School belonging and attainment at key stage 4: linked data study

School belonging and attainment at key stage 4: linked data study

Research report

August 2026

Pallavi Banerjee: University of Cambridge

Contents

• List of tables
• List of figures
• Acknowledgements
• Executive summary
• Background, purpose and data
• Headline findings
• Interpretation and policy relevant insights
• Introduction
• Aims
• Literature review
• Summary of existing evidence
• Data
• PISA data
• PISA study design and sampling
• National pupil database (NPD)
• Data linkage and matching
• Variables of interest
• Analysis
• Sensitivity analysis
• Interpreting results
• Results
• Distribution of the school belonging index
• Distribution of the bullying index
• Reported school belonging
• Attainment 8
• Prior attainment
• Pupil background characteristics
• Variation in belonging across groups
• Weighted mean bullying index by pupil background characteristics
• Multiple linear regression analysis
• Conclusion
• References

List of tables

• Table 1 Belonging index (weighted)
• Table 2 Bullying index (weighted)

Table 3 Student responses to school belonging: I feel like I belong at school 28

• Table 4 Attainment 8
• Table 5 KS2 prior attainment
• Table 6 FSM eligibility
• Table 7 Ethnicity
• Table 8 Special educational needs
• Table 9 Gender
• Table 10 Belonging index by gender
• Table 11 Belonging index by ethnicity
• Table 12 Belonging index by FSM eligibility
• Table 13 Belonging index by SEN
• Table 14 Bullying index by gender
• Table 15 Bullying index by ethnicity
• Table 16 Bullying index by SEN
• Table 17 Bullying index by FSM eligibility

Table 18 Regression coefficients for models 1 and 2 41

Table 19 Regression coefficients for models 3 and 4 42

List of figures

• Figure 1 Density plot for belonging index (unweighted)
• Figure 2 Density plot for bullying index (unweighted)
• Figure 3 Belonging index by gender
• Figure 4 Belonging index by ethnicity
• Figure 5 Belonging index by FSM eligibility
• Figure 6 Belonging index by SEN
• Figure 7 Reported school belonging by gender
• Figure 8 Reported school belonging by ethnicity
• Figure 9 Reported school belonging by FSM eligibility
• Figure 10 Reported school belonging by SEN
• Figure 11 Bullying by gender
• Figure 12 Bullying by ethnicity
• Figure 13 Bullying by SEN
• Figure 14 Bullying by FSM

Acknowledgements

The report draws on linked survey and administrative data from the Programme for International Student Assessment (PISA) and the National Pupil Database (NPD). I acknowledge the contributions of the Organisation for Economic Co-operation and Development (OECD) and the Department for Education (DfE) responsible for the collection, management and provision of these datasets.

These powerful datasets allow us to move beyond snapshots of inequality to generate evidence that can lead to meaningful change in the lives of young people.

I would like to thank all the students and teachers in schools who took part in PISA 2022 and consented to their survey responses to be linked to their NPD individual level attainment data from key stages 2 and 4.

I would also like to thank colleagues in DfE for their valuable feedback during different stages of this research and analysis: John Rolfe, Dan Richards, Patrick Searle, Daryl Gowlett, Grace Taylor-Joyce, Michael Dale, Sam Ramsay and Jessica Hunt.

Executive summary

In recent years there has been growing recognition of the role of belonging as being foundational for a range of young people’s outcomes - from health and wellbeing to attendance. In response, a cross-DfE programme of work, to develop our understanding of belonging in England, highlighted that evidence is needed for policy development. Belonging continues to grow in interest, especially given the government’s ambitions set out in the recent Schools White Paper (2026), Every Child Achieving and Thriving.

To build an evidence base, this report examines the relationship between inequalities in school belonging and attainment among secondary school pupils using linked administrative data. It responds to growing policy interest in pupil wellbeing and addresses a key priority area: to understand variation in school belonging and attainment and how these are shaped by pupil background characteristics and psychosocial risk factors like bullying.

Background, purpose and data

We examine the relationship between school belonging and attainment of 16-year-olds in England using the Organisation for Economic Cooperation and Development’s (OECD) Programme for international student assessment (PISA) 2022 survey data, linked to the National Pupil Database (NPD). Using this linked longitudinal data, we first assess whether school belonging varies by pupil background characteristics. Following this, we assess the extent to which bullying varies by the same set of characteristics highlighting inequalities in pupil experience. We then estimate multiple linear regression models to examine whether measures of school belonging are associated with attainment, controlling for pupil characteristics such as socioeconomic background, ethnicity, gender, special educational needs and the negative experience of being bullied.

Attainment 8, the department’s headline performance measure, has been used as the measure of attainment to summarise a pupil’s overall GCSE achievement across eight qualifications taken at the end of key stage 4.

The report uses two different measures of belonging from the PISA student questionnaire. One of these is PISA’s composite measure of belonging, a standardised index which we refer to as the belonging index in this report. It captures responses to six different items: “I feel like an outsider (or left out of things at school)”, “I make friends easily at school”, “I feel like I belong at school”, “I feel awkward and out of place in my school”, “other students seem to like me” and “I feel lonely at school”.

While the OECD belonging index provides a reliable composite measure, it aggregates multiple dimensions that may not uniformly capture a student’s subjective sense of school belonging. We therefore also examine the single item measure based on pupil responses to the statement: “I feel like I belong at school”. Student responses to this statement have

been recorded on a four-point Likert scale: strongly agree, agree, disagree, strongly disagree. The latter is a more direct and interpretable indicator for school belonging. We refer to it as reported school belonging.

Finally, the report also investigates self-reported experiences of being bullied in school and how it is associated with attainment and belonging. As the bullying index is based on self-reported responses to being bullied, it is difficult to draw firm conclusions on whether certain subgroups were less likely to be bullied or were bullied but were less likely to report being bullied. The index however helpfully captures repeated peer aggression, consistent with international definitions of bullying, based on student responses to questions on how often the following happened to them: “other students made fun of them”, “spread nasty rumours about them”, “left them out of things on purpose”, “other students took away or destroyed their belongings”, “they were hit or pushed around”, “they were forced to do things they didn’t want to”.

The headline findings report conclusions drawn from the analyses which made use of reported school belonging, first, followed by the OECD belonging index. Though both measures provide a useful and complementary metrics for belonging, their interpretations vary. As discussed earlier, the single item construct on reported school belonging directly measures the core concept without conflating this with distinct constructs such as bullying or wider school relationships. It is also free from aggregation bias and does not hide heterogeneity in experiences and is therefore a more direct measure.

Note, any references to the OECD average refers to the average for the entire population, not the average for any specific sub-group across the OECD.

Headline findings[footnote 1]

Reported school belonging varies by pupil background characteristics

• Boys were more likely to report feeling they belonged at school with 66.40%

(SE=1.39) of all boys and 58.09% (SE=1.41) of all girls saying they felt they belonged at their current school. Male students were statistically significantly more likely than female students to agree with school belonging (p<0.001).

• Students eligible for FSM were less likely to report belonging at school. 58.20%

(SE=2.61) of all FSM eligible pupils and 63.18% (SE=1.09) of all non-FSM pupils agreed or strongly agreed that they felt like they belonged at their current school. However, the difference was not statistically significant between both subgroups at the 5% level (p=0.06).

• SEN pupils typically reported poorer belonging at school. 56.62% (SE=3.10) of all SEN pupils and 62.98% (SE=1.16) of non-SEN pupils agreed that they felt like they

belonged at their current school. However, the subgroup differences were not quite statistically significant at the 5% level (p=0.05).

• Belonging varies by ethnicity, with White students less likely to report they belong at

school. 60.26% (SE=1.25) of White students, 70.17% (SE=2.88) of Asian students, 72.13% (SE=2.52) of Black students and 65.16% (SE=3.84) of all students from Mixed ethnic backgrounds who responded felt they belonged at their current school. The subgroup differences were statistically significant between White and Asian (p<0.01), and White and Black students (p<0.001) but not for Mixed and White.

These differences in school belonging are also mirrored when measured using the composite school belonging index

• Boys (score= -0.06, SE=0.02) had higher scores on the belonging index than girls

(score= -0.37, SE=0.02). This difference was statistically significant (p<0.001). Belonging scores for boys and girls were also statistically significantly lower than the OECD average for all pupils.

• Non-SEN pupils (score= -0.20, SE=0.02) had significantly higher scores (p<0.01)

than SEN pupils (score= -0.36, SE=0.05). Both Non-SEN and SEN pupils had statistically significantly lower scores than the OECD average for all pupils.

• Non-FSM pupils (score= -0.19, SE=0.02) had higher scores than FSM pupils

(score= -0.34, SE=0.03). This difference was statistically significant (p<0.001). FSM and non-FSM pupils both had statistically significantly lower belonging index scores than the OECD average for all pupils.

• Asian students (score= -0.13, SE=0.04) and Black students (score= -0.15.

SE=0.06) had higher belonging scores than White students (score= -0.23, SE=0.02). Mixed ethnicity students (score= -0.21, SE=0.06) had similar scores to White students. However, these subgroup differences between White students and those belonging to Asian (p=0.05), Black (p=0.24) or Mixed (p=0.83) ethnic groups were not statistically significant (p>0.05). White, Asian and Mixed ethnicity students had statistically significantly lower scores than the OECD average for all pupils.

Bullying varies by pupil background characteristics

• Girls (score=0.04, SE=0.04) report higher levels of bullying than the OECD

average, whereas boys (score= -0.01, SE=0.03) report marginally lower levels of bullying than the OECD average of all pupils. This difference in scores between both groups and the OECD average was not statistically significant (p>0.05). The subgroup difference between boys and girls was not statistically significant (p>0.05).

• SEN pupils (score=0.23, SE=0.07) report statistically significantly higher levels of bullying than the OECD average of all pupils. Non-SEN pupils report lower bullying

than the OECD average (score= -0.01, SE=0.03). The difference in bullying scores between SEN and non-SEN pupils was statistically significant (p<0.001).

• FSM pupils (score=0.09, SE=0.05) reported higher bullying than the OECD average

of all pupils, though this was not significant at the 5% level. Non-FSM pupils had weighted means (score=0.00, SE=0.03) the same as the OECD average. The differences in scores between FSM and non-FSM pupils was not statistically significant (p>0.05).

• White students report statistically significantly higher bullying (0.06, SE=0.03) than

the OECD average of all pupils whereas Asian (score= -0.22, SE= 0.05) students report significantly lower bullying than the OECD average. Black (score = -0.07, SE=0.08) and Mixed ethnicity students (score= -0.02, SE=0.08) also report bullying scores lower than White students. The subgroup differences between White and Asian students were statistically significant (p<0.001) but not so for others.

School belonging is associated with improved attainment, once we account for differences in pupil background characteristics, prior attainment and experience of bullying, though the strength and significance depend on how belonging is measured

• Pupils who agreed they belonged at school (based on the single-item measure of

belonging) had significantly higher attainment than those who didn’t, after controlling for prior attainment, demographics and bullying.

• Removing bullying from the model had little impact on the estimated coefficients - the

relationship remains strong, positive and significant which suggests the relationship between school belonging and attainment is not driven by bullying alone

• The model shows moving from strongly agreeing to strongly disagreeing (that an

individual belongs at school) is associated with a statistically significant 8-point decrease in attainment 8 score after accounting for wider demographic factors, prior attainment and bullying. It represents a profound shift in a student’s academic profile, equivalent to dropping around a full GCSE grade across eight different subjects.

• However, when we measure belonging using the broader belonging index, we do not

observe a statistically significant relationship between belonging and attainment, after adjusting for prior attainment, demographics and reported experience of bullying. When bullying is excluded from the model, we observe a modest positive association between belonging and attainment, though this remains statistically insignificant.

• Because the belonging index conflates a range of concepts into a single index, some

of which are related to bullying, we are unable to detect a strong effect on attainment in the remaining variation in the belonging index. For this reason, we conclude that the single item measure of belonging is preferred as a cleaner and more precise metric.

• We conclude there is a complex relationship between belonging, bullying and

attainment. Bullying appears to play some role in driving the positive relationship between belonging and attainment, but this is not the only mechanism.

Interpretation and policy relevant insights

Although the research design does not intend to establish causality, the analysis shows that attainment is higher on average among pupils with stronger reported school belonging. This holds for the single item measure of reported school belonging even after we control for prior attainment, demographics and bullying, suggesting that this effect is not entirely driven by pupil characteristics. We therefore conclude higher school belonging is associated with higher attainment.

The patterns observed are consistent with existing evidence from studies which suggest students’ sense of school belonging is positively associated with academic attainment indicating a consistent relationship. The PISA survey was administered in the same academic year as these young people took their GCSE exams, so the findings need to be interpreted with caution as we do not make any claims about how school belonging influences attainment in the much longer term.

Empirical research shows school belonging is associated with a range of positive outcomes for children and young people, particularly in relation to health and wellbeing. It also shows a significant and consistent correlation with attainment - an outcome influenced by many different factors - making this relationship especially meaningful. Evidence from the literature further indicates that belonging is linked to engagement in learning, motivation, and school attendance.

Thriving and belonging are valuable goals, and current policy encourages schools to prioritise them. Recent DfE research has explored pupils’ experiences of school belonging, safety, and enjoyment (DfE, 2025). Our understanding of how belonging relates to attainment, based on the evidence presented in this report, suggests that these aims are mutually reinforcing.

Introduction

In recent years, there has been increasing recognition that a sense of belonging at school is fundamental to a wide range of outcomes for young people from health and wellbeing to attendance.

The Schools White Paper, “Every Child Achieving and Thriving” (2026) aims to boost wellbeing and sense of belonging, with a requirement for schools to monitor this by 2029. To support the development of an evidence base, for policymaking in DfE and the wider sector, this report explores the relationship between inequalities in school belonging and academic attainment among secondary school pupils. The analysis uses linked administrative and survey data to understand how school belonging and attainment vary, and how these patterns are influenced by pupils’ backgrounds and psychosocial risk factors such as bullying.

Aims

The aim of this report is (1) to examine inequalities in school belonging and bullying across pupil groups, and (2) to assess the extent to which these psychosocial factors contribute to differences in attainment beyond background characteristics and prior achievement.

To address these aims the following research questions have been answered:

• Does school belonging vary across different subgroups of young people by their

background characteristics such as FSM eligibility, ethnicity, gender and special educational needs?

• Which student subgroups report higher bullying by pupil background characteristics?
• To what extent is school belonging associated with attainment, after controlling for other observed factors, including bullying?

Literature review

Educational attainment is a key determinant of employment, income and health (OECD, 2023). Despite policy commitments to equality of opportunity, attainment gaps persist across socioeconomic background (SEB) and demographic factors such as ethnicity and gender. Recent research increasingly recognises that these disparities are shaped not only by structural inequalities or demographics but also by psychosocial factors impacting pupil wellbeing such as bullying (Samara et al., 2021) and sense of belonging in school (Goodenow & Grady, 1993; OECD, 2019, Korpershoek et al., 2019). This review synthesises the literature on these relationships between pupil characteristics and educational attainment.

Summary of existing evidence

Socioeconomic background

Mechanisms linking socioeconomic background (SEB) and attainment include differences in parental education, home learning environment, young person’s perceptions of safety in the area where they live and behavioural factors (Banerjee, 2025). The relationship between SEB and the development of cognitive and non-cognitive skills is a highly contested area of research. While the connection is documented the causal mechanisms, measurement biases and the true malleability of these skills are debated (Feinstein, 2003; Wang et al. 2026). However, socioeconomic background continues to remain one of the most consistent predictors of educational outcomes (Sirin, 2005; OECD, 2018) internationally (Reardon 2011). DfE uses eligibility for free school meals (FSM) as a proxy indicator of SEB and education statistics show the attainment gap between FSM pupils and their peers emerges early and widens through schooling (DfE, 2022).

Ethnicity

Significant attainment gaps exist across ethnic groups in the UK. White British, Black Caribbean and Gypsy/Roma students experience the largest attainment gaps frequently compounded by disadvantage. In the 2023/24 school year there were large variations in attainment between FSM eligible pupils in these ethnic groups (SMC 2025). For example, 76% of FSM eligible pupils from the Chinese ethnic group got a grade 5 or above compared with 6% of Gypsy or Roma pupils. FSM eligible pupils from Indian and Bangladeshi ethnic groups (both 50%) were more likely to get a grade 5 or above compared with White British (19%) and Black Caribbean (19%) pupils (DfE 2024).

Gender

DfE data show in 2024/25 by the end of primary schooling, 66% of girls met the expected standard in English reading, writing and Maths, compared with 59% of boys (DfE 2025). In

2022/23 the attainment gap at GCSE level between girls and boys tightened to its smallest difference for 14 years, but with 24.9% of girls achieving grade 7 compared with 19.1% of boys, there was still a significant difference of 6 percentage points. DfE data also show that among children eligible for free school meals, girls outperform boys in all ethnic groups at GCSEs (DfE 2025). Within the A level and applied general cohort, female students continue to achieve higher average point scores compared to male students while at ages 16-24 NEET rates are higher for men (DfE 2026, Youth Futures Foundation 2026).

Gender gaps in attainment have shifted over time, yet gendered patterns persist in some subject choices and post-16 pathways (DfE 2024). Explanations include differences in motivation, classroom behaviour, and peer culture (Francis & Skelton, 2005). However, gender interacts with other characteristics: for instance, White working-class boys and Black Caribbean boys often experience particularly poor outcomes (DfE, 2022). Understanding gender differences therefore requires a broader intersectional lens that considers how social identities, psychosocial factors such as school belonging and expectations interact.

Special educational needs

In 2024/25, only 9% of those pupils with an Education, health and care plan (EHCP) met the expected reading, writing and Maths standard at the end of key stage 2 (DfE, 2025). In 2024/25, 18.7% of pupils at the end of key stage 4 in state funded schools had a special educational need. The gap for the percentage achieving grades 5 and above in English and Maths was 14.9 percentage points and the gap for attainment 8 was 19 points for 2024/25. SEN pupils, if undiagnosed, are also more likely to experience bullying (Rose et al., 2011) and exclusion (Parsons & Platt, 2017), both of which negatively affect attainment. However, studies also note wide variation within the SEN population, with outcomes shaped by type and severity of need, school support, and family socioeconomic background (Dyson & Gallannaugh, 2008).

Bullying

Bullying has a negative association with academic outcomes (Nakamoto & Schwartz, 2010). Victims of bullying often experience anxiety, reduced concentration, absenteeism and lower academic motivation (Juvonen & Graham, 2014). Bullying also exacerbates inequalities as often students from minority groups, undiagnosed SEN, or low-SES backgrounds are disproportionately targeted (Tippett et al., 2010). Longitudinal studies show that chronic victimisation predicts lower grades and increased dropout risk (Gomes et al., 2020). Furthermore, bullying contributes to diminished school belonging, which mediates its impact on learning (Swearer et al. 2010).

School belonging

A strong sense of school belonging - feeling accepted, respected, and included in the school community - is associated with higher motivation and engagement (Goodenow & Grady, 1993; OECD, 2019). A meta-analysis of 51 studies (N=67,378) shows low belonging predicts absenteeism, disengagement, and dropout (Allen et al., 2018). Belonging is shaped by teacher-student relationships, peer acceptance, and inclusive school climates (Booker, 2006). Importantly, experiences of discrimination, bullying, or marginalisation reduce belonging and more so among minority and disadvantaged students than their peers. Interventions promoting inclusivity and student voice have been shown to improve both belonging and attainment (Juvonen et al., 2012).

Drawing on 82 correlational studies published in peer reviewed journals between 2000 and 2018, a meta-analytic review found a small positive relationship between school belonging and academic achievement (Korpershoek et al., 2019). Pupils in England who agreed or strongly agreed that they felt like they belonged at school had an average PISA mathematics score of 510, significantly higher than the average scores of 478 for pupils who either disagreed or strongly disagreed with the statement (Ingram et al., 2023). Employing structural equation modelling and TIMSS data, Liu and Huang (2024) found school belonging directly affected Maths achievement in the United States, but it did not in Japan and Korea.

The evidence exploring the link between belonging and attainment is complex. A literature review and new analysis of OECD data, reported in a 2025 briefing paper from the Social Market Foundation, argues that there is no inherent trade-off between wellbeing and attainment (SMF 2025). The paper shows, by 2022, the UK had seen the largest fall in pupils’ sense of school belonging amongst all OECD countries, though its relative academic performance in PISA improved. In comparison, countries such as Japan achieved the largest increase in belonging, whilst achieving the highest mathematics score (SMF 2025).

Research gaps

DfE data shows pupil characteristics correlate with attainment. Disadvantaged students often face compounded disadvantage due to systemic bias and resource constraints (DfE 2022) while psychosocial factors such as bullying and belonging may mediate or moderate these relationships, suggesting a complex web of interactions. OECD reports have highlighted the need to explore these pathways (OECD, 2019), building on integrated approaches that combine social and emotional variables with structural inequalities.

Existing research demonstrates that socioeconomic background, ethnicity, gender, SEN and bullying influence attainment. However, few studies have examined these factors simultaneously using large-scale data, particularly for England. Robust analysis is also required to make inferences and investigate the relationship between belonging and

attainment. This study addresses these gaps by examining the relationships using linked survey and administrative data through multivariate analysis.

Theoretical framework

Belonging is theoretically conceptualised as a fundamental human need closely linked to motivation, engagement, and ultimately, academic attainment. Three complementary frameworks - Maslow’s Hierarchy of Needs (1943), Self-Determination Theory (Deci & Ryan, 1985), and Goodenow’s Psychological Sense of School Membership (1993) provide the foundation for understanding how belonging operates within educational contexts.

According to Maslow’s Hierarchy of Needs (1943), belonging represents a core human requirement situated between basic physiological and safety needs, and higher-order needs of esteem and self-actualisation. Within educational settings, this hierarchy suggests that a sense of belonging is a prerequisite for cognitive engagement and learning. Students who feel socially connected and emotionally secure are more likely to participate actively in learning and achieve higher academic outcomes. Conversely, the absence of belonging can hinder motivation and self-esteem, thereby constraining attainment.

Building on this, Self-Determination Theory (SDT) proposed by Deci and Ryan (1985) emphasises that human motivation is driven by the fulfilment of three innate psychological needs: autonomy, competence, and relatedness. Within this framework, belonging aligns with the need for relatedness - the experience of being connected to, cared for, and valued by others. In the school context, students who experience positive relationships with teachers and peers are more likely to develop intrinsic motivation, sustained engagement, and a stronger commitment to learning. SDT thus positions belonging not only as an emotional state but also as a key motivational mechanism that supports persistence and performance. At the same time, when negative experiences such as bullying occur this need is thwarted.

Complementing these broad psychological perspectives, Goodenow’s (1993) Psychological Sense of School Membership (PSSM) provides a school-specific operationalisation of belonging. Goodenow defines belonging as “the extent to which students feel personally accepted, respected, included, and supported by others in the school social environment.” This conceptualisation emphasises students’ subjective perceptions of their membership within the school community. Empirical research grounded in the PSSM framework has consistently demonstrated strong associations between school belonging, engagement, and academic attainment, highlighting that students who perceive themselves as valued members of their school environment tend to exhibit higher motivation and achievement.”

Together, these frameworks situate belonging as a multidimensional construct that underpins both the emotional and motivational foundations of learning. They collectively

suggest that fostering students’ sense of belonging within schools is integral to supporting pupil wellbeing and engagement and thereby improving educational attainment.

In the context of this study, the integration of these psychological frameworks informs the operationalisation of school belonging. The PISA measures of school belonging- capturing students’ feelings of acceptance, inclusion, and respect within their school environment - align closely with Goodenow’s (1993) conception of the Psychological Sense of School Membership and reflect the relatedness component of Self-Determination Theory (Deci & Ryan, 1985). These constructs provide empirical indicators of the motivational and emotional dimensions identified by Maslow (1943) as foundational to learning.

Based on this theoretical foundation, the linked PISA-NPD administrative data allows for an empirical test of the theoretical construct “school belonging” which has been previously difficult to measure, making the patterns in the administrative data meaningful. The report summarises our findings from the examination of how variations in students’ psychological and social experiences at school relate to their educational performance when controlling for background characteristics.

Data

To examine the relationship between school belonging and attainment of 16-year-olds in England we use the Organisation for Economic Cooperation and Development’s (OECD) Programme for International Student Assessment (PISA) 2022 survey data, linked to the national pupil database (KS2 and KS4 attainment). Using this linked dataset, we first assess whether school belonging varies by pupil background characteristics. We then examine which subgroups have higher instances of reported bullying highlighting inequalities in pupil experience. Finally, we then estimate multiple linear regression models to examine whether school belonging is associated with KS4 attainment, controlling for pupil background characteristics such as socioeconomic background, ethnicity, gender, special educational needs and experience of being bullied.

PISA data

The PISA is the world’s largest international comparative survey of 15-year-old students organised by the OECD. It assesses the extent to which they have acquired key knowledge and skills essential for full participation in social and economic life (OECD 2025). Pisa 2022 was administered in November and December 2022 and most pupils who took part completed their GCSE exams in 2023. In England, 4,763 pupils from 165 schools completed the PISA assessments and the student questionnaire. The data for this report comes from the student questionnaires which collects information from pupils about their attitudes, experiences and feelings in education settings. This data is the most recent available PISA study. Although OECD conducted a PISA survey in 2025, results are due to be published later in 2026, after this analysis was conducted.

PISA study design and sampling

PISA takes a sophisticated, and therefore technically complex, approach to the design and administration of surveys. It is impractical for the PISA questionnaires to be administered to every single pupil in each participating country. Participating countries therefore assess a sample of their eligible pupils. PISA uses a two-stage stratified sampling design. Schools are selected randomly within explicit strata. Students are then randomly sampled within those schools. The goal is to obtain a nationally representative sample of 15-year-olds in education. Countries participating in PISA are required to adhere to strict international sampling procedures which facilitate comparability.

In line with this procedure for PISA 2022, a sample of secondary schools representing different geographical regions in England, the different types of schools that pupils may attend and a range of academic achievement at the school level based on the GCSE performance of previous pupil cohorts in these schools were selected. Within each participating school, a random sample of 40 eligible pupils were selected to take the assessment. Most of these pupils were in year 11.

However, because of differential non-response and the way school participation works some bias can appear. Some sub populations may be under or overrepresented once the final data is obtained. It is however important to note that PISA design counteracts this issue, as far as possible, through statistical methods. For further details on PISA design see PISA 2022 Technical Standards (OECD, 2022). To mitigate the identified problems through design the analysis presented in this report makes use of a weighted sample.

National pupil database (NPD)

Key stage 4 attainment data cover the GCSE results of pupils in 2022/23 who attended state funded schools in England. The data provides the main headline measure of attainment 8. The data also provides information about pupil characteristics such as prior attainment in key stage 2 (KS2) Maths, eligibility for free school meals, special educational needs, gender and ethnicity. For more details see key stage 4 performance data guidance[footnote 2] (DfE 2024).

Data linkage and matching

The linked dataset created for this analysis is based on an initial sample of 4,763 15-year- olds who answered the PISA questionnaire. The linked data was created by matching the PISA survey responses to individual level data from the NPD. Only records of those pupils who had consented to matching their PISA responses to NPD were used.

Checks were conducted to ensure that the datasets had merged correctly at each stage. This involved comparing variables across Pisa 2022, KS2-PISA data and the KS2-PISA- KS4 datasets. To verify the integrity of the linkage, patterns of missing data were also examined across a set of key variables in all three datasets. Some variables appear in all three datasets, others in only one or two datasets. Consistency checks focussed on variables present across overlapping datasets. The distribution of missing values and the continuity of records across these stages indicated that the linkage had been performed accurately.

Variables of interest

Attainment 8

Attainment 8 measures the average achievement of pupils in up to 8 qualifications. This includes: English language, English literature (double weighted providing both English language and English literature are taken), Maths (double weighted), three further qualifications that count in the English Baccalaureate (EBacc), and three further qualifications that can be GCSE qualifications (including EBacc subjects) or any other non-

GCSE qualifications on the DfE approved list. This variable was used as the attainment measure for headline findings reported from the models in this report.

Prior attainment

In Npd, KS2 Maths score is reported as a scaled score and has been used as a measure of prior attainment. These scores are converted from raw marks to ensure fair comparison over time, with specific raw scores needed for each scaled score. The lowest scaled score that can be awarded on a KS2 test is 80 and the highest scale score is 120. The expected standard (average) score is 100 which means pupils scoring at least 100 will have met the expected standard in the test. Scores below 100 mean that the pupil has not met the expected standard in the test.

Gender

In the NPD, gender is recorded as a binary variable derived from school census returns. For KS4 attainment tables, it is typically coded as male and female, reflecting the legal sex recorded by the school rather than self-identified gender.

Free school meals

Young people in state-funded schools in England are entitled to receive free school meals if a parent or carer are in receipt of a range of qualifying benefits. Free school meals (FSM) is a binary variable that states whether a pupil is eligible for, and claiming, free school meals at the time of the annual spring school census. Pupils not eligible for free school meals are marked as not FSM. FSM eligibility has been routinely used as a proxy indicator for socioeconomic disadvantage as it is based on family’s financial circumstances and household income deprivation.

Ethnic group

Ethnicity in the NPD is broken down into two main variables: a minor grouping variable and a major grouping variable. We used the classification for major ethnic groups. This census data item is provided for all pupils aged five and over as at the previous 31 August. Where the information has not yet been collected then this is recorded as not yet obtained. If a pupil or parent has refused to give the information, then ‘refused’ is recorded and returned. Major ethnic groups used in the analysis are: White, Asian, Black, Chinese, Mixed and Any Other Ethnic Group (AOEG). For the analysis Chinese, AOEG and those with an unclassified ethnicity have been removed from bivariate analysis because of very small sample sizes.

Special educational needs

The special educational needs (SEN) variable indicates whether a pupil has learning difficulties or disabilities that make it harder for them to learn than most children of the same age. The analysis uses a binary variable for SEN, flagging students with special educational needs (capturing those with SEN support or those with an Education, Health and Care plan) and non-SEN pupils.

Bullying

The bullying index in PISA 2022 represents students self-reported exposure to bullying, operationalised as a standard latent index derived from the student questionnaire. The index is constructed from a set of items asking how often during the previous 12 months, they experienced specific bullying behaviours in school. These items are designed to capture repeated peer aggression, consistent with international definitions of bullying. Students are asked in the last 12 months how often the following happened to them:

• Other students made fun of them
• Other students spread nasty rumours about them
• Other students had left them out of things on purpose
• They were threatened by other students
• Other students took away or destroyed their belongings
• They were hit or pushed around
• They were forced to do things they didn’t want to

Responses to each item are provided on an ordered frequency scale including never, a few times a year, a few times a month, once a week or more. In the composite index higher values indicate more frequent experience of being bullied.

Following PISA’s standard scaling procedures, the bullying items are combined using an item response theory (IRT) approach (a Rasch partial credit model) which estimates a single latent trait parameter representing each student’s underlying level of exposure to bullying. Item parameters are estimated from the pooled international sample to ensure cross-national comparability. Student level estimates of the latent trait are then standardised across OECD countries to have a mean of 0 and a standard deviation of 1.

As a result the bullying index is a continuous variable that is not bound by the raw response options; observed values typically range between approximately -3 and +4, with higher values indicating above average exposure to bullying relative to the OECD distribution.

There are some important considerations for interpretations. Bullying in PISA is school- based peer bullying. The index reflects self-perception of being bullied, not externally

verified incidents. The analysis is based on bullying reported by students, and it is possible there were some students who were bullied but may not have reported. In applied research the PISA bullied variable is used to examine associations with mental health, wellbeing and sense of belonging.

School belonging

School belonging is a complex construct. We use two different measures of school belonging.

1. School belonging index

The school belonging index in PISA 2022 consists of a six-item scale generated using pupil responses to the following items:

My school is a place where:

• I feel like an outsider (or left out of things) at school
• I make friends easily at school
• I feel like I belong at school
• I feel awkward and out of place in my school
• Other students seem to like me and
• I feel lonely at school

For each of these items, pupil responses were captured on a 4-point Likert scale ranging from strongly agree, agree, disagree and strongly disagree. The 6-item measure referred to as belonging index in this report has been treated as a continuous standardised index because after aggregation and standardisation the result is a continuous scale. The index is scaled so that the OECD average is standardised with mean zero and standard deviation of one and takes a wide range of numeric values rather than distinct categories.

However, the construct validity of the belonging index is often questioned as it does not measure ‘belonging’ as much as a combination of belonging and related factors. Additionally, the belonging index is a composite scale built from 6 different questionnaire items. It is constructed by OECD using IRT and scaling and is designed to maximise usable data. This value is already provided in PISA data. Crucially, this scale can still be computed by OECD even if a student has some missing responses. The scaling procedure allows partial information to contribute so a larger sample is retained because students don’t need to answer every item to get an index score.

2. Reported school belonging

While the OECD belonging index measures belonging, it aggregates multiple dimensions that may not uniformly capture a student’s subjective sense of belonging. We therefore

also examine the single item measure based on pupil responses to the statement: “I feel like I belong at school”.

Student responses to this statement have been recorded on a four-point Likert scale: strongly agree, agree, disagree, strongly disagree. For this variable if a student did not answer the question, they are coded as data missing. We therefore get a smaller sample because it requires a valid response to that specific item. This variable is more accurate and therefore our preferred measure as it is the most precise and cleanest measure of belonging. Throughout this report we use the term ‘belonging index’ to represent the 6- item measure and the single item measure: student agreement with the statement I feel like I belong at school, is referred to as ‘reported school belonging’.

Missing data

In PISA surveys different students get asked a different subset of questions so the sample for each bivariate analysis is based on those pupils in the sample who answer the relevant question for that analysis. PISA 2022 was administered during the COVID-19 pandemic and some of the missing data is also due to absenteeism, non-response or invalid responses. Cases with missing data for PISA variables used in the analysis (items related to belonging and bullying) have been excluded. Missing data for pupil background characteristics was 11% in the NPD.

The final models include only cases with valid responses for all the variables included in the model. The number of cases for each analysis has been provided alongside tables and figures. The sample differs slightly for each measure, and when demographic breakdowns are introduced, missing data is excluded, which means breakdowns may not be directly comparable (see analysis section for more detail on how missing data have been treated). However, as descriptive statistics are presented using weighted estimates they attempt to provide an accurate reflection of population level patterns.

Analysis

The analytical approach proceeds in stages.

First, we present descriptive statistics including summary measures of the belonging index, school belonging, bullying, attainment 8, prior attainment and pupil background characteristics.

Second, we examine how the school belonging index, reported school belonging and the bullying index vary by pupil characteristics by reporting weighted mean values by gender, FSM eligibility, SEN and ethnicity and bar charts where appropriate.

The final stage of the analysis involves estimating four sequential multiple linear regression models to identify patterns in the data and assess whether the relationships persist after accounting for multiple factors simultaneously. The models isolate the effect of the belonging index and school belonging on attainment 8 while controlling for pupil background characteristics. Models 1 and 2 use the belonging index whereas models 3 and 4 use reported school belonging.

Bullying is included in the main model specification because it is plausibly a confounder in the relationship between school belonging and attainment. Experiences of bullying could directly depress educational attainment, while also negatively affecting pupils’ sense of belonging at school, which may in turn influence attainment. Failing to account for bullying therefore risks attributing effects to belonging that are more directly driven by bullying. However, bullying is also conceptually and empirically very closely related to school belonging, particularly for the composite school belonging index used by PISA which captures feelings of social inclusion, loneliness, and peer relationships. Controlling for bullying may therefore “remove” a component of belonging that is part of the explanatory pathway rather than an independent source of confounding.

For this reason, analyses are presented both with (Models 1 and 3) and without bullying (Models 2 and 4) included, to assess how robust the estimated association between belonging and attainment is to alternative modelling assumptions and to avoid over- interpreting results that may depend on a specific treatment of this overlapping construct.

Sensitivity analysis

Sensitivity analyses were conducted to assess the robustness of the findings to alternative analytical assumptions, sample definitions and model specifications.

First, to examine the impact of missing data, models were run with samples including and excluding cases with missing values and the outputs were compared. The findings were broadly consistent indicating that results are not driven by differences in sample composition.

Second, alternative models were run with different attainment variables and the same set of predictors. The analysis reported here uses attainment 8. Alternative models included use of the binary indicator (strong pass) achieving grades 9-5 in level 2 qualifications. Very similar results were obtained for the likelihood of predicting attainment using school belonging, bullying and pupil background characteristics.

Third, the sensitivity of results to weighting was examined by comparing weighted and unweighted estimates (where appropriate). The overall pattern of findings remained consistent suggesting that results are not driven by the application of survey weights.

Fourth, additional checks were conducted using alternative model specifications including variations in coding of predictors (such as for SEND and ethnic classifications) these analysis did not materially alter the key findings.

Finally, subgroup analysis were undertaken to explore whether patterns differed across pupil groups for example by ethnicity, gender, SEN or FSM eligibility. The results were broadly consistent across subgroups reinforcing the generalisability of the findings. The consistency of results across these sensitivity checks indicates that the findings are robust to reasonable variations in analytical choices and sample definition.

Interpreting results

An important caveat is that PISA questionnaire was administered to 15-year-olds, so we are working with a specific subset of English pupils. The findings based on the analysis of this dataset will only apply to this age group and the relationship between belonging and attainment for other age groups may vary.

A source of uncertainty is measurement error which arises from questionnaire data. Self- reported variables are subject to social desirability bias, differential interpretation across cultures and recall error. Measurement error also occurs because a pupil’s responses may be influenced by factors such as their interpretation of the items that they respond to or their level of motivation on the day of the test.

Results

The results are presented in three stages. We begin with summary statistics describing the sample for the variables of interest in this report. Weighted estimates along with number of cases and missing data for each variable are provided first. This is then followed by descriptive analyses of key patterns across belonging index, reported school belonging and bullying. Finally, regression results examining adjusted associations to study the link between belonging and attainment are discussed via alternate model specifications.

Distribution of the school belonging index

Figure 1 shows the unweighted distribution of the belonging index. The density plot shows an approximately continuous distribution ranging from -3.25 to 2.75, with a mean of -0.21 (Sd=0.76, Se=0.02). Quartiles are well separated from the minimum and maximum values, and the distribution shows no evidence of pronounced floor or ceiling effects supporting the treatment of the belonging index as a continuous variable in subsequent regression analysis. From the PISA sample of 4,763, data was missing for 21% of the cases largely caused either by unprecedented challenges because of the COVID-19 pandemic as the study was carried out in November and December 2022 leading to skipped or unrecorded questionnaires or due to item level missing data as students were free to decline answering specific questions. Table 1 provides weighted estimates for the 3,769 respondents who had answered the questions and were included in the index.

Figure 1 Density plot for belonging index (unweighted)

Table 1 Belonging index (weighted)

Distribution of the bullying index

Figure 2 shows the unweighted distribution of the bullying index. The values range from - 1.23 to 4.69 with higher values indicating higher reported bullying than the OECD average for all pupils. The distribution is bimodal and right skewed, with a clear concentration of observations at lower and mid-range values. This indicates that most individuals report relatively low levels of bullying, however experiences are not evenly distributed across the group. The long right tail suggests that a smaller subset of individuals experienced substantially higher levels of bullying than the OECD average. Although this group has fewer students, their experiences are qualitatively different from those of the majority, highlighting heterogeneity in bullying reports.

Table 2 is weighted and shows that the distribution has a mean of 0.02 (Sd=0.93, Se=0.03). A substantial proportion of observations clustered at the lower bound of the scale. In contrast the upper quartile is well below the maximum value, providing no evidence of a ceiling effect. While the index displays a floor effect, variability across the upper range remains substantial, and the distribution supports its use as a continuous measure. This continuous nature of the index allows it to be used in correlational analyses, group comparisons, regression models after accounting for PISA’s complex survey design by including student weights. As with other IRT based indices, the bullying index reflects a model-based estimate.

Consistent with PISA conventions, missing or invalid responses that prevent scale construction for the bullying index are treated as missing data during analysis. The analysis is based on 3,763 observations with 1,000 missing values (20%) excluded from the original PISA sample of 4,763 cases. Missing data in the variable is due to deliberate item skipping, absenteeism, or invalid responses.

Figure 2 Density plot for bullying index (unweighted)

Table 2 Bullying index (weighted)

Reported school belonging

Table 3 summarises pupil responses to the statement “I feel like I belong at school”. From the initial PISA sample (N=4,763), data was missing for 1,620 cases, and in PISA 2022 these are coded into three different categories. These were: not applicable (635), no response (114) and system missing (871). The missing data was mainly due to response being invalid, multiple responses or no response.

The analysis presented in this report excludes all these three categories of missing data. Missing data treatment can be done in different ways. The approach taken by the PISA national report (DfE 2023) provides headline findings for only valid responses excluding all missing data and working with a similar sample (total sample N=3,107). They show school belonging as 63% nationally. We follow a similar approach for missing data treatment as in the PISA 2022 national report. All analyses are based on weighted percentages of valid responses (N=3,143), with missing data excluded. Reported school belonging summarised in Table 3 from our weighted analysis is 63.08% which aligns with existing published analysis.

Table 3 Student responses to school belonging: I feel like I belong at school

In all tables and analytical outputs that follow, the raw frequencies have been converted into weighted percentages, in line with the PISA national report. However, this may mean the final headline outputs differ from other analysis such as the Health Behaviour in School-aged Children survey report (HBSC 2023) which does not exclude missing data when calculating percentages, or other analyses which do not use weights.

Attainment 8

Distribution of Attainment 8 scores range from 0 to 90. The weighted sample has a mean of 50.60 (Table 4). The distribution is centred around this value and is approximately symmetric. The similarity between the mean and median (51) suggests that extreme values are unlikely to exert a strong influence on the distribution. The analysis is based on 4,233 cases with 530 missing values. Although the variable originates from the NPD, the values are observed only for pupils selected into the PISA sample and therefore inherit the PISA sampling design.

Table 4 Attainment 8

Prior attainment

KS2 prior attainment was measured by using end of key stage 2 Maths scores. Values range from 80 to 120, with a mean of 105.04 and a median of 106, indicating that the distribution is centred around this variable and is broadly symmetric (Table 5). The analysis includes 4,001 valid cases, with 530 missing values excluded.

Table 5 KS2 prior attainment

Pupil background characteristics

All NPD variables had missing data of 11.1% (N=4,763). Table 6 shows the weighted distribution of the sample (N=4,233) by FSM eligibility. 18.04% of respondents were FSM eligible and 81.96% were not on FSM.

Table 6 FSM eligibility

Ethnicity classified into major ethnic groups was used for the analysis and has been summarised in table 7. The subgroups were White, Asian, Black, Chinese, Mixed, any other ethnic group (AOEG) and Unclassified (N=4,233). Data was missing for 530 students

(11.1%). Because the sample sizes for Chinese, any other ethnic group (AOEG) and Unclassified are very small, these groups are excluded from bivariate analyses.

Table 7 Ethnicity

There were 11% cases which were unclassified due to information being refused or not obtained on SEN (Table 8). Information on SEN provision has two major categories (N=4,233). These are: No identified SEN (89.17%) and SEN (those with SEN support and those with an EHCP, 10.83%).

Table 8 Special educational needs

Table 9 shows the weighted percentages by gender. From the sample N=4,233, 49.63% of respondents were girls and 50.37% were boys.

Table 9 Gender

Variation in belonging across groups

To examine the robustness of associations across alternative operationalisations of belonging, we estimate the relationship between the belonging index and pupil background characteristics first, followed by equivalent estimations using reported school belonging. Estimating parallel models allows us to assess whether observed associations

with pupil background are consistent across measures or whether results vary depending on how the construct is defined.

Belonging index by pupil background characteristics

The charts below show the mean belonging index by sub-group. A score of zero represents the OECD average for all pupils. Error bars represent 95% confidence intervals, so where the error bars overlap with zero on the y-axis, the results are not statistically significantly different to the OECD average.

Gender differences

Boys (score= -0.06, SE=0.02) had higher scores on the belonging index than girls (score= -0.37, SE=0.02). This difference was statistically significant (p<0.001). As Figure 3 shows belonging scores were statistically significantly lower than the OECD average for both boys and girls. The relatively small standard errors for both groups indicate that these estimates are precise and that the difference is unlikely due to sampling variability (Table 10)

Table 10 Belonging index by gender

Figure 3 Belonging index by gender

Differences by ethnicity

Asian students (score= -0.13, SE=0.04) and Black students (score= -0.15. SE=0.06) had higher belonging scores than White students (score= -0.23, SE=0.02). Mixed ethnicity

students (score= -0.21, SE= 0.06) had similar scores to White students (Table 11). The subgroup differences were not statistically significant (p>0.05). However, White, Asian and Mixed ethnicity students had significantly lower scores than the OECD average for all pupils.

Chinese and AOEG students were much smaller subgroups so are not shown here due to smaller sample sizes (Figure 4). These results should be interpreted with caution, as ethnic group classifications used here are broad and may mask heterogeneity in experiences within groups, for example Asian students include multiple subgroups and may not necessarily have similar scores.

Table 11 Belonging index by ethnicity

Figure 4 Belonging index by ethnicity

Differences by FSM eligibility

FSM eligible pupils had a weighted mean of -0.34 (SE=0.03) whereas those not eligible for FSM had a weighted mean of -0.19 (SE=0.02). This difference of 0.15 points was significant (p<0.001). See table 12 and figure 5. FSM and non-FSM pupils both had statistically significantly lower belonging index scores than the OECD average for all pupils (p<0.001).

Table 12 Belonging index by FSM eligibility

Figure 5 Belonging index by FSM eligibility

Differences by SEN

Non-SEN pupils (score = -0.20, SE=0.02) had significantly higher scores (p<0.01) than SEN pupils (score = -0.36, SE=0.05). See table 13. Non-SEN pupils had statistically significantly lower scores than the OECD average for all pupils. SEN pupils also had statistically significantly lower scores than the OECD average (Figure 6)

Table 13 Belonging index by SEN

Figure 6 Belonging index by SEN

Reported school belonging by pupil background characteristics

Variations in reported school belonging by gender

Male students were significantly more likely than female students to agree with school belonging. 66.40% (SE=1.39) boys said they felt they belonged at their current school while 33% said they did not feel they belonged at school. Similarly, 58.09% (SE=1.41) girls felt they belonged at school while 42% did not feel they belonged at school (Figure 7).

The subgroup differences in reported school belonging by gender between boys and girls who agreed that they felt they belonged at their current school were statistically significant (p<0.001). These results are descriptive and do not account for other student or school characteristics that may be associated with both variables. The analysis excludes all cases with missing data for gender and/or student responses (N = 2,766).

Figure 7 Reported school belonging by gender

Variations in reported school belonging by ethnicity

Belonging varies by ethnicity, with White students less likely to report they belong at school. 60.26% (SE=1.25) White students, 70.17% (SE=2.88) Asian, 72.13% (SE=2.52) Black students and 65.16% (SE=3.84) of all students from mixed ethnic backgrounds who responded felt they belonged at their current school (Figure 8). The subgroup differences were statistically significant (p<0.01) between White and Asian, and White and Black students (p<0.001) but not for other subgroups.

Figure 8 Reported school belonging by ethnicity

Variations in reported school belonging by FSM eligibility

FSM eligible pupils were less likely to report belonging at school. 58.20% (SE=2.61) of all FSM eligible pupils and 63.18% (SE=1.09) of all non-FSM pupils agreed or strongly agreed that they felt like they belonged at their current school (Figure 9). However, the subgroup differences in reported school belonging between FSM and non-FSM students who agreed that they felt they belonged at their current school was not statistically significant (p>0.05)

Figure 9 Reported school belonging by FSM eligibility

Variations in reported school belonging by SEN

SEN pupils reported poorer belonging at school. 56.62% (SE=3.10) of all SEN pupils and 62.98% (SE=1.16) of non-SEN pupils agreed that they felt like they belonged at their current school (Figure 10). The subgroup differences were not significant (p=0.05).

