Research and analysis

Rough Sleeping Questionnaire 2025: Methodology and technical annex

Updated 26 February 2026

Applies to England

A report of the fieldwork and data processing methodology for the Rough Sleeping Questionnaire 2025.

1. Methodology

1.1 Summary

The 2025 Rough Sleeping Questionnaire had a broadly consistent approach with that of the 2019/20 survey. Prior to the fieldwork, MHCLG reviewed and updated the 2019/20 RSQ for use in 2025. Local authorities were contacted to explain the purpose of the research and identify suitable locations to conduct the fieldwork.

Fieldwork was conducted at 77 support service sites, such as hostels and day centres, in 32 local authorities across England. At each service site, staff and interviewers invited service users to take part in the survey. The teams targeted invitations at service users who had slept rough within the past 12 months. Verian interviewers spoke to 1,292 people in total, of which 966 had slept rough in the previous 12 months.

1.2 Questionnaire design

The RSQ was designed in-house by MHCLG researchers with input from survey experts, academics, government analysts, people with lived experience of sleeping rough, and frontline homelessness staff. The topics and questions were designed to provide a comprehensive understanding of respondents’ backgrounds, histories of homelessness, support needs, and public service use.

MHCLG commissioned Verian to deliver the updated survey, including questionnaire scripting, fieldwork logistics and interviewing. The previous questionnaire (which was designed in 2019) was reviewed by MHCLG and updated question phrasing and added new questions in line with the latest policy interests. Verian undertook a detailed review of the updated questionnaire, which focussed on practical scripting points and clarifying filters. They also reviewed interviewer instructions, text and instructions in the self-completion section, presentation of questions, and the potential risk of mode-related measurement effects.

Where possible, standardised questions were adopted to ensure comparability with other data sources, for example, validated questions and scales were used for demographics, wellbeing and general health. Respondents also had the opportunity to express their experiences in their own words, captured in open text boxes. Respondents were routed through the questionnaire, with certain questions asked depending on previous answers.

The survey included several free-text questions to give participants the opportunity to share their experiences in their own words. Open text boxes were included at the end of each section of the survey, asking if there was anything else participants wanted to say about that topic. There were also open questions about triggers of sleeping rough, barriers to finding accommodation and participants hopes for the future.

An ethical review of the approach was conducted by the Sheffield Hallam University Ethics Committee. The project was classified as Level 3 risk on the Sheffield Hallam University’s Ethics Committee scale as it included:

• Direct engagement with vulnerable populations
• Face-to-face data collection methods
• Sensitive content (for example, trauma, substance use, housing instability)

In addition, Verian conducted an internal ethical review, resulting in a list of recommendations for interviewers when conducting the fieldwork. Details of these are included in section 1.4

1.3 Local authority participation

To ensure the sample was as representative and as large as possible, MHCLG selected 32 local authorities to participate. Local authorities were chosen based on:

• Distribution across the English regions representing urban, suburban, and rural areas
• Historical data from the 2019 RSQ and 2024 Rough Sleeping monitoring information from local authorities
• Local authority readiness and capacity to support research

MHCLG and Verian engaged with the Rough Sleeping Leads in each local authority to explain the survey delivery process and to discuss any local adaptations to the fieldwork approach. Prior to conducting the fieldwork, Verian visited each support service to discuss the practicalities of conducting interviews at each site. This included:

• health and safety and safeguarding arrangements
• the best times for interviewers to attend
• identifying a safe, private space for interviewing
• conducting a risk assessment

The 32 participating local authorities were grouped into four clusters to allow for an efficient management of fieldwork logistics, engagement, and quality assurance. If the initial allocation was unsuitable for the local authority or the support services, Verian reassigned the local authority to a later cluster. The first cluster was used as a live pilot, with learnings from this cluster implemented in future clusters.

1.4 Conducting the fieldwork

The 2025 survey was conducted between January and April 2025. This took place at 77 support service sites, such as hostels and day centres, in 32 local authorities across England.

Interviews were conducted by face-to-face interviewers, using computer assisted in-person interviewing (CAPI) and computer assisted self-interviews. Fieldwork in each cluster was conducted over a two to three week period.

Respondents had the option to complete the RSQ independently or assisted by a researcher.  

Verian interviewers spoke to 1,292 service users in total. Of these, 1,225 were both eligible to take part and agreed to participate in the survey. Reasons for not interviewing users included insufficient time to take part, language barriers, or being under the age of 18.

