([<reflink idref="bib50" id="ref57">50</reflink>]) did not find teacher expectations interacted with the characteristics of the classroom population; however, they found expectations were generally higher for classrooms characterised by high-performing and advantaged children. This latter finding was echoed by Ready and Wright ([<reflink idref="bib44" id="ref58">44</reflink>]) who found that student socioeconomic status was more strongly related to teacher inaccuracy in classrooms with more disadvantaged students. The wider school environment could also play a role in differences between teacher estimates/predictions and achieved test scores with studies having indicated that grading deviations may vary by school type (Dhillon, [<reflink idref="bib8" id="ref59">8</reflink>]; Marcenaro-Gutierrez & Vignoles, [<reflink idref="bib30" id="ref60">30</reflink>]). Explicit differences between individual schools are less often given attention; however, given the role schools played in the generation of CAGs in 2020 it is important to investigate potential school differences in differential grade inflation in this study.</p> <hd id="AN0188054556-7">Previous analyses of CAGs</hd> <p>Ofqual has investigated equality concerns for the 2020 grades (Lee et al., [<reflink idref="bib26" id="ref61">26</reflink>]; Stratton et al., [<reflink idref="bib47" id="ref62">47</reflink>]). Lee et al. ([<reflink idref="bib26" id="ref63">26</reflink>]) compared relationships between student sociodemographic characteristics and the 2020 grades (calculated, CAGs, and final grades) with these same relationships for examination results from 2018 and 2019. They concluded there is little evidence that the 2020 grades disadvantaged student groups based on their characteristics, with the strongest overall difference between years being the large increase in mean grades. Stratton et al. ([<reflink idref="bib47" id="ref64">47</reflink>]) focused more specifically on the CAGs and included student, subject, and school features when considering differing relationships between years. They evidenced a ceiling effect whereby the highest prior attainers saw smaller GCSE grade increases over previous years. Additionally, they suggested that CAGs may have closed the gap somewhat between candidates from more versus less deprived areas and that grades for independent schools and small cohorts may have shown more grade inflation in 2020. Stratton et al. ([<reflink idref="bib47" id="ref65">47</reflink>]) also explored a limited number of interaction effects between student characteristics; however, these were not the central focus of the analysis, were restricted to two-way interactions, and showed few notable changes in 2020 compared to the previous years. Indeed, their overall conclusion was that whilst the CAGs were higher on average, most relationships with student, subject, and school characteristics had not substantially changed in 2020.</p> <p>These Ofqual reports utilise a new linked administrative dataset called GRADE (GRading and Admissions Data for England) (Office for National Statistics, [<reflink idref="bib35" id="ref66">35</reflink>]), which is the only available resource to assess the 2020 grading situation in England. A fundamental issue with using GRADE to explore potential teacher bias is the identification of a suitable benchmark on which to base comparisons to isolate unusual change outside of normal year-on-year variation. Lee et al. ([<reflink idref="bib26" id="ref67">26</reflink>]) and Stratton et al. ([<reflink idref="bib47" id="ref68">47</reflink>]) take the approach of using the last two 'normal' years (the 2017/18 and 2018/19 academic years) as the schema of grading (using the reformed 9–1 system) is the same. However, two years is not a long enough time period to establish 'typical' variation as you cannot rule out noise or abnormal variation in one particular year. Therefore, researchers must rely on subjective judgements of what should be considered a substantively important change. Moreover, even where notable differences are identified, this does not constitute proof of bias. Real substantive changes in grading gaps year-on-year can be confounded with any systematic changes in bias. Additionally, recall that teacher judgements are not the only form of assessment susceptible to bias (Lee & Newton, [<reflink idref="bib25" id="ref69">25</reflink>]). This is not to say that we should not evaluate the situation in 2020, but rather that any analysis of these data must acknowledge upfront the limitations of the data resource and appreciate the necessarily cautious nature of any claims of change in 2020. Furthermore, given the subjective nature of judgements, it is doubly important for researchers to provide independent verification of the claims of no differential grade inflation drawn by Ofqual.</p> <p>Analyses of the 2020 grades arising outside of official Ofqual reports are still rare. One exception is a recent study by Magowan ([<reflink idref="bib29" id="ref70">29</reflink>]) who examined the differences between the 2020 CAGs and the predicted grades derived from fitting prediction models to the 2018 and 2019 data. They found that whilst relative bias across different student characteristics was small, when three-way interactions were considered, the differences were more substantial. For example, the largest intra-group range Magowan ([<reflink idref="bib29" id="ref71">29</reflink>]) identified in terms of main effects was just over a grade in a total grade score (where students were taking at least eight GCSEs) (1.097, for their deprivation measure), whereas the largest intra-group range for studied three-way interactions was almost four grades (3.771) for ethnicity</p> <p>Graph</p> <p> <ephtml> <math xmlns="http://www.w3.org/1998/Math/MathML"><mo>×</mo></math> </ephtml> deprivation</p> <p>Graph</p> <p> <ephtml> <math xmlns="http://www.w3.org/1998/Math/MathML"><mo>×</mo></math> </ephtml> prior attainment combinations. Magowan ([<reflink idref="bib29" id="ref72">29</reflink>]) therefore highlights the importance of considering interaction effects, and we expand on these multidimensional perspectives using a novel quantitative approach to studying intersections of student characteristics.</p> <hd id="AN0188054556-8">Intersectionality</hd> <p>The evaluation of potential biases in teacher assessments according to interactions of student characteristics warrants further examination. Neglecting interactions can mean important differences across student groups are missed, whilst investigating interactions of student characteristics can provide a more nuanced portrayal of inequality. The study of interactions links to the conceptual background of intersectionality (Crenshaw, [<reflink idref="bib7" id="ref73">7</reflink>]), which focuses on how social systems of oppression (e.g. racism, classism, sexism) are inherently interrelated. The characteristics of individuals position them at the intersections of these mutually constituted social systems, giving rise to heterogeneous experiences of (dis)advantage which are beyond what may be discerned from a purely unidimensional understanding of identity (Green et al., [<reflink idref="bib15" id="ref74">15</reflink>]). Thus, stereotypes and expectations for students may develop differently for different intersections of student characteristics. For example, teachers may perceive the behaviour of Black boys more negatively than other student groups, potentially influencing judgements of ability (Wint et al., [<reflink idref="bib53" id="ref75">53</reflink>]). Interlocking systems of (dis)advantage may also influence student performance, for instance, disadvantaged White students are considered a group 'forgotten' by the UK educational system (House of Commons Education Committee, [<reflink idref="bib20" id="ref76">20</reflink>]). Therefore, in evaluating potential bias in the 2020 grades, we can draw upon intersectional perspectives to provide a richer understanding of the dimensions of inequality and to reveal potentially hidden marginalisation in grading practices.</p> <hd id="AN0188054556-9">This study</hd> <p>In this study, we explored intersectional and school-level variation in the 2020 final grades for GCSE students in England, comparing this to variation found for examination grades in 2018 and 2019. By evaluating intersectional interaction effects and explicitly considering individual school effects for possible changes in 2020, we addressed key gaps in current studies of the 2020 CAGs and studies of teacher biases more widely. This study also serves as an important independent investigation to verify Ofqual's claims of little to no differential grade inflation resulting from the switch to CAGs in 2020.</p> <hd id="AN0188054556-10">Data</hd> <p>This study used data from the new GRADE linked administrative dataset (Office for National Statistics, [<reflink idref="bib35" id="ref77">35</reflink>]). This dataset is a joint venture, combining information from Ofqual, the DfE National Pupil Database (NPD), and the Universities and Colleges Admissions Service (UCAS). We utilised data on student GCSE grades from 2018 and 2019 (the last 'normal' years pre-pandemic) and final grades from 2020. We chose to use final grades in 2020 rather than CAGs as the final grades awarded to students are those that ultimately mattered to students. However, we refer to these 2020 final grades as CAGs to differentiate them from typical examination years and as CAGs were used in most cases (only 4.8% of individual grades in our sample represent calculated grades). In supplementary analyses, we found that these 4.8% were distributed similarly across our student characteristics to cases where the CAG was the same or higher than the calculated grade (see Supplementary Table S1). In the interests of space, we did not evaluate differential grade inflation at A-level; however, we encourage similar studies of these higher-level qualifications.