What goes with school absence

A first look at the causes of school-level absence in England’s secondary schools, framed by the research and tested on published data for every school.

NoteWork in progress

This is the first cut of the absence analysis on the roadmap. It describes how absence varies between schools, and with what. It does not yet say why. Every figure comes from published school-level data, so every result is an association between schools, not an effect on pupils.

Why this matters for value added

Absence is the largest single term in Value Added v2.1’s expectation. The model takes each school’s absence as given: a school with more absence is expected to get lower results. That is fair to schools whose absence is outside their control. It is unfair if some of that absence is something the school could have prevented. To tell the two apart, we need to know what absence goes with. This page makes a start.

What the research says

A research map of UK work since 2016 (summarised below; sources at the foot of the page) says absence is not one behaviour with one cause. Health, family, peer, school and wider circumstances all bear on it, and they interact. It is higher in secondary schools, rises with age, and is concentrated among pupils facing poverty, special educational needs, health difficulties and other disadvantage. Absence harms attainment whether or not it is authorised, but it is also a marker of difficulties that were already there.

The post-pandemic rise is real and not fully explained. The leading accounts are:

  • more illness;
  • anxiety and avoiding school;
  • broken routines and weaker attendance norms;
  • unmet SEND needs;
  • strained relations between home and school;
  • slow access to support.

No single cause has been shown to dominate.

Each strand of that research suggests something we can measure for every school in England:

The research points to What we can measure for each school Source
Physical illness, the most common recorded reason Illness as a share of possible sessions, as its own outcome DfE absence data
Mental health, anxiety, school-related distress Pupils whose main SEN need is social, emotional and mental health DfE SEN data, January 2026
SEND, and gaps that have not closed SEN support, EHC plans, autism, learning difficulties DfE SEN data
Poverty and family circumstances Free school meals; young carers School census, January 2026
Differences by ethnicity and sex White British, Gypsy, Roma and Traveller, EAL; share of girls School census
Prior attainment and disengagement KS2 prior attainment; neighbourhood education Value Added inputs
School experience: climate, belonging, fit Only proxies: size, sixth form, admissions, governance, faith Census, GIAS
Place Region, local authority, rural or urban Census
Post-COVID norms The trend by reason; not testable across schools in one year DfE absence data

Ofsted grades are left out on purpose. Attendance is part of what an inspection judges, so a grade would partly be absence measured again.

How absence has changed, by reason

Figure 1: Absence by reason in England’s state-funded secondary schools, % of possible sessions, full years, 2016/17 to 2024/25 (sessions-weighted; DfE school-level absence data). 2019/20 was not collected; 2020/21 was a year of partial closure.
Figure 2: Persistent absence (missing 10% or more of sessions) and severe absence (50% or more), % of pupils, same schools and years.

The reasons have not moved together:

  • Illness jumped from 2.8% of sessions in 2018/19 to 5.2% in 2021/22. By 2024/25 it had fallen back to 3.7%, most of the way home.
  • Unauthorised absence other than holidays doubled, from 1.3% to 2.9% in 2023/24. It has barely come down since.
  • Severe absence rose every year, from 1.2% of pupils to 3.6%, nearly three times as many.

Persistent absence is falling, but the tail is still growing. That matches the research: the average can improve while the worst cases get worse.

What absence is made of

Every missed session carries a register code, and DfE publishes them for each school. In the autumn and spring terms of 2025/26, illness and unauthorised absence together were most of it.

Table 1: Reasons for absence in England’s state-funded secondary schools, 2025/26 autumn and spring terms. Reasons under 1% of absence are left out.
Reason Share of all absence % of possible sessions
Illness 45.9% 3.67
Unauthorised, other 29.4% 2.35
Authorised, other 4.8% 0.38
Medical appointments 4.5% 0.36
Reduced timetable 3.6% 0.29
Unauthorised holiday 3.6% 0.29
Late 3.5% 0.28
Suspended (code E) 3.1% 0.25

Suspension is a small part of it. Code E, a pupil suspended with no alternative provision, is 3.1% of absence, or 0.25 points. This is counted from the register, not estimated.

Schools that suspend more do have more absence of other kinds, but most of that is who their pupils are. Each extra suspension per 100 pupils goes with 0.043 points more non-suspension absence between schools, and 0.012 points when a school is compared with itself across years. It adds little to value added either. Absence adds 0.11 to the R² of Attainment 8 on top of suspension; suspension adds 0.005 on top of absence. So the model has no need of a suspension term. The full analysis is in Exclusions and Attendance.

What goes with absence, school by school

The model covers 3,147 state-funded secondary schools with full data. It uses their absence in the autumn and spring terms of 2025/26 and their intake as recorded in the January 2026 school census and SEN data. For each of seven outcomes, it is a multilevel model of the outcome (logged) on all the characteristics at once, with region and local authority effects.

Each number below is the difference in that kind of absence that goes with one standard deviation more of a characteristic, all else equal. A standard deviation is roughly the gap between a typical school and one noticeably above it. For yes-or-no characteristics, the number compares the two kinds of school. Purple means more absence, green less. Blank cells are differences that could be zero.

Figure 3: Difference in each kind of absence (%) with one standard deviation more of each characteristic (yes vs no for the categories), all else equal: multilevel models with region and local authority effects, 2025/26 autumn and spring terms. Purple, more absence; green, less. Blank, the 95% interval includes zero.

What stands out:

  • Poverty, prior attainment and mobility go with the absence a school has least claim to. With one SD more pupils on free school meals, unauthorised absence is +37% and severe absence +22%. Illness, once everything else is allowed for, is -9%. Lower KS2 prior attainment and more pupils joining mid-phase point the same way.
  • Among the types of need, social, emotional and mental health need is the one that matters. One SD more of it goes with +10% severe and +8% other authorised absence. Once the types of need are in the model, the share of pupils on SEN support or with EHC plans adds nothing more. This fits the research’s emphasis on anxiety and distress.
  • Schools with more White British pupils have more absence of every kind, especially severe (+30% per SD) and unauthorised (+12%). EAL goes the other way. This is a comparison between schools, not pupils. It matches the national statistics, where absence is lower among most minority ethnic groups. It does not say why.
  • Girls. Schools with more girls have more medical-appointment (+3%) and severe absence (+6%). This matches the research finding that girls are absent more than boys.
  • The school.
    • Selective schools have far less unauthorised (-37%) and severe absence (-48%) than intake alone predicts.
    • Schools with a religious character have somewhat less (-11% unauthorised).
    • Schools with a sixth form have more absence among their 11-16 pupils (+5% overall). This echoes Value Added v2.1, where 11-16 schools do better.
    • Being in a multi-academy trust makes little difference.
  • Place. Rural schools record +12% absence for medical appointments and -8% unauthorised absence, against urban schools near a city. A plausible reading is that getting to an appointment from a rural school costs a whole session rather than part of one. That bears on the distance-to-school question.
  • Young carers barely register. This is more likely a measurement problem than a finding: schools differ widely in how many young carers they identify.