Figure 10 Reported school belonging by SEN

Weighted mean bullying index by pupil background characteristics

The charts below show the mean bullying index by subgroup. A score of zero represents the OECD average for all pupils. Error bars represent 95% confidence intervals, so where the error bars overlap with zero on the y-axis, the results are not statistically significantly different to the OECD average

Bullying index and gender

Girls (score=0.04, SE=0.04) report higher levels of bullying than the OECD average, whereas boys (score= -0.01, SE=0.03) report lower levels of bullying than the OECD average (table 14 and figure 11). This difference in scores between boys and the OECD average was not statistically significant (p>0.05). The subgroup difference between boys and girls was not statistically significant.

Table 14 Bullying index by gender

Figure 11 Bullying by gender

Bullying by ethnicity

Weighted mean scores on the bullying index for White (score=0.06, SE=0.03), Black (score= -0.07, SE=0.08) and Mixed ethnicity students (score= -0.02, SE=0.08) were close to the OECD average while for Asian students (score= -0.22, SE=0.05) it was lower. Compared to the OECD average the differences in bullying scores for Asian students were

statistically significant. The ethnic subgroup difference between White and Asian students (p<0.001) was statistically significant but not so for others. See table 15 and figure 12.

Table 15 Bullying index by ethnicity

Figure 12 Bullying by ethnicity

Bullying by SEN

SEN pupils (score=0.23, SE=0.07) report statistically significantly higher levels of bullying than the OECD average for all pupils. Non-SEN pupils report lower bullying than the OECD average (score= -0.01, SE=0.03). The difference in bullying scores between SEN and non-SEN pupils was statistically significant (p<0.001). See table 16 and figure 13.

Table 16 Bullying index by SEN

Figure 13 Bullying by SEN

Bullying index by FSM

FSM pupils (score=0.09, SE=0.05) reported higher bullying than the OECD average for all pupils. Non-FSM pupils had weighted means (score = 0.00, SE=0.03) closer to the OECD average and the difference between the scores for non-FSM pupils and the OECD average was not statistically significant (Table 17). The differences in scores between FSM and non-FSM pupils (Figure 14) was not statistically significant (p>0.05).

Figure 14 Bullying by FSM

Table 17 Bullying index by FSM eligibility

Multiple linear regression analysis

Multiple linear regression models were estimated to examine the relationship between attainment 8, belonging, bullying, and pupil background characteristics. The belonging index, bullying index and KS2 prior attainment have been treated as continuous variables. FSM eligibility, gender, ethnicity and SEN have been treated as categorical variables with reference categories set to not-FSM, Male, White and no-SEN. All models are weighted and standard errors have been reported. Regression coefficients are interpreted as the expected change in attainment 8 associated with a one-unit change in the predictor, holding other variables constant.

There are four multivariate models. Attainment 8 is the outcome measure used in all models. The first two models use the belonging index, and the last two models use reported school belonging. Bullying is included in models 1 and 3 and dropped from models 2 and 4.

Estimating association between Attainment 8 and belonging index

Table 18 presents results of the multiple linear regression models fitted to examine predictors of attainment 8. Model 1 has an adjusted R2 of 0.49. Model 1 is a full model estimating the association between attainment 8 and the belonging index (N=3,120). After adjusting for covariates, the belonging index was weakly negatively associated with attainment 8 (B= -0.10, SE=0.33) however this was statistically insignificant (p>0.05).

Model 2 is a reduced model. We wanted to see how results change when bullying is removed from the model. We show excluding bullying, the association between belonging index and attainment 8 becomes stronger and positive (B=0.42, SE=0.33), however the coefficient for belonging index remains insignificant. The coefficients for all other variables remain similar and the adjusted R2 does not meaningfully change. The change suggests that bullying shares explanatory power with belonging index, indicating overlapping or confounding effects.

Estimating association between Attainment 8 and reported school belonging

Table 19 shows the coefficients and standard errors for model 3 and 4 fitted using reported school belonging, bullying and pupil background indicators (N=2,605). Model 3 had an adjusted R2 of 0.50.

Compared to those who “strongly agreed” that they felt like they belong at their current school, students who agreed on average had attainment scores 1.11 points lower (SE=1.10), controlling for other variables. Likewise, students who disagreed had 4.02 points (SE=1.13) lower attainment scores while those who strongly disagreed have 7.72 points lower attainment scores (SE=1.30).

A clear gradient was therefore observed across response categories for school belonging, with “disagree” and “strongly disagree” being statistically significantly different to “strongly agree”. Lower levels of school belonging were associated with lower attainment 8 scores. As reported school belonging changed from strongly agree to strongly disagree, we found a drop in attainment 8 scores by 7.72 points equivalent to almost a full GCSE grade lower across all eight GCSEs.

A one unit increase in bullying index led to reduction of 0.21 points (SE = 0.27) in attainment 8 scores, however in neither model 3 or 4 is this significant. Results from this alternative specification are broadly consistent with model 2 for all other predictors. Model 4 is a reduced model estimated to see how results change when bullying is removed from the model (N=2,632, Adjusted R2=0.50).

Broadly, the estimates are consistent between models. However, when excluding bullying, the coefficient for school belonging demonstrating association with attainment 8 becomes slightly stronger, but the overall association remains clear and consistent with a strong, positive and significant relationship between belonging and attainment, that is not driven by bullying alone.

These models show that the single item construct used in models 3 and 4 is a more direct and precise measure of school belonging and that stronger school belonging is associated with higher attainment 8 scores.

Table 18 Regression coefficients for models 1 and 2

Table 19 Regression coefficients for models 3 and 4

Conclusion

The report set out to assess the relationship between school belonging and attainment and answer three research questions. The overall conclusions of the analysis are summarised below.

Does school belonging vary across different subgroups of young people by their background characteristics such as FSM eligibility, ethnicity, gender and special educational needs?

Across almost all sub-groups in this analysis, belonging index scores were below the OECD average for all pupils, and the belonging gap varied significantly between subgroups by gender (p<0.001), SEN (p<0.01) and FSM (p<0.001). Girls (p<0.001), SEN pupils (p<0.01), and FSM-eligible pupils (p<0.001) report significantly lower belonging scores than boys, non-SEN, and non-FSM pupils. By ethnicity, Asian pupils have belonging scores closest to the OECD average, while White, Black, and Mixed ethnicity pupils show larger gaps (negative). This difference between White and other ethnic groups were not statistically significant. White, Asian, Black and Mixed ethnicity students had statistically significantly lower scores than the OECD average for all pupils. Overall, this highlights consistent inequalities, with particularly low belonging among girls and disadvantaged groups (FSM and SEN).

Reported sense of belonging at school varies significantly across pupil groups, but the patterns are broadly similar to the belonging index analysis. Boys, non-FSM pupils, and non-SEN pupils are more likely to report feeling a higher sense of belonging, while girls (p<0.001), FSM-eligible pupils (p>0.05), and SEN pupils (p=0.05) report lower levels though some differences were not quite significant at the 5%. By ethnicity, White pupils are less likely to feel they belong to their current school compared to Asian (p<0.01) and Black pupils (p<0.001), with these differences being statistically significant.

Findings highlight clear disparities in school belonging, with lower belonging concentrated among disadvantaged (FSM) pupils, SEN pupils and girls, though the strength and significance depend on the measure used. Generally, White pupils had lower belonging scores, but some differences were statistically insignificant.

Which student subgroups report higher experiences of bullying?

Bullying experiences are uneven across pupil groups in England relative to the OECD average for all pupils. Higher levels of bullying are reported by girls, pupils with SEN, FSM- eligible pupils, and White pupils, with these differences being statistically significant for SEN pupils. In contrast, boys, non-SEN and non-FSM pupils report levels similar to the OECD average for all pupils. Asian (p<0.001) pupils report significantly lower levels of bullying than the OECD average and also when compared with White students (p<0.001).