The questionnaire and research information was translated into other languages to encourage more non-UK nationals to take part. The vast majority of the 1,225 questionnaires were completed in English (98%). A minority were self-completed in another language, most of which were in Polish (2%). The other languages were Arabic, Romanian and Lithuanian.

After de-duplication, 1,204 people had completed the survey. Of these, 966 had slept rough within the previous 12 months and 238 were classified as ‘other service users’.

Carrying out research that representatively samples people who sleep rough is inherently difficult, given the transient and hidden nature of rough sleeping. In practice, there was an element of convenience sampling of participants given the nature of the research. Interviewing people in services centres (rather than those on visibly rough sleeping on streets) meant that those who were less engaged with rough sleeping services were less likely to be interviewed.

All researchers and service staff undertook ethical training on conducting research with vulnerable groups and in gaining valid informed consent. Where respondents did not speak English, an on-call translation service was available to answer participants questions. Any instances where it was believed there was insufficient understanding of the research to establish valid informed consent, individuals did not take part in the fieldwork.

To reduce any risk of harm Verian worked with the interview sites to ensure support was available for anyone who needed it. They also ensured participants knew they could stop the interview at any time and were instructed to stop if a participant became noticeably distressed.

In some instances, researchers were proactive and interviewed people who were available at that time, to not exclude service users who were keen to participate. For example, if someone had last slept rough more than 12 months previously and was living in a hostel.

1.5 Data management

Using the first cluster of local authorities, Verian checked the application of question routing and corrected a small number of errors before proceeding with the rest of the clusters. The raw survey data was provided in an SPSS data file, which was cleaned by Verian to ensure that:

• values matched frequency counts produced from the ‘raw’ data
• base numbers were consistent with filter definitions
• missing values were coded correctly
• labelling was correct and consistent.

They also removed 21 duplicate responses identified through matching personal details, keeping only the first instance of each.

The cleaned data set was transferred to MHCLG via a secure file transfer protocol.

1.6 Analysis and reporting

Descriptive statistics were run for each topic of interest, taking the frequency, proportion, or mean of groups across each outcome. This analysis is descriptive and does not control for other factors. Where relevant, statistics were also calculated by subgroups, including gender and nationality.

Not all variables were available for all respondents, due to either the question not being asked to all respondents (based on their responses to previous questions), or from participants choosing not to answer. Where respondents did not answer the question (by either skipping the question entirely or answering ‘Don’t know’ or ‘Don’t want to say’), this was coded as a non-response. Those who gave a non-response are still included in the base sample size when calculating the descriptive statistics. Where the question was not asked of respondents, the base sample size is reduced, indicated by “n =”  in the tables and captions.

To ease interpretation of the results, some responses are not reported in the descriptive statistics. These may include non-responses, responses from an exhaustive list of possible answers for that question, or those that accounted for a low proportion of the total who answered.

Statistical tests were performed on variables of interest to determine which responses showed significant differences between two groups (e.g. women and men). Answers that displayed significance are denoted by * or mentioned otherwise. Further details can be found in the technical annex section.

Thematic analysis was applied to the free text responses to the questionnaire. One analyst read the responses and assigned initial granular codes. These codes where then grouped into initial themes. A sub-sample of the data was coded by a second analyst and this coding compared to the first coder’s, to quality assure the analysis. These themes were then refined in team discussions. Quotes were edited to remove spelling and typing errors. Participants were given the option of completing the questionnaire alone, or with the help of a researcher. It should be noted that often the interviewer operated the tablet on which the questionnaire was completed, and typed responses for participants. Due to this, and the need to succinctly distil complex personal stories, the responses are therefore often not direct verbatim quotes of the participants’ words, but instead a summary or slight rewording.

All data management and quantitative analysis was predominantly conducted in R version 4.1.2. The data management and analysis were quality assured internally and the report has been externally peer reviewed.

2. Technical annex

Annex 1: Comparison with Rough Sleeping Snapshot demographics

The questionnaire was answered on a voluntary basis and conducted in services centres such as hostels and day centres. The this makes it difficult to accurately reflect the true demographic profile of rough sleepers from the respondents. Rough sleeping experiences across different demographics are known to be more hidden, transient and intermittent, than others. Therefore, the findings from the RSQ should not treated as a statistical representation of the rough sleeping population in England.

Comparisons with the most recent Rough Sleeping Snapshot data are shown below.

Table 1.1 Gender profiles for the RSQ 2025 and Snapshot 2024

Gender	RSQ 2025 (n = 966)	Snapshot 2024 (n = 4,667)
Female	16%	15%
Male	84%	83%
Other or not known	< 1%	3%

Table 1.2 Age profiles for the RSQ 2025 and Snapshot 2024

Age	RSQ 2025 (n = 966)	Snapshot 2024 (n = 4,667)
Under 18	n/a	< 1%
18-25	6%	5%
26 and over	94%	86%
Not known	0%	9%

Table 1.3 Nationality profiles for the RSQ 2025 and Snapshot 2024

Nationality	RSQ 2025 (n = 966)	Snapshot 2024 (n = 4,667)
UK	81%	63%
EU	7%	16%
Non-EU	11%	11%
Not known	0%	10%

Annex 2: Statistical analysis

In the expanded findings reports, statistical significance was used to assess which responses had meaningful differences between 2 groups (e.g. women and men, UK and non-UK nationals, etc.). In most cases a Chi-Square Test was conducted, with Fisher’s Exact Test used if there were any expected values which had less than 5 counts. Significance was highlighted when the resulting p-value was less than 0.05

The tables below are examples of the actual counts between 2 groups for answers with statistical significance.

Table 2.1: Mental health needs, by gender

Gender	Mental health need	No mental health need
Women	138	13
Men	656	153

P-value < 0.01

Table 2.2: Victim of crime in past  6 months, by gender

Gender	Victim of crime in past  6 months	Not a victim of crime in past  6 months
Women	36	115
Men	325	484

P-value < 0.001

Table 2.3: Victim of domestic abuse (at any point since the age of 16), by gender

Gender	Victim of domestic abuse	Not a victim of domestic abuse
Women	104	47
Men	252	557

P-value < 0.001

Table 2.3: Ever spent time in prison, by gender

Gender	Spent time in prison	No time in prison
Women	57	94
Men	413	396

P-value < 0.01

Annex 3: Fiscal unit costs

The data set includes participants who have slept rough within the last year. The RSQ data records which services individuals used and the number of times individuals have accessed these. All questions measure experiences in either the last 3 or 12 months, except for questions relating to:

• The number arrests (12 months only)
• Time spent in prison (12 months only)

Participants are more likely to answer accurately when providing information in the last quarter, compared to the last year. As such, using this data and extrapolating to a year was deemed more robust.

3.1 Unit cost per year calculations

3.1.1 Health services and substance treatments

The total number of individuals who used a service was multiplied by the mean frequency of service usage and the unit cost. The unit cost was derived from the Unit Cost Database (e.g. GP cost per hour). This was subsequently converted into an annual cost by multiplying the total service visitors in the past 3 months by 4 (to get an annual estimate), the mean number visits and unit cost. Next, it was divided by the total number of survey participants (excluding those with missing data for the relevant question) to give an annual cost per individual who slept rough.

3.1.2 Criminal justice services

The average number of arrests in the past year, the total number of people in prison in the last year and the average length of time in prison over the past year was calculated from the questionnaire responses. These were multiplied by their respective unit costs and added together to form the total cost to criminal justice services. This was divided by the total number of survey participants to get the average cost.

3.1.3 Rough sleeping services

To derive the cost of rough sleeping services, the average number of nights slept rough over the last three months was estimated from questionnaire responses. This number was uprated to an annual figure and combined with an uprated average weekly cost of support for accommodation based homelessness services incurred by local authorities, which was sourced from the 2015 joint publication by Crisis and University of York.

3.1.4 Unit costs and data sources

This section provides a detailed breakdown of the individual unit costs used to calculate the cost per average rough sleeper per year. These cost estimates were sourced from the Greater Manchester Combined Authority Unit Cost Database published in November 2025.

