Foreword

The Ministry of Housing, Communities and Local Government (MHCLG) is committed to following an evidence-informed approach to reducing homelessness and rough sleeping. This includes building up an evidence base to help us better understand the people who become homeless as well as what interventions work to reduce homelessness.

This study presents a statistical analysis of repeat homelessness, a growing area of policy interest for MHCLG. The research was conducted by the Homelessness Pilot team as part of the Better Outcomes through Linked Data (BOLD) programme led by Ministry of Justice. The pilot aims to develop MHCLG’s evidence base by linking its data on homelessness and rough sleeping to other datasets. In this study, the Homelessness Case Level Information Collection (H-CLIC) has been linked to personal information gathered from local authorities to enable data linking research projects.

Linked datasets have the potential to enhance existing data, providing a richer source of information and more context into the ongoing issues faced by vulnerable people aided by government departments. This evidence base is essential to help MHCLG make data driven decisions to improve the lives of those affected by homelessness and rough sleeping.

The linked dataset contained a total of over 68,000 applications across 52 local authorities in England from 2018 to 2023. Through this linked data, MHCLG analysts were able to identify substantially more households that experienced repeat homelessness than the original data contained. The analysis details the characteristics that were associated with repeat homelessness households, such as their demographics, additional support needs and time spent in temporary accommodation.

I would like to thank the BOLD Homelessness Pilot team for conducting the research, analysts in the H-CLIC team for overseeing the collection of homelessness data and personal information, colleagues at the ONS for their hard work linking the datasets together, and the policy and analytical colleagues at MHCLG for reviewing the outputs. I would also like to thank the participating local authorities whose contributions have allowed us to do this work.

Finally, I would like to thank to those leading the BOLD programme. This work has given MHCLG the opportunity to collaborate on exciting areas of research and improve our evidence base, for which I am especially grateful.

Stephen Aldridge

Director for Analysis and Data & Chief Economist

Ministry of Housing, Communities and Local Government

Executive summary

This study shows the findings from the analysis of repeat homelessness conducted by the Homelessness Pilot Team as part of the Better Outcomes through Linked Data (BOLD) programme.

The Homelessness Pilot Team, based in the Ministry of Housing, Communities and Local Government (MHCLG), oversaw the data linkage and conducted the subsequent analysis relating to repeated experiences of homelessness.

The definitions of key terminology can be found at the end of the Executive Summary.

Background

The 2017 Homelessness Reduction Act (HRA) reformed England’s homelessness legislation by placing duties on local authorities (LAs) to intervene at earlier stages to prevent homelessness in their areas. MHCLG compiles statutory homelessness statistics from Homelessness Case Level Information Collection (H-CLIC) data provided by local authorities in England to report on activities undertaken under statutory homelessness duties.

Challenges arise when using homelessness data records to monitor households who have become homeless again over time. These events might not be identified from H-CLIC, as the household may seek support from different geographic areas and may not report previous applications.

The BOLD project aims to fill this gap and help to develop our understanding of people who are repeatedly homeless, including those who seek support across local authority boundaries. Identifying and understanding repeat homelessness will help MHCLG in its commitment to reduce homelessness and can provide other benefits such as:

• providing valuable information about the cycle of homelessness and its impact on a range of outcomes for homeless households
• assessing the effectiveness of the HRA
• improving services provided to those at risk of becoming homeless

The following research questions were proposed to help better understand the trends and characteristics associated with repeat experiences of homelessness:

1. What proportion of applications feature repeat homelessness?
2. How do the demographics of people experiencing repeat homelessness compare to those homeless for the first time?
3. Are people with self-identified support needs and groups with common experiences such as recently leaving prison more likely to experience repeat homelessness?
4. What is the history of housing and support activities for those experiencing repeat homelessness?
5. Is there a difference in use of temporary accommodation for those experiencing repeat homelessness compared to those presenting as homeless for the first time?

There are several ways to identify households who have experienced repeat homelessness in the HCLIC dataset, the details of which are explained below:

History of Repeat Homelessness Support Need: Local authorities record if any member of the household had been homeless prior to their current experience of homelessness, which is flagged as a “history of repeat homelessness” support need. However, this does not specify the previous number of homeless applications made, when they were made, or which household members had been affected or the length of time they were homeless.

Previous Case Reference Numbers: An applicant’s previous case reference number may be recorded if they have previously engaged with homelessness services in that specific local authority.

Person IDs: The datasets may be searched for the ID numbers of people who have appeared (either as main applicants or dependents) in more than one homelessness application. There may be reasons why this information is not accurately recorded for all applications, for example people being allocated new person ID numbers by their local authority on subsequent applications.

Since 2020, MHCLG has collected personal identifier data as part of the Homelessness Data England (HDE) project. As at January 2024, 52 local authorities had submitted personal identifiers. Linking this information to homelessness case data is intended to better identify individuals who appear in H-CLIC.

Due to the limitations of the dataset used in this study, the findings are neither representative of all local authorities in England or intended to be extrapolated to non-participating authorities.

What can this report be used for?

It can be used for:

• A case study of the potential usefulness of linked data
• Identifying features associated with households in repeat homelessness

It cannot be used for:

• Estimating the prevalence of repeat homelessness for local authorities in England
• Finding causes of repeat homelessness

Summary of key findings

A summary of the analytical findings by theme are provided below. Further details of the results are provided in Section 3 of this report. The definitions of key terminology can be found at the end of the Executive Summary.

Prevalence of repeat homelessness

Linking homelessness case data to an ID Spine of personal information led to more instances of repeat homelessness to be identified. The linked dataset contained applications from 68,328 households, almost entirely made between 2018 Q2 and 2023 Q1. Of these, 16% (10,852 applications) were identified as involving a household member who had been previously homeless at least once prior to the application, either through: multiple appearances in the dataset, a history of repeat homelessness support need or a previous case reference number.

Out of the 10,852 repeat homelessness applications, 7,340 were identified as such solely due the data linking process. Although the proportion of applications became increasingly identifiable as repeat homelessness over time, this appears to be an artefact of data linking as more data was available for linking in later quarters following further developments of H-CLIC since 2018, along with the increasing number of local authorities participating in HDE since 2020.

