Hidden homelessness: international evidence review

4. Methods for identifying hidden homeless populations

In the following chapter methods for identifying and counting homeless and hidden homeless populations that emerged from the literature will be presented in four sections. The first section (4.1) groups primary data collection methods. Section 4.2 details secondary data collection methods and their specific sources. Section 4.3 introduces innovative data collection strategies which present two novel approaches to homeless data collection. Finally, section 4.4 discusses the use of a combination of multiple methods and data sources to triangulate homelessness estimations.

Some of the methods discussed in this chapter refer to the improvement of the enumeration of a specific population group, while others are presented more generally as ways of improving the counts of less visible populations. Their contextual specificities are accounted for in this section. Where possible, case studies with examples of the application of each method to hidden homelessness are included while also reflecting on their strengths and limitations.

4.1 Primary research methods

This section focuses on methods designed to collect data directly from people with first-hand experience of homelessness and hidden homelessness. This includes methods such us surveys, interviews, focus groups and observations, among others.

4.1.1 Overnight Point in Time (PiT) Counts

Overnight point in time (PiT) counts are the data collection method that most frequently appeared in the literature on homeless counts. PiT counts are a cross-sectional observational survey method used to measure unsheltered types of homelessness (Bretherton & Pleace, 2018). They are one of the most often used methods in England, North America and parts of Europe (Bretherton & Pleace, 2018; Rabinovitch, 2015; Schneider, Brisson, & Burnes, 2016). It is worth noting that Wales stopped conducting rough sleeping overnight counts in 2020 due to the COVID-19 pandemic and later decided to replace this method with monthly council estimations.

PiT counts consist mostly of overnight counts in urban areas, covering previously mapped areas considered 'hotspots' (areas where there is an expectation to find people sleeping rough). This means a count of people sleeping rough conducted during a specific day/night by a group of trained volunteers. A brief survey is sometimes also conducted with individuals if they consent. This method can also include the enumeration of people sleeping in homeless shelters on one given night.

PiT counts can be repeated periodically to establish trends or count over an extended period of time (Bretherton & Pleace, 2018). For example, in England, rough sleeping overnight street counts are conducted annually over a single night during the autumn (Department for Levelling Up Housing and Communities, 2023). In Canada, the frequency of conducting PiT counts varies in different cities. Vancouver conducts annual PiT counts, Edmonton and Calgary conduct them every two years and metropolitan Vancouver and Toronto conduct them every three years (Rabinovitch, 2015). This has resource implications and may require a balanced consideration of available resource versus the frequency required to obtain robust data. PiT counts are also conducted in Italy, France and Spain, however, they are used in specific urban areas rather than nationwide and on a less frequent basis than is observed in the Canadian context, with counts taking place several years apart (Bretherton & Pleace, 2018).

Challenges and limitations of overnight Point in Time (PiT) counts

As previously established, not all homelessness involves rough sleeping. It is known that, due to the reliance of PiT counts on visibility of people experiencing homelessness, they are likely to underestimate the overall homeless population and overrepresent specific sub-groups like those who are visibly homeless and people with complex needs (Agans et al., 2014; Baptista, Benjaminsen, Pleace, & Busch-Geertsema, 2012; Rabinovitch, 2015; A. Smith, 2015).

The enumeration of only those who are 'bedded-down' (sleeping on the street) or about to bed-down, might miss anyone who is walking around using, for example, all-night cafes, shops, bars or restaurants for shelter. People sleeping rough may also be in hospitals, train stations, riding on public transport, and sometimes in police custody when counts are being conducted and may be missed for those reasons (Bretherton & Pleace, 2018). One of the main criticisms of PiT counts present in the literature is that they often undercount women and other vulnerable populations who stay purposely out of sight due to physical safety concerns, as well as those who adopt other strategies such as sofa surfing (Bretherton & Pleace, 2018; Pleace, 2016). Further, PiT counts frequently do not include rural areas, where people experiencing homelessness may be more dispersed (Busch-Geertsema, Culhane, & Fitzpatrick, 2016).

Additionally, counting every person sleeping rough, or in a space not fit for human habitation, over a 24-hour period is likely not achievable, especially in large urban areas and/or sparsely populated rural areas (Rabinovitch, 2015). PiT counts are not designed to cover the entire geography of a particular area and tend to focus on places which are known to host people experiencing homelessness or places familiar to the volunteers/organisations to allow counts to be efficient and cost-effective. This often means focusing on urban spaces where there is an expectation of finding clusters of people experiencing homelessness. This can lead to an undercount or a misrepresentation of the characteristics of the total homeless population (Weare, 2019). The characteristics of people sleeping rough may differ from other groups experiencing other types of homelessness circumstances. This undermines the potential for this method to provide an accurate representation of the overall homeless population (Rabinovitch, 2015).

PiT counts are also very sensitive to the environmental and social context of the specific date on which it is conducted (Busch-Geertsema et al., 2016; Hall, 2017). This could mean, for example, the weather on the day/night or major events happening in the surroundings where the count is taking place. For example, there could be more police presence if there is a large concert or social gathering occurring nearby- such as a protest- and this can deter people from staying overnight in the area due to fears of criminalisation.

Finally, volunteers' training on identification of people sleeping rough for the purpose of including them in the count also plays a central role. The assessment that volunteers make of who is and who is not sleeping rough, and even their demographic characteristics (such as their gender, ethnicity or age), can skew the count or hinder its quality (Busch-Geertsema et al., 2016). Sometimes a further survey is conducted with people who are awake during the count, which can provide more accuracy, but as most counts happen overnight that is not always possible.