</p> <p>We used data on student sociodemographic characteristics from the NPD. Our sample comprises 'typical' GCSE students: those in state schools receiving their grades age 16, who took five or more GCSEs including English and mathematics, and for whom we had complete sociodemographic information. We excluded students from independent schools as these were largely not captured in the NPD and would lead to unacceptably high rates of missingness. In 2018, the sample consisted of 398,181 students within 3,328 schools; for 2019 the sample was 435,599 students in 3,406 schools, and in 2020, it was 425,031 students in 3,437 schools.</p> <p>For each year, we calculated an average GCSE score to serve as our outcome, pooling across all subjects taken. This accounted for the differing number of GCSEs that students took (minimum 5, average 7 for all years; see Supplementary Table S2 for full breakdown); however, we did not consider subject specific biases. In supplementary analyses, we explored results for a set of individual subject grades but found similar results to those for the combined score in regard to changes year to year (see Supplementary Table S3, and Supplementary Figures S1 and S2). We focused on the following student sociodemographic characteristics: sex (Male, Female), ethnicity (White, Black, Asian, Chinese, Mixed, Other), and Income Deprivation Affecting Children Index (IDACI) score split into tertiles (High, Medium, Low). Additionally, we used a combined English and mathematics score from Key Stage 2 (KS2; age 11) split into deciles to capture prior attainment differences. We combined information on student sex, ethnicity, deprivation, and prior attainment to create intersectional strata representing the combination of these social identities, giving 360 intersections in total (10</p> <p>Graph</p> <p> <ephtml> <math xmlns="http://www.w3.org/1998/Math/MathML"><mo>×</mo></math> </ephtml> 2</p> <p>Graph</p> <p> <ephtml> <math xmlns="http://www.w3.org/1998/Math/MathML"><mo>×</mo></math> </ephtml> 6</p> <p>Graph</p> <p> <ephtml> <math xmlns="http://www.w3.org/1998/Math/MathML"><mo>×</mo></math> </ephtml> 3). We chose these characteristics to balance capturing the most salient proxies for prominent systems of marginalisation, and what has previously been identified as of potential importance for biases, with manageable interpretation. Preliminary descriptive analyses summarising students' average grades across these student characteristics are detailed in the supplementary material (see Supplementary Figure S3). Additional student characteristics (Special Educational Needs [SEN], Free School Meal status [FSM], and speaking English as an Additional Language [EAL]) were also explored in these preliminary analyses to help identify which characteristics might exhibit differential grade inflation and were ruled out on this basis (Supplementary Figure S4).</p> <p>We also included school type and school size as these characteristics were highlighted in other reports of differential grade inflation (Stratton et al., [<reflink idref="bib47" id="ref78">47</reflink>]). School type was split into academies (schools funded directly by the government with more control over how they are run), comprehensives (schools run by the Local Authority), selective (grammar schools that actively select students based on high achievement), sixth form colleges, and other (covering all remaining school types, such as further education and tertiary establishments). School size was the number of students attending the school grouped as: 50 or fewer students; between 50 and 100 students; between 100 and 200; and over 200 students. Descriptive summaries of student grades by school type and size are available in the supplementary material (Supplementary Figure S3).</p> <hd id="AN0188054556-11">Methods</hd> <p>We drew upon a novel technique utilising multilevel linear regression to quantitatively assess intersectional variation in student average grades: multilevel analysis of individual heterogeneity and discriminatory accuracy (intersectional MAIHDA) (Evans et al., [<reflink idref="bib10" id="ref79">10</reflink>], [<reflink idref="bib9" id="ref80">9</reflink>]; Merlo, [<reflink idref="bib33" id="ref81">33</reflink>]). This method has so far received little application in educational settings (Keller et al., [<reflink idref="bib21" id="ref82">21</reflink>]; Prior et al., [<reflink idref="bib41" id="ref83">41</reflink>]). It involves treating intersectional social identities as contexts in which individuals are situated. Under the MAIHDA approach we can parsimoniously evaluate many intersections of multiple social dimensions simultaneously and directly quantify the power of intersectional strata to classify individuals according to their outcomes (Evans et al., [<reflink idref="bib10" id="ref84">10</reflink>], [<reflink idref="bib9" id="ref85">9</reflink>]). A technical appendix of the basic MAIHDA approach is provided in the supplementary materials.</p> <p>The treatment of intersectional strata as contexts may seem unusual to those familiar with applications of multilevel modelling to naturally aggregated units such as schools or neighbourhoods. However, the use of sociodemographic characteristics as clusters in multilevel models is not without precedent (see work by Gelman & Hill, [<reflink idref="bib11" id="ref86">11</reflink>]). The argument for this approach relies on the conceptual understanding that the practices, structures, and cultures by which and through which social categories gain meaning also create similarities of experience by which individuals can be clustered. In this application we may consider how students occupying the same intersecting positionalities (i.e. Black disadvantaged girls or White advantaged boys) may share social experiences with peers and teachers, and more widely how they are framed by discourse, policy, or stereotypes. This gives rise to potential dependencies in observations between students sharing an intersectional identity, which we account for through the multilevel structure. The MAIHDA approach to intersectional positionalities is argued as a key benefit as it helps to avoid 'blaming the victim': the categories are features of strata, not individuals (Evans et al., [<reflink idref="bib10" id="ref87">10</reflink>]).</p> <p>Our students were therefore nested within intersectional strata defined by combinations of student characteristics. We were also interested in school-level variation and the assessment of differences in average grades across individual school effects, so we extended the MAIHDA two-level model by cross-classifying students (level-1) as simultaneously but separately nested within their intersectional identities (level-2) and their schools (also conceptually at level-2; Leckie, [<reflink idref="bib23" id="ref88">23</reflink>]).</p> <p>The analytical strategy involved fitting a set of three multilevel models, repeated separately for each of the three years (2018, 2019 and 2020). Model 1 was an unadjusted two-way cross-classified model with no covariates. From this model, we assessed to what degree intersectional strata and schools explained overall variation in student average grades. It is important to note that baseline variation for the intersectional strata represents the action of both the main effects of the sociodemographic components as well as their interactions. In Model 2, we added the sociodemographic components of the intersectional strata into the fixed portion of the model. This accounted for the main effects of these characteristics, with any remaining stratum-level variation representing the action of two- and higher-way interactions between components. At the school level, the control for sociodemographic main effects accounted for differences in school mean grades which are predicted by school variation in these factors. Thus, the model moves closer to isolating the effect of school practices and policies on student average grades. In our final model (Model 3) we further adjusted the models by entering the school characteristics as main effect covariates. We fit all models by maximum likelihood estimation using the mixed command in Stata (Leckie, [<reflink idref="bib24" id="ref89">24</reflink>]).</p> <hd id="AN0188054556-12">Results</hd> <p></p> <hd id="AN0188054556-13">Intersectional stratum and school variation</hd> <p>The intercepts and variance components from our separate models predicting average GCSE grades in 2018, 2019 and 2020 are provided in Table 1. The overall grade inflation arising from the use of CAGs in 2020 over the 'normal' examination years was evident from the intercepts of the unadjusted model (Model 1): the average grade in 2018 was 5.13, 5.07 in 2019, and 5.59 in 2020, representing an increase of approximately half a grade on average. The increase in 2020 over 2019 was clearly substantively important. For example, a half-grade difference on average would be the equivalent of a student achieving one grade higher in four out of eight subjects. This overall grade inflation is well known and was a major impetus for the initial use of calculated grades in place of CAGs.