This suggests inequalities in bullying prevalence across demographic and socio-economic groups, highlighting vulnerability particularly for SEN pupils where we detect a statistically significant difference. The analysis shows that FSM eligible and White pupils are also more likely to report being bullied at school, though this was not always statistically significant at the 5% level.

To what extent is school belonging associated with attainment, after controlling for other observed factors, including bullying?

Taken together, the analysis provides evidence of a positive association between school belonging and attainment, but that the nature and strength of this association depend on how belonging is measured and how bullying is treated in the model.

When belonging is measured using a single, direct measure about whether pupils feel that they belong at school (reported school belonging), there is a clear and statistically significant association with attainment. This is the preferred measure of belonging as it is the most precise metric available in the data. Pupils who report not feeling that they belong at their current school have substantially lower attainment, and this relationship is stable whether bullying is included in the model or not, after conditioning on other demographic and prior attainment variables. This suggests that the single-item measure captures a broader and more intrinsic dimension of school belonging that is not fully explained by bullying alone.

When belonging is measured using the composite belonging index that includes items closely related to peer relationships and social exclusion, the association with attainment is weaker and not statistically significant (whether bullying is controlled for or not). The change in magnitude and sign of the coefficient when bullying is removed from the model suggests that much of the apparent relationship between this composite measure and attainment is shared with or driven by bullying experiences.

Overall, the analysis supports the conclusion that bullying is an important factor in understanding the relationship between school belonging and attainment and likely represents one mechanism through which low belonging is associated with poorer outcomes (attainment 8 scores). However, the results also indicate that pupils’ general sense of belonging at school is associated with attainment over and above reported bullying experiences. These findings underline the importance of being explicit about how belonging is conceptualised and measured, and of interpreting results considering the overlap between belonging, bullying, and wider experiences.

Future work should build on this analysis by longitudinal tracking and exploring additional pathways through which belonging could impact attainment (such as attendance) to ensure that policy responses remain evidence informed.

References

Banerjee, P. (2025). The home learning environment and attainment. Research report Department for Education RR1502 Isbn: 978-1-83870-629-6 www.gov.uk

Booker, K.C. (2006) School Belonging and the African American Adolescent: What Do We Know and Where Should We Go? The High School Journal, 89, 1-7. http://dx.doi.org/10.1353/hsj.2006.0005

Deci, E. L., & Ryan, R. M. (1985). Intrinsic motivation and self-determination in human behaviour. Springer New York: Plenum Press.

DfE (2024). Key stage 4 performance data guidance academic year 2022/23 - Explore education statistics. Department for Education www.gov.uk Accessed March 2026

DfE (2025). Pupil experiences in school: academic year 2024 to 2025 - Gov.UK Department for Education www.gov.uk Accessed March 2025

DfE (2025). Secondary accountability measures: 2025 guidance for maintained secondary schools, academies and free schools. Department for Education www.gov.uk Accessed November 2025

DfE (2025). Key stage 2 attainment. Academic year 2024/25 revised www.gov.uk Accessed February 2026

DfE (2026). Every child achieving and thriving. Schools White paper 2026 Department for Education policy paper. ISBN 978-1-5286-6032-7, E03470295 www.gov.uk

Department for Education (2026), A level and other 16 to 18 results, Academic year 2024/25, Explore Education Statistics.

Dyson, A. and Gallannaugh, F. (2008). Disproportionality in special needs education in England. The Journal of Special Education, 42(1), p.36-46

Feinstein, L. (2003). Inequalities in the early cognitive development of British children in the 1970 cohort. Economica 70(277), p. 73-97 Doi: 10.1111/1468-0335.t01-1-00272

Francis, B. & Skelton, C. (2005). Reassessing Gender and Achievement: Questioning Contemporary Key Debates. Routledge 1-200. Doi: 10.4324/9780203412923

Gomes, A. M., Martins, M. C., Farinha, M., Silva, B., Ferreira, E., Caldas, A. C., & Brandão, T. (2020). Bullying’s negative effect on academic achievement. International Journal of Educational Psychology, 9(3), 243–268.

Goodenow, C. (1993). The psychological sense of school membership among adolescents: Scale development and education correlates. Psychology in the Schools, 30(1), 79-90.

Ingram, J., Stiff, J., Cadwallader, S., Lee, G., & Kayton, H. (2023). PISA 2022: National report for England. Department for Education. Reference: RR1392 Isbn: 978-1-83870- 522-0

Juvonen, J., & Graham, S. (2014). Bullying in schools: The power of bullies and the plight of victims. Annual Review of Psychology, 65, 159–185. https://doi.org/10.1146/annurev- psych-010213-115030

Korpershoek, H., Canrinus, E. T., Fokkens-Bruinsma, M. and Boer, H. de. (2019). The relationships between school belonging and students’ motivational, social-emotional, behavioural and academic outcomes in secondary education: a meta-analytic review. Research papers in education, 35 (1), p.1-40

Liu, Y. and Huang, J. (2025). School belonging and students’ math achievement among secondary students: the role of math attitude and disorderly behaviour. Psychology in the schools, 62(1), p. 167-184, Doi: 10.1002/pits.23316

Maslow, A. H. (1943). A theory of human motivation. Psychological review, 50(4), 370- 396.

Nakamoto, J., & Schwartz, D. (2010). Is peer victimization associated with academic achievement? A meta‑analytic review. Social Development, 19(2), 221–242.

OECD (2019). Pisa 2018 results www.oecd.org OECD (2022) Pisa publications: Technical standards www.oecd.org Accessed March 26 OECD (2023). The state of learning and equity in education. Report www.oecd.org Accessed March 2026 Parsons, S. and Platt, L. (2017). The early academic progress of children with special educational needs. BERJ, 43: 466-485 Reardon, S. F. (2011). The Widening Academic Achievement Gap between the Rich and the Poor: New Evidence and Possible Explanations. In R. Murnane, & G. Duncan (Eds.), Whither Opportunity? Rising Inequality, Schools, and Children’s Life Chances (pp. 91-116). Russell Sage Foundation.

Rose, C. A., Espelage, D. L., Aragon, S. R. and Elliott, J. (2011). Bullying and victimisation among students in special education and general education curricula. Exceptionality Education International, 21(2), pp. 2-14

Samara, M., Nascimento, B. D., El-Asam, A., Hammuda, S. and Khattab, N. (2021). How can bullying victimisation lead to lower academic achievement? A systematic review and meta-analysis of the mediating role of cognitive motivational factors. International Journal of Environmental Research Public Health, 18(5): 2209, Doi: 10.3390/ijerph18052209

Sirin, S. R. (2005). Socioeconomic status and academic achievement: A meta‑analytic review of research. Review of Educational Research, 75(3), 417–453.

Social Mobility Commission (2025). Attainment at age 16. State of the nation www.gov.uk SMF (2025). We used to have it all: is there a trade-off between wellbeing and attainment in schools. Social market foundation Briefing paper www.smf.co.uk Accessed March 2026 Swearer, S. M., Espelage, D. L., Vaillancourt, T., & Hymel, S. (2010). What can be done about school bullying? Linking research to educational practice. Educational Researcher, 39(1), 38–47 Tippett, N., Houlston, C., & Smith, P. K. (2010). Prevention and response to identity-based bullying among local authorities in England, Scotland, and Wales. Equality and Human Rights Commission.

Wang, D., Shen, M. & Wu, X. (2026). Socioeconomic status and inequalities in early development of non-cognitive skills: Evidence from China. China Economic Review, 95, 102586

Youth Futures Foundation (2026, based on ONS Annual Population Survey), NEET statistics dashboard.

© Department for Education copyright 2026 This publication is licensed under the terms of the Open Government Licence v3.0, except where otherwise stated. To view this licence, visit nationalarchives.gov.uk/doc/open- government-licence/version/3. Where we have identified any third-party copyright information you will need to obtain permission from the copyright holders concerned. Reference: RR1645 Isbn: 978-1-83870-811-5 For any enquiries regarding this publication, contact www.gov.uk/contact-dfe.

This document is available for download at www.gov.uk/government/publications.