Policy paper

Local Government Outcomes: Statistical Neighbours model

Updated 1 July 2026

1. Introduction

MHCLG uses a local authority statistical neighbours model to support comparisons between local areas and aid interpretation of data within the Local Outcomes Framework. 

A statistical neighbours model MHCLG uses identifies which local authorities are most similar to one another based on a set of contextual characteristics. Such models are widely used across local government to enable meaningful comparison and benchmarking.

This MHCLG model compares local authorities across a defined set of indicators, including urbanicity and levels of deprivation. Local authorities with the most similar profiles across these indicators are designated as statistical neighbours.

Developing a bespoke model allows the indicator set to be tailored to the factors most relevant to the outcomes and metrics captured in the Framework. MHCLG has worked closely with the Office for National Statistics’ (ONS) Local Statistics and Analysis Division, building on their nearest neighbours method and code to produce a model whose underlying data and methods are best suited to the Framework. As a result, the statistical neighbours used here will differ from those used elsewhere, for example, in the Department for Education’s Children’s social care dashboard.

Through the Local Outcomes Framework tool, users will be able to compare the outcomes and metrics for a given local area with those of its statistical neighbours. This document sets out the methodology behind the statistical neighbours model and details of its development. The statistical neighbours will be updated on an annual basis as new data becomes available.

For further information, see the Local Outcomes Framework.

How we developed the model

The model identifies neighbours by comparing each local authority’s profile across a set of contextual indicators. The indicators were chosen to reflect the contextual factors most relevant to the priority outcomes in the Local Outcomes Framework and were tested for data quality and statistical robustness.

A full description of the data, method and checks is set out in the methodology.

Statistical neighbours for English local authorities

The table below lists the 15 statistical neighbours for each local authority in England, alongside brief notes on the methodology and full data sources.

Download the Statistical Neighbours model list for English local authorities.

How to read the table

Each row corresponds to a local authority. The columns list that authority’s 15 statistical neighbours, ordered from most similar to least similar. Neighbour 1 is the most similar local authority and Neighbour 15 is the least similar of the 15.

An illustrative extract is shown below.

Local authority	Neighbour 1	Neighbour 2	Neighbour 3	… through to Neighbour 15
Area A	Most similar area to Area A	Second most similar area to Area A	Third most similar area to Area A	… Fifteenth most similar area to Area A
Area B	Most similar area to Area B	Second most similar area to Area B	Third most similar area to Area B	… Fifteenth most similar area to Area B

Interpreting the neighbours

Neighbours are listed in order of similarity, but the differences between successive neighbours are not evenly spaced. Two points are worth keeping in mind:

• How similar any authority’s neighbours are to it varies from place to place. For some local authorities, all neighbours may be very similar matches; for others, even the most similar neighbours may be relatively different.
• The list is ordered by rank.

Comparisons are generally more meaningful against the full set of 15 neighbours than against any single one.

2. Methodology

This section sets out the methodology behind the Statistical Neighbours model developed by MHCLG for use with the Local Outcomes Framework. It explains the purpose of the model, the data and methods used, the checks that have been applied, and how local authorities and other users can give feedback.

Purpose and context

The Local Outcomes Framework sets out the outcomes central and local government are working together to achieve, measured through a consistent set of published metrics. When the Framework’s metrics are brought together in a digital tool on gov.uk, users will be able to compare a local authority’s outcomes and metrics with those of other authorities that share a similar context.

Local authorities operate in very different circumstances. Population size, levels of deprivation, the rural or urban character of an area, and the structure of the local economy all shape the context in which services are delivered. Comparing authorities against the national average, or against the authorities nearest to them geographically, can obscure these differences and make it harder to interpret outcomes fairly.

Statistical neighbours models address this by identifying, for each local authority, a set of other authorities that are most similar to it across a defined set of contextual indicators. These models are well established in local government. They support benchmarking, self-assessment, and sector-led improvement by giving a more balanced basis for comparison.

MHCLG’s Statistical Neighbours model has been built specifically to complement the Local Outcomes Framework. It uses contextual indicators chosen for their relevance to the Framework’s priority outcomes, and it is designed to work alongside the Framework rather than to replace other models used elsewhere in the public sector.

Determining statistical neighbours

Statistical neighbours are obtained by calculating how similar each local authority is to every other local authority across the range of contextual factors. Specifically: 

• Each local authority is assigned a profile consisting of its associated values across the chosen contextual indicators, creating a multi-dimensional snapshot of each authority. These contextual indicators are pre-processed, with outlier handling and standardisation applied.
• Similar local authorities are then identified by calculating the Euclidean distance between these profiles, which is a mathematical measure of how similar the authorities are across all the contextual indicators simultaneously. The neighbours are restricted at this stage to only obtain neighbours from the same tier.
• Authorities with the smallest Euclidean distances between them have the most similar profiles and are considered statistical neighbours. This is limited to the fifteen most similar authorities.