How much each group explains

Figure 4: Share of the variation between schools in each kind of absence (logged) explained by each group of characteristics: Shapley values of R² from OLS, region in Place. Each column sums to the total R² shown at the foot.

The kinds of absence split cleanly in two:

  • Unauthorised, persistent and severe absence are mostly about who the pupils are. Intake and school characteristics account for between 62% and 68% of the variation. Poverty, prior attainment and the school itself carry most of it.
  • Illness and medical appointments are mostly not. Only 30% and 24% is accounted for, and Place is the largest single group for both. Two schools with the same intake can record very different amounts of illness.

That is the case for v2.1’s sensitivity test that counts illness separately from other absence (Methodology).

The school’s part

“The school” carries 13 to 14 points of R² for unauthorised, persistent and severe absence above. That does not mean schools are to blame for that much absence. The group is five measured features:

  • size;
  • whether the school has a sixth form;
  • its admissions (selective, or non-selective);
  • who runs it;
  • whether it has a religious character.

What schools do is not measured at all. Splitting the group into its five features shows where the credit goes:

Figure 5: The school group’s share of R², split into its five features: Shapley values with each feature as a player of its own and the other seven groups as before (OLS, region in Place), 2025/26 autumn and spring.

Most of it is selection. Admissions accounts for 9.2 of the points for unauthorised absence, 10.9 for persistent and 10.8 for severe. Selective schools choose their pupils. Prior attainment and free school meals measure some of what selection does, but not all of it, and admissions picks up the rest. That is a fact about who comes to the school, not about how it is run. Who runs it (2.6 points), size, sixth form and religious character each add little.

What a school does shows up somewhere else: in how far apart similar schools are. Take schools with the same intake, needs and setting, in the same local authority. What is left over (the models’ school-level residuals) is what the measured characteristics cannot explain. That includes how a school runs attendance, and also anything about its pupils the data does not see.

Table 2: How far apart schools with the same intake, needs and setting, in the same local authority, are on each kind of absence (2025/26 autumn and spring), and whether the differences repeat (the same models on 2024/25 full-year absence).
Outcome Quarter from top vs quarter from bottom Tenth from top vs tenth from bottom Correlation with 2024/25 Top quarter in 2025/26 also top quarter in 2024/25
Overall absence +24% +52% 0.84 70%
Illness +33% +76% 0.81 68%
Medical appointments +36% +82% 0.78 68%
Other authorised +61% +157% 0.73 63%
Unauthorised +53% +133% 0.81 70%
Persistent absence (10%+) +31% +73% 0.84 68%
Severe absence (50%+) +67% +174% 0.70 62%
  • The gaps are large. Among otherwise similar schools in the same authority, the one a quarter of the way from the top has 53% more unauthorised absence than the one a quarter of the way from the bottom. For severe absence the gap is 67%; for overall absence, 24%.
  • They repeat. A school’s unexplained absence in 2025/26 correlates 0.81 with its own in 2024/25 for unauthorised absence, and 0.84 overall. Of the schools in the top quarter for unexplained unauthorised absence in 2025/26, 70% were in the top quarter the year before; chance would give 25%. These are stable features of schools or their communities, not noise.
  • That stability is not proof of a school effect. The same 2026 characteristics are used for both years, and pupils the data does not see are as stable as a school’s practice. Only following schools whose practice changed would separate the two.
TipFor heads and governors
  1. Judge absence against what your intake predicts, not against the national average. The area pages give each Sussex school’s absence by kind, its prediction, and how it compares with similar schools in the same authority over two years.
  2. Structure is not the lever. Being in a multi-academy trust, its size, and its religious character make little difference once intake is allowed for. Most of what looks like a “school” effect in the averages is selective admissions, which is about who is admitted.
  3. Practice may well be the lever. Schools with the same intake differ by half again or more in unauthorised and severe absence, and the differences persist. A school well above similar schools, year after year, should ask what its look-alikes below the line are doing differently.
  4. Look at the kind of absence. Unauthorised, persistent and severe absence are where similar schools differ most. Illness differs mostly by place, and partly by how readily it is authorised. Comparing schools on unauthorised absence alone falls hardest on those with poorer pupils.
  5. The research points to where to look. It is consistent on:
    • early identification;
    • relationships and belonging;
    • support for social, emotional and mental health needs (the type of need that goes most with absence here);
    • work with families. Parental engagement is the one approach with consistent support from randomised trials, though most were in the United States.
  6. Find a twin finds the schools in England most like yours in intake and setting. It also finds anti-twins: the look-alikes with much lower absence, or much higher value added. These are the schools worth calling.
  7. Schools with a sixth form have +5% more absence among their 11-16 pupils than comparable 11-16 schools. That is worth a look in 11-18 schools.

Place, once the intake is allowed for

Table 3: Share of the variation between schools at the region and local authority levels, before and after the characteristics are in the model (multilevel models, 2025/26 autumn and spring).
Outcome Region and authority, before Region and authority, after Explained by the characteristics
Overall absence 14% 5% 59%
Illness 31% 14% 26%
Medical appointments 20% 9% 20%
Other authorised 13% 12% 25%
Unauthorised 19% 4% 68%
Persistent absence (10%+) 11% 3% 65%
Severe absence (50%+) 18% 4% 60%

For most kinds of absence, a school’s region and authority matter little once its intake is known. Illness and other authorised absence are the exceptions: about 14% and 12% of the variation between schools is still at the region and authority levels after allowing for intake. Both are kinds of absence a school chooses to authorise. Some places’ schools record more illness than their intakes predict. That could be because their pupils really are ill more often. It could also be because schools in those places authorise absence as illness more readily, so that the same absence elsewhere would be recorded as unauthorised. The data cannot tell these apart.