Case study: PiT count of unsheltered youth based on a local collaborative partnership

Trawver and Aguiniga (2016) present the strategies employed to adapt a PiT count of unaccompanied youth in the US to better suit the characteristics of that population after the previous 'Youth Count! Initiative' in 2013 was proven to have significantly undercounted this population. In order to improve this outcome a collaborative partnership was created in 2014 between academics of the University of Anchorage (Alaska), community youth services (a homeless youth shelter and services, a drop-in centre and medical clinic for youth and a teen club and resource centre) and local students. Their aim was to conduct a successful youth PiT count, with better outreach, that more accurately reflected the number of homeless youths in their area.

A key aspect of the collaboration was that each partner took on their specific role based on their skills and resources. Community organisations led on the logistics and organisation of the count, faculty members produced the survey used to collect the data and a local student led on the volunteer recruitment.

The survey included questions that collected the necessary data to meet the official requirements of the Department for Housing and Urban Development (HUD), but also included bespoke questions to inform the service provision for local organisations. The city-wide street count with outreach took place over a 24-hour period. It collected the information on 70+ unsheltered unaccompanied young people in the area, which is almost double the number of the previous Youth Count! in 2013 (Trawver & Aguiniga, 2016, p. 262). The overarching lesson from this study is that making the most of local resources is central to the success of this count. Combining local resources and knowledge allowed to tailor the design and data collection strategy improving the outcome of the count and quality of the data.

4.1.2 Prevalence surveys - longitudinal and cross-sectional

Prevalence surveys measure the presence of a condition, in this case homelessness, across the general population of an area. This method can provide a reference point against which other methods for counting people sleeping rough and homelessness can be compared. This method consists of drawing a representative sample of households in a population and conducting a survey to ask residents if they have anyone staying in their home who cannot stay indefinitely (Agans et al., 2014). This question helps differentiate between young people or dependents cohabitating with someone presumed to be able to stay indefinitely, and someone staying without an invitation to stay indefinitely. Only a small percentage of respondents are required to answer beyond the first screening question- whether they have someone couch surfing in the property or not (Lohmann, 2021). Based on the results an estimate is developed for that area.

This method is particularly valued for being able to enumerate types of hidden homelessness such as sofa surfing (Rabinovitch, 2015). One of the most widely known examples of the use of this method for homelessness enumeration was in the US in Greater Los Angeles, California as part of a wider homelessness enumeration exercise. It was later replicated in Canada, France and the UK (Bretherton & Pleace, 2018).

Challenges and limitations of Prevalence surveys

One limitation of prevalence surveys is that, since people experiencing hidden homelessness are a relatively small population compared to a wider national population, finding cases of hidden homelessness in this way is a 'rare event' statistically. Because of this, very large samples will be required in order to produce a robust estimation. Such large samples are more likely present in national surveys or in a national census. Smaller samples can lead to very large confidence intervals, with estimations for a whole nation based on only a handful of respondents (Bretherton & Pleace, 2018). In addition, some harder to reach hidden homeless populations, like recent migrants, might require oversampling to appear in sufficient numbers (Lohmann, 2021).

To conduct a survey by phone, phone numbers which are geographically bound would be required to draw a sample in a particular area. For some countries there is only a landline registry with enough data to create a sample and for others this could also mean mobile telephone lines. Nevertheless, it is likely that these communication methods will cover certain parts of the hidden homeless population, specifically those staying with family and friends, and that needs to be accounted for when drawing conclusions.

A final limitation of this method is that surveys with household-based data collection exclude people without settled housing at the time the survey takes place. This might mean those experiencing long-term homelessness, those rough sleeping and those repeatedly homeless will be less likely to be identified (Lohmann, 2021). However, there are other methods such as PiT counts and service-based surveys that would be better suited for those circumstances. The novelty of this method is that it offers the possibility of estimating the prevalence of some concealed types of homelessness within a wider population.

Two case studies: Telephone prevalence surveys

A study was conducted in 2009 to estimate the total homeless population of Greater Los Angeles, California. The cross-sectional prevalence survey was based on a randomised sample of 4,288 households with landlines. It aimed to find those who were homeless but were currently living with someone housed, either sofa surfing or doubled-up. Participants were asked if someone who was homeless was staying with them. Only 16 people experiencing hidden homelessness were identified in this count, which produced an estimate of 10,800 people experiencing hidden homelessness in the total population with a large standard error of 3,421; producing wide intervals at the 95 percent confidence level (HC11 ^ 6705) (Agans et al., 2014, p. 225).

Due to the low proportion of cases identified and the imprecise estimation produced, in the following count in 2011 the researchers took a 'Multiplicity-based approach'. This approach would consider hidden homelessness as a statistically rare event. This approach was previously used by health care providers to estimate the prevalence of rare conditions in a population. It requires researchers to increase the sample's coverage and therefore make the rare event 'less rare' to allow for more robust estimations. The researchers did this by adding a follow-up question to the survey which asked respondents not only to report if someone experiencing homelessness was staying with them but also if they knew of someone experiencing homelessness staying in their neighbour's property. Respondents were finally asked how confident they were of this, only those cases who responded to be 'very confident' and 'quite confident' were included in the count.

This new approach produced a more reliable estimation of 18,622 people experiencing hidden homelessness in Greater Los Angeles, California at the time. This meant a reduction of the relative standard error from 32% to 15% (Agans et al., 2014, p. 225). However, it was not significant based on a normal distribution test (p=0.06) (Agans et al., 2014, p. 224).