These steps are discussed in further detail below.

Building a profile for each local authority

Each local authority is assigned a profile made up of its values across the 21 contextual indicators. This creates a multi-dimensional snapshot of the authority.

Calculating similarity

Similarity between authorities is calculated using the Euclidean distance between their profiles. Euclidean distance is a standard mathematical measure of how close two points are in multi-dimensional space, where authorities whose profiles are close together across all 21 indicators have the smallest Euclidean distance and are identified as statistical neighbours.

Pre-processing the indicators

Before calculating distances, two pre-processing steps are applied to ensure no single indicator dominates the model:

• Outlier treatment (Winsorisation). Values below the 1st percentile are capped at the 1st percentile value, and values above the 99th percentile are capped at the 99th percentile value. This reduces the influence of a small number of extreme values without removing any local authority from the analysis.
• Standardisation. Each indicator is rescaled to have a mean of zero and a standard deviation of one, so that indicators measured on very different scales (for example, population size in the millions, and percentages between 0 and 100) enter the calculation on equal terms. All indicators are given the same weight in the calculation; their relative influence on the final neighbour matches emerges from the patterns in the data rather than from any weighting applied in advance.

Tier matching

Local authorities in England are structured into different tiers, with different responsibilities for service delivery. To make sure neighbours are comparable in both context and service delivery, the model applies matching restrictions by authority type:

• county councils are only matched with other county councils
• non-metropolitan district councils are only matched with other non-metropolitan district councils
• single-tier authorities (unitary authorities, London boroughs and metropolitan boroughs) are only matched with other single-tier authorities

Aggregating data for counties

Four of the 21 indicators are not published at county level, but are available at the lower tier. In these cases, county-level values are constructed by aggregating district-level data using a weighted average:

Value_County  =  Σ ( ω_i × Value_{District i} )  ÷  Σ ω_i

where the sum runs across all districts i within the county, and ωᵢ is a weighting factor appropriate to the indicator. The weighting factor used for each indicator follows the guidance published with the underlying dataset, so that the aggregated county value is consistent with how the measure is defined at other geographies.

The indicators aggregated in this way, and the weighting factor used, are:

Indicator	Weighting factor	Source of underlying data
Gross value added per hour worked	District productive hours worked	ONS
Percentage of population in rural areas	District population (2021)	ONS
Percentage of children in relative poverty	District population of children	DWP
Taxbase per household	Number of chargeable dwellings per district	MHCLG

Indicator selection and testing

The starting point for indicator selection was the set of variables used in the ONS nearest neighbours model. From that starting point, the list was refined through a combination of technical testing and engagement with policy leads.

Three technical tests were used:

• Correlation analysis. Pairwise correlations between indicators were examined, and indicators with very high correlations were considered for removal to avoid over-weighting any underlying factor. In a small number of cases, correlated indicators were retained where each was judged to capture a distinct aspect of context relevant to the Local Outcome Framework’s outcome areas.
• Variance analysis. Indicators with very low variation across local authorities were considered for removal, on the basis that they add little to the identification of differences between authorities. The indicators with the highest variance across England are deprivation, the percentage of children in relative poverty, and the proportion of the population identifying as white.
• Principal component analysis (PCA). PCA was used to identify which indicators contribute most to the overall spread of the data. Indicators that contributed little to the main components were considered for removal.

Alongside this technical testing, the indicator list was iterated with policy and analytical experts for each of the priority outcomes in the Framework. The aim was to balance two concerns by keeping the model brief enough to be interpretable, while making sure it reflects the contextual factors most relevant to the outcomes it is designed to support. Final decisions on inclusion and exclusion were made with input from technical experts both within and outside MHCLG.

Specific authorities

Because the Isles of Scilly and the City of London have unusual contexts, their statistical neighbours are less close matches than for other authorities. We will continue to review whether these authorities should be shown in the same way in future versions of the model.

Indicators used in the model

The model uses 21 indicators. The full list is set out below, with the description used in the model and the source of the underlying data.