Demographics

Lead applicants that were identified as experiencing repeat homelessness were predominantly:

• in one adult households with no children (74% of repeatedly homeless lead applicants)
• male (58%)
• out of work (74%), meaning those not in full or part time employment, study, or training

Each demographic variable was studied separately, as opposed to co-occurrences of these, e.g. female and unemployed, male with children, etc.

Support needs

Support needs are areas of additional needs that mean the household requires support to acquire and sustain accommodation, as recorded in their homelessness application. Of the six support needs investigated (physical ill health and disability, alcohol dependency, domestic abuse, drug dependency, mental health problems and offending history), all were associated with households identified as experiencing repeat homelessness. The highest prevalence of repeat homelessness (identified through all means described above) was observed in households that contained:

• drug dependency needs (repeat homelessness identified in 51% of these households)
• a history of offending (50%)
• alcohol dependency needs (46%)

It is unknown to what extent these observations are driven by the presence of co-occurring support needs within the households.

Support from local authorities

Local authorities provide support through activities to prevent or relieve households from homelessness. The split of duties owed (following the initial assessment) in the dataset showed that 67% repeat homeless households were found to be owed a relief duty. For first time homelessness, the split of prevention and relief duties was almost equal with each other.

First time homeless households appeared to be more successful in securing long term accommodation. For example, when looking at relief activity outcomes, 18% of first time households owed a relief duty secured accommodation for 12 months or more, whereas for repeat homelessness, this was 11%. Interviewing these households would provide qualitative data to help understand relief outcomes for first time and repeat homeless households.

Temporary accommodation

Single households were placed in temporary accommodation more often if they had experienced repeat homelessness (40% of repeat homeless single households)^{[footnote 1]}. Overall, repeat homeless households tended to have shorter stays (median per stay of 84 days) in temporary accommodation than those homeless for the first time (131 days).

Implications and recommendations

Repeat homelessness was found to be increasingly identifiable after linking personal information to attribute data of homelessness applications. Further investigation is required to identify the underlying causalities. This would also benefit from a more complete dataset covering a greater number of local authorities over longer time periods.

Definitions

Statutory Homelessness: Households owed a homelessness prevention, relief or main duty by a local authority. This is often used to describe households that are already homeless or threatened with homelessness within the next 56 days. Further details are available in the official statistics technical note.

Application: A record of an approach from the household seeking help from the local authority and the activities undertaken by the local authority as part of their statutory duties. The application is closed when the local authority duty ends. Each application has a main applicant and records other members of the household.

Where appropriate, the term “application” is used interchangeably with the term “household” throughout this report.

Household: A household can be made up of single people, multiple adults, or adults with children. If that household, or a member of that household, later experiences homelessness again, and applies for help from a local authority, a new application will be opened whether it is the same local authority or a different one. In the instance of a new application, the applicant may choose not to disclose their history of homelessness.

Prevention Duty: Local authorities may deliver their prevention duty through any activities aimed at preventing a household threatened with homelessness within 56 days from becoming homeless. This would involve activities to enable an applicant to remain in their current home or finding them alternative accommodation. The duty lasts for up to 56 days but may be extended if the local authority is continuing with efforts to prevent homelessness.

Relief Duty: The relief duty is owed to households that are already homeless on approaching a local authority and require help to secure settled accommodation. The duty lasts 56 days and can only be extended by a local authority if the household is not owed the main homelessness duty.

Main Duty: The ‘main’ homelessness duty describes the duty a local authority has towards an applicant who is unintentionally homeless, eligible for assistance and has priority need. Applicants who have priority need include households with dependent children or a pregnant woman; those who are homeless due to fire, flood or other emergency; those who are particularly vulnerable due to ill health, disability or old age; those having been in custody or care; or those who have become homeless due to violence or the threat of violence. These households are only owed a main duty if they did not secure accommodation in the prevention or relief stage, and so is not owed to those ‘threatened with homelessness’. In addition, a minimum of 56 days of assistance must have elapsed from a household approaching the local authority to being owed a main duty.

Support Needs: Areas of additional needs that mean the household requires support to acquire and sustain accommodation, giving an indication of the additional services local authorities need to provide. The local authority may report co-occurring support needs within the household. Support needs recorded in homelessness datasets are self-reported by the applicant at the time they contacted the local authority and might not be verified.

Repeat Homelessness: In this study, repeat homelessness will be used to describe distinct periods of homelessness. It was identified when either of the following applied to any household member:

• the person appears in a previous application recorded in the dataset, as identified using personal identifiers
• their household has a support need for having history of repeat homelessness recorded in H-CLIC
• they had a record of a previous case reference number relating to a previous homeless application with the local authority

If no household members meet any of the criteria, they are assumed to be first time homeless.

Further details on defining the repeat and first time homelessness applications are outlined in the methodology section.

1. Introduction

1.1 Better Outcomes through Linked Data (BOLD)

The BOLD programme is led by Ministry of Justice in partnership with the Ministry of Housing, Communities and Local Government (MHCLG), Department of Health and Social Care, Welsh Government and Public Health Wales. It was created to demonstrate how people with complex needs can be better supported by linking and improving the government data held on them in a safe and secure way. BOLD has initially focused on four pilot areas: reducing homelessness, substance misuse, re-offending and supporting victims of crime.

This study is the first statistical analysis of a dataset created by linking homelessness attribute data from H-CLIC with homelessness personal identifiers from local authorities. It is an ad-hoc MHCLG publication and serves as a proof of concept within the wider BOLD programme.

1.2 Statutory homelessness

The Homelessness Reduction Act 2017 reformed England’s homelessness legislation by placing duties on local authorities to intervene at earlier stages to prevent homelessness in their areas. This includes providing support if a household is threatened with homelessness within 56 days, known as a prevention duty, and to provide support to households that are homeless, known as a relief duty.

The legislative changes have been reflected in an enhanced Homelessness Case Level Information Collection (H-CLIC) data specification for local authorities to follow since April 2018.

MHCLG produces statistics from data returns by local authorities to monitor levels of statutory homelessness across England, and to report on the activities carried out by local authorities to meet their statutory homelessness duties.

The data returns indicate whether a household has reported a support need of a history of repeat homelessness, based on the information and understanding of the local authority and the assessor. However, the information collected does not specify the previous number of applications made, which household members had been affected or verify this against previous occurrences in H-CLIC records. The reference numbers of the applicants’ previous homelessness case may also be recorded in H-CLIC, but this is optional, and usage may vary between local authorities. These sets of information may still be useful to the authority in supporting those households but may not always be a complete reflection of any repeat homeless experiences.

Additionally, when a person makes a homelessness application that is not their first, the history of repeat homelessness support need is not always recorded. For instance, they may choose not to disclose their history or may not be found in local authority records. If they present in a different local authority, their approach will be recorded as a new application.

1.3 Aims and objectives

The repeat homelessness project aims to demonstrate what factors and characteristics are associated with individuals at risk of repeatedly experiencing homelessness. The intention of this analysis was to:

• understand the trends and characteristics associated with repeat experiences of homelessness
• in alignment with the wider BOLD objectives, demonstrate the opportunities for analysis when linking datasets, namely:
  • Homelessness Case Level Information Collection (H-CLIC)
  • An ID spine of local authority IDs and personal identifiers compiled from the Homelessness Data England (HDE) Project

To meet these objectives, the following research questions have been proposed:

1. What proportion of applications feature repeat homelessness?
2. How do the demographics of people experiencing repeat homelessness compare to those homeless for the first time?
3. Are people with self-identified support needs and groups with common experiences such as recently leaving prison more likely to experience repeat homelessness?
4. What is the history of housing and support activities for those experiencing repeat homelessness?
5. Is there a difference in use of temporary accommodation for those experiencing repeat homelessness compared to those presenting as homeless for the first time?

2. Datasets and methodology

2.1 Homelessness Case Level Information Collection

The Homelessness Case Level Information Collection (H-CLIC) is a mandatory data collection for local authorities in England for the monitoring of the HRA 2017. MHCLG uses H-CLIC to collect detailed case-level information on households each quarter as they progress through each of the prevention, relief and main duty decision stages of the statutory homelessness system. The data has been collected since the second quarter of 2018.

The official statistics are published on a quarterly basis with an additional annual report and have acquired Accredited Official Statistics status.

2.2 Homelessness Data England project

The Homelessness Data England (HDE) project was set up to collect personal identifiers from local authorities to allow for linking of H-CLIC data across local authorities and linking to other administrative datasets. A sub-set of local authorities first submitted personal data for the HDE project in 2020 Q3. Since then, a growing number have completed quarterly submissions. As of January 2024, 52 local authorities have participated in this data collection.

The submission window for personal data is at the end of the following quarter, i.e. Quarter 1 data (April to June) is collected in September.

To preserve applicants’ privacy and adhere to data protection laws, the identifiers are collected separately and processed by the Office for National Statistics (ONS). They have created a linked ID spine and unique ID codes to identify where multiple local authority person or case IDs are related to the same individual in the H-CLIC data and enable linking across time (within H-CLIC itself) and with other government departmental datasets.

2.3 Methodology

2.3.1 Data linking

The first stage of the data linking was the creation of the ID Spine (dataset S1 in Figure 2.1 below). This is a lookup table consisting of unique identifiers codes and H-CLIC identifiers. ONS created this by:

• collating H-CLIC personal information (such as names, date of births, etc., represented by H2 in Figure 2.1), and IDs for each person and case
• using this data to identify previous cases for the same person, by linking using personal identifiers e.g. name, date of birth, etc.
• allocating a new unique ID number to each distinct person

Figure 2.1: Overview of the data linking process. Local authorities submit attribute data for use in official statistics, and personal information for the HDE project

After MHCLG received the ID Spine, it was joined to the H-CLIC attribute data (H3, which contains information such as support needs, accommodation provided, etc.) by matching the case reference and H-CLIC person ID numbers. Not all of this data was able to be linked to the ID Spine. Further details of the data linking results are provided in Annex B.

The resulting linked dataset (L1) does not contain H-CLIC personal information (H2), meaning that the data handled is de-identified. This ensures that it can be used safely and responsibly by MHCLG for the BOLD programme.

2.3.2 Identifying repeat homelessness

Records were removed from the dataset where the initial assessment stated that the household had either withdrawn or found to be not threatened with homelessness within 56 days. Next, each household in the dataset was classified as experiencing either repeat or first time homelessness for each of their applications, which was used to describe distinct periods of homelessness. An application was deemed as a repeat experience when any of the following applied to any household member:

• the person featured in an older case in the dataset, as identified through the ID Spine information
• their household has a support need for having a history of repeat homelessness
• the person has a previous case reference number relating to a previous homeless application with the local authority

If all household members did not meet either of these criteria, the household was assumed to be experiencing homelessness for the first time. Examples of how each household was identified as either Repeat or First Time are provided in Annex C.

2.3.3 Data analysis

After identifying the repeat status of each household, the dataset was subsequently analysed to identify the characteristics of those experiencing repeat homelessness. For each characteristic, the proportion of those experiencing repeat and first time homelessness was calculated.

2.4 Limitations of this study

This is the first statistical analysis of a dataset created by linking homelessness attribute data from H-CLIC with homelessness personal identifiers from local authorities. It is a proof of concept and whilst the findings are interesting, the analysis has several limitations:

2.4.1 Time period

The analysis primarily covers applications made from 2018 Q2 to 2023 Q1, which accounts for over 99.5% of the data. Some applications from 2004 Q2 to 2018 Q1 and 2023 Q2 to 2023 Q4 are also included, as detailed in Annex E.

As the analysis is limited to data largely from 2018 onwards, there is a reduced likelihood of identifying all repeat homelessness cases.

2.4.2 Self-selection

Fifty-two local authorities, self-selected to participate, submitted personal data. A list of the participating authorities can be found in Annex A. This self-selection means that the trends observed may not be representative of non-participating authorities as only a subset of local authorities participated. The findings are not intended to be extrapolated to non-participating authorities.

The linked dataset does not capture all repeat homelessness applications made in the participating local authorities. This limitation arises from the self-selection of local authorities, the restriction of data to H-CLIC submissions (with the HDE project starting in 2018 and 2020 respectively), and the challenges in matching individuals by their ID Spine in H-CLIC.