A similar study was conducted in Germany which explored the feasibility of collecting data on hidden homelessness in population surveys. In a population-wide multi-topic telephone survey, respondents were asked if they had hosted friends, family or other persons over the last 12 months who had no accommodation of their own (Lohmann, 2021). This study was conducted in three different areas using a sample of landlines of approximately 1,000 for each area. The author concluded that a sample of that size was not large enough to calculate robust estimations. They recommended using a larger sample in future studies to improve the quality of the estimation (Lohmann, 2021).

Case study: The rough sleeping census for women in London

The aim of this study was to pilot a methodology tailored to women experiencing homelessness, specifically rough sleeping (Young & Hodges, 2022). It also looked to establish the circumstances and characteristics of this group and their prior journey into homelessness. The method used was a Period Prevalence Count (PPC) in the form of a short survey conducted by outreach practitioners from local organisations during the daytime. This survey was purposely short (10 questions) to avoid burdening the participants and maximising the chances of full completion. A voucher was provided as an incentive to the participants.

The study was conducted across 21 of London's boroughs over a 5-day period in October 2022. Initially it was conceived to collect data from multiple data points which included local authorities' data and service user's data. However, this proved to be challenging due to low levels of response from organisations and, ultimately, the data collected came from the PPC survey.

Although the study looked to engage a wide variety of support services working with women, not all kinds of services had the same response levels. The researchers thought this could have been because not all services conduct the same levels of outreach as it might not be in line with the kind of support they provide. For example, this was the case of services for women experiencing domestic or sexual abuse which rely on women contacting them more than being contacted through outreach.

Another challenge noted was identifying women that might be rough sleeping during the daytime, as they would not be bedded down. Outreach workers reported finding it hard to identify and engage women to conduct the survey. They noted sometimes being mistaken for being fundraisers rather than outreach workers when approaching them. Additionally, some of the women that normally interacted with the practitioners did not participate. Although some women that had never interacted with outreach workers did participate, the census is still likely to have underrepresented the total number of women rough sleeping in the area.

There was a total of 154 valid responses to the survey after removing invalid and duplicated ones (Young & Hodges, 2022). One of the main conclusions was that women were found to rough sleep in places that are more varied than the current definition of rough sleeping used in England considers, so incorporating this information could help better include women's rough sleeping in the official statistics. Second, although most of participants reported accessing housing support services, not all of them did. This reinforced the understanding multiple methods and strategies are needed to capture the complexity of the circumstances faced by women sleeping rough.

Case study: 'Journeys Home' longitudinal survey

'Journeys Home' was a national longitudinal survey of people living in Australia who were either homeless or at high risk of becoming homeless. It was launched in September 2011 and collected information on homelessness through a longitudinal survey tracking the same 1,700 people over two and half years. It was funded by the Australian Government through the Department of Social Services and The Melbourne Institute was responsible for the design and content of the survey (Bevitt et al., 2015).

Australia's current definition of homelessness includes people who are living in culturally inappropriate or severely over-crowded accommodation (Johnson et al., 2017). Thus, when designing the data collection instrument researchers were sensitive to collecting data that was not constrained by a definition of homelessness that focused exclusively on rooflessness. Data was collected on the type of accommodation in which people lived, the stability of their housing arrangements, the security of their tenure and the quality of the accommodation.

The sample was drawn from social security records (Centrelink) over the previous 10 years. This database allowed service providers to flag clients who were homeless or at risk of homelessness and provided unique identifiers. Using these records, a sample was drawn and respondents contacted. For those who opted in, data was then collected face-to-face or by telephone according to what was preferable by the respondent (Wooden et al., 2012).

The fieldwork took part over six waves, between 2011 and 2014. In each wave respondents were asked detailed questions about their housing, personal and family circumstances in the previous year. A cash incentive was provided with each agreed interview. The main challenges to this approach are the cost of maintaining a large sample of a very mobile population over the extent of the research project and managing drop out levels. Nevertheless, researchers reported sustained high rates of engagement with 84% of the initial sample participating in all the waves of the study (Bevitt et al., 2015).

Case study: US youth survey

A study conducted by Curry, et al. (2017) looked to understand the prevalence of sofa surfing among homeless youth in US and the implications it has for this population regarding vulnerability and needs for support. The study used mixed methods. Data was collected through a national survey for homeless youth conducted via mobile and landline phones. In-depth interviews were then conducted to contextualise the findings.

This study reached some similar conclusions to those of Petry, et al. (2022). First, that the prevalence of sofa surfing in the US increases with age for young people between the ages of 13 to 25 (Curry et al., 2017). Second, that there are differences between youth who report multiple types of homelessness and youth who report only having sofa surfed. Young people with experience of multiple types of homelessness were more likely to be LGBTI, Black, multiracial or Latino than those who just had sofa surfing experiences. Third, that most young people experiencing homelessness had experience of sofa surfing at some point.

One of the main limitations of this study is that the survey was self-reported for young people from 18 to 25 years, while for those aged between 13 to 17 years old the survey was answered by an adult member of the household. The authors suspect that this could have led younger participants to answer differently to avoid disclosing information to adults responding for them.

4.1.3 Service-Based Surveys (SBS)

Service-based surveys (SBS) are a subtype of prevalence surveys that were prevalent in the literature as part of large-scale counts in Canada, in the form of PPC, and as part of the methods used in Nordic countries to enumerate homelessness.