Indicator ID	Indicator	Description	Source
1	Deprivation	English Indices of Deprivation (2025)	MHCLG
2	Proportion of population identifying as white	Proportion of population identifying their ethnic group within the high-level “White” category	ONS
3	Population size	Mid-year population estimates	ONS
4	Percentage of population in rural areas	Proportion of population living in rural output areas	ONS
5	Employment rate	Percentage of people aged 16 to 64 in employment	ONS
6	Education rate	People aged 19 and above participating in further education (including apprenticeships), per 100,000 people	DfE
7	Qualification rate	Percentage of population with Level 3 or above qualifications	ONS
8	Digital connectivity	Percentage of premises with access to gigabit capable broadband coverage	Ofcom
9	Gross value added per hour worked	Estimated value of goods and services produced per hour worked (£)	ONS
10	Business births	New enterprises as a share of active enterprises	ONS
11	High growth businesses	Percentage of enterprises classed as “high growth” out of active enterprises with more than ten employees	ONS
12	Ratio of non-working age to working age population	Ratio of population at non-working age (15 and under or 64 and over) to population at working age (16-63)	ONS
13	Income	Gross median weekly pay (£)	ONS
14	Proportion of population renting privately	Percentage of households classified as “private rented”	ONS
15	Percentage of children in relative poverty	Percentage aged under 16 living in relatively low-income families	DWP
16	Percentage of pupils with SEN or EHCP support	Percentage of pupils in receipt of Special Educational Needs (SEN) support or with an Education, Health and Care Plan (EHCP)	DfE
17	Housing repossession rate	Private landlord possession claims per 100,000 households owned by a private landlord	MoJ
18	Proportion of population without a disability	Percentage of population not disabled under the Equality Act 2010 (age-standardised proportion)	ONS
19	Housing affordability	Ratio of median house price to median gross annual workplace-based earnings	ONS
20	Taxbase per household	Average council tax per dwelling (£)	MHCLG
21	Area cost adjustment	Measure of local variation in the cost of delivering public services	MHCLG

Strengths and limitations

Strengths

• Tailored to the Local Outcomes Framework. The model has been built specifically to support the Local Outcomes Framework, with an indicator set informed by the factors most relevant to the Framework’s priority outcomes.
• Transparent and reproducible. The methodology, inputs and indicator definitions are published in full. The model is built on the ONS nearest neighbours method and code, which is itself publicly documented.
• Publicly available data only. The model uses no unpublished or commercially restricted data. Users can trace any result back to its underlying source.
• Structured engagement with policy leads. The indicator set has been developed with input from policy and analytical experts for each priority outcome, combining technical testing with subject-matter expertise.
• Comparability by authority type. Tier matching ensures neighbours are comparable in both context and service responsibilities.

Limitations

• Context, not judgement. Statistical neighbours are a tool for comparison, not a ranking. A similar context does not necessarily mean similar outcomes should be expected. Differences between neighbours may reflect genuine variation in delivery, local priorities, or factors not captured in the indicator set.
• Reliance on publicly available data. The model depends on the timing and coverage of published data. Indicators are drawn from data releases spanning a range of reference periods, and updates will lag behind real-world change. Where data is not published at county level, values are constructed through weighted aggregation from district data.
• Indicator coverage. No fixed set of indicators can capture every aspect of local context. Some factors that matter to outcomes, for example, aspects of health, migration, or community characteristics, are only partially represented, or are represented through proxies.
• Correlation between indicators. Some indicators in the model remain correlated with one another, because each was judged to capture a distinct and relevant aspect of context. Users should interpret neighbour relationships in the round, rather than attributing them to any single indicator.
• Sensitivity to Local Government Reorganisation (LGR). The model is based on the current structure of local authorities in England. Planned changes to local government structures will require the model to be updated as new authorities come into existence and others are replaced.
• First iteration. This is the first version of the model. It will evolve in response to feedback, further peer review, and changes to the Framework itself.
• Local Area Benchmarking: This model is designed for broad comparison and is intended to complement, not replace, nearest neighbour’s models used for specific topic areas, such as the Department for Education’s Children’s social care dashboard.

Data standards and governance

The model has been developed in line with established data and analytical standards, including the principles of the Code of Practice for Statistics (Trustworthiness, Quality and Value).

Data standards

The indicators in the model are:

• drawn from published, publicly available sources where no unpublished, restricted, or commercially licensed data is used
• produced by established official or accredited sources, including the Office for National Statistics (ONS), MHCLG, the Department for Education (DfE), the Department for Work and Pensions (DWP), the Ministry of Justice (MoJ) and Ofcom
• required to have valid data for approximately 90% of local authorities in England, so that the model can be applied consistently across the country
• largely outside the direct control of a local authority, so that comparisons focus on the context within which outcomes are delivered, not on performance itself

Governance and quality assurance

The model has been developed in close collaboration with the ONS Local Statistics and Analysis Division. It builds on the ONS nearest neighbours method and code, with ONS providing technical input and support throughout development.