Figure 6: Each local authority’s illness absence against what its schools’ intakes predict, 2025/26 autumn and spring (multilevel model, region and authority effects together). 152 authorities, lowest on the left; Sussex’s three are named.

Brighton and Hove records +27% more illness absence than its schools’ intakes predict, the 2nd highest of 152 authorities. Its unauthorised absence is -5% against intake (110th). The city’s high absence, which the Geographical effects page finds costs its pupils so little attainment, is illness absence, and its intake does not explain it. That leaves two readings: an unusual amount of illness, or a habit of authorising absence as illness that elsewhere would go unauthorised. Which it is matters for value added. If it is recording, the city’s absence is not the kind that costs attainment, and the model’s absence term over-forgives it. The next two sections take that further.

Illness or recording?

Authorities differ in how the same absence gets recorded, and the pattern follows disadvantage.

Table 4: Absence by reason across local authorities, grouped by disadvantage, % of possible sessions, 2025/26 autumn and spring terms. Authorities with more than a million possible sessions.
Local authorities Illness Unauthorised, other All absence
Least disadvantaged quarter 4.00 1.79 7.73
2nd quarter 3.99 2.11 8.00
3rd quarter 3.35 2.65 7.99
Most disadvantaged quarter 3.01 3.26 8.35
  • More disadvantaged authorities record less illness and more unauthorised absence. Total absence barely moves. Across authorities, illness and unauthorised absence correlate -0.46.
  • There is some real trade-off within authorities too. Following each authority over time, after allowing for national trends, one point more illness goes with 0.28 points less unauthorised absence (p = 0.024). If absence were simply being re-labelled, the figure would be close to one. It suggests that recording explains part of the difference between places, not all of it.

The same holds between pupils. In 2024/25, full year, illness was 32% of disadvantaged pupils’ absence in England and 52% of non-disadvantaged pupils’. Disadvantaged pupils still have more illness-coded absence, but their unauthorised absence is about four times as high.

Judging a school on unauthorised absence therefore falls hardest on schools with poorer pupils, whether or not those pupils miss more school.

Brighton and Hove, code by code

Brighton and Hove’s secondary absence was 9.84% in 2025/26, autumn and spring, against 8.01% for England. Illness is 1.74 points of that 1.83-point gap, or 95%.

Figure 7: Illness absence, Brighton and Hove and England, % of possible sessions, autumn and spring terms. 2019/20 is not in the published series.
Table 5: Brighton and Hove’s rank among 150 local authorities on each kind of absence, 2025/26 autumn and spring (1 is the highest).
Reason % of possible sessions Rank
Authorised, other 0.61 1
Illness 5.42 2
All absence 9.84 6
Unauthorised, other 2.42 65
Medical appointments 0.35 80
Unauthorised holiday 0.21 114
Suspended (code E) 0.13 134
  • It is the whole city. All ten schools are above England’s 3.67 on illness; the lowest is at 4.59.
  • The gap grew after the pandemic. It was 0.50 points in 2018/19 and is 1.74 now.
  • It is high for pupils of every kind. Disadvantaged pupils’ illness absence was 6.55% in 2024/25, full year, against 4.22% in England. For non-disadvantaged pupils it was 4.65% against 3.34%, and for all pupils 5.18% against 3.61%. Those rank 3rd, 2nd and 2nd of 150.
  • Re-labelling cannot be the whole story. Unauthorised absence is mid-table (65th of 150) and only slightly below what the city’s intake predicts, while illness is far above it. If schools here were turning truancy into illness, unauthorised absence would be low. For disadvantaged pupils it is high: 6.85% against 5.35%.

Code I means the school accepted the explanation given. It covers a chronic condition, a heavy cold, anxiety that makes the school gate impossible, and a parent saying “not well today”. The published data cannot separate them. The research’s emphasis on anxiety and the shape of the city’s pattern both point towards emotionally based school avoidance, but that is a hypothesis. Only the council’s and schools’ own records could test it.

Is it a coastal problem?

The authorities with the largest illness excess look coastal: the Isle of Wight, South Tyneside, Brighton and Hove, Torbay. Census data from The Engagement Platform also reports weaker belonging and enjoyment among coastal secondary pupils, and a steeper fall in engagement between Year 7 and Year 8. That comes from a commercial provider, gives no sample sizes and has no attendance data, but belonging is exactly what this page cannot measure.

A school counts as coastal here if it is within 10 km of the sea, and an authority if at least half its sessions are in such schools. That makes 20 of 150 authorities coastal.

Table 6: How much more of each kind of absence coastal schools have than others, with 95% intervals: the multilevel models above with a coastal term added, 2025/26 autumn and spring.
Outcome Free school meals only Without ethnicity and language Everything on this page
Overall absence +2.5% (-0.6 to +5.7) +3.6% (+1.0 to +6.2) +1.4% (-1.0 to +3.9)
Illness +6.4% (+2.3 to +10.7) +6.5% (+2.5 to +10.6) +4.7% (+0.8 to +8.7)
Medical appointments +4.1% (+0.5 to +7.7) +3.0% (-0.4 to +6.5) +0.7% (-2.6 to +4.1)
Other authorised +1.5% (-4.0 to +7.3) +0.9% (-4.3 to +6.5) -0.9% (-6.0 to +4.5)
Unauthorised -2.2% (-7.8 to +3.8) +2.4% (-2.5 to +7.6) -0.2% (-5.0 to +4.8)
Persistent absence (10%+) +3.4% (-0.9 to +7.8) +5.9% (+2.5 to +9.3) +3.4% (+0.2 to +6.7)
Severe absence (50%+) -0.5% (-7.9 to +7.5) +5.1% (-1.2 to +11.9) -0.7% (-6.3 to +5.3)
Figure 8: Illness absence above or below what disadvantage predicts, against the share of each authority’s secondary sessions in schools within 10 km of the sea. Authorities with more than a million possible sessions, 2025/26 autumn and spring.
  • Illness is the one kind of absence clearly left. With everything on this page in the model, coastal schools have +4.7% illness absence (+0.8% to +8.7%), and persistent absence is +3.4%. Overall and unauthorised absence are indistinguishable from other schools. That is real, but small: about a sixth of a point of illness.
  • The wider gap is mostly who and where. Comparing whole authorities on disadvantage alone, coastal ones have 1.04 points more absence of all kinds. The table is stricter. Its models compare schools within regions and authorities, so a difference between whole authorities counts for less. That may go too far, and so may allowing for language, because few pupils with English as an additional language is partly what a coastal intake is. Taking language out, in the middle column, raises every figure a little.
  • It differs by pupil group. In 2024/25, allowing for each authority’s disadvantage, the coastal gap in all absence is 1.89 points for disadvantaged pupils, 0.50 for non-disadvantaged pupils and 1.04 for all pupils. For disadvantaged pupils most of it is unauthorised (1.06 points).
  • It does not explain Brighton and Hove. Adding a coastal term only takes the city’s excess from 1.72 points to 1.33, still 5th highest. Coastal authorities also vary hugely: illness is 5.70% on the Isle of Wight and 2.31% in Blackpool. Camden, with no coast, is among the highest.