The data collection tool used to undertake these counts can take the form of a short survey which is administered when people access the community services. This screening tool standardises the collection of basic information about participants like age, gender, ethnicity and most recent accommodation, among other information. It also can collect case identifier information to avoid double counting (David Robinson, 2002).

SBSs often take place over a longer period than a PiT count, such as seven or more consecutive days to establish the prevalence of homelessness in a community. The type of support service administering the survey is not restricted to homelessness services but can include other services that homeless populations might attend in their community, such as a food bank or a church meal program. Authors emphasise that aspects such as planning and communication with the participating organisations are essential to its success. It is also important to ensure that services have the capacity and resources to participate (Hall, 2017).

Collecting data from a wide range of organisations providing front-line services has the potential to identify those in more hidden forms of homelessness who might come in contact with services that are not directly related to homelessness (Hall, 2017; Kauppi et al., 2020). The wider the range of organisations participating in an area, the better the chance of reaching hidden homeless populations that might not engage with the traditional homeless support services.

For that reason, SBSs provide the opportunity to count people that do not approach local authorities or homelessness support services (Kauppi et al., 2020; David Robinson, 2002). They can also be used in conjunction to street counts conducted in an area to be able to identify people experiencing homelessness that might have been missed by the overnight count. This can be done by surveying the services the day following the night a PiT count was conducted to ask people where they spent the previous night. Then, compare both the daytime and overnight counts to calculate people who might have been missed in the overnight count (Busch-Geertsema et al., 2016).

Challenges and limitations of service-based surveys

The main challenge for SBSs is ensuring the participation from services, especially those that are smaller and have less resources to participate in data collection (Busch-Geertsema, Benjaminsen, Filipovic Hrast, & Pleace, 2014). A wide coverage of services, beyond those providing homelessness support, is central to capturing those in hidden homelessness. Something to bear in mind is that the collection of data is restricted to the extent of service provision in an area. This can be particularly challenging in rural and more sparsely populated areas where there might be fewer services present in the community.

Secondly, the estimates produced through SBSs are restricted to people who visit agencies participating in the count during the collection period and some people who do not use services frequently may be missed. Additionally, some services only operate on specific days of the month, and again, people may be missed in the enumeration if the count is scheduled during a period where some services do not operate. To tackle this limitation, the longer the period the count is in place the more accurate it can be (Hall, 2017). Although chances of duplication increase with longer counts, this could be avoided with robust deduplication processes.

Thirdly, because the data collection is done by people whose main priority might be service provision rather than undertaking the count, there might be difficulties guaranteeing the rigour of data collection procedures from participating services. Capacity of the support services should also be considered when recruiting them to participate in the count as smaller organisations might struggle to find sufficient staff to administer the surveys (David Robinson, 2002). That being said, careful consideration should be given to avoid excluding smaller organisations that could have contact within populations experiencing hidden homelessness.

Fourthly, and connected to the previous point, these types of counts that occur over a specific time period are liable to underestimate sub-groups within the homeless population that have shorter periods of homelessness. Inversely, those who are homeless for longer periods of time are more likely to be identified by these methods (David Robinson, 2002).

Lastly, it is possible that some individuals experiencing homelessness will not be identified using this method, as they may choose not to use any services or may not see themselves as homeless. For example, some women experiencing homelessness were reported to avoid accessing mixed-gender services for concerns over their safety based on previous experiences of gendered violence. This was also true for LGBTI people. Additionally, the literature mentioned that some women who had experienced domestic abuse feared the perpetrators could track their presence back to services or someone could inadvertently reveal their presence there (Bretherton & Pleace, 2018).

Case study: The enumeration of the homeless population in the Cochrane District in Ontario, Canada

Kauppi et al. (2020) conducted a study comparing PiT counts and PPC as way of counting the local homeless populations in the Cochrane District in Ontario, Canada. This enumeration project is particularly relevant to this research as it specifically compared these methods of enumeration in terms of their usefulness for identifying hidden homeless populations. It is worth noting that the term 'Aboriginal peoples' is used in these studies to refer to the three legally defined culture groups in Canada: Métis, Inuit, and First Nations.

The study compared data collected simultaneously by a PPC and PiT count in the city of Timmins, in the Cochrane District of Ontario. The participants were asked about the different circumstances they found themselves in while homeless. The study found that staying with family and friends was the most prevalent experience among the surveyed population of the Cochrane District of Ontario, with three quarters of respondents having been in that situation at some point (Kauppi et al., 2020, p. 39). This was followed in half the cases by having slept outdoors, and in over 40% of cases by having stayed in a motel or rented a room, a third reported having slept in vehicles or having squatted and a quarter reporting having stayed at an institution or offering services in exchange of accommodation including sex (Kauppi et al., 2020, p. 39). These results are useful to put into perspective the range of circumstances that are often consistent with homelessness and the proportions in which they may present.

The study concluded that the PiT count had produced a low count in Timmins, identifying only 111 people experiencing homelessness, while the PPC identified another 431 people experiencing homelessness (Kauppi et al., 2020, p. 13). Additionally, the PiT count did not fully capture the demographic makeup of the homeless population. The PPC found an overrepresentation of Aboriginal people among the people experiencing hidden forms of homelessness in the count (Kauppi et al., 2020). The results also pointed to the fact that those that were currently experiencing unsheltered types of homelessness had, during their housing trajectory, also been in circumstances of hidden homelessness.

The authors concluded that PPCs were particularly valuable to produce homelessness counts in rural and remote areas in Canada. While it was more costly than a PiT count, it could also be conducted less frequently and targeted to rural and harder to reach populations (Kauppi et al., 2020).