</p> <p>Table 1. Model 1, 2 and 3 results for 2018, 2019, and 2020, including intercepts plus strata, school and student variances, variance partitioning coefficients (VPCs), and proportional changes in variance (PCVs).</p> <p> <ephtml> <table><thead><tr><td /><td /><td /><td>2018</td><td>2019</td><td>2020</td></tr><tr><td /><td /><td /><td>Est.</td><td>S.E</td><td>Est.</td><td>S.E</td><td>Est.</td><td>S.E</td></tr></thead><tbody><tr><td>Model 1: Unadjusted</td><td>Intercept</td><td>5.13</td><td>0.063</td><td>5.07</td><td>0.068</td><td>5.59</td><td>0.068</td></tr><tr><td>Strata variance</td><td>1.39</td><td>0.105</td><td>1.60</td><td>0.121</td><td>1.65</td><td>0.124</td></tr><tr><td>School variance</td><td>0.19</td><td>0.005</td><td>0.18</td><td>0.005</td><td>0.13</td><td>0.004</td></tr><tr><td>Student variance</td><td>1.22</td><td>0.003</td><td>1.23</td><td>0.003</td><td>1.14</td><td>0.002</td></tr><tr><td>VPC</td><td>Strata</td><td>49.7%</td><td /><td>53.1%</td><td /><td>56.4%</td><td /></tr><tr><td /><td>School</td><td>6.8%</td><td /><td>6.0%</td><td /><td>4.6%</td><td /></tr><tr><td>PCV</td><td>Strata</td><td>–</td><td /><td>–</td><td /><td>–</td><td /></tr><tr><td /><td>School</td><td>–</td><td /><td>–</td><td /><td>–</td><td /></tr><tr><td>Model 2: Student characteristics</td><td>Intercept</td><td>4.37</td><td>0.023</td><td>4.28</td><td>0.022</td><td>4.82</td><td>0.022</td></tr><tr><td>Strata variance</td><td>0.01</td><td>0.001</td><td>0.01</td><td>0.001</td><td>0.01</td><td>0.001</td></tr><tr><td>School variance</td><td>0.19</td><td>0.005</td><td>0.18</td><td>0.005</td><td>0.13</td><td>0.004</td></tr><tr><td>Student variance</td><td>1.22</td><td>0.003</td><td>1.23</td><td>0.003</td><td>1.14</td><td>0.002</td></tr><tr><td>VPC</td><td>Strata</td><td>0.4%</td><td /><td>0.4%</td><td /><td>0.5%</td><td /></tr><tr><td /><td>School</td><td>13.4%</td><td /><td>12.8%</td><td /><td>10.4%</td><td /></tr><tr><td>PCV</td><td>Strata</td><td>99.6%</td><td /><td>99.6%</td><td /><td>99.6%</td><td /></tr><tr><td /><td>School</td><td>0.2%</td><td /><td>0.1%</td><td /><td>0.2%</td><td /></tr><tr><td>Model 3: School characteristics</td><td>Intercept</td><td>4.44</td><td>0.024</td><td>4.35</td><td>0.024</td><td>4.85</td><td>0.024</td></tr><tr><td>Strata variance</td><td>0.01</td><td>0.001</td><td>0.01</td><td>0.001</td><td>0.01</td><td>0.001</td></tr><tr><td>School variance</td><td>0.17</td><td>0.005</td><td>0.16</td><td>0.004</td><td>0.13</td><td>0.004</td></tr><tr><td>Student variance</td><td>1.22</td><td>0.003</td><td>1.23</td><td>0.003</td><td>1.14</td><td>0.002</td></tr><tr><td>VPC</td><td>Strata</td><td>0.4%</td><td /><td>0.4%</td><td /><td>0.5%</td><td /></tr><tr><td /><td>School</td><td>11.8%</td><td /><td>11.4%</td><td /><td>9.9%</td><td /></tr><tr><td>PCV</td><td>Strata</td><td>0.1%</td><td /><td>0.0%</td><td /><td>0.0%</td><td /></tr><tr><td /><td>School</td><td>12.8%</td><td /><td>11.9%</td><td /><td>5.5%</td><td /></tr></tbody></table> </ephtml> </p> <p>The unadjusted model also shows that student-level variation was smaller in 2020 (1.14) than the previous years (1.22 in 2018; 1.23 in 2019). The CAGs also showed smaller school-level variation (0.13 in 2020 versus 0.19 in 2018 and 0.18 in 2018). In contrast, stratum-level variation was highest in 2020 (1.65), however this was closer to the 2019 figure (1.60) than the variation in 2019 was to that in 2018 (1.39). Therefore, we could not conclude with confidence that this was evidence of a greater reliance on student intersectional sociodemographic characteristics for the 2020 CAGs.</p> <p>In Model 1, there was a similar amount of stratum-level variation (as a percentage of total variation) in all three years, with slightly less in 2018 (49.7%) and slightly more in 2019 and 2020 (53.1% and 56.4% respectively). Therefore, a strong degree of variability in average grades was associated with the intersections of sociodemographic characteristics to which a student belonged, likely driven primarily by student prior attainment. Therefore, more broadly for the study of educational inequalities, it may be worthwhile to consider intersections of key characteristics as a context of interest for policy, monitoring and interventions for student achievement.</p> <p>In the unadjusted model, this variation represented the action of both main and interactional effects. However, it is important to recall that approximately 40 per cent of overall variation remained at the student level (within stratum and within school) representing the action of other student characteristics not used in defining our stratum. In contrast, there was a much smaller degree of variation positioned between schools, the highest percentage being 6.8% in 2018, followed by 6.0% in 2019, and just 4.6% in 2020. This suggests that schools were of lesser importance to students' average grades in 2020 than in the previous years (as much as we could establish with only three years to evaluate).</p> <p>In Model 2, any remaining stratum-level variation corresponded to the action of any two- and higher-way interaction effects, as we controlled for the sociodemographic component main effects. Stratum-level variance dropped dramatically in Model 2 for all three years: the sociodemographic main effects explained the same proportion of stratum-level variation (99.6%) between the unadjusted and adjusted models. This left a similar small proportion of stratum-level variation in all three years (0.4% in 2018 and 2019, and 0.5% in 2020). Therefore, the strength of the intersectional strata lay in the main effects of the constituent characteristics with very little attributable to interaction effects above and beyond these main effects. It is also notable that this patterning of variation was remarkably stable across all three years studied, suggesting that the intersectional characteristics analysed had similar explanatory power over grades despite the unusual situation in 2020. The breakdown of the relative power of additive versus multiplicative effects for student outcomes may appear as a key benefit of intersectional MAIHDA. However, researchers should not fall victim to thinking that finding additive effects predominant invalidates intersectional thinking for educational inequalities. It merely suggests that for this particular outcome, in this specific subset of students, inequality patterns were operating more consistently across other axes of comparison.</p> <p>Additionally, by controlling for student characteristics in Model 2, we accounted for any element of school-level variation by cohort differences between schools in these characteristics, with remaining variation more likely to reflect the action of school practices and context on student learning and assignment of CAGs. The controls in Model 1 were much less powerful at explaining school-level variation than stratum-level (the percentage explained between Model 1 and Model 2 was 0.2% in 2018 and 2020, and 0.1% in 2019). That the student sociodemographic characteristics had similar explanatory power in 2020 as they did in previous years could be seen as a positive result regarding equalities considerations. If the explanatory power of these sociodemographic variables were to be much higher at the school level in 2020 versus previous years this might have suggested that teachers relied heavily on these characteristics in constructing the student CAGs. As a result of the dramatic drop in the stratum-level variance, the percentage of remaining CAG variation seen at the school level was higher in Model 2 than in Model 1 (13.4% in 2018, 12.8% in 2019 and 10.4% in 2020). We also continued to see smaller school-level variation in 2020 than the previous normal examination years.</p> <p>In Model 3, we were particularly interested in seeing how our school-level factors (type and size) impact on the remaining school-level variation. Our school characteristics explained more school-level variance in 2018 (12.8%) and 2019 (11.9%) than in 2020 (5.5%). That these school-level factors were not as powerful in 2020, and by a notable margin (the percentage change was half that seen in the previous years) suggested the process of producing CAGs may have worked to reduce mean school differences by these school characteristics.</p> <hd id="AN0188054556-14">Coefficient comparison</hd> <p>Figure 1 (top) plots the regression coefficients from Model 2 along with their 95% confidence intervals (full results given in Supplementary Table S4). We will briefly note any divergent main effects relationships in 2020, following Stratton et al. ([<reflink idref="bib47" id="ref90">47</reflink>]) in deeming differences greater than 0.10 of a grade (and where the difference between 2020 and 2019 is greater than that between 2018 and 2019) as notable. One standard deviation in the overall mean grade scores across all three years was 1.7 grade points; therefore, our threshold criteria is highly cautious (less than one tenth of a standard deviation), allowing us to identify even marginal effects.</p> <p>Graph: Figure 1. Regression coefficients and their 95% confidence intervals for the student sociodemographic characteristics from Model 2 (top row) and for the school characteristics from Model 3 (bottom row) for 2018, 2019 and 2020.