Guidance

Contextual GLD score methodology

Published 27 January 2026

Applies to England

About the contextual GLD score

A school’s contextual good level of development (GLD) score is an estimate of GLD, with certain cohort characteristics (such as the proportion of children eligible for free school meals (FSM) or with special educational needs (SEN)), considered.

Contextual GLD scores have been calculated using a linear regression model. This linear regression model is a statistical technique that uses actual GLD scores and contextual cohort information from schools across England to establish the general relationship between all these variables. This then allows us to estimate what a school’s GLD score would be, given the specific characteristics of that school’s cohort, if that school followed the pattern seen in the data as a whole. That estimate is what we call the “contextual GLD score”.

For example, summer-born children (children born between April 1st and August 31st) are less likely to achieve GLD. Imagine there are 2 schools with an identical cohort, except that School 1 has a relatively younger cohort. The model will predict that School 1 will have a lower contextual GLD score than School 2.

It’s important to remember that this score is an estimate, not a precise measurement. Schools are complex, and not all influencing factors can be captured in the model. Always triangulate data in this report with other information, such as observations of and conversations with children and teachers.

A school’s contextual GLD score is not a target they are predicted to meet, a benchmark to compare against, or a minimum standard. It is a tool to help reflect on a school’s outcomes with their context taken into account.

Data sources

For the 2024 to 2025 reports, we used data from the following sources:

• internal early years foundation stage profile (EYFSP) pupil-level data (from 2022 onwards)
• school-level data from the:
  • schools, pupils and their characteristics (academic year 2024 to 2025) statistics
  • special educational needs in England (academic year 2024 to 2025) statistics
  • Get Information about Schools database

Scope

Only data relating to pupils included in published national and local authority EYFSP figures were used. We excluded data relating to pupils from independent schools, special schools, and state-funded nurseries.

Only schools that met all the criteria below received a report:

• state-funded primary and secondary schools with 2024 to 2025 EYFSP data for more than 5 pupils
• schools that were in the January 2025 census and did not close or change their unique reference number before the 2024 to 2025 EYFSP return
• schools that, for each variable feeding into the GLD model, have at least one pupil who is not missing data for that field (such as SEN status or ethnicity)

Schools who met any of the following criteria did not receive a report:

• state-funded nursery schools (also known as maintained nursery schools), special schools, alternative provision or independent schools
• schools with fewer than 6 children in their reception class for 2024 to 2025
• newly opened or academised schools
• schools missing 100% of data for any field

Trusts or local authorities that have only one school in scope of receiving a report will not have a responsible body or local area report generated.

For the latest iteration of the reports published in November 2025, 15,208 schools were in scope and received a report. An additional 514 schools are included in the final model, but these were removed from receiving a report due to not having 2024 to 2025 data.

Cohort-level variables

We included pupil characteristics at cohort level using EYFSP data. These were the:

• proportion of children in the reception cohort on FSM
• proportion of children in the reception cohort with SEN support
• proportion of children in the reception cohort with an Educational Health Care plan (EHCP)
• proportion of female children in the reception cohort
• proportion of children in the reception cohort with English as an additional language (EAL)
• mean age at the start of the academic year (in months) of the reception cohort - this takes into account that younger children (those born in the summer) are less likely to reach GLD than those born in the autumn
• mean Income Deprivation Affecting Children Index (IDACI) score of the reception cohort – this is currently based on the 2019 IDACI index, but we will look to update this with the 2025 IDACI index in future iterations
• major ethnic groups, including % Asian, % Black, % Mixed, % White of the reception cohort
• mean urban/rurality score for the reception cohort, with 1 representing children in the most urban areas and 10 representing children living in the most rural areas. To calculate this, each child has a rural-urban classification code which categorises their home location. The pupils are grouped by their school, and then the mean urban/rurality score is calculated.
• proportion of children in the reception cohort who were not present at the start of the year. Children who joined the school where they were assessed on or before 30 September are considered not mobile. Children who joined on or after 1 October are considered mobile. This includes children without census data for that school, as they are assumed to have arrived after the summer census. The proportion of pupils who were not present at the start of the year is calculated from this.
• spread of attainment of early learning goals in the reception cohort. To calculate this, the number of Early Learning Goals (ELGs) each pupil achieves at the expected level is counted. The variation or spread of the number of ELGs reached is measured within each school using the standard deviation.

School-level variables

We included the following pupil characteristics at school level using the ‘Schools, pupils and their characteristics’ and ‘Special educational needs in England’ publications:

• proportion of children in the school on FSM
• proportion of children in the school with SEN support
• proportion of children in the school with an EHCP

These were used to supplement cohort-level data, especially where younger children in the reception cohort may not yet be identified for FSM or SEN support.

Variables not included

In a future iteration of this model, we are hoping to include the:

• proportion of children in the reception cohort who have received pre-reception formal childcare

There are several variables that we have not included in the calculation for contextual GLD scores due to those variables closely aligning with other variables already included in the model, a lack of available data, or those variables not relating to cohort characteristics. For example, we have not included:

• minor ethnic groups
• local authority
• absence rate
• school funding per pupil
• immigration status
• proportion of children in the reception cohort who are Children of Service Families
• proportion of children in need in the reception cohort, including those on child in need plans, child protection plans, children looked after by local authorities, care leavers and disabled children

Given the complexity of schools and the multitude of factors influencing outcomes, not all relevant variables can be captured within the model. As such, contextual GLD scores should be interpreted as estimates rather than absolute measures.

Modelling approach

We used a linear regression model to predict the percentage of pupils achieving GLD based on pupil characteristics.