Case study: Nordic countries' approach to homelessness counts

Sweden, Norway and Denmark all conduct periodical two-step counts, first carrying out a mapping of support services and then asking these services to conduct a survey of their users over a specific week (Benjaminsen, Dhalmann, Dyb, Knutagård, & Lindén, 2020). A very important aspect of this enumeration strategy is that the services reached are not just those targeting people experiencing homelessness but a much wider range of agencies and social services. This includes job-centres, drop-in cafes, parts of the health system, addiction treatment centres, NGOs, among others (A. Smith, 2015). This is called 'extended service-based count' and was found to be an effective way of obtaining information about people who are often missed in other forms of homeless counts in these countries. More on Denmark's approach to homelessness counts is discussed in Section 4.4.

Austria also carries out a count via data supplied through support services. However, in the paper outlining the Austrian approach, it was pointed out by the authors that services surveyed are limited and access to these services is also restricted to those born in Austria (Hermans, Dyb, Knutagard, Novak-Zezula, & Trummer, 2020), therefore, limiting the count of immigrant and other harder-to-reach populations.

4.1.4 Multiplier estimations: Capture-Recapture, Plant Capture and Multiple list methods

This section will explore the multiplier methods discussed across the literature: Capture-Recapture, Plant Capture and Multiple lists. All these methods use the comparison of two or more lists/observations of the same population to check their overlap and estimate the total size of a population. They do this by using a formula to multiply the results to calculate the incidence of a phenomenon in a wider population. This formula is known as a 'multiplier'.

Capture-Recapture, Plant Capture and the Multiple lists methods are underpinned by three main assumptions. First, that the size of the total population remains stable during the count. Second, that the probability of all members of the population to be included in one of the counts is the same. Third, that the probability of being included in each count/observation is different (Weare, 2019).

Capture-Recapture

The Capture-Recapture method, also referred to as 'mark and re-capture', has been used in the past to estimate the size of homeless populations as well as other harder to reach populations such as injecting drug users. This method involves fieldwork which consists of conducting an observation and 'tagging' an individual when counted, to record their singular presence. This technique is based on the conducting of two or more independent observations of the same population. These observations can be simultaneous, from two sources that represent approximately the same population, or they can be from the same source at two points in time (Williams, 2010). The rationale behind this method is that the more times a single individual of the population is tagged the smaller the population must be. The longer it takes to recapture an individual the larger the population is estimated to be (Bretherton & Pleace, 2018).

Capture-Recapture is discussed in the literature as having the potential to produce a more accurate estimate than PiT overnight counts provided the assumptions about the total population are met (Bretherton & Pleace, 2018; Weare, 2019). These counts can also be rerun periodically to produce trend data.

Plant capture

Plant Capture is an alternative use of the Capture-Recapture method which was only mentioned briefly in one identified study and does not appear to be widely used in homelessness counts. It consists of 'planting' a group of people passing as homeless or rough sleepers across the area where the count will take place. Then, the rate at which the 'plants' are counted or not is used to estimate the accuracy of the count (Bretherton & Pleace, 2018).

This method raises several issues; first, is the ethical consideration of disguising people to pass as homeless in an area where people are experiencing homelessness which can be seen as highly insensitive. In addition to that, there is the issue of the lack of clarity about what would it mean to make someone 'look' homeless, this in itself is a point of contention (Bretherton & Pleace, 2018). Finally, in the literature reviewed this method was not shown to add value to that of the original Capture-Recapture method but it did come with additional ethical complexity and fieldwork costs.

Multiple lists

The Multiple lists method, instead of collecting data through observations, uses two or more lists (often administrative records) obtained separately of the same population. These lists are compared to establish their overlap, which is then used to calculate the estimated size of the total population. Although it is a secondary method it is included here as it shares the assumptions and use of multipliers of Capture-Recapture and Plant capture methods (Weare, 2019).

This approach has the benefit of having a relative low cost as it does not involve fieldwork. However, the way the 'lists' are produced often violates the necessary assumptions. First, the lengthier the period over which data were collected the more changes the size of the population could have, violating the first assumption. Second, two lists they might have different collection periods which are not comparable. Lastly, not all population members have the same chance to be included as they might not interact with all the organisations providing the records used as lists (David Robinson, 2002).

Challenges and limitations of multiplier methods

Many of the same limitations as with PiT counts and SBS arise with multiplier methods, as they frequently rely on observation. Particularly those in relation to the extent and nature of the coverage and the time period during. which data is collected. The reason why someone is only counted once might be down to geographical coverage and the selection of the area where the survey looks for people experiencing homelessness.

In addition, there are inherent problems associated with counting people who move around physically, who conceal themselves for safety, who move in and out of homelessness or who simply do not have a fixed abode that are not entirely addressed by these methods. For example, women rough sleeping may be more likely to be missed by methods using observation for the same reasons they may be missed by street counts, because they remove themselves from sight or conceal their gender due to safety concerns (Bretherton & Pleace, 2018). (See section 5.1.1).

Relying on external observations as the main approach to counting the homeless population introduces multiple sources of bias. This can be somewhat mitigated by the level of training and knowledge of the observers, yet not fully removed. Turning observations into standardised short surveys/interviews can also mitigate this risk but this introduces new costs including the need for further logistical and analytical resources (Berry, 2007).

Another challenge of the use of multiplier methods is that they produce estimates with large confidence intervals. However, confidence intervals can be narrowed with repeated resampling, for example, by redoing the procedure monthly over a six-month period (Busch-Geertsema et al., 2016).