Within MHCLG, the model has been subject to internal peer review by analysts outside the project team. The choice of indicators has been developed through structured engagement with policy leads responsible for each of the priority outcomes in the Local Outcomes Framework, so that the indicators reflect the factors most relevant to those outcomes. The methodology behind the model has also undergone external peer review from analysts within several English local authorities, and the methodology will continue to evolve in response to feedback from the local government sector and statistical peers.

Future updates

The model will be updated on an annual basis to ensure neighbours are determined from the most recent available data. Additional updates, such as those due to local government restructuring or boundary changes may be applied on an ad-hoc basis.

3. Commitment to Trustworthiness, Quality and Value

The Code of Practice for Statistics set standards for organisations in producing and publishing official statistics and ensure that statistics serve the public good. Whilst the Statistical Neighbours model is not designated as official statistics, MHCLG are committed to work in a way that builds confidence in data, statistics and analysis, and facilitate transparency and accountability. As this statement demonstrates, we have aimed to comply with the Code and are on the Office for Statistics Regulation (OSR)-led Register of TQV Voluntary Application.

Trustworthiness

1. Show integrity

The data has been selected from existing government statistics, tabulated and then checked by MHCLG officials, all who comply with the Civil Service Code and the Seven Principles of Public Life.

2. Lead responsibly

The model has been produced and checked by analysts who work under the supervision of the Department’s Head of Profession for Statistics. Decisions on which contextual indicators and methods feed into the model have been made with analyst and policy teams across MHCLG. Feedback from ONS and engagement with the local government sector has been incorporated. The methods have been tested with users and will continue to develop through continued user research and engagement. 

3. Be transparent

The underlying (input) data is all already public domain information, and details of the methodology have been made available. Data used in the model (and model results) will be updated annually. 

4. Manage data responsibly

Data is collated from existing publications with all data processing carried out to department standards ensuring accuracy and review before release. No sensitive data is included.

Quality

5. Prioritise quality

Data is compiled from existing statistical publications, ensuring that only high-quality data sources are used. Each indicator is reviewed for suitability, including its coverage and release schedule across local authorities to confirm it is fit for use in the model. Links to original sources are provided with methodology statements. As the model develops, we may make changes to the methodology in response to user feedback, new data releases, and sector engagement. Any such changes will subject to testing and be documented and communicated to users.

6. Be rigorous 

Data processing pipelines have been developed to automate processing and efficiently collate and store data. 

The model builds on the established Statistical Neighbours approach developed by the ONS subnational data team, with the methodology adapted such that the indicators align with the contexts relevant to the Local Outcomes Framework. We have engaged with the ONS team through peer review to ensure our approach is methodologically sound while remaining fit for Local Outcomes Framework purposes.

All aspects of data production are quality-assured prior to publication. This includes external quality assurance of code, underlying data and final outputs visible in the model, comparing against original sources. Additionally the model has undergone peer review from technical colleagues within ONS.

7. Be open about quality

Links to methods and data sources are indicated in the accompanying methodology. We encourage users to consider quality implications before making local comparisons, and direct users to the strengths and limitations section in the methodology.

Value

8. Be relevant

The Statistical Neighbours model has been developed to support the Framework by providing local authorities and central government with a consistent, transparent way to identify comparable areas. This enables more meaningful benchmarking of outcomes, supports local decision-making, and helps users interpret performance data in context while filling a gap for a standardised comparison tool tailored to the Framework.

Users from central and local government have contributed to the development of the model, through quality assurance and invited feedback. We encourage all users with feedback to contact us to report any potential inaccuracies or make suggestions. Email localoutcomesframework@communities.gov.uk to get in touch.

9. Be clear

The model has been designed to present all comparisons in a standardised way to make it as simple as possible to identify and interpret statistical neighbours. Users are signposted to original sources and are provided with context and methodology statements to ensure information is presented clearly at various levels of technical expertise, and avoid misleading conclusions. The model design ensures that only meaningful comparisons between equivalent levels of local government are encouraged.

10. Be accessible

The model and underlying data are available to all, with outputs presented in clear tables to be accessible to screen readers and different users. We have conducted an accessibility audit of the site and provide an accessibility statement. We are always trying to improve the accessibility of our data and welcome any suggestions.

4. How to give feedback

Feedback on the model is being invited through an open call running until 31 July 2026, and MHCLG is grateful to those who have already contributed. This publication is intended to make it easier to provide informed feedback by setting out the methodology in detail. Feedback will continue to be welcomed up to, and indeed beyond, 31 July 2026 as the model develops.

Where possible, please include:

• which aspects of the model your feedback relates to (for example, indicator choice, aggregation method, presentation in the digital tool)
• the perspective you are writing from (for example, local authority, sector body, researcher, member of the public)
• any suggested changes or additional evidence to be considered

Email localoutcomesframework@communities.gov.uk to get in touch.