2.4.3 Double counting

Households identified as either first-time or repeat homeless may include the same individuals, leading to double counting. This double counting allows for observations of changes in characteristics, for example if a person spent time in prison between their first and second homelessness applications

2.4.4 Comparison with Official Statistics

The analysis in Section 3 is not comparable to observations of repeat homelessness found in the official statistics, which only count households flagged with the history of repeat homelessness support need.

2.4.5 Analytical constraints

The analysis is descriptive in nature, rather than a causal inference study. In addition, the methodology does not include statistical significance testing of variables. It is assumed that the dataset captures all the journeys of households entering, leaving, and re-entering the homelessness support system have been captured. However, it was not possible to quantify the uncertainty of this, or the impact on repeat homelessness prevalence rates.

Support needs appeared to be strongly linked to repeat homelessness. However, their effects may have been amplified by co-occurring support needs. Further analysis is required to study each support need in isolation.

3. Results

3.1 What the results cover

The results in this section cover two primary areas. Firstly, in Sections 3.2 and 3.3, a summary of the data linking results and an overview of repeat homelessness identified from this. Further detailed observations are then made in Sections 3.4 to 3.7 by comparing characteristics between repeat and first time homeless households.

3.2 Linking results

The dataset returned 68,328 applications, raised by 63,417 distinct people identified as lead applicants. There were 121,355 distinct household members in total, identified as either lead applicants or other members such as partners and children.

3.3 What proportion of applications feature repeat homelessness?

Out of the 68,328 applications, 10,852 (16%) contained people experiencing repeat homelessness, identified by the methodology outlined in Section 2.3.2.

Table 3.1: Around 1 in 6 applications were identified as repeat homeless. The linked dataset allowed over a third of these to be identified as repeat homelessness

	Repeat (with linking)	First Time (with linking)
Repeat (without linking)	7,340	464
First Time (without linking)	3,512	57,012
Total applications (n = 68,328)	10,852	57,476

^{Matrix of the breakdown of Repeat and First Time applications identified from the linked dataset, compared to the same applications without the use of linking to the ID Spine (52 participating local authorities, Q1 2018 - Q3 2023).}

There were 3,512 repeat applications that would have otherwise been labelled as first time, had the process been applied to them without the use of the ONS Spine ID (Table 3.1). Further details explaining how repeat homelessness would have been identified without the use of the Spine ID is explained in Annex D.

Those 3,512 repeat applications are labelled as “identified by linking” in Figure 3.1 below. The other 7,340 that would have been identified as repeats regardless of linking are also shown separately (in dark blue).

The dataset contained 464 repeat homelessness applications that would have been categorised as first time homelessness if data linking was not used, potentially illustrating the limitations of linking by deterministic matching. Examples may include differences in spelling or special characters within applicant’s names between their first and repeat applications. The analysis in this study is based on 10,852 as the number of repeat applications, which assumes a relatively small level of uncertainty from data linking process.

Figure 3.1: The number of applications identified as first time homelessness and repeat homelessness rose over time

^{Number of applications of homelessness available in the HDE project per quarter (52 participating local authorities, 2018 Q2 – 2023 Q1).}

The upward trend of applications from the dataset, particularly from 2020 Q1 onwards, is mainly attributed to:

• the transition to H-CLIC in 2018, and subsequent development of existing and new variables, resulting in better quality data returns
• more local authorities submitting personal information to HDE over time since 2020
• better data matching in quarters with higher numbers of applications and personal information

Therefore, low volumes of repeat homeless households in the early quarters are to be expected. Realistically, almost all of these could only be identified as such by the support need or have a previous case reference. For example, an application made in 2018 that had a previous case reference for an application made by the same person in 2015.

There was a peak in 2021 Q3 at 9,235 applications, dropping to 7,430 in 2023 Q1. The number of repeat homeless applications also increased over time, peaking at 1,688 applications in 2022 Q3, accounting for 18% of the applications found from that quarter.

In addition, 323 lead applicants were identified as people who made homelessness applications in more than one local authority. These applications could only be identified as repeat applications using the linked dataset, as each local authority assigns person reference numbers individually.

3.4 How do the demographics of people experiencing repeat homelessness compare to those homeless for the first time?

3.4.1 Sex of main applicant

Between 2018 Q2 and 2023 Q1, 58% of repeat homeless households had male lead applicants, and 41% of repeat homeless households had female lead applicants. In contrast, for first time homelessness, male lead applicants accounted for 43% of households, and 57% of lead applicants were female^{[footnote 2]}.

3.4.2 Household type

Households were most often one adult households. This category accounted for 74% of those identified as repeatedly homeless and 54% the first time homeless population. The second most frequent household type were lone parent households (one adult with dependent children), at 15% of repeat homeless households and 27% of first time homeless households.

3.4.3 Employment status of main applicant

The employment status types most common among repeat homeless lead applicants can be attributed with a lack of security, i.e. being unemployed, not working due to illness, etc. These employment status types accounted for 74% of lead applicants in repeat homeless applications, compared to 61% for first time homelessness, as shown in Figure 3.2.

Figure 3.2: The percentage of repeat homelessness lead applicants not in employment, study, or training was 74%, compared with 61% of first time homelessness applicants

^{Bar chart showing the proportions of lead applicant employment status, for repeat and first time homelessness applications. The chart is based on 52 participating local authorities across the time period 2018 Q2 – 2023 Q1.}

3.5 Are people with self-identified support needs and groups with common experiences such as recently leaving prison more likely to experience repeat homelessness?

This question explores if repeat homelessness is prominent within certain groups of people with common characteristics or experiences. The groups were firstly identified based on their support needs. Then, the analysis was repeated for specific groups: LGBTQ+ individuals, people who have left care, hospital, or prison, and veterans.

3.5.1 Support needs

To choose from an extensive list of support needs, a random forest model was used to shortlist those that presented as important features that associated with repeat homelessness^{[footnote 3]}. Six support needs (shown in Figure 3.3) were selected: physical ill health and disability, offending history, history of mental health problems, drug dependency needs, risk or experience of domestic abuse and alcohol dependency needs.

Out of all the households, 46% had at least one of the support needs mentioned above. Households can also have co-occurring support needs.

Figure 3.3: For every one of the support needs investigated, repeat homelessness was more prevalent among households containing that support need, compared to those that did not have the support need

^{Bar chart displaying prevalence of the 10,852 repeat homeless applications by specified support needs, split by the presence of that need within the household. The chart is based on 52 participating local authorities across the time period 2018 Q2 – 2023 Q1.}

For each of these support needs, repeat homelessness was more prevalent among households that had that support need, than those that did not.

The greatest difference in repeat homelessness prevalence was seen in the offending history group, where it was identified in 50% of those households. For those households without an offending history, repeat homelessness affected 12%. The next biggest difference was seen when comparing the drug dependency needs group (51%) to those not in the drug dependency needs group (14%). Similarly, nearly half of the households with alcohol dependency needs were identified as experiencing repeat homelessness (46%).

Out of those with at least one registered support need, 45% (14,378 households) had two or more support needs. Co-occurring support needs indicate complex circumstances and understanding their impact on repeat homelessness (specific combinations) requires further problem structuring and hypothesis testing methods.

3.5.2 Groups with common experiences

The following groups were selected for analysis:

• LGBTQ+
• care leavers
• hospital leavers
• prison leavers
• veterans

These groups (shown in Figure 3.4) were approximated using H-CLIC variables including self-reported support needs, referral agency, sexual orientation, lead applicant sex and accommodation type. Out of 68,328 applications, 13% contained household members belonging to at least one of the groups mentioned above.

Figure 3.4: For each group examined, repeat homelessness was more prevalent among these groups compared to applicants not belonging to these groups

^{Bar chart displaying prevalence of the 10,852 repeat homeless applications by specified groups, split by whether or not any household members belong to that group. The chart is based on 52 participating local authorities across the time period 2018 Q2 – 2023 Q1.}

The greatest difference in repeat homelessness prevalence was seen in the prison leavers group. Of the 2,443 households that included people identified as a prison leaver, 37% experienced repeat homelessness. Of the 65,885 households not including a prison leaver, repeat homelessness affected 15%.

Whilst it is possible for households to contain more than one of these groups, this only occurred in 534 (less than 1%) households.

3.6 What is the history of housing and support activities for those experiencing repeat homelessness?

This question compares types of accommodations and duties between repeat and first time homelessness.

3.6.1 Housing at the time of application

At the time the application was made, 25% of the total repeat homeless applications had a referral, compared to 15% for first time homelessness. Referrals most commonly came from the National Probation Service^{[footnote 4]}. Out of the 2,664 repeat homeless households that had a referral, 30% came from the National Probation Service. Whereas the corresponding figure for first time homelessness was 15%.

The most frequent reason for the loss of home (shown in Figure 3.5 below) was family no longer willing or able to accommodate. It accounted for 19% of those identified as repeatedly homeless.

Figure 3.5: Family no longer willing or able to accommodate was the most common reason for loss of home for repeat homeless households

^{Bar chart showing the proportions of households by reason for loss (or threat of loss) of last settled home, for repeat and first time homelessness applications. The chart is based on 52 participating local authorities across the time period 2018 Q2 – 2023 Q1.}

For first time homelessness applications, the most common reason was end of private rented (assured shorthold) tenancy, with a rate of 25%. Evictions from supported housing were noticeably more prevalent for repeat homelessness (12%) than first time homelessness (3%), with a 9 percentage point difference between the share of applications.

Around half of the first time homeless households were predominantly living either with family or in the private rented sector at the time they sought help from their local authority. These were also the most common types for repeat homeless households, but to a lesser extent at just under one third. There was a noticeable contrast in the prevalence of those homeless after being held in custody at the time of application, which was 3.4 times more prevalent among repeat homeless households than first time homelessness. Similarly, rough sleeping at the time of application was found to be 3.1 times more prevalent among repeat homeless households.

3.6.2 Prevention activities

The analysis in Sections 3.6.4 to 3.6.6 relates to the 47% (31,929 applications) of all households in the dataset owed a prevention duty^{[footnote 5]}. Out of these, 11% (3,553 applications) were repeat homelessness applications.

For activities that led to homelessness being prevented, the most frequent activity was accommodation secured by the local authority or organisation delivering housing options service. This was observed in 19% of repeat homeless households owed a prevention duty and 23% for first time homelessness. Similar rates were also observed for households owed that had no prevention activity, but information provided to them. Unsuccessful prevention activities were comparable for both repeat (15%) and time households (14%).

For both repeat and first time households owed a prevention duty, 53% secured accommodation (for all accommodation types in total) and just under one quarter remained homeless when the activity ended. The main distinction was seen when looking at specific types of accommodation secured, as shown in Figure 3.6 below.

Figure 3.6: Remaining homeless was the most common reason for the prevention duty ending for repeat homeless households

^{Bar chart showing proportions of households by reason the prevention duty ended, for repeat and first time homelessness applications. The chart is based on 52 participating local authorities across the time period 2018 Q2 – 2023 Q1.}

First time households were more successful than repeats in securing alternative accommodation for 12 or more months (20% and 15% respectively). Whereas households that secured existing accommodation for 6 months was more prevalent among households experiencing repeat homelessness, rather than first time homelessness (18% and 13% respectively) after a prevention activity ended.

Following a prevention duty, the most common accommodation type was self-contained private rented sector. This was 18% of repeat households owed a prevention duty, and 24% for first time homelessness. Social rented supported accommodation was 1.6 times more prevalent among repeat homelessness applications than first time homelessness applications.

3.6.3 Relief activities

53% (36,259 applications) of all applications were owed a relief duty^{[footnote 6]}, upon their initial assessment. The analysis below in Sections 3.6.7 to 3.6.9 concerns these households only. From these, 7,290 were for repeat homelessness, meaning that two thirds of all the repeat homelessness households in the dataset were owed a relief duty.

For activities that led to homelessness being relieved, the most frequent activity was accommodation secured by the local authority or organisation delivering housing options service. This was observed in 21% of repeat homeless households owed a relief duty and 27% from the first time households owed.

Households owed a relief duty may have no activity taken, for example if contact was lost and 56 days had passed. This was comparable between repeat (28%) and first time (29%) households owed. However, unsuccessful relief activities were more prevalent for repeat homelessness (23%) than for first timers (18%).

Figure 3.7: 56 days elapsing was the most common reason the relief duty ended for repeat homeless households

^{Bar chart showing proportions of households by reason the relief duty ended, for repeat and first time homelessness applications. The chart is based on 52 participating local authorities across the time period 2018 Q2 – 2023 Q1.}

Relief duties ended most often because 56 days had elapsed (see Figure 3.7 above). It accounted for 38% of those repeatedly homeless owed a relief duty and 40% for that of first time homelessness owed. Like prevention duties, first time homeless households were more successful in securing accommodation for 12 months or more than those experiencing repeat homelessness (18% and 11% respectively). Whereas on a shorter term basis, accommodation secured for 6 months was higher among repeat households compared to first timers (23% and 19% respectively).

Activities that relieved homelessness led to proportionally more social rented accommodation among the repeat households (18%) than the first timers (13%) following a relief duty. Whereas private rented sector (2% point difference) and registered provider tenancy (5% point difference) accommodation were more associated with those relieved from first time homelessness.

3.7 Is there a difference in use of temporary accommodation for those experiencing repeat homelessness compared to those presenting as homeless for the first time?

Temporary Accommodation (TA) describes accommodation secured by a local housing authority under their statutory homelessness functions. This section explores TA use from 2018 Q2 to 2023 Q1 out of both the repeat and first time homelessness groups. Households that entered TA but did not have an exit date as of 1st January 2024 were categorised as those “Still in TA” at the time this analysis was conducted.

Historical trends from the official statistics show that households placed in TA are predominantly those with dependent children. However, the majority (63%) of households in the linked dataset were single households (a term for households without children, including couples and those with two or more adults), which is also reflected in the official statistics. Therefore, the observations on TA use in this section is largely driven by single households. However, it should be noted this is not representative of the TA system in general as single households are not owed a main duty unless they are in priority need and therefore will not be in TA.

Those who experienced repeat homelessness appeared to be more reliant on TA than first time homeless applicants, as suggested by Figure 3.8.

Figure 3.8: Repeat homelessness had no noticeable impact on TA use for families, but showed slightly more TA use for single households

^{Bar chart displaying proportion of TA use by repeat homelessness status. The chart is based on 52 participating local authorities across the time period 2018 Q2 – 2023 Q1.}

Out of all repeat homeless applications, 40% resulted in the household entering TA, compared to 33% for all first time homeless applications. Based on the limited number of observations in the linked dataset, this difference is largely driven by households owed a relief duty.

On average, the repeat homelessness group spent less time in TA. This is reflected in both household composition types in Figure 3.9 below.

Figure 3.9: When households stayed in TA, those experiencing repeat homelessness generally had shorter stays than first time homelessness

^{Box plot showing distribution of stay durations for those who have left TA, (52 participating local authorities, 2018 Q2 – 2023 Q1). The box widths reflect repeat and first time homeless population sizes. Outliers have been removed from the figure.}

From those who had entered and left temporary accommodation in the linked dataset, overall repeat homelessness applicants had a median of 84 days for each stay. For first time homelessness, this figure was 131 days. The duration of stays was also more variable for first time homeless households (with the lower and upper quartiles of 49 and 291 days respectively).

Looking at a subset of the data where repeat homeless lead applicants were identified by multiple rows only, 5% of those placed in TA on their first application went on to experience repeat homelessness. For lead applicants not placed in TA at their first identified application, this was only slightly more, at 8%.

4. Recommendations

This study helps form the evidence base of MHCLG and BOLD by linking H-CLIC to other datasets. For the first time, homelessness application data can show:

• people that otherwise would have not been identified as experiencing repeat homelessness
• opportunities potentially missed by local authorities to identify repeat homelessness
• people who made homelessness applications in more than one local authority

The main finding is that repeat homelessness was found to be increasingly identifiable after linking personal information to attribute data of homelessness applications. Around 1 in 6 applications were identified as repeat homeless. The linked dataset allowed over a third of these to be identified as repeat homelessness. It was found to have associations with a variety of household characteristics that included having additional support needs or belonging to groups with lived experiences, such as prison leavers.

4.1 Improvements in data

This study has detailed some of the characteristics associated with repeat homelessness. If MHCLG are to better understand repeat homelessness through a data led approach, further investigations are required. Such analysis would need to be aimed at the causality of the characteristics resulting in repeat homelessness.

A deeper understanding of repeat homelessness may also require looking at the long term outcomes of such households. Data improvements are recommended to enable this, outlined sections 4.1.1 to 4.1.5 below.

4.1.1 Submissions from more local authorities

A greater number of participatory local authorities would help facilitate detailed analysis of those who apply for support in more than one local authority.

4.1.2 Longer data collection periods

Coupled with a greater number of local authorities, this would allow data to mature sufficiently to observe journeys into repeat homelessness.

4.1.3 Further data linkage with H-CLIC

Additional datasets may include educational outcomes, health conditions, employment history, etc. This would also identify long-term outcomes for people who did not have further H-CLIC records because they avoided repeat homelessness. In addition to government data, this could involve data held by organisations such as housing providers and charities.

4.1.4 Other analytical methods

Different techniques may include causal inference and significance testing (e.g. logistic regression). Approaches orientated towards causality could quantify the extent that repeat homelessness happened due to the presence of a characteristic, such as specific or co-occurring support needs. Such techniques are more effective with larger samples (via more local authorities and longer time series data), which in turn would allow inferences to be more meaningful.

4.1.5 Collaboration with the third sector

Organisations in direct contact with the households, such as charities and housing associations that may have front line knowledge that could supplement the evidence provided in this study, and potentially present opportunities to link with their datasets.

4.2 Enhancing the support available

The findings in this study suggest that households experiencing repeat homelessness have more complex needs than those that are first time homeless. With improved data collection and analysis to help identify and support the journeys of those who experience repeat homelessness, the findings in the study central and local government could provide better targeted and proactive support. Sections 4.2.1 and 4.2.2 outline some suggested recommendations to this.

4.2.1 Tailored support

A more tailored approach to supporting households experiencing repeat homeless through policies, funding, or duties from central government. Local authorities should focus on helping those who have found housing after their first homelessness experience to prevent future homelessness. This can include financial education and support for individuals leaving institutions such as the care or prison systems.