Lastly, the assumptions that the homeless population size remains stable during the data collection period and that each person has equal chances of being identified are somewhat undermined by the mobile and fluid nature of the homeless population and this needs to be taken into account and mitigated by the design of the study (David Robinson, 2002).

Case study: Capture-Recapture in Plymouth and Torbay

An enumeration study was conducted in Plymouth and Torbay, England using a hybrid model that combined a simple Capture-Recapture approach with longitudinal count. This involved capturing two samples of a population taken three times over a one-year period, leading to a total of six counts in each location. This longitudinal model was adopted on pragmatic grounds to improve the reliability of the counts as opposed to conducting them over a short period of time.

The homeless population was counted through the records held by local services like hostels, hospitals and soup kitchens, among others. When the records were not sufficiently robust or detailed a monitoring form was then used for the count. Individuals would be tagged using four identifiers: sex, date of birth, where they were staying at present and length of time spent in either study location. Complete data was available for 90% of the cases and for the remaining it was possible to draw conclusions from the incomplete data (Williams, 2010, p. 55). Additionally, there were random data collection quality checks in some locations.

The author of the study concluded that this method was suitable to the geography of both localities. Both being urban areas bounded by countryside and sea with a low population density which made movement in and out of the areas easier to control for. The data collected allowed the calculation of three measures of homelessness: the estimated total population derived from each set of counts; longitudinal data on change and a mean of all three sets of counts. The longitudinal nature of this study provided important data on trends as well as a comparison of each count to the overall mean which could be interpreted as an indication of a level of reliability (Williams, 2010).

4.2 Secondary research methods

This section focuses on secondary research methods used to identify and count hidden homeless populations. This means research methods which use secondary data sources, which are sources where data was collected with purposes other than the research on hidden homelessness or homelessness.

The most frequently used secondary data sources include censuses, national health records and administrative databases used by homelessness support services, also known as Homeless Management Information Systems (HMIS). Johnson (2017) argues that the integration of multiple sorts of administrative data produced by public services can offer a valuable data source for homelessness research. Most governments are already collecting increasing quantities of administrative data to be able to provide services for their constituents including data on housing, welfare, justice and health among others, so there is an opportunity to make the most of this resource.

The methods discussed below are indirect statistical estimations, predictive analysis and spatial techniques. They all have in common the use of different databases of administrative or routinely collected data to estimate or identify the presence/risk of homelessness. While care needs to be taken when drawing conclusions based on data not collected for the purpose of identifying homelessness, it can also be a cost-effective way to produce valuable insights with already available resources.

4.2.1 Indirect estimations using administrative data

Indirect estimation is a statistical method used to estimate the size of the homeless population in an area or larger population by using client records from organisations such as homelessness support service providers, social security agencies, public health services, etc. Data is collected by these services as a result of service provision that may or may not be related to homelessness. Some forms of administrative and official data used in the past to estimate the size of homeless and hidden homeless populations are homelessness support services' client databases, social security/benefits data, council tax records and household surveys (Rabinovitch, 2015).

The logic behind the use of public records or administrative service data is that people experiencing homelessness may be in contact with multiple services or organisation even if they do not approach their local authorities or specialised homelessness support services. By capturing other forms of data from multiple sources a wider picture can be painted of the prevalence of homelessness and hidden forms of homelessness, as well as the pathways into and out of homelessness (Benjaminsen et al., 2020).

One form of service records used in estimations comes from Homeless Management Information Systems (HMIS). HMIS are administrative databases used by homelessness support services to collect information from individuals on the multiple occasions they come into contact with them. These are frequently shared by homelessness services in an area, building a large pool of data about the demographic makeup and circumstances of the homeless population in it. In order for HMIS to be useful data sources, most services in the community must use it. HMIS are often suitable sources of data for longitudinal analysis of homelessness because their records are continuously updated and dated (Rabinovitch, 2015).

Challenges and limitations of re-processing administrative data

There are some limitations related to re-processing administrative data. Firstly, data sharing agreements are often needed- though not always- for these kinds of studies to take place. Secondly, while longitudinal studies are much needed to balance out the (often employed) cross-sectional data collection methods in homelessness research, monitoring of individuals through health and housing systems can be ethically complex and ethical considerations should be explored within each legal context. Thirdly, caution is required to clarify those groups who might not be included in the data pool. There are limits to their generalisability to their overall population.

Fourthly, the definition of homelessness used affects who is entitled to receive support for certain services, which in turn affects the profile characteristics of the service users. This should be considered when selecting service records as secondary data sources. The literature also suggests that it is essential to compare the makeup of the population of service users against the wider population for which the estimation is intended, to avoid producing unreliable or biased results (Deleu, Schrooten, & Hermans, 2021).

Additionally, studies that use HMIS as their main source rely on support services to detect people experiencing homelessness, but not all people experiencing homelessness approach support services. Therefore, using solely this type of data introduces the risk of excluding people who are experiencing homelessness who do not engage with homeless services and may be more likely to be experiencing a hidden form of homelessness (Bretherton & Pleace, 2018).

The strength of this approach, however, is that the existence of a large-scale national database with potentially better coverage than homeless counts can still further our knowledge on people experiencing homelessness (Roncarati, Byrne, & McInnes, 2021). By pooling data on different areas of a person's life it may be possible to spot patterns between behaviours, experiences and outcomes that can help make pathways in and out of homelessness more visible (Richard et al., 2019).