The Year 7 to 8 pattern shows up in unauthorised absence, not illness. DfE publishes absence by year group for each authority, so the same pupils can be followed for a year, in the two cohorts up to 2024/25.

Table 7: Rise in absence from Year 7 to Year 8 among the same pupils, percentage points, mean across authorities, and the coastal difference allowing for disadvantage, EAL, SEN and region. Standard errors clustered on authority.
Absence Coastal Other Brighton and Hove, two cohorts Coastal difference p
All absence 1.73 1.27 2.80, 2.01 +0.17 0.026
Illness 0.19 0.15 0.98, 0.40 +0.01 0.859
Unauthorised, other 0.92 0.62 1.36, 0.85 +0.15 0.016

That echoes TEP’s finding in a different measure. Absence rises faster after Year 7 in coastal authorities, but the rise is in unauthorised absence (+0.15, p = 0.016), and illness shows no difference (+0.01). So whatever is weakening engagement on the coast does not look like the source of the coastal illness excess. Brighton and Hove differs here too: its rise in illness is well above both groups, and its rise in unauthorised absence is near the coastal figure. It also starts high, with Year 6 absence already 5th highest. These figures have no split by disadvantage.

Being on the coast accounts for only a small part of the city’s illness excess. The rest needs another explanation.

The full models

Each of the seven outcomes has the same multilevel model. For school \(i\) in local authority \(j\) in region \(k\):

\[\log(y_{ijk} + 0.1) = \beta_0 + \sum_p \beta_p\, x_{p,ijk} + u_k + v_{jk} + e_{ijk}\]

\[u_k \sim N(0, \sigma^2_{\text{region}}), \quad v_{jk} \sim N(0, \sigma^2_{\text{LA}}), \quad e_{ijk} \sim N(0, \sigma^2_{\text{school}})\]