4.2.2 Digital tools

As opportunities to identify repeat homelessness at the time of application can be missed, this could be mitigated by automating the identification of repeat homelessness at the start of an application. Digital solutions using H-CLIC data could potentially improve the accuracy of the process and help deliver timely targeted interventions.

A more proactive management approach could involve the use of early warning systems. Digital applications that utilise statistical methods (mentioned in Section 4.1.4) could quantify the risk that a first time homelessness person has of falling into repeat homelessness, based on their circumstances.

5. Further Information

5.1 Research reports

MHCLG’s statutory homelessness statistics, which also delivers insights from H-CLIC, are labelled as Accredited Official Statistics. Further information on Accredited Official Statistics is available via the UK Statistics Authority website.

5.2 Enquiries or feedback

The work by the BOLD pilot teams could be useful to people who want to ensure that provisions for homelessness support applications are operating as intended.

If you have any enquiries or feedback about this report, please contact hrsresearch@communities.gov.uk and ask for the BOLD Homelessness Pilot Team.

Annex A: List of participating local authorities

The following local authorities have submitted personal data as part of the Homelessness Data England (HDE) project:

• Arun District Council
• Ashford Borough Council
• Babergh District Council
• Barrow-in-Furness Borough Council*
• Bath and North East Somerset Council
• Blackburn with Darwen Borough Council
• Bristol City Council
• Broadland District Council
• Canterbury City Council
• Chelmsford City Council
• Crawley Borough Council
• Darlington Borough Council
• Dartford Borough Council
• Dover District Council
• East Hertfordshire District Council
• Folkestone and Hythe District Council
• Great Yarmouth Borough Council
• Kensington and Chelsea Royal Borough
• Leicester City Council
• Liverpool City Council
• Maidstone Borough Council
• Mid Devon District Council
• Mid Suffolk District Council
• Newcastle-under-Lyme Borough Council
• North Kesteven District Council
• North Somerset Council
• North Tyneside Council
• Northumberland County Unitary Authority
• Oadby and Wigston Borough Council
• Preston City Council
• Reigate and Banstead Borough Council
• Sevenoaks District Council
• Slough Borough Council
• South Kesteven District Council
• South Lakeland District Council*
• South Norfolk Council
• Southwark London Borough
• Spelthorne Borough Council
• Stoke-on-Trent City Council
• Sunderland City Council
• Surrey Heath Borough Council
• Sutton London Borough
• Tandridge District Council
• Thanet District Council
• Tonbridge and Malling Borough Council
• Tunbridge Wells Borough Council
• Walsall Metropolitan Borough Council
• West Lancashire Borough Council
• West Lindsey District Council
• West Suffolk Council
• Wiltshire County Unitary Authority
• Worcester City Council

The local authorities are self-selected; therefore, trends may not be representative of non-participating local authorities.

*In April 2023 two of the participating local authorities, Barrow-in-Furness and South Lakeland, were both replaced by Cumberland Council. The linked dataset was unaffected by previous boundary changes.

Table A1: The number of local authorities for each of the total homelessness applications in each quarter

Date of application	Total number of local authorities	Total applications in dataset
2004 to 2017	12	106
2018 Q1	5	11
2018 Q2	21	303
2018 Q3	25	330
2018 Q4	25	361
2019 Q1	23	367
2019 Q2	26	407
2019 Q3	31	424
2019 Q4	38	506
2020 Q1	42	856
2020 Q2	41	929
2020 Q3	44	2,039
2020 Q4	48	2,672
2021 Q1	45	3,806
2021 Q2	49	4,079
2021 Q3	49	5,097
2021 Q4	51	5,892
2022 Q1	51	8,293
2022 Q2	51	8,233
2022 Q3	51	9,235
2022 Q4	50	7,430
2023 Q1	44	6,791
2023 Q2 to 2023 Q4	14	161

Table A1 shows that although the applications came from 52 distinct local authorities in total, they featured at various quarters in the linked dataset. The table does not show the number of local authorities that participated or number of homeless applications in each quarter. Instead, the table indicates that matching the ID Spine and case attribute datasets returned higher volumes from 2020 Q3 to 2023 Q1.

Annex B: Data linking results

The resulting linked dataset (L1 from Figure 2.1) returned 68,328 applications, which had been raised by 63,417 distinct people identified as lead applicants. There were 121,355 distinct household members in total, identified as either lead applicants or other members such as partners and children.

Linking the ID Spine to the H-CLIC dataset showed that there were more people in the H-CLIC data extract than there were local authority issued person ID codes. After controlling for local authorities using the same labelling formats (see Annex D), there would have been 126,984 people (instead of 121,355) had linking not been applied, i.e. using the H-CLIC person ID rather than those from the ID Spine).

Instances were found where multiple H-CLIC Person IDs were allocated to the same ID Spine. This fulfils the expectation that data linking would uncover a smaller number of distinct household members than previously indicated by the 68,328 applications without linking.

Conversely, other instances were found where multiple ID Spines were allocated to the same H-CLIC Person ID, illustrated in Table B1 below.

Table B1: Conceptual example how 5 people (SP1 to SP5) may be interpreted as 3 people if the H-CLIC ID was used instead of the ONS ID

Local Authority	Case Reference	H-CLIC Person ID	ONS ID Spine
X	X1	HP1	SP1
X	X2	HP1	SP2
Y	Y1	HP2	SP3
Z	Z1	HP3	SP4
Z	Z2	HP3	SP5

Reasons for this may be wide ranging, for example:

• A change in spouse in the household being misidentified as the same partner from a previous application
• Special characters in a person’s name being used in their separate applications, resulting 2 separate ID Spines for that person.

The latter of these examples would highlight a limitation of deterministic matching when the data linking was carried out.

The analytical findings assume that every match that allocated ID Spines to household members list in a homeless application was accurate. This assumption is based on the process using personal identifiers and accounted for local authority location. For this reason, repeat homelessness was determined using ID Spine, as well as the previous case references and history of repeat homelessness support need flagged from H-CLIC attribute data. repeat homelessness figures may still be under-reported however, as the ID spine only covers a limited time span and number of local authorities.

Not all homeless application data could be matched during data linking:

• 97 % of people (all household members including partners, children and other dependents) from the personal information dataset (i.e. the ID Spine) were matched to at least one household from the attribute data
• Only 2% of the ID Spines were matched to more than one household
• 43% of households were matched to more than one ID Spine

This means that the 68,328 applications are an undercount of actual applications that happened in the participating local authorities during 2018 Q2 to 2023 Q1.

Annex C: Examples of identifying repeat homelessness

The first method to identify repeat experiences was at a person level (for lead applicants and other household members) in the dataset by counting the instances their person ID (which is usually the ID Spine after linking, but can be applied to the H-CLIC Person ID as mentioned in Annex D) appeared more than once:

Table C1: Example of counting repeat status by number of appearances in the dataset

Case Reference	Person ID	Interaction Count	Repeat person
A	P1	1	No
A	P2	1	No
B	P3	1	No
C	P1	2	Yes
D	P1	3	Yes
D	P4	1	No
E	P5	1	No

In Table C1, person P1 has been found to experience repeat homelessness in applications C and D (their second and third experiences respectively).

A person’s previous case reference may also be used to indicate if they have approached homelessness services in their local authority before. In Table C2, this would apply to person P3.

Table C2: Example of counting repeat status by number of previous case references

Case Reference	Person ID	Previous Case	Repeat PCR
A	P1	-	No
A	P2	-	No
B	P3	X	Yes
C	P1	-	No
D	P1	-	No
D	P4	-	No
E	P5		No

On a household (case reference number) level, the support need for a history of repeat homelessness could also be identified, as depicted in Table C3.

Table C3: Example of counting repeat status by support need flag

Case Reference	Repeat SN flag
A	Yes
B	No
C	No
D	Yes

These tables can be joined together (using an inner join), which can be summarised further at a case level.

Table C4: Joined table by combining E1, E2 and E3 at a person level

Case Reference	Person ID	Repeat person	Repeat PCR	Repeat SN flag	Status
A	P1	No	No	Yes	Repeat
A	P2	No	No	Yes	Repeat
B	P3	No	Yes	No	Repeat
C	P1	Yes	No	No	Repeat
D	P1	Yes	No	Yes	Repeat
D	P4	No	No	Yes	Repeat
E	P5	No	No	No	First Time

Table C4 shows that applications A to D are examples of repeat homelessness applications, whereas case E meets the criteria of first time homelessness. This methodology was applied to the linked dataset, before further statistical analysis was conducted.

Annex D: Linked vs without linking approaches

In Section 3.2 the impact of data linking was shown by comparing repeat homelessness identification applied to the linked dataset with that applied to a dataset without linking. The link approach involved looking at ONS Spine IDs observed multiple times in the dataset, whereas without linking this would look at Person IDs from H-CLIC. The findings in Section 3.3 onwards are based on the linked approach using ONS Spine IDs.

Challenges arose initially when this methodology was applied to H-CLIC Person IDs for a comparison. There were observations in the dataset where the local authority allocated a person ID number using the same syntax. This meant that some person ID numbers in H-CLIC were duplicated for separate people, as demonstrated in Table D1 below.

Table D1: Example to illustrate the concept of duplicated Person IDs

ONS Spine ID	H-CLIC Person ID	Lead Applicant Sex	Lead Applicant Nationality	Case Reference	Local Authority
SP1	HP1	Male	Other	C1	LA1
SP2	HP1	Female	UK	C2	LA2
SP3	HP2	Male	UK	C3	LA1
SP3	HP3	Male	UK	C4	LA2

In this example, it is assumed that there are 4 separate people recorded (which the linking process uncovers to find that HP2 and HP3 are the same person), where local authorities LA1 and LA2 have both allocated the ID of “HP1” to separate people making a homelessness application. Identifying repeat homelessness using this information alone would lead to the finding that HP1 referred to a person who experienced repeat homelessness. This is likely an artefact from duplicated ID numbers, as the ONS IDs would indicate 2 separate people from different local authorities (and even more likely if they were lead applicants on their applications).

To mitigate the against the effect this would have on repeat homelessness in this study, other attribute data was concatenated to create an alternative ID for the approach without linking. When applied to Table D1, this represents the 4 separate IDs desired as shown in Table D2:

Table D2: Concatenation of attributes from Table D1

ONS Spine ID	Alternative ID	Case Reference
SP1	HP1-M-O-LA1	C1
SP2	HP1-F-UK-LA2	C2
SP3	HP2-M-UK-LA1	C3
SP3	HP3-M-UK-LA2	C4

This assumes that the following attributes do not change when people make a homelessness application:

• Their lead applicant’s recorded sex (regardless of if the row refers to a lead applicant or a dependent)
• Their lead applicant’s recorded nationality
• Local authority of the application

The identification of repeat homelessness applications was applied after the creation of the alternative ID. In the example of Table D2, the process would find no repeat homeless applications under the approach without linking, whereas case C4 would have been found to be repeat homelessness (specifically, one identified because of linking), due to more than one application involving the person with the Spine ID SP3.

Annex E: Summary of linked dataset

Table E1: Summary of households from the linked dataset, by quarter of application data and the identified repeat status. The columns containing the totals of repeat homeless applications are exclusive to one another

Date of Application	First Time Homeless Households	Repeat Homeless Households (excluding those identified by linking)	Repeat Homeless Households (identified by linking)
2004 to 2017	104	<5	<5
2018 Q1	11	<5	<5
2018 Q2	281	19	<5
2018 Q3	312	17	<5
2018 Q4	352	9	<5
2019 Q1	340	19	8
2019 Q2	372	30	<5
2019 Q3	379	36	9
2019 Q4	450	51	<5
2020 Q1	742	104	10
2020 Q2	782	133	14
2020 Q3	1,794	223	22
2020 Q4	2,270	362	40
2021 Q1	3,231	493	82
2021 Q2	3,505	463	111
2021 Q3	4,407	502	188
2021 Q4	5,003	594	295
2022 Q1	7,071	846	376
2022 Q2	6,897	863	473
2022 Q3	7,547	1,033	655
2022 Q4	6,032	825	573
2023 Q1	5,467	710	614
2023 Q2 to 2023 Q4	127	6	28
Total Applications (n= 68,328)	57,476	7,340	3,512