Before modelling, variables were standardised to ensure they were comparable and to prevent differences in scale from biasing coefficient estimates. For example, in the November 2025 version of the reports, mean age in months varies from 49.8 to 56.8, while mean urban/rurality score from 1.0 to 10.0. Standardising ensures different variables are comparable.

We included interaction terms where relationships between variables were expected to influence outcomes, allowing the model to capture more nuanced relationships.

Model selection

We compared models using Akaike’s Information Criterion to balance model fit and complexity. The final model included predictors that consistently appeared in the best-performing models.

All the cohort-level and school-level variables listed earlier are included in the final model, as well as the following interactions:

• mean age at the start of reception and the proportion of children in the reception cohort on FSM
• mean age at the start of reception and the proportion of children in the reception cohort with SEN support
• proportion of children in the reception cohort with EAL and the proportion of children in the reception cohort on FSM

The refined model explained 45.1% of the variation in GLD outcomes. This means that just under half of the differences in children’s GLD results can be accounted for by the variables included in the model. The remaining 54.9% is influenced by other unmeasured or external factors not included in the model. Models that explain more than 40% of the variation are often considered strong in explanatory power.[1]

We used the model to estimate a contextual GLD score for each school. This represents the GLD rate expected based on the school’s context.

Comparing contextual scores to average actual results

Contextual GLD scores are then compared with a school’s average actual GLD percentage so they can see whether their results are:

• better than predicted
• at the predicted level
• lower than predicted

We used a hybrid approach to define these thresholds, based on either the percentage point difference one pupil makes or the 95% confidence interval, depending on cohort size.

A school is considered ‘as predicted’ if their actual GLD is within a certain distance of their contextual GLD score. This distance is unique to each school. For most schools, this distance is based on the percentage point difference that one child makes to the overall score, known as the child’s difference. However, if a school’s 95% confidence interval around the predicted score is wider than the child’s difference, then the confidence interval is used instead.

For example, if a school has 10 children in the cohort, the child’s difference would be 100/10 = 10ppt. If a school has 25 children in the cohort, the child’s difference would be 100/25 = 4ppt. The larger the cohort, the smaller the child’s difference. Most schools (99.9%) used a child’s difference.

Below, we have included the number and percentage of schools nationally in each category for the November 2025 iteration of reports that include the 2024/25 EYFSP data:

• better than predicted: 5,044 (32.1%)
• at the predicted level: 6,203 (39.5%)
• lower than predicted: 4,475 (28.5%)

A school’s contextual GLD is measured relative to the performance of schools across England. This means that, even as national outcomes increase, the proportion of schools that are ‘better than predicted’, ‘at the predicted level’, or ‘lower than predicted’ may not change substantially.

Even if a school’s average actual GLD is better than their contextual GLD, there may still be areas for improvement. The Department for Education defines ‘better than predicted’ based on current national performance, so thresholds can change over time. This means that a school performing well now may not remain above predicted as national results improve.

Therefore, all schools should regularly review their outcomes and focus on continuous improvement, even if their current performance exceeds their contextual GLD.

Assumptions

Missing data: we have assumed that known pupil data is representative of the full cohort. For example, in a cohort where 10% of pupils are missing information about their FSM eligibility, we assume the same proportion of these pupils are FSM eligible as in the 90% of the cohort with known FSM information. This assumption avoids the need for schools that have any missing information to be removed from the model. However, please note that, as a result, in Annex B of the reports:

• ethnicity inputs will not sum to 100% as this is calculated based on pupils with only known ethnicities and ‘any other ethnic group’ values (which were not included in the model) are not shown, however, ‘% unknown’ is calculated using the total number of pupils
• percentages for children with GLD, eligible for FSM, with SEN support, an EHCP, EAL and each known ethnicity category have been calculated by removing children with missing data from the total denominator

Linear relationships: we have assumed predictors have a linear effect on GLD.

Under-identification: we have assumed reception cohort-level FSM, SEN support and EHCP may underestimate true rates, and have supplemented with school-level data.

Mobility data: when a school undergoes a change of governance (for example, joining a new academy trust) part way through the year, it is challenging to determine mobility accurately. To ensure these schools can be included in the model, we assign them the national median mobility figure for the year in which the change occurred. We are currently developing an improved approach for future report publications.

Limitations

Some variables were excluded due to data availability or safeguarding reasons (for example, children in need and service family status). Publishing detailed breakdowns could risk identification, which may have serious consequences. Their exclusion ensures confidentiality and safety.

We decided to include variables that are closely related to each other because this helps the model make better overall predictions. However, this means the influence of each individual variable is harder to separate and explain.

Trust or local authority contextual score

A trust or local authority’s overall contextual GLD score combines all the contextual GLD scores for the schools in the trust or local authority into one figure.

Each school’s score is weighted by its size (the proportion of children in that school). To do this, we multiply each school’s contextual GLD score by its weight (share of pupils), then add these together to get an overall weighted score.

This overall score is then compared to the 3 year average GLD percentage for the trust or LA to show the difference. Please note, only schools with a contextual GLD score are included in this calculation and in the 3 year average. This means schools who have recently become part of the trust or local authority will be excluded. It’s also worth noting that rapid improvement will be masked by the three-year average.

Again, it’s important to remember that this score is an estimate, not a precise measurement. Always triangulate data in this report with other information, such as observations of and conversations with children and teachers. Even if your average actual GLD is better than your contextual GLD, there may still be areas for improvement.

As with the school reports, a trust or local authority’s overall contextual GLD score is in line with their average actual GLD if it is within a ‘child’s difference’. This is based on the number of children in the trust or local authority cohort.

---

[1] Cohen, J. (1988). Statistical Power Analysis for the Behavioural Sciences (2nd ed.). Lawrence Erlbaum Associates.