</p> <p>For the student characteristics, none of the 2020 coefficients met both elements of our criteria for notable divergences. The 2020 coefficient for the highest KS2 decile was 0.10 grade points lower than in 2019, which could indicate the action of a ceiling effect (you cannot predict above the top grades, limiting possible grade inflation). However, this was not out of bounds of 'normal' variability given the difference to the coefficient in 2018 was greater (0.30). Notably, excepting decile 10, every 2020 regression coefficient was greater in absolute magnitude than in 2018 or 2019, so KS2 appeared to be a stronger predictor in 2020 than in the previous years. In contrast, the coefficients for ethnicity and deprivation were smaller in magnitude in 2020.</p> <p>Figure 1 (bottom) provides the regression coefficients related to the school characteristics from Model 3 (see Supplementary Table S5 for full results). Selective schools showed smaller grade inflation in 2020 (coefficient was 0.37 in 2020 compared to 0.48 in 2019), likely reflecting the impact of the ceiling effect. There was a very large difference (1.01 grade points) between the 2020 and 2019 coefficients for sixth form colleges. However, the 2020 coefficient was not significant, likely due to the very small number of sixth forms in our GCSE sample (these institutions typically cater for higher education levels), so we will not overemphasise this result. Additionally, the 2020 coefficient for other school types (−0.51) was substantially smaller than in 2019 (−0.62): a 0.11 difference compared with a 0.09 difference between 2018 and 2019. However, this seemed to be a continuation of a trend rather that a particular divergence for 2020. Finally, the smallest schools (<50 students) showed a less negative relationship with average grade in 2020 (−0.09) than in 2019 (−0.28) or 2019 (−0.26). This difference was just over one tenth of a standard deviation in the overall grade distribution at 0.18 points.</p> <p>Our coefficient comparisons revealed that the predictive power of student characteristics appeared to have shifted towards prior attainment in 2020. However, the model results generally revealed a pattern of relative stability through time, particularly regarding intersectional variation. A closer look at the stratum effects is warranted to evaluate whether the apparent consistency holds for the individual intersectional strata, particularly those in the adjusted models representing the action of two-way or higher interactions.</p> <hd id="AN0188054556-15">Stratum effects</hd> <p>Figure 2 provides scatterplots comparing the estimated stratum effects for the unadjusted models (top row) and the model adjusted for prior attainment and sociodemographic characteristics (bottom row) across all year combinations (Supplementary Figure S5 provides the plot additionally controlling for school characteristics). In Model 1 the correlations between years were extremely high (<emph>r</emph> = 0.99) evidencing a strong degree of consistency from 2018 through to 2020, driven by the dominant main effects. Therefore, the rank ordering of the strata was very stable despite the overall grade inflation in 2020. In Model 2, where the stratum effects represented the action of any two- or higher-way interactions between the student characteristics, the correlations were lessened (<emph>r</emph> = 0.66 comparing 2018 and 2019, and <emph>r</emph> = 0.68 comparing 2019 and 2020). There was generally less stability year-on-year in interactional effects. However, we saw neither a substantially lower correlation between 2020 and 2019 than between 2019 and 2018 (suggesting that sociodemographic interactions played out differently in 2020 than in previous years) nor a dramatically higher correlation (suggesting that there could have been excessive use of the 2019 results in assigning CAGs).</p> <p>Graph: Figure 2. Scatterplots of stratum effects for the unadjusted model (Model 1, top row) and model adjusted for student sociodemographic characteristics (Model 2, bottom row) comparing 2018 and 2019 (left) and 2019 and 2020 (right).</p> <p>Given our unusual operationalisation of what would typically be individual characteristics as higher-level contexts, readers may wonder at the impact of stratum size on the estimated stratum effects and any sensitivities to the normality assumptions of the random effects. Though several smaller strata were present, the average stratum size was between approximately 1100 and 1300 depending on the year (see Supplementary Table S6) and we correspondingly find the overall degree of shrinkage to be small (see Supplementary Figure S6). Under MAIHDA the stratum effects are assigned values post-estimation using empirical Bayes prediction, which shrinks the effects towards the overall average in inverse proportion to their size. This 'shrinkage' is a conservative approach which helps protect against overinterpretation of extreme predictions and Type 1 errors of inference under multiple testing (Bell et al., [<reflink idref="bib2" id="ref91">2</reflink>]). However, estimation of the model parameters, including the intersectional variation, remains unbiased as the shrinkage is applied post-estimation, and these parameters are also unaffected by the normality assumptions (Rabe-Hesketh & Skrondal, [<reflink idref="bib43" id="ref92">43</reflink>]). In contrast, the empirical Bayes predictions can be sensitive to the normality assumptions of the random effects where cluster sizes are small (Rabe-Hesketh & Skrondal, [<reflink idref="bib43" id="ref93">43</reflink>]) but we find reasonable agreement with normal distributions for our stratum and school effects across all models (see Supplementary Figures S7 and S8 respectively).</p> <p>The correlations between schools in Model 1 (see Supplementary Figure S9) were still high, but lower than those for the intersectional strata (<emph>r</emph> = 0.80 comparing 2018 and 2019; <emph>r</emph> = 0.82 comparing 2019 and 2020). There was generally less stability year-on-year in school effects on average grades, which aligned with the literature on the instability of school performance over time (Prior et al., [<reflink idref="bib42" id="ref94">42</reflink>]). In Model 3 (the model more pertinent to the examination of school effects), the correlation remained the same for the 2019 to 2020 comparison (<emph>r</emph> = 0.82). This corroborated our earlier findings, suggesting our school characteristics had lower explanatory power in 2020 than in the previous years, and suggested there could have been excessive reliance on the 2019 results in generating CAGs.</p> <hd id="AN0188054556-16">Intersectional interactions</hd> <p>To further evaluate the intersectional interactional effects, we plotted the predicted stratum effects from Model 2 using the sociodemographic components of the strata (Figure 3). These stratum effects represent to what degree the average grades of students composing an intersectional group differ on average from that predicted by their combination of main effects. Positive effects show stratum groups who tend to score more highly on average than their main effects predict, and negative effects those who score lower. For the most part, we saw relative consistency in how these interactions played out across the three years, or where there was variation, this was present between all three years rather than the 2020 year appearing particularly divergent. Additionally, the general scale of these interactional stratum effects was small.</p> <p>Graph: Figure 3. Model 2 predicted stratum effects for the model adjusted for student sociodemographic characteristics for 2018, 2019 and 2020, by KS2 group, gender, ethnicity, and IDACI tertile.</p> <p>Through application of the same criteria as before regarding notable effects (a shift of 0.1 of a grade or more from the previous year, where the 2020 to 2019 change is greater in scale than the 2019 to 2018 change) we found only 11 (3.1%) notable effects from the 360 intersectional strata (see Supplementary Table S7). These strata represented a mix of characteristics, though all the identified groups feature ethnic minority students with only the Chinese and White groups not appearing. The most extreme differences seen between 2020 and 2019 were not very substantial, being just over one tenth of the standard deviation in overall grades across all three years (0.17) at 0.18 (other ethnicity, female, high deprivation, KS2 decile 6) and −0.20 (Black, female, mid deprivation, KS2 decile 4). However, we will not place too much weight on these 'notable' stratum changes as similarly sized differences were also observed between 2018 and 2019, where 3.9% (14 strata) meet the 0.1 grade difference threshold (see Supplementary Table S8).</p> <p>Plots such as Figure 3 demonstrate the utility of the MAIHDA approach in facilitating examination of multiple interactions simultaneously. Whilst many of those here were close to zero – an indication that these groups performed closely to what we expected from their combination of main effects, some patterns were notable. For instance, the u-shape shown by White male students where those with low or high prior achievement exhibited higher mean grades on average than what would be expected from their main effects. A pattern not replicated for White female students who showed a more linear increasing pattern across the prior achievement distribution, particularly for less deprived students. We focused here on the interaction effects in isolation as we were primarily concerned with year-on-year comparisons in their manifestation in student grades. However, where student inequalities specifically are of interest, it is preferable to examine the total predicted stratum effects, accounting for both the main effects and interactions, in order to understand the context of a student's score. Two different strata could both show positive interaction effects of similar size, but one may be overperforming from an initially high position, whilst another may be working to compensate against a generally low scoring standpoint. Examination of stratum effects in this manner would help to identify which student groups are suffering academically or achieving more highly, information that could then be used to target further investigation to reveal the processes that may give rise to these differing experiences.