Table 3.1: Summary of unit cost descriptions and sources used

Service	Unit cost description	Unit cost database original source
A&E attendance	Average cost of A&E attendance across all incidents.	National Schedule of Reference Costs 2023–24 for NHS trusts and NHS foundation trusts
Alcohol dependency treatment	Average cost of intervention for mild alcohol dependency	NICE Clinical Guideline 115: Alcohol Use Disorders – alcohol dependence, costing report.
Ambulance call out	Average cost per ambulance call out regardless of treatment outcome.	National Schedule of Reference Costs 2023–24 for NHS trusts and NHS foundation trusts
Arrests	Average cost of detained arrests. Made up of police and duty solicitor costs.	Salford: Police costs 2006/7.
Drug treatment	Average annual cost of effective drug treatment (at least 12 weeks).	Estimating the crime reduction benefits of drug treatment and recovery (National Treatment Agency for Substance Misuse 2012)
GP visit	Average cost per GP visit, adjusted for a 15‑minute appointment.	Unit Costs of Health & Social Care 2024 (Jones).
Mental health appointment	Average cost of counselling	Unit Costs of Health & Social Care 2014 (Curtis).
Mental health hospital stay	This cost is composed of an initial mental health assessment and a cost per bed per day.	National Schedule of NHS Costs 2020-21 for NHS trusts and NHS foundation trusts and National Schedule of Reference Costs 2023–24 for NHS trusts and NHS foundation trusts
Physical health appointment (last 3 months)	Average cost per hospital outpatient attendance.	National Schedule of Reference Costs 2023–24 for NHS trusts and NHS foundation trusts.
Physical health hospital stay (last 3 months)	Average cost of a period admitted in hospital.	National Schedule of Reference Costs 2023–24 for NHS trusts and NHS foundation trusts.
Prison (last year)	Average cost of one prisoner per year adjusted for average time spent in prison.	HM Prison & Probation Service Annual Report and Accounts 2023-24.
Rough sleeping services (last 3 months)	Support costs for accommodation-based homelessness services (median cost across all intensity levels)	Crisis UK: An estimation of the financial costs of single homelessness in the UK (Pleace, 2015)

3.2 Data manipulation

To determine the unit costs of some services, additional data manipulation was required. This is summarised below, and relates to the sources listed above.

3.2.1 GP services

For this analysis, an assumption of 15 minutes per GP appointment was made. The unit cost was multiplied by the average number of GP appointments reported per respondent.

3.2.2 Alcohol and drug treatment services

The questionnaire asked whether respondents had accessed drug or alcohol treatment services within the last 3 months. In the absence of a data on the frequency of service use in the questionnaire, it has been estimated that if a respondent had reported treatment in the last three months, it was on one occasion. The unit cost for treatment was then multiplied by the number of respondents who had reported service use.

3.2.3 Mental and physical health services

Mental and physical health service unit costs consist of costs from hospital appointments and stays, which were calculated separately and summed together. Costs associated with staying in hospital as a result of a mental health issue used two units: a cost per hospital stay and a cost per initial mental health assessment. For respondents who had been an inpatient in hospital for a mental health need, the mean number of days spent in hospital in the past three months was 9 days, which was multiplied by the number of respondents who had reported a stay for a mental health issue and the unit cost. This was assumed to give the total cost for three months. This was converted to an annual figure and divided by respondents to the survey to generate the average cost.

3.2.4 Arrests and convictions

Respondents were asked to provide the number of times they had been arrested in the last year. If a range was selected from the answer options (e.g. 5-9 times), the mid-point was used. By multiplying the number by the frequency of individuals who were in that category and adding them altogether, an estimate for the total number of arrests in the 12 months prior to completing the survey was provided. The number of respondents who reported an arrest was multiplied by the unit cost to give a yearly cost and divided by the number of respondents to generate an average cost.

3.2.5 Prison

All respondents were asked to report whether they had ever been to prison. Anyone who had been to prison within the last year was asked about the number of separate times and the total length of time spent in prison over the last year. Similar to arrests, ranges were provided for the length of stay in prison. The midpoint was used for each range, excluding the ‘a week or less’ option where a week was assumed and ‘more than 9 months’ where a year was assumed. The aggregate length of time in prison was calculated and multiplied by the number of respondents who had spent time in prison in the last year alongside the annual cost and divided by survey responses to generate an average cost.

3.2.6 Rough sleeping services

The unit cost provided by the source was from 2015/16, and was subsequently uprated to 2024/25 prices. Respondents were asked to specify (or provide their best guess if unsure) the number of nights they had spent sleeping rough over the last three months. The number of respondents was multiplied by four to obtain an annual figure then multiplied by the average number of nights slept rough in weeks alongside the average cost to local authorities to obtain an annual figure.