Case study: US Homeless Management Information System

A recent study by Petry, et al. (2022) used administrative data obtained from the Homeless Management Information System databases of 16 communities across the US to create a convenience sample of young people (under the age of 25). The researchers used multinomial logistic regression to assess whether demographic characteristics, homeless history, risk and victimisation, among other indicators, were associated with the place where young people spent their homeless nights (sofa surfing, sleeping on the streets or in a shelter).

Some of the main findings of this study were, firstly, that minority ethnic and LGBTI youth were at higher risk of any kind of homelessness. LGBTI youth were also more likely to sleep on the streets compared to young people staying in an overnight shelter. As youth approached legal age (18+ years) they were more likely to sofa surf. These three findings are consistent with the overall literature reviewed. Lastly, this research found a positive correlation between sofa surfing or sleeping rough and having an income and having unmet basic needs, compared to youth sleeping in shelters (Petry, Hill, Milburn, & Rice, 2022, p. 746). The researcher interpreted this finding as likely due to the role that shelters play in satisfying basic needs.

The authors point out that the need to voluntarily opt in by the communities that were included in the sampling introduced a source of bias. In addition, the data was collected with the purpose of allocating housing resources which the authors believe could have biased the responses relating to aspects like mental health, illegal substance use and risky behaviours as people might want to minimise them to increase their chances of allocation. The authors also highlight that the fear of involvement of child protection services might have affected the responses from younger participants (Petry et al., 2022).

Case study: The CHAIN database

CHAIN (Combined Homelessness and Information Network) is an HMIS multi-agency database which has detailed and comprehensive data on rough sleeping in London. It records service use by individual people experiencing homelessness over time, including services used by people sleeping rough and people in the 'wider street population'- meaning people who inhabit the streets at all times of day. The data is collected for administrative and monitoring purposes by professionals working these populations. However, CHAIN collects data based on service contacts, which means it is not a census or representative survey of all people experiencing homelessness in London (Bretherton & Pleace, 2018).

A 2018 study produced a multivariate analysis on women's homelessness using the CHAIN database. This was done by pulling the demographic characteristics and circumstances surrounding homelessness episodes for women between 2012 and 2017 in London (Bretherton & Pleace, 2018). This study focused on the trends for women experiencing homelessness and in their interactions with services. It helped widen the understanding of the complexity of the needs of women sleeping rough, including mental health needs and domestic abuse support.

It concluded that Black British women represented 20% of homeless women that were UK citizens, although that demographic only made up 3.4% of the wider British population (based on the 2011 census) (Bretherton & Pleace, 2018, p. 7). In addition, women sleeping rough were found to be more likely than men to be in a younger age bracket of 25 years or less (Bretherton & Pleace, 2018, p. 9). This is consistent with a trend reported in 2008 that the demographic composition of the homeless population of UK with it becoming younger and increasingly female (Quilgars, Johnsen, & Pleace, 2008).

Case study: The use of health and other records in predictive analytics in Canada and Scotland

One study used large-scale health databases to retrospectively predict homelessness outcomes in Ontario, Canada (Richard et al., 2019). The authors make the case for using this approach because it provides a lower cost alternative to primary data collection at national level, as well as allowing a longitudinal perspective of the trajectories of people experiencing homelessness and hidden homelessness (Richard et al., 2019).

The authors discussed the potential use of predictive analysis of routinely collected data to anticipate housing outcomes, particularly homelessness. This would be done with the intention of being able to predict and prevent negative housing outcomes before they happen based on other publicly available data. This kind of data processing technique has the capacity to spot patterns in high volumes of data, exponentially growing its processing capacity and complexity.

This approach is further supported by a study about health and homelessness in Scotland which matched and compared health data from a 15-year cohort of households that had experienced homelessness to individuals that had not in the most and least deprived areas (The Scottish Government, 2018). Comparisons of interactions with health services between these cohorts were made by looking at the number of times people appeared in these various datasets.

This study used six health datasets from the National Health Service (NHS) together with information about deaths from National Records of Scotland. There was a particular focus on mental health, drug-related health conditions and alcohol-related health conditions as those are considered to have links with homelessness. This study found that prior to a homelessness episode there were increased interactions with health services and that a peak in interactions was seen around the time of the first homelessness assessment and right after (The Scottish Government, 2018). This is a promising finding as it could lead to ways to identify people experiencing homelessness who have not approached their local authorities or support services.

Case study: Re-processing of administrative secondary data sources

Another example of research produced using existing administrative data is the research conducted by (Bramley, Fitzpatrick, McIntyre, & Johnsen, 2022) on homelessness among Black and minority ethnic communities in the UK. In this case the researchers reprocessed 10 databases consisting of multiple secondary data sources such as administrative data, previous cohort studies conducted with administrative data and surveys to produce a report on the experience of homelessness of these communities in the UK. This report concluded that when keeping other socio-economic and demographic factors constant, being Black, of a minority ethnic group and/or having a migration background still increased the chances of experiencing homelessness in the UK (Bramley et al., 2022).

This study also found that Black British communities were three and a half times more likely to experience statutory homelessness than White British people in England. People of Asian descent were found to have lower rates of statutory homelessness than Black people, yet they were at a higher risk of hidden forms of homelessness like overcrowding and living in double-up households (Bramley et al., 2022). This study also points to discrimination from landlords, both social and private, as one of the contributing factors to the higher levels of homelessness among Black and minority ethnic people (Belanger, 2013; Bramley et al., 2022; Shankley & Finney, 2020).