</p> <p>The differences between years in the predicted school effects from our fully adjusted model (Model 3) were considerably larger than those of the stratum effects (see Supplementary Figure S10). Whilst most school differences were close to zero (representing relative stability between years), some schools shifted by half a grade or more, and a small number shifted by one or more grades in average GCSE score. The overall variation was smaller for changes between 2020 and 2019 (standard deviation 0.22 for 2020 to 2019, compared with 0.26 for 2019 to 2018), with a greater frequency of schools appearing to make little change between the two years. This aligned with our previous findings regarding the smaller variability between schools in 2020 and that there was likely considerable reliance on the school 2019 results in producing the CAGs.</p> <hd id="AN0188054556-17">Discussion</hd> <p>In this study, we contribute to the growing body of work analysing the student CAGs awarded in England in lieu of examination grades when GCSE examinations were cancelled in response to Covid-19 in 2020. Drawing upon the notion of intersectionality, we focused on addressing the need for further examination of interaction effects in assessments of whether the switch to CAGs brought about any new or divergent relationships with student background characteristics. Overall, the picture that emerged is one of stability in intersectional relationships over time. Our intersectional MAIHDA analysis revealed that the combination of social and demographic identities to which a student belongs has the same explanatory power in all three years considered. Additionally, correlations between intersectional effects were similar through time, suggesting the use of CAGs did not dramatically alter the rank ordering of students. Moreover, only 3.1 per cent of interactional stratum effects in 2020 showed notable (>0.10 grade points) differences from 2019, with a similar percentage of such differences also found between 2018 and 2019. It appears that the move to CAGs in 2020 largely did not introduce any substantial divergences from previous years in relationships with student characteristics, even when we considered intersectional interaction effects. Therefore, this independent investigation draws similar conclusions to the Ofqual equalities analyses (Lee et al., [<reflink idref="bib26" id="ref95">26</reflink>]; Stratton et al., [<reflink idref="bib47" id="ref96">47</reflink>]).</p> <p>As with the previous analyses (Stratton et al., [<reflink idref="bib47" id="ref97">47</reflink>]), our results indicate a ceiling effect resultant from the overall grade inflation, and we highlight a shift in predictive power towards KS2 scores in 2020. This suggests heavy reliance on prior attainment by teachers and schools when assigning CAGs, rather than on individual achievements or performance in the intervening years. This may have reduced the accuracy of CAGs as a reflection of a student's current ability. The higher predictive power of prior attainment may also reflect the fact that CAGs factored out some of the typical unpredictable sources of variability in student outcomes, such as last minute 'cramming' for exams, shock events, or exam question variation. Teachers gave students the 'benefit of the doubt' (Holmes et al., [<reflink idref="bib19" id="ref98">19</reflink>]). There may also have been a conscious or unconscious effort by teachers not to base judgements on student demographics to avoid bias, lessening their predictive power in favour of prior attainment.</p> <p>The most notable differences arising in 2020 were between schools. Selective schools showed less grade inflation. As selective school students are expected to be multiply advantaged, in ways beyond that captured by our student characteristics, this result likely reflected the impact of the ceiling effect on highest grades. Additionally, smaller schools appeared to benefit more from the move to CAGs than schools with larger cohorts. Smaller schools, and therefore smaller classes, could represent a different mix of subjects than at larger institutions, and these small-class subjects may have been marked particularly optimistically under the 2020 CAGs. Elsewise, teachers in smaller schools may have felt indirect pressure towards greater optimism due to potentially closer relationships with their students, and subsequently feeling more invested in student outcomes. There may also have been fewer resources for CAG moderation processes at smaller schools, and Holmes et al. ([<reflink idref="bib19" id="ref99">19</reflink>]) indicate smaller centres tended to take less data-driven approaches to CAG generation, relying more on subjective evidence.</p> <p>Again, these results are similar to those identified by Stratton et al. ([<reflink idref="bib47" id="ref100">47</reflink>]). Though these shifts were generally small, they highlighted potential sources of unfairness in the awarding of the 2020 CAGs and avenues for further research. Whilst we controlled for important student background characteristics, and thus many cohort differences between schools, there are likely to be other student factors (and thus potential sources of bias) associated with attendance at selective or small-sized schools that we did not capture. Additionally, the distribution of different school types is not evenly distributed around England, meaning the exploration of geographic dimensions to potential inequalities arising from the use of CAGs in 2020 will be an important opportunity for future work.</p> <p>However, we demonstrate that school-level variation was lower in 2020 than the previous 'normal' examination years and that our studied school characteristics had substantially less explanatory power in 2020. Where between-school differences were smaller, this could suggest schools had similar approaches to producing CAGs, although it could also reflect the naturally smaller variation arising from the overall grade inflation and the associated ceiling effect. Moreover, our results highlighted that there may have been heavy reliance on the 2019 results when determining 2020 grades at the school level. Teachers, particularly those at larger centres using more data-driven approaches (Holmes et al., [<reflink idref="bib19" id="ref101">19</reflink>]), may have relied on knowledge built up from past cohorts with the most recent (2019) being the most prominent.</p> <p>As highlighted in the introduction, this study was necessarily limited by the data available from GRADE (Office for National Statistics, [<reflink idref="bib35" id="ref102">35</reflink>]). Most notably, our ability to determine what constituted 'normal' year-on-year variation was limited to just two years where we could maintain comparability in grading scales. We chose a relatively modest criteria for identifying 'notable' shifts and changes at 0.1 of a grade (less than one tenth of a standard deviation in overall average grades). That very few variables or intersectional stratum effects met this criterion means our overall conclusion remains that the switch to CAGs in 2020 largely did not noticeably widen or narrow pre-existing inequalities according to student, school, or intersectional characteristics. This result could be used to give confidence to the use of teacher assessments in any future scenarios of examination disruption. However, the results here, whilst representative of a cohort of state school students in England, are not necessarily generalisable to other contexts or situations. For instance, we cannot speak to potential biases at independent schools and the reliance that could be made on prior achievement by the CAGs may not be possible under widespread use of teacher assessments in place of examinations.</p> <p>Additionally, it is important to recall that CAGs were an unusual teacher assessment in that they were partly a predicted grade (how well a student would have done had examinations gone ahead). This contrasts with the teacher-assessed grades that were issued in 2021, which represented the grades at which students had evidenced achievement through other assessments (Ofqual, [<reflink idref="bib39" id="ref103">39</reflink>]). Previous work by McManus et al. ([<reflink idref="bib32" id="ref104">32</reflink>]) has shown that performance at university correlates better with attained rather than predicted grades at A-level. Given the general optimism displayed by the 2020 CAGs, this could suggest that some of the 2020 university entrants may under attain at university relative to other cohorts. Fairness in relation to other cohorts, particularly those who took traditional examinations, is another issue that must be considered before endorsement of the CAG approach, and these aspects will be important for future work to explore. Further issues such as the burden of work for teachers in compiling assessments to make grading estimates, and the treatment of candidates without a school placement, should also be deliberated in weighing up the relative merit of teacher assessments as a replacement for examinations.</p> <hd id="AN0188054556-18">Acknowledgements</hd> <p>This work contains statistical data from ONS which is Crown Copyright. The use of ONS statistical data and data from other owners in this work does not imply the endorsement of the ONS or other data owners in relation to the interpretation or analysis of the statistical data. This work uses research datasets which may not exactly reproduce National Statistics aggregates. Analysis was carried out in the Secure Research Service, part of the Office for National Statistics.</p> <hd id="AN0188054556-19">Disclosure statement</hd> <p>No potential conflict of interest was reported by the authors.</p> <hd id="AN0188054556-20">Data availability statement</hd> <p>The dataset used in this analysis is available on application from the Office for National Statistics, doi: https://doi.org/10.57906/k68n-bt74.</p> <hd id="AN0188054556-21">Supplementary materials</hd> <p>Supplemental data for this article can be accessed online at https://doi.org/10.1080/03054985.2024.2407624</p> <ref id="AN0188054556-22"> <title> References </title> <blist> <bibl id="bib1" idref="ref5" type="bt">1</bibl> <bibtext> BBC News. (2020, August 13). A-levels: Anger over 'unfair' results this year. https://<ulink href="http://www.bbc.co.uk/news/education-53759832">www.bbc.co.uk/news/education-53759832</ulink></bibtext> </blist> <blist> <bibl id="bib2" idref="ref91" type="bt">2</bibl> <bibtext> Bell, A., Holman, D., & Jones, K. (2019). Using shrinkage in multilevel models to understand intersectionality. Methodology, 15 (2), 88 – 96. https://doi.org/10.1027/1614-2241/a000167</bibtext> </blist> <blist> <bibl id="bib3" idref="ref11" type="bt">3</bibl> <bibtext> Boliver, V. (2011). Expansion, differentiation, and the persistence of social class inequalities in British higher education. Higher Education, 61 (3), 229 – 242. https://doi.org/10.1007/s10734-010-9374-y</bibtext> </blist> <blist> <bibl id="bib4" idref="ref43" type="bt">4</bibl> <bibtext> Boone, S., & Van Houtte, M. (2013). Why are teacher recommendations at the transition from primary to secondary education socially biased? A mixed-method research. British Journal of Sociology of Education, 34 (1), 20 – 38. https://doi.org/10.1080/01425692.2012.704720</bibtext> </blist> <blist> <bibl id="bib5" idref="ref34" type="bt">5</bibl> <bibtext> Burgess, S., & Greaves, E. (2013). Test scores, subjective assessment, and stereotyping of ethnic minorities. Journal of Labor Economics, 31 (3), 535 – 576. https://doi.org/10.1086/669340</bibtext> </blist> <blist> <bibl id="bib6" idref="ref18" type="bt">6</bibl> <bibtext> Campbell, T. (2015). Stereotyped at seven? Biases in teacher judgements of pupils' ability and attainment. Journal of Social Policy, 44 (3), 517 – 543. https://doi.org/10.1017/S0047279415000227</bibtext> </blist> <blist> <bibl id="bib7" idref="ref73" type="bt">7</bibl> <bibtext> Crenshaw, K. (1989). Demarginalizing the intersection of race and sex: A Black feminist critique of antidiscrimination doctrine, feminist theory and antiracist politics. University of Chicago Legal Forum, 1 (8), 139 – 167.</bibtext> </blist> <blist> <bibl id="bib8" idref="ref50" type="bt">8</bibl> <bibtext> Dhillon, D. (2005). Teachers' estimates of candidates' grades: Curriculum 2000 advanced level qualifications. British Educational Research Journal, 31 (1), 69 – 88. https://doi.org/10.1080/0141192052000310038</bibtext> </blist> <blist> <bibl id="bib9" idref="ref80" type="bt">9</bibl> <bibtext> Evans, C. R., Leckie, G., & Merlo, J. (2020). Multilevel versus single-level regression for the analysis of multilevel information: The case of quantitative intersectional analysis. Social Science & Medicine, 245, 112499. https://doi.org/10.1016/j.socscimed.2019.112499</bibtext> </blist> <blist> <bibtext> Evans, C. R., Williams, D. R., Onnela, J.-P., & Subramanian, S. V. (2018). A multilevel approach to modelling health inequalities at the intersection of multiple social identities. Social Science & Medicine, 203, 64 – 73. https://doi.org/10.1016/j.socscimed.2017.11.011</bibtext> </blist> <blist> <bibtext> Gelman, A., & Hill, J. (2006). Data analysis using regression and multilevel/hierarchical models. Cambridge University Press.</bibtext> </blist> <blist> <bibtext> Gibbons, S., & Chevalier, A. (2008). Assessment and age 16+ education participation. Research Papers in Education, 23 (2), 113 – 123. https://doi.org/10.1080/02671520802048638</bibtext> </blist> <blist> <bibtext> GOV.UK. (2020). Further details on exams and grades announced. https://<ulink href="http://www.gov.uk/government/news/further-details-on-exams-and-grades-announced">www.gov.uk/government/news/further-details-on-exams-and-grades-announced</ulink></bibtext> </blist> <blist> <bibtext> GOV.UK. (2023). The national curriculum. https://<ulink href="http://www.gov.uk/national-curriculum">www.gov.uk/national-curriculum</ulink></bibtext> </blist> <blist> <bibtext> Green, M. A., Evans, C. R., & Subramanian, S. V. (2017). Can intersectionality theory enrich population health research? Social Science & Medicine, 178, 214 – 216. https://doi.org/10.1016/j.socscimed.2017.02.029</bibtext> </blist> <blist> <bibtext> The Guardian. (2020a, August 13). A-level results: Almost 40% of teacher assessments in England downgraded. https://<ulink href="http://www.theguardian.com/education/2020/aug/13/almost-40-of-english-students-have-a-level-results-downgraded">www.theguardian.com/education/2020/aug/13/almost-40-of-english-students-have-a-level-results-downgraded</ulink></bibtext> </blist> <blist> <bibtext> The Guardian. (2020b, August 13). England A-level downgrades hit pupils from disadvantaged areas hardest. https://<ulink href="http://www.theguardian.com/education/2020/aug/13/england-a-level-downgrades-hit-pupils-from-disadvantaged-areas-hardest">www.theguardian.com/education/2020/aug/13/england-a-level-downgrades-hit-pupils-from-disadvantaged-areas-hardest</ulink></bibtext> </blist> <blist> <bibtext> Harlen, W. (2005). Trusting teachers' judgement: Research evidence of the reliability and validity of teachers' assessment used for summative purposes. Research Papers in Education, 20 (3), 245 – 270. https://doi.org/10.1080/02671520500193744</bibtext> </blist> <blist> <bibtext> Holmes, S., Churchward, D., Howard, E., Keys, E., Leahy, F., & Tonin, D. (2021). Centre assessment grades: Teaching staff Interviews, Summer 2020. Ofqual.</bibtext> </blist> <blist> <bibtext> House of Commons Education Committee. (2021). The forgotten: How white working-class pupils have been let down, and how to change it. First report of session 2021–22. House of Commons. Retrieved July 17, 2023, from https://committees.parliament.uk/publications/6364/documents/70802/default/</bibtext> </blist> <blist> <bibtext> Keller, L., Lüdtke, O., Preckel, F., & Brunner, M. (2023). Educational inequalities at the intersection of multiple social categories: An introduction and systematic review of the Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy (MAIHDA) approach. Educational Psychology Review, 35 (31), 1 – 37. https://doi.org/10.1007/s10648-023-09733-5</bibtext> </blist> <blist> <bibtext> Kelly, A. (2021). A tale of two algorithms: The appeal and repeal of calculated grades systems in England and Ireland in 2020. British Educational Research Journal, 47 (3), 725 – 741. https://doi.org/10.1002/berj.3705</bibtext> </blist> <blist> <bibtext> Leckie, G. (2013a). Cross-classified multilevel models – Concepts. LEMMA VLE Module, 12, 1 – 60. <ulink href="http://www.bristol.ac.uk/cmm/learning/online-course/">http://www.bristol.ac.uk/cmm/learning/online-course/</ulink></bibtext> </blist> <blist> <bibtext> Leckie, G. (2013b). Cross-classified multilevel models – Stata practical. LEMMA VLE Module, 12, 1 – 52. http://www/bristol.ac.uk/cmm/learning/online-course/</bibtext> </blist> <blist> <bibtext> Lee, M. W., & Newton, P. (2021). Systematic divergence between teacher and test-based assessment: Literature review. Ofqual. Retrieved May 22, 2023, from https://<ulink href="http://www.gov.uk/government/publications/systematic-divergence-between-teacher-and-test-based-assessment/systematic-divergence-between-teacher-and-test-based-assessment-literature-review">www.gov.uk/government/publications/systematic-divergence-between-teacher-and-test-based-assessment/systematic-divergence-between-teacher-and-test-based-assessment-literature-review</ulink></bibtext> </blist> <blist> <bibtext> Lee, M. W., Stringer, N., & Zanini, N. (2020). Student-level equalities analyses for GCSE and A level. Ofqual.</bibtext> </blist> <blist> <bibtext> Lee, M. W., & Walter, M. (2020). Equality impact assessment: Literature review. Ofqual. Retrieved June 27, 2023, from https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment%5fdata/file/879605/Equality%5fimpact%5fassessment%5fliterature%5freview%5f15%5fApril%5f2020.pdf :</bibtext> </blist> <blist> <bibtext> Lindhal, E. (2016). Are teacher assessments biased? Evidence from Sweden. Education Economics, 24 (2), 224 – 238. https://doi.org/10.1080/09645292.2015.1014882</bibtext> </blist> <blist> <bibtext> Magowan, L. (2023). Centre assessment grades in 2020: A natural experiment for investigating bias in teacher judgements. Journal of Computational Social Science, 6 (2), 609 – 653. https://doi.org/10.1007/s42001-023-00206-x</bibtext> </blist> <blist> <bibtext> Marcenaro-Gutierrez, O., & Vignoles, A. (2015). A comparison on teacher and test-based assessment for Spanish primary and secondary students. Educational Research, 57 (1), 1 – 21. https://doi.org/10.1080/00131881.2014.983720</bibtext> </blist> <blist> <bibtext> Martínez, J. F., Stecher, B., & Borko, H. (2009). Classroom assessment practices, teacher judgements, and student achievement in mathematics: Evidence from the ECLS. Educational Assessment, 14 (2), 78 – 102. https://doi.org/10.1080/10627190903039429</bibtext> </blist> <blist> <bibtext> McManus, I. C., Woolf, K., Harrison, D., Tiffin, P. A., Paton, L. W., Cheung, K. Y. F., & Smith, D. T. (2021). Predictive validity of A-level grades and teacher-predicted grades in UK medical school applicants: A retrospective analysis of administrative data in a time of COVID-19. BMJ Open, 16 (11), e047354. https://doi.org/10.1136/bmjopen-2020-047354</bibtext> </blist> <blist> <bibtext> Merlo, J. (2018). Multilevel analysis of individual heterogeneity and discriminatory accuracy (MAIHDA) within an intersectional framework. Social Science & Medicine, 203, 74 – 80. https://doi.org/10.1016/j.socscimed.2017.12.026</bibtext> </blist> <blist> <bibtext> Murphy, R., & Wyness, G. (2020). Minority report: The impact of predicted grades on university admissions of disadvantaged groups. Education Economics, 28 (4), 333 – 350. https://doi.org/10.1080/09645292.2020.1761945</bibtext> </blist> <blist> <bibtext> Office for National Statistics. (2021). ONS SRS metadata catalogue, GRading and admissions data England-Ofqual-DfE-UCAS, Dataset. Released December 14, 2021 https://doi.org/10.57906/k68n-bt74</bibtext> </blist> <blist> <bibtext> Ofqual. (2020a). Awarding GCSE, AS, A level, advanced extension awards and extended project qualifications in summer 2020: Interim report. https://<ulink href="http://www.gov.uk/government/publications/awarding-gcse-as-a-levels-in-summer-2020-interim-report">www.gov.uk/government/publications/awarding-gcse-as-a-levels-in-summer-2020-interim-report</ulink></bibtext> </blist> <blist> <bibtext> Ofqual. (2020b). GCSE and A level students to receive centre assessment grades. Retrieved August 17, 2020, from https://<ulink href="http://www.gov.uk/government/news/gcse-and-a-level-students-to-receive-centre-assessment-grades">www.gov.uk/government/news/gcse-and-a-level-students-to-receive-centre-assessment-grades</ulink></bibtext> </blist> <blist> <bibtext> Ofqual. (2020c, April 3). How GCSEs, AS & A levels will be awarded in summer 2020. Retrieved March 20, 2023, from https://<ulink href="http://www.gov.uk/government/news/how-gcses-as-a-levels-will-be-awarded-in-summer-2020">www.gov.uk/government/news/how-gcses-as-a-levels-will-be-awarded-in-summer-2020</ulink></bibtext> </blist> <blist> <bibtext> Ofqual. (2022). Teacher assessed grades in summer 2021: Surveys. Retrieved July 12, 2022, from https://<ulink href="http://www.gov.uk/government/publications/teacher-assessed-grades-in-summer-2022-interviews-and-surveys/teacher-assessed-grades-in-summer-2021-surveys">www.gov.uk/government/publications/teacher-assessed-grades-in-summer-2022-interviews-and-surveys/teacher-assessed-grades-in-summer-2021-surveys</ulink></bibtext> </blist> <blist> <bibtext> Plewis, I. (1997). Inferences about teacher expectations from national assessment at key stage one. The British Journal of Educational Psychology, 67 (2), 235 – 247. https://doi.org/10.1111/j.2044-8279.1997.tb01240.x</bibtext> </blist> <blist> <bibtext> Prior, L., Evans, C., Merlo, J., & Leckie, G. (2023). Sociodemographic inequalities in student achievement: An intersectional multilevel analysis of individual heterogeneity and discriminatory accuracy (MAIHDA) with application to students in London, England. Sociology of Race & Ethnicity. arXiv :2211.06321v2[stat.AP]. https://doi.org/10.1177/23326492241267251</bibtext> </blist> <blist> <bibtext> Prior, L., Jerrim, J., Thomson, D., & Leckie, G. (2021). A review and evaluation of secondary school accountability in England: Statistical strengths, weaknesses and challenges for 'progress 8' raised by COVID-19. Review of Education, 9 (3), e3299. https://doi.org/10.1002/rev3.3299</bibtext> </blist> <blist> <bibtext> Rabe-Hesketh, S., & Skrondal, A. (2022). Multilevel and longitudinal modelling using Stata (fourth edition) volume 1: Continuous responses. Stata Press.</bibtext> </blist> <blist> <bibtext> Ready, D. D., & Wright, D. L. (2011). Accuracy and inaccuracy in teachers' perceptions of young children's cognitive abilities: The role of child background and classroom context. American Educational Research Journal, 48 (2), 335 – 360. https://doi.org/10.3102/0002831210374874</bibtext> </blist> <blist> <bibtext> Reeves, D. J., Boyle, W. F., & Christie, T. (2001). The relationship between teacher assessments and pupil attainments in standard test tasks at key stage 2, 1996–98. British Educational Research Journal, 27 (2), 141 – 160. https://doi.org/10.1080/01411920120037108</bibtext> </blist> <blist> <bibtext> Snell, M., Thorpe, A., Hoskins, S., & Chevalier, A. (2008). Teachers' perceptions and A-level performance: Is there any evidence of systematic bias? Oxford Review of Education, 34 (4), 403 – 423. https://doi.org/10.1080/03054980701682140</bibtext> </blist> <blist> <bibtext> Stratton, T., Zanini, N., & Noden, P. (2021). An evaluation of centre assessment grades from summer 2020. Ofqual.</bibtext> </blist> <blist> <bibtext> Tenenbaum, H. R., & Ruck, M. D. (2007). Are teachers' expectations different for racial minority than for European American students? A meta-analysis. Journal of Educational Psychology, 99 (2), 253 – 273. https://doi.org/10.1037/0022-0663.99.2.253</bibtext> </blist> <blist> <bibtext> Thomas, S., Madaus, G. F., Raczek, A. E., & Smees, R. (1998). Comparing teacher assessment and standard task results in England: The relationship between pupil characteristics and attainment. Assessment in Education Principles, Policy & Practice, 5 (2), 213 – 246. https://doi.org/10.1080/0969594980050205</bibtext> </blist> <blist> <bibtext> Timmermans, A. C., Kuyer, H., & van der Werf, G. (2015). Accurate, inaccurate, or biased teacher expectations: Do Dutch teachers differ in their expectations at the end of primary education? The British Journal of Educational Psychology, 85 (4), 459 – 478. https://doi.org/10.1111/bjep.12087</bibtext> </blist> <blist> <bibtext> UCAS. (2023). Predicted grades – what you need to know for entry this year. https://<ulink href="http://www.ucas.com/advisers/managing-applications/predicted-grades-what-you-need-know-entry-year">www.ucas.com/advisers/managing-applications/predicted-grades-what-you-need-know-entry-year</ulink></bibtext> </blist> <blist> <bibtext> Urhahne, D., & Wijnia, L. (2021). A review on the accuracy of teacher judgements. Educational Research Review, 32, 100374. https://doi.org/10.1016/j.edurev.2020.100374</bibtext> </blist> <blist> <bibtext> Wint, K. M., Opara, I., Gordon, R., & Brooms, D. R. (2022). Countering educational disparities among black boys and black adolescent boys from pre-K to high school: A life course-intersectional perspective. The Urban Review, 54 (2), 183 – 206. https://doi.org/10.1007/s11256-021-00616-z</bibtext> </blist> <blist> <bibtext> Wyness, G. (2016). Predicted grades: Accuracy and impact: A report for University and College Union. University and College Union. https://<ulink href="http://www.ucu.org.uk/media/8409/Predicted-grades-accuracy-and-impact-Dec-16/pdf/Predicted%5fgrades%5freport%5fDec2016.pdf">www.ucu.org.uk/media/8409/Predicted-grades-accuracy-and-impact-Dec-16/pdf/Predicted%5fgrades%5freport%5fDec2016.pdf</ulink></bibtext> </blist> </ref> <aug> <p>By Lucy Prior and George Leckie</p> <p>Reported by Author; Author</p> <p></p> <p>Lucy Prior is a senior research associate in Quantitative Methods in Education at the Centre for Multilevel Modelling and School of Education, University of Bristol. Her research interests include multilevel modelling, social inequalities, school effects and performance, health geography, and neighbourhood deprivation. Her research employs quantitative techniques, such as multilevel modelling, mediation analysis and regression techniques.</p> <p>George Leckie is a Professor of Social Statistics and Co-Director of the Centre for Multilevel Modelling at the University of Bristol, UK. His methodological interests include the development and dissemination of multilevel modelling. His substantive expertise involves applying this method to study institutional and geographical influences on sociodemographic inequalities, particularly school effects on student learning and debates surrounding government school league tables. George's work has been funded by 12 grants from UK and European funding councils. He has authored over 100 journal articles and book chapters and delivered more than 75 multilevel short courses worldwide.</p> </aug> <nolink nlid="nl1" bibid="bib13" firstref="ref1"></nolink> <nolink nlid="nl2" bibid="bib38" firstref="ref2"></nolink> <nolink nlid="nl3" bibid="bib22" firstref="ref3"></nolink> <nolink nlid="nl4" bibid="bib36" firstref="ref4"></nolink> <nolink nlid="nl5" bibid="bib16" firstref="ref6"></nolink> <nolink nlid="nl6" bibid="bib17" firstref="ref7"></nolink> <nolink nlid="nl7" bibid="bib37" firstref="ref9"></nolink> <nolink nlid="nl8" bibid="bib34" firstref="ref12"></nolink> <nolink nlid="nl9" bibid="bib18" firstref="ref13"></nolink> <nolink nlid="nl10" bibid="bib51" firstref="ref14"></nolink> <nolink nlid="nl11" bibid="bib14" firstref="ref15"></nolink> <nolink nlid="nl12" bibid="bib29" firstref="ref16"></nolink> <nolink nlid="nl13" bibid="bib52" firstref="ref17"></nolink> <nolink nlid="nl14" bibid="bib44" firstref="ref19"></nolink> <nolink nlid="nl15" bibid="bib12" firstref="ref20"></nolink> <nolink nlid="nl16" bibid="bib28" firstref="ref21"></nolink> <nolink nlid="nl17" bibid="bib46" firstref="ref22"></nolink> <nolink nlid="nl18" bibid="bib30" firstref="ref23"></nolink> <nolink nlid="nl19" bibid="bib25" firstref="ref24"></nolink> <nolink nlid="nl20" bibid="bib27" firstref="ref25"></nolink> <nolink nlid="nl21" bibid="bib40" firstref="ref28"></nolink> <nolink nlid="nl22" bibid="bib50" firstref="ref30"></nolink> <nolink nlid="nl23" bibid="bib48" firstref="ref33"></nolink> <nolink nlid="nl24" bibid="bib45" firstref="ref40"></nolink> <nolink nlid="nl25" bibid="bib49" firstref="ref41"></nolink> <nolink nlid="nl26" bibid="bib54" firstref="ref47"></nolink> <nolink nlid="nl27" bibid="bib19" firstref="ref48"></nolink> <nolink nlid="nl28" bibid="bib31" firstref="ref56"></nolink> <nolink nlid="nl29" bibid="bib26" firstref="ref61"></nolink> <nolink nlid="nl30" bibid="bib47" firstref="ref62"></nolink> <nolink nlid="nl31" bibid="bib35" firstref="ref66"></nolink> <nolink nlid="nl32" bibid="bib15" firstref="ref74"></nolink> <nolink nlid="nl33" bibid="bib53" firstref="ref75"></nolink> <nolink nlid="nl34" bibid="bib20" firstref="ref76"></nolink> <nolink nlid="nl35" bibid="bib10" firstref="ref79"></nolink> <nolink nlid="nl36" bibid="bib33" firstref="ref81"></nolink> <nolink nlid="nl37" bibid="bib21" firstref="ref82"></nolink> <nolink nlid="nl38" bibid="bib41" firstref="ref83"></nolink> <nolink nlid="nl39" bibid="bib11" firstref="ref86"></nolink> <nolink nlid="nl40" bibid="bib23" firstref="ref88"></nolink> <nolink nlid="nl41" bibid="bib24" firstref="ref89"></nolink> <nolink nlid="nl42" bibid="bib43" firstref="ref92"></nolink> <nolink nlid="nl43" bibid="bib42" firstref="ref94"></nolink> <nolink nlid="nl44" bibid="bib39" firstref="ref103"></nolink> <nolink nlid="nl45" bibid="bib32" firstref="ref104"></nolink>
Header DbId: eric
DbLabel: ERIC
An: EJ1500707
AccessLevel: 3
PubType: Academic Journal
PubTypeId: academicJournal
PreciseRelevancyScore: 0
IllustrationInfo
Items – Name: Title
  Label: Title
  Group: Ti
  Data: Student Intersectional Sociodemographic and School Variation in GCSE Final Grades in England Following COVID-19 Examination Cancellations
– Name: Language
  Label: Language
  Group: Lang
  Data: English
– Name: Author
  Label: Authors
  Group: Au
  Data: <searchLink fieldCode="AR" term="%22Lucy+Prior%22">Lucy Prior</searchLink> (ORCID <externalLink term="https://orcid.org/0000-0001-6557-9082">0000-0001-6557-9082</externalLink>)<br /><searchLink fieldCode="AR" term="%22George+Leckie%22">George Leckie</searchLink> (ORCID <externalLink term="https://orcid.org/0000-0003-1486-745X">0000-0003-1486-745X</externalLink>)
– Name: TitleSource
  Label: Source
  Group: Src
  Data: <searchLink fieldCode="SO" term="%22Oxford+Review+of+Education%22"><i>Oxford Review of Education</i></searchLink>. 2025 51(5):763-784.
– Name: Avail
  Label: Availability
  Group: Avail
  Data: Routledge. Available from: Taylor & Francis, Ltd. 530 Walnut Street Suite 850, Philadelphia, PA 19106. Tel: 800-354-1420; Tel: 215-625-8900; Fax: 215-207-0050; Web site: http://www.tandf.co.uk/journals
– 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="%22Secondary+Education%22">Secondary Education</searchLink><br /><searchLink fieldCode="EL" term="%22Higher+Education%22">Higher Education</searchLink><br /><searchLink fieldCode="EL" term="%22Postsecondary+Education%22">Postsecondary Education</searchLink>
– Name: Subject
  Label: Descriptors
  Group: Su
  Data: <searchLink fieldCode="DE" term="%22Foreign+Countries%22">Foreign Countries</searchLink><br /><searchLink fieldCode="DE" term="%22COVID-19%22">COVID-19</searchLink><br /><searchLink fieldCode="DE" term="%22Pandemics%22">Pandemics</searchLink><br /><searchLink fieldCode="DE" term="%22Secondary+Education%22">Secondary Education</searchLink><br /><searchLink fieldCode="DE" term="%22Exit+Examinations%22">Exit Examinations</searchLink><br /><searchLink fieldCode="DE" term="%22Colleges%22">Colleges</searchLink><br /><searchLink fieldCode="DE" term="%22Public+Schools%22">Public Schools</searchLink><br /><searchLink fieldCode="DE" term="%22Secondary+Schools%22">Secondary Schools</searchLink><br /><searchLink fieldCode="DE" term="%22Small+Schools%22">Small Schools</searchLink><br /><searchLink fieldCode="DE" term="%22State+Schools%22">State Schools</searchLink><br /><searchLink fieldCode="DE" term="%22Selective+Admission%22">Selective Admission</searchLink><br /><searchLink fieldCode="DE" term="%22Intersectionality%22">Intersectionality</searchLink><br /><searchLink fieldCode="DE" term="%22Scores%22">Scores</searchLink><br /><searchLink fieldCode="DE" term="%22Student+Characteristics%22">Student Characteristics</searchLink><br /><searchLink fieldCode="DE" term="%22Grade+Inflation%22">Grade Inflation</searchLink><br /><searchLink fieldCode="DE" term="%22Test+Bias%22">Test Bias</searchLink><br /><searchLink fieldCode="DE" term="%22Prediction%22">Prediction</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: 10.1080/03054985.2024.2407624
– Name: ISSN
  Label: ISSN
  Group: ISSN
  Data: 0305-4985<br />1465-3915
– Name: Abstract
  Label: Abstract
  Group: Ab
  Data: In 2020, COVID-19 forced the cancellation of all student end-of-school examinations in England. Schools were asked to provide centre assessment grades (CAGs), offering their best estimates for what students would have achieved had they sat their examinations. Although initially replaced in favour of grades calculated via an algorithm, students were eventually awarded their CAGs following widespread public outcry over the calculated grades. Whether CAGs were unfairly awarded across different student groups and schools in 2020 compared to previous years is a key question. However, existing analyses of bias in CAGs are limited by a lack of attention to potential interactions between student characteristics and thus to hidden differential grade inflation across intersectional groups. We addressed this by examining student GCSE performance in 2018, 2019 and 2020 via a Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy (MAIHDA) analysis of intersectional sociodemographic variation which we cross-classified with schools given their role in generating CAGs. Overall, a picture of stability emerged where, despite substantial overall grade inflation in 2020, the use of CAGs did not appear to have generated new or divergent intersectional relationships in comparison to previous years, suggesting CAGs showed a similar susceptibility to bias as normal examinations.
– Name: AbstractInfo
  Label: Abstractor
  Group: Ab
  Data: As Provided
– Name: DateEntry
  Label: Entry Date
  Group: Date
  Data: 2026
– Name: AN
  Label: Accession Number
  Group: ID
  Data: EJ1500707
PLink https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=eric&AN=EJ1500707
RecordInfo BibRecord:
  BibEntity:
    Identifiers:
      – Type: doi
        Value: 10.1080/03054985.2024.2407624
    Languages:
      – Text: English
    PhysicalDescription:
      Pagination:
        PageCount: 22
        StartPage: 763
    Subjects:
      – SubjectFull: Foreign Countries
        Type: general
      – SubjectFull: COVID-19
        Type: general
      – SubjectFull: Pandemics
        Type: general
      – SubjectFull: Secondary Education
        Type: general
      – SubjectFull: Exit Examinations
        Type: general
      – SubjectFull: Colleges
        Type: general
      – SubjectFull: Public Schools
        Type: general
      – SubjectFull: Secondary Schools
        Type: general
      – SubjectFull: Small Schools
        Type: general
      – SubjectFull: State Schools
        Type: general
      – SubjectFull: Selective Admission
        Type: general
      – SubjectFull: Intersectionality
        Type: general
      – SubjectFull: Scores
        Type: general
      – SubjectFull: Student Characteristics
        Type: general
      – SubjectFull: Grade Inflation
        Type: general
      – SubjectFull: Test Bias
        Type: general
      – SubjectFull: Prediction
        Type: general
      – SubjectFull: United Kingdom (England)
        Type: general
    Titles:
      – TitleFull: Student Intersectional Sociodemographic and School Variation in GCSE Final Grades in England Following COVID-19 Examination Cancellations
        Type: main
  BibRelationships:
    HasContributorRelationships:
      – PersonEntity:
          Name:
            NameFull: Lucy Prior
      – PersonEntity:
          Name:
            NameFull: George Leckie
    IsPartOfRelationships:
      – BibEntity:
          Dates:
            – D: 01
              M: 01
              Type: published
              Y: 2025
          Identifiers:
            – Type: issn-print
              Value: 0305-4985
            – Type: issn-electronic
              Value: 1465-3915
          Numbering:
            – Type: volume
              Value: 51
            – Type: issue
              Value: 5
          Titles:
            – TitleFull: Oxford Review of Education
              Type: main
ResultId 1