  • The outcome \(y\) is a percentage: of possible sessions for the five kinds of absence, of pupils for persistent and severe. The 0.1 lets severe absence be zero.
  • The measured characteristics \(x\) are standardised to mean 0 and SD 1, so each \(\beta\) is the change in log absence for one SD, and \(100\,(e^{\beta} - 1)\) is the % difference shown in Figure 3.
  • The categories are compared with a reference: 11-16, non-selective in a selective area, maintained, no religious character, and urban near a major town or city.
  • Estimation is REML in lme4. The log-likelihoods, AIC and BIC are from the same models refitted by maximum likelihood. The marginal R² is the variance of the fixed-effect predictions over the total; the conditional R² adds the region and authority effects (Nakagawa and Schielzeth). “Explained by the characteristics” is the fall in total variance from the model with no characteristics.
Overall absence: fixed effects. Estimates are on the log scale, per SD for a measured characteristic, against the reference for a category.
Term Estimate SE t % difference (95% CI)
Intercept (reference school, mean characteristics) 2.059 0.025 81.4
Free school meals (%) 0.102 0.006 16.0 +10.7 (+9.3 to +12.1)
SEN support (%) -0.015 0.009 -1.7 -1.5 (-3.1 to +0.2)
EHC plans (%) -0.006 0.005 -1.2 -0.6 (-1.5 to +0.4)
Social, emotional and mental health need (%) 0.048 0.006 7.6 +4.9 (+3.6 to +6.2)
Autism (%) 0.016 0.005 3.3 +1.6 (+0.6 to +2.5)
Learning difficulty, moderate or specific (%) 0.006 0.006 0.9 +0.6 (-0.6 to +1.8)
Young carers (%) 0.002 0.004 0.4 +0.2 (-0.6 to +0.9)
English as an additional language (%) -0.027 0.008 -3.3 -2.6 (-4.2 to -1.1)
White British (%) 0.089 0.010 9.2 +9.3 (+7.2 to +11.4)
Gypsy, Roma and Traveller (%) 0.019 0.004 5.3 +1.9 (+1.2 to +2.7)
Girls (%) 0.016 0.003 4.8 +1.6 (+0.9 to +2.3)
Joined mid-phase (%) 0.045 0.004 11.5 +4.6 (+3.8 to +5.4)
KS2 prior attainment -0.064 0.007 -8.7 -6.2 (-7.5 to -4.8)
Neighbourhood education 0.001 0.005 0.1 +0.1 (-0.9 to +1.1)
Size (log pupils) -0.006 0.004 -1.5 -0.6 (-1.4 to +0.2)
Has a sixth form (vs 11-16) 0.051 0.009 6.0 +5.3 (+3.5 to +7.0)
Non-selective elsewhere (vs same) 0.014 0.022 0.6 +1.4 (-2.8 to +5.8)
Selective (vs non-selective, selective area) -0.130 0.030 -4.4 -12.2 (-17.2 to -7.0)
Multi-academy trust (vs maintained) -0.024 0.010 -2.5 -2.4 (-4.2 to -0.5)
Single-academy trust (vs maintained) -0.027 0.012 -2.2 -2.7 (-5.0 to -0.3)
Religious character (vs none) -0.044 0.009 -4.9 -4.3 (-6.0 to -2.6)
Urban, further from a city (vs nearer) 0.012 0.015 0.8 +1.2 (-1.7 to +4.2)
Rural (vs urban, nearer a city) -0.005 0.012 -0.4 -0.5 (-2.7 to +1.8)
Random effects: the variance at each level, with no characteristics and in the full model.
Level Variance, no characteristics Variance, full model Share of the starting total, after
Region 0.0066 0.0006 1%
Local authority 0.0057 0.0034 4%
School 0.0767 0.0323 36%
Data and fit.
Measure Value
Schools 3,147
Local authorities 152
Regions 9
Mean outcome (%) 8.14
Explained by the characteristics 0.591
Marginal R² (fixed effects) 0.594
Conditional R² (with region and authority) 0.639
OLS R² (Shapley total, region fixed) 0.601
Log-likelihood (ML) 861.7
Log-likelihood, no characteristics (ML) -500.0
AIC -1669.3
BIC -1505.9
Illness: fixed effects. Estimates are on the log scale, per SD for a measured characteristic, against the reference for a category.
Term Estimate SE t % difference (95% CI)
Intercept (reference school, mean characteristics) 1.356 0.045 30.1
Free school meals (%) -0.094 0.010 -9.5 -9.0 (-10.7 to -7.2)
SEN support (%) -0.015 0.013 -1.2 -1.5 (-4.0 to +1.0)
EHC plans (%) -0.000 0.007 -0.1 -0.0 (-1.5 to +1.4)
Social, emotional and mental health need (%) 0.024 0.010 2.5 +2.4 (+0.5 to +4.3)
Autism (%) 0.017 0.007 2.4 +1.7 (+0.3 to +3.2)
Learning difficulty, moderate or specific (%) 0.010 0.009 1.1 +1.0 (-0.7 to +2.9)
Young carers (%) 0.008 0.006 1.4 +0.8 (-0.4 to +1.9)
English as an additional language (%) -0.029 0.012 -2.4 -2.9 (-5.2 to -0.5)
White British (%) 0.052 0.015 3.5 +5.3 (+2.3 to +8.4)
Gypsy, Roma and Traveller (%) 0.001 0.005 0.1 +0.1 (-1.0 to +1.2)
Girls (%) 0.018 0.005 3.7 +1.8 (+0.8 to +2.8)
Joined mid-phase (%) 0.015 0.006 2.6 +1.5 (+0.4 to +2.7)
KS2 prior attainment -0.030 0.011 -2.7 -3.0 (-5.1 to -0.9)
Neighbourhood education 0.004 0.008 0.5 +0.4 (-1.2 to +1.9)
Size (log pupils) -0.012 0.006 -2.0 -1.2 (-2.4 to -0.0)
Has a sixth form (vs 11-16) 0.033 0.013 2.6 +3.4 (+0.8 to +6.0)
Non-selective elsewhere (vs same) -0.068 0.035 -1.9 -6.5 (-12.8 to +0.2)
Selective (vs non-selective, selective area) -0.274 0.045 -6.1 -24.0 (-30.4 to -17.0)
Multi-academy trust (vs maintained) -0.046 0.015 -3.2 -4.5 (-7.2 to -1.8)
Single-academy trust (vs maintained) -0.037 0.019 -2.0 -3.7 (-7.1 to -0.1)
Religious character (vs none) -0.011 0.014 -0.8 -1.1 (-3.8 to +1.5)
Urban, further from a city (vs nearer) 0.062 0.023 2.7 +6.4 (+1.8 to +11.2)
Rural (vs urban, nearer a city) 0.039 0.018 2.2 +4.0 (+0.4 to +7.6)
Random effects: the variance at each level, with no characteristics and in the full model.
Level Variance, no characteristics Variance, full model Share of the starting total, after
Region 0.0135 0.0046 4%
Local authority 0.0243 0.0128 11%
School 0.0833 0.0726 60%
Data and fit.
Measure Value
Schools 3,147
Local authorities 152
Regions 9
Mean outcome (%) 3.71
Explained by the characteristics 0.257
Marginal R² (fixed effects) 0.218
Conditional R² (with region and authority) 0.369
OLS R² (Shapley total, region fixed) 0.296
Log-likelihood (ML) -440.5
Log-likelihood, no characteristics (ML) -700.7
AIC 935.0
BIC 1098.5
Medical appointments: fixed effects. Estimates are on the log scale, per SD for a measured characteristic, against the reference for a category.
Term Estimate SE t % difference (95% CI)
Intercept (reference school, mean characteristics) -0.869 0.037 -23.5
Free school meals (%) -0.043 0.009 -4.9 -4.2 (-5.9 to -2.6)
SEN support (%) -0.007 0.012 -0.6 -0.7 (-3.0 to +1.6)
EHC plans (%) 0.007 0.007 1.0 +0.7 (-0.6 to +2.0)
Social, emotional and mental health need (%) 0.006 0.009 0.7 +0.6 (-1.1 to +2.4)
Autism (%) 0.007 0.007 1.0 +0.7 (-0.6 to +2.0)
Learning difficulty, moderate or specific (%) 0.009 0.008 1.0 +0.9 (-0.8 to +2.5)
Young carers (%) 0.002 0.005 0.4 +0.2 (-0.8 to +1.2)
English as an additional language (%) -0.007 0.011 -0.7 -0.7 (-2.9 to +1.5)
White British (%) 0.071 0.013 5.3 +7.4 (+4.6 to +10.2)
Gypsy, Roma and Traveller (%) 0.001 0.005 0.2 +0.1 (-0.9 to +1.1)
Girls (%) 0.033 0.005 7.3 +3.4 (+2.5 to +4.3)
Joined mid-phase (%) 0.004 0.005 0.7 +0.4 (-0.7 to +1.5)
KS2 prior attainment 0.012 0.010 1.2 +1.2 (-0.8 to +3.2)
Neighbourhood education 0.021 0.007 2.9 +2.1 (+0.7 to +3.5)
Size (log pupils) -0.041 0.006 -7.1 -4.0 (-5.1 to -2.9)
Has a sixth form (vs 11-16) 0.063 0.012 5.4 +6.5 (+4.1 to +9.0)
Non-selective elsewhere (vs same) 0.038 0.029 1.3 +3.8 (-1.9 to +9.9)
Selective (vs non-selective, selective area) -0.013 0.041 -0.3 -1.3 (-8.9 to +7.0)
Multi-academy trust (vs maintained) -0.032 0.013 -2.4 -3.2 (-5.7 to -0.6)
Single-academy trust (vs maintained) -0.006 0.017 -0.3 -0.6 (-3.8 to +2.8)
Religious character (vs none) -0.003 0.013 -0.2 -0.3 (-2.7 to +2.2)
Urban, further from a city (vs nearer) 0.039 0.021 1.9 +3.9 (-0.2 to +8.2)
Rural (vs urban, nearer a city) 0.112 0.016 7.0 +11.9 (+8.4 to +15.5)
Random effects: the variance at each level, with no characteristics and in the full model.
Level Variance, no characteristics Variance, full model Share of the starting total, after
Region 0.0032 0.0028 3%
Local authority 0.0148 0.0056 6%
School 0.0702 0.0619 70%
Data and fit.
Measure Value
Schools 3,147
Local authorities 152
Regions 9
Mean outcome (%) 0.37
Explained by the characteristics 0.203
Marginal R² (fixed effects) 0.207
Conditional R² (with region and authority) 0.301
OLS R² (Shapley total, region fixed) 0.238
Log-likelihood (ML) -156.6
Log-likelihood, no characteristics (ML) -406.7
AIC 367.2
BIC 530.6
Other authorised: fixed effects. Estimates are on the log scale, per SD for a measured characteristic, against the reference for a category.
Term Estimate SE t % difference (95% CI)
Intercept (reference school, mean characteristics) 0.046 0.062 0.7
Free school meals (%) 0.123 0.014 8.9 +13.1 (+10.1 to +16.3)
SEN support (%) -0.026 0.019 -1.4 -2.6 (-6.1 to +1.0)
EHC plans (%) 0.030 0.010 2.9 +3.0 (+0.9 to +5.2)
Social, emotional and mental health need (%) 0.080 0.014 5.9 +8.3 (+5.4 to +11.2)
Autism (%) 0.005 0.010 0.5 +0.5 (-1.5 to +2.5)
Learning difficulty, moderate or specific (%) 0.014 0.013 1.1 +1.5 (-1.1 to +4.1)
Young carers (%) 0.014 0.008 1.7 +1.4 (-0.3 to +3.0)
English as an additional language (%) -0.047 0.017 -2.7 -4.6 (-7.8 to -1.3)
White British (%) 0.053 0.021 2.5 +5.4 (+1.1 to +9.9)
Gypsy, Roma and Traveller (%) 0.022 0.008 2.8 +2.2 (+0.6 to +3.7)
Girls (%) -0.002 0.007 -0.3 -0.2 (-1.6 to +1.2)
Joined mid-phase (%) 0.047 0.008 5.6 +4.8 (+3.1 to +6.6)
KS2 prior attainment -0.052 0.016 -3.3 -5.0 (-7.9 to -2.0)
Neighbourhood education 0.023 0.011 2.1 +2.4 (+0.2 to +4.6)
Size (log pupils) -0.028 0.009 -3.1 -2.7 (-4.4 to -1.0)
Has a sixth form (vs 11-16) 0.061 0.018 3.3 +6.3 (+2.5 to +10.2)
Non-selective elsewhere (vs same) 0.021 0.048 0.4 +2.1 (-7.0 to +12.1)
Selective (vs non-selective, selective area) -0.053 0.064 -0.8 -5.2 (-16.4 to +7.5)
Multi-academy trust (vs maintained) -0.064 0.021 -3.1 -6.2 (-10.0 to -2.4)
Single-academy trust (vs maintained) -0.041 0.026 -1.5 -4.0 (-8.8 to +1.1)
Religious character (vs none) -0.062 0.020 -3.2 -6.0 (-9.6 to -2.4)
Urban, further from a city (vs nearer) 0.039 0.032 1.2 +4.0 (-2.4 to +10.8)
Rural (vs urban, nearer a city) -0.058 0.025 -2.3 -5.6 (-10.1 to -0.8)
Random effects: the variance at each level, with no characteristics and in the full model.
Level Variance, no characteristics Variance, full model Share of the starting total, after
Region 0.0110 0.0089 4%
Local authority 0.0186 0.0186 8%
School 0.2066 0.1489 63%
Data and fit.
Measure Value
Schools 3,147
Local authorities 152
Regions 9
Mean outcome (%) 1.06
Explained by the characteristics 0.253
Marginal R² (fixed effects) 0.268
Conditional R² (with region and authority) 0.383
OLS R² (Shapley total, region fixed) 0.307
Log-likelihood (ML) -1553.4
Log-likelihood, no characteristics (ML) -2065.3
AIC 3160.9
BIC 3324.4
Unauthorised: fixed effects. Estimates are on the log scale, per SD for a measured characteristic, against the reference for a category.
Term Estimate SE t % difference (95% CI)
Intercept (reference school, mean characteristics) 0.929 0.052 17.9
Free school meals (%) 0.313 0.013 24.3 +36.7 (+33.3 to +40.2)
SEN support (%) 0.002 0.017 0.1 +0.2 (-3.1 to +3.7)
EHC plans (%) -0.024 0.010 -2.4 -2.4 (-4.2 to -0.5)
Social, emotional and mental health need (%) 0.064 0.013 5.1 +6.6 (+4.0 to +9.3)
Autism (%) 0.021 0.010 2.2 +2.1 (+0.2 to +4.0)
Learning difficulty, moderate or specific (%) -0.016 0.012 -1.3 -1.5 (-3.9 to +0.8)
Young carers (%) -0.003 0.008 -0.4 -0.3 (-1.8 to +1.2)
English as an additional language (%) -0.020 0.016 -1.2 -1.9 (-5.0 to +1.2)
White British (%) 0.114 0.019 5.8 +12.1 (+7.9 to +16.4)
Gypsy, Roma and Traveller (%) 0.015 0.007 2.1 +1.5 (+0.1 to +3.0)
Girls (%) 0.021 0.007 3.1 +2.1 (+0.8 to +3.4)
Joined mid-phase (%) 0.079 0.008 9.9 +8.2 (+6.5 to +9.9)
KS2 prior attainment -0.130 0.015 -8.9 -12.2 (-14.7 to -9.6)
Neighbourhood education -0.022 0.010 -2.1 -2.2 (-4.1 to -0.2)
Size (log pupils) 0.021 0.008 2.6 +2.2 (+0.5 to +3.9)
Has a sixth form (vs 11-16) 0.056 0.017 3.3 +5.8 (+2.3 to +9.4)
Non-selective elsewhere (vs same) 0.068 0.044 1.5 +7.0 (-1.9 to +16.7)
Selective (vs non-selective, selective area) -0.469 0.060 -7.9 -37.5 (-44.4 to -29.7)
Multi-academy trust (vs maintained) 0.005 0.019 0.3 +0.5 (-3.2 to +4.4)
Single-academy trust (vs maintained) -0.066 0.025 -2.7 -6.4 (-10.8 to -1.7)
Religious character (vs none) -0.113 0.018 -6.2 -10.7 (-13.8 to -7.4)
Urban, further from a city (vs nearer) -0.055 0.030 -1.8 -5.3 (-10.7 to +0.4)
Rural (vs urban, nearer a city) -0.086 0.023 -3.7 -8.2 (-12.3 to -3.9)
Random effects: the variance at each level, with no characteristics and in the full model.
Level Variance, no characteristics Variance, full model Share of the starting total, after
Region 0.0332 0.0025 1%
Local authority 0.0544 0.0159 3%
School 0.3694 0.1297 28%
Data and fit.
Measure Value
Schools 3,147
Local authorities 152
Regions 9
Mean outcome (%) 3.00
Explained by the characteristics 0.676
Marginal R² (fixed effects) 0.664
Conditional R² (with region and authority) 0.706
OLS R² (Shapley total, region fixed) 0.679
Log-likelihood (ML) -1331.4
Log-likelihood, no characteristics (ML) -3004.1
AIC 2716.8
BIC 2880.2
Persistent absence (10%+): fixed effects. Estimates are on the log scale, per SD for a measured characteristic, against the reference for a category.
Term Estimate SE t % difference (95% CI)
Intercept (reference school, mean characteristics) 3.021 0.031 96.3
Free school meals (%) 0.140 0.008 17.0 +15.1 (+13.2 to +17.0)
SEN support (%) -0.015 0.011 -1.3 -1.5 (-3.7 to +0.8)
EHC plans (%) -0.019 0.006 -2.9 -1.8 (-3.1 to -0.6)
Social, emotional and mental health need (%) 0.054 0.008 6.5 +5.6 (+3.9 to +7.3)
Autism (%) 0.017 0.006 2.8 +1.7 (+0.5 to +3.0)
Learning difficulty, moderate or specific (%) 0.002 0.008 0.2 +0.2 (-1.4 to +1.8)
Young carers (%) 0.000 0.005 0.0 +0.0 (-1.0 to +1.0)
English as an additional language (%) -0.021 0.011 -2.0 -2.1 (-4.1 to -0.1)
White British (%) 0.097 0.013 7.7 +10.2 (+7.5 to +13.0)
Gypsy, Roma and Traveller (%) 0.019 0.005 4.0 +1.9 (+1.0 to +2.9)
Girls (%) 0.018 0.004 4.2 +1.9 (+1.0 to +2.7)
Joined mid-phase (%) 0.054 0.005 10.3 +5.6 (+4.5 to +6.7)
KS2 prior attainment -0.100 0.010 -10.4 -9.6 (-11.2 to -7.8)
Neighbourhood education -0.012 0.007 -1.8 -1.2 (-2.5 to +0.1)
Size (log pupils) -0.017 0.006 -3.1 -1.7 (-2.8 to -0.6)
Has a sixth form (vs 11-16) 0.071 0.011 6.3 +7.4 (+5.1 to +9.8)
Non-selective elsewhere (vs same) 0.018 0.027 0.7 +1.8 (-3.4 to +7.3)
Selective (vs non-selective, selective area) -0.414 0.039 -10.5 -33.9 (-38.8 to -28.6)
Multi-academy trust (vs maintained) -0.040 0.013 -3.1 -3.9 (-6.3 to -1.5)
Single-academy trust (vs maintained) -0.030 0.016 -1.8 -3.0 (-6.0 to +0.2)
Religious character (vs none) -0.063 0.012 -5.2 -6.1 (-8.3 to -3.9)
Urban, further from a city (vs nearer) 0.030 0.020 1.5 +3.0 (-0.9 to +7.1)
Rural (vs urban, nearer a city) 0.011 0.015 0.7 +1.1 (-1.9 to +4.3)
Random effects: the variance at each level, with no characteristics and in the full model.
Level Variance, no characteristics Variance, full model Share of the starting total, after
Region 0.0082 0.0006 0%
Local authority 0.0114 0.0041 2%
School 0.1571 0.0579 33%
Data and fit.
Measure Value
Schools 3,147
Local authorities 152
Regions 9
Mean outcome (%) 21.97
Explained by the characteristics 0.646
Marginal R² (fixed effects) 0.648
Conditional R² (with region and authority) 0.675
OLS R² (Shapley total, region fixed) 0.650
Log-likelihood (ML) -37.1
Log-likelihood, no characteristics (ML) -1625.2
AIC 128.1
BIC 291.6
Severe absence (50%+): fixed effects. Estimates are on the log scale, per SD for a measured characteristic, against the reference for a category.
Term Estimate SE t % difference (95% CI)
Intercept (reference school, mean characteristics) 1.098 0.064 17.2
Free school meals (%) 0.199 0.015 12.9 +22.0 (+18.3 to +25.7)
SEN support (%) -0.030 0.021 -1.4 -3.0 (-6.9 to +1.1)
EHC plans (%) -0.007 0.012 -0.6 -0.7 (-2.9 to +1.7)
Social, emotional and mental health need (%) 0.091 0.015 6.0 +9.6 (+6.3 to +12.9)
Autism (%) 0.031 0.011 2.7 +3.1 (+0.8 to +5.4)
Learning difficulty, moderate or specific (%) 0.001 0.015 0.1 +0.1 (-2.8 to +3.0)
Young carers (%) 0.008 0.009 0.9 +0.8 (-1.0 to +2.7)
English as an additional language (%) -0.010 0.020 -0.5 -1.0 (-4.7 to +2.9)
White British (%) 0.259 0.024 11.0 +29.5 (+23.7 to +35.6)
Gypsy, Roma and Traveller (%) 0.026 0.009 2.9 +2.6 (+0.9 to +4.4)
Girls (%) 0.056 0.008 6.9 +5.7 (+4.1 to +7.4)
Joined mid-phase (%) 0.090 0.010 9.4 +9.4 (+7.4 to +11.5)
KS2 prior attainment -0.161 0.018 -9.1 -14.9 (-17.8 to -11.9)
Neighbourhood education 0.046 0.012 3.7 +4.7 (+2.2 to +7.3)
Size (log pupils) 0.020 0.010 2.0 +2.0 (+0.0 to +4.1)
Has a sixth form (vs 11-16) 0.047 0.021 2.3 +4.8 (+0.6 to +9.1)
Non-selective elsewhere (vs same) 0.060 0.050 1.2 +6.2 (-3.7 to +17.1)
Selective (vs non-selective, selective area) -0.658 0.072 -9.1 -48.2 (-55.1 to -40.3)
Multi-academy trust (vs maintained) 0.002 0.023 0.1 +0.2 (-4.3 to +4.9)
Single-academy trust (vs maintained) -0.061 0.030 -2.0 -5.9 (-11.3 to -0.2)
Religious character (vs none) -0.093 0.022 -4.2 -8.8 (-12.7 to -4.8)
Urban, further from a city (vs nearer) 0.041 0.036 1.1 +4.2 (-2.9 to +11.8)
Rural (vs urban, nearer a city) -0.029 0.028 -1.0 -2.9 (-8.1 to +2.7)
Random effects: the variance at each level, with no characteristics and in the full model.
Level Variance, no characteristics Variance, full model Share of the starting total, after
Region 0.0530 0.0083 2%
Local authority 0.0434 0.0149 3%
School 0.4468 0.1929 36%
Data and fit.
Measure Value
Schools 3,147
Local authorities 152
Regions 9
Mean outcome (%) 3.65
Explained by the characteristics 0.602
Marginal R² (fixed effects) 0.588
Conditional R² (with region and authority) 0.632
OLS R² (Shapley total, region fixed) 0.622
Log-likelihood (ML) -1938.2
Log-likelihood, no characteristics (ML) -3285.2
AIC 3930.5
BIC 4093.9

What one standard deviation is, in each characteristic’s own units, for the 3,147 schools in the models, and how many schools fall in each category:

Table 8: Measured characteristics: mean and standard deviation across the schools in the models. KS2 prior attainment is centred and neighbourhood education is already a z-score; size is the log of pupils on roll.
Characteristic Mean One SD
Free school meals (%) 28.76 14.66
SEN support (%) 14.61 5.84
EHC plans (%) 3.68 2.11
Social, emotional and mental health need (%) 5.00 2.71
Autism (%) 3.02 2.01
Learning difficulty, moderate or specific (%) 5.60 3.14
Young carers (%) 1.49 1.85
English as an additional language (%) 19.06 17.89
White British (%) 60.41 28.97
Gypsy, Roma and Traveller (%) 0.27 0.86
Girls (%) 49.56 16.33
Joined mid-phase (%) 4.49 3.87
KS2 prior attainment 4.48 2.72
Neighbourhood education -0.00 1.00
Size (log pupils) 6.92 0.41
Table 9: Categories: schools in each, with the reference category first.
Characteristic Category Schools
Sixth form No 1178
Sixth form Yes 1969
Admissions Non-selective in a selective area 203
Admissions Non-selective elsewhere 2781
Admissions Selective 163
Governance Maintained 526
Governance Multi-academy trust 2129
Governance Single-academy trust or not recorded 492
Religious character No 2578
Religious character Yes 569
Rural or urban Urban, near a major town or city 2519
Rural or urban Urban, further from one 222
Rural or urban Rural 406

What this cannot tell us

  • Schools, not pupils. A school with more pupils on free school meals having more absence does not show that its pupils on free school meals are the ones absent.
  • Association, not cause. Some of the school characteristics may follow from absence rather than lead to it, such as a school’s size or who runs it.
  • Two terms of one year. Autumn and spring carry more illness than summer, and one year is a snapshot. The trend figures are full years.
  • Measured needs only. SEN data counts identified need. The research warns that unmet or unidentified need sits outside it, and so do young carers not known to their school.
  • School climate is unmeasured. Belonging, bullying, relationships and fit come out of the research as important, and nothing here measures them.
  • Reasons are published by place, not by pupil group within a school. The disadvantaged and non-disadvantaged figures are for whole authorities, for full years up to 2024/25, so they do not line up exactly with the 2025/26 school figures.
  • Coastal is a judgement. The 10 km line treats estuaries narrower than 10 km as land, which leaves Southampton inland. Large counties such as Devon have many coastal schools but fall below the half-way mark.

Next steps

  1. Within-school change. Started: How school absence changes follows every school from 2016/17 to 2025/26, tests whether absence moves when intake does, and looks at academy conversion and changes of trust.
  2. Illness or recording? Started: Illness or recording? finds a partial trade-off. Separating the two for good needs records that say what lies behind code I, such as anxiety against infection, and whether a few pupils or many are behind it.
  3. Distance to school, from the roadmap. The rural appointments result gives a first reason to expect it to matter.
  4. The severe tail. Severe absence is rising while the average falls. It needs its own model.
  5. Back into value added. Absence against intake, split by kind, could replace raw absence in a future version of the model.

Sources

The research summary draws on a research map of UK work since 2016 (compiled with Elicit), including:

The coastal section also cites Fitzpatrick (2026), What engagement data reveals about coastal schools, The Engagement Platform, which is not peer reviewed.

Data: DfE school-level absence (2016/17 to 2024/25, full years; 2025/26, autumn and spring), DfE absence by pupil characteristic (authorities, full years to 2024/25), DfE suspensions and permanent exclusions, the January 2026 school census and SEN school-level data, GIAS, ONS county and unitary authority boundaries, and the Value Added v2.1 inputs. Code: R/08_absence_causes.R and R/17_absence_recording.R, with R/15_exclusions.R and R/15b_coastal_absence.R in school_attainment_tool.