Case study: An estimation of hidden homelessness prevalence in London, UK

The New Policy Institute published a report that used multiple official housing records and statistics to produce an estimate of the prevalence of hidden homeless and housing need in London during the last quarter of 2003 (Palmer, 2004). This exercise focused specifically on people who lived in London and who fitted England's legal definition of homelessness but who had not been provided with accommodation by their local authority (Palmer, 2004). This was one of the first attempts to produce an estimation of hidden homelessness in England.

However, the authors emphasise the many limitations that gaps in the data from public records when trying to calculate the number of people living in overcrowded households; people at risk of eviction; people squatting and people leaving institutional care with no accommodation to go to after (Palmer, 2004).

4.2.2 Indirect estimations using census data

The use of census data to produce indirect estimations of the prevalence of homelessness in a population appeared consistently across the evidence reviewed. This could be because census databases are uniquely comprehensive of the national population and provide consistent collection of variables that allows comparison, with only a few exceptions. They are also generally available, cost-effective and have high quality standards.

Two main uses of census data were found in the evidence. The first is Australia's approach to producing a census that actively looks to include people experiencing homelessness. Although it is not a homeless population census and there is still the danger of undercounting, it has proven a successful way to include homeless and hidden homeless populations in the census data which will later make it a valuable source to estimate homelessness prevalence across the country.

The second approach is the one present in two studies from Canada, where already collected census data was re-processed to estimate housing needs and overcrowding as a form of homelessness in specific populations. These two approaches are described in more detail below.

Challenges and limitations of using census data

There are some limitations to using the national census as a secondary source of data to produce an estimate of the hidden homeless population. The first, and most obvious, is that people experiencing homelessness at the time of the census might not live in officially recognised or accessible places, so they could be missed by census counts. The second, that homelessness is in general a dynamic circumstance which the census is not designed to fully capture (Baptista et al., 2012). Additionally – even when a household could be identified as homeless, variables that would enable the identification of practices like sofa surfing and the scale of sofa surfing are not currently included in the questionnaire.

Case study: Estimating homelessness through the national census

The Australian national 2021 census of population and housing conducted by the Australian Bureau of Statistics (ABS) is used to estimate the prevalence of homelessness in Australia at the time. Homelessness is not a characteristic that is directly measured in the census, so estimates of those experiencing homelessness were derived using analytical techniques and statistical assumptions. The relevance of this case is the use of a specific strategy to ensure the participation of all people experiencing homelessness in the census, this includes people that often would be classed as experiencing hidden homelessness.

Some key elements of this approach were, first, the use of 'place of usual residence' to count people experiencing homelessness and not just the place where they had spent the night of the census to be able to understand the location and living circumstances separately. Second, the homelessness enumeration strategy focussed on early engagement and building strong relationships with all levels of government and key stakeholders. In many cases census staff was recruited directly from people in the communities. This meant that the census questionnaire was delivered by trusted members of the communities giving higher chances of engagement with the census and better quality answers. Third, support with filling the census was provided (in-person, online and via telephone) and tailored communication materials was provided and adapted to local circumstances and specific populations like youth (Australian Boureau of Statistics, 2023b).

Lastly, there were different approaches taken to reach people experiencing homelessness depending on where they were staying. The strategy included additional collection and support mechanisms for people living in three broad situations on census night, plus one residual category:

• people living 'not in a dwelling' (i.e., 'people living in improvised dwellings, tents or sleeping out' also known as 'people sleeping rough')

For people in these circumstances the collection period was extended to six days starting in the night of the census on 10^th August 2021. Specific adjustments to this collection period were made in areas affected by covid restrictions as homelessness support services operated on a different schedule then.

This population group was interviewed by staff with previous experience working with homeless populations and in locations where services are provided for people experiencing homelessness to get the most accurate accounts.

In most circumstances, people 'not in a dwelling' were counted through interview using a paper shortened version of the main census questionnaire which was designed to identify people experiencing homelessness. The ABS undertook a quality assurance process at the end of count to remove any duplicate forms, given the mobility of some people experiencing homelessness and the extended enumeration period.

• people living 'in a private dwelling' (i.e., 'people staying temporarily with other households' or 'people living in 'severely' crowded dwellings')

Most people in private dwellings completed their census using either the paper or online general questionnaire. This included people staying temporarily with other households and those who were living in overcrowded private accommodation.

• people living 'in a non-private dwelling' (i.e., people in supported accommodation for the homeless, people living in boarding houses or people in other temporary accommodation in hostels or motels paid for by the local authorities)

The Census count for non-private dwellings took two main approaches. Where the presence of the institution or temporary accommodation residence was publicly known, residents were counted as within any other institutions like hospitals or residential care. For those establishments where the location is not publicly known, such as female youth refuges, their location was not disclosed to the wider census staff and specialist staff conducted the count.

• other marginally housed groups (people marginally housed but not classified as homeless: people living in other crowded dwellings; people in other improvised dwellings; people housed in caravan parks; humanitarian migrants; housing with major structural problems or where residents are in constant threat of violence)

Previous homelessness records were used to inform data imputation of key demographic information for people in each of these four main categories. The census concluded that there were 122,494 people estimated to be experiencing homelessness on the night of the 2021 census in Australia. 55,9% of people experiencing homelessness are still some population groups who are underestimated in the census and homelessness estimates. This includes: youth experiencing homelessness; Aboriginal and Torres Strait Islander people; and people displaced from domestic and/or family violence (Australian Boureau of Statistics, 2023a).

Case study: The relationship between minority ethnic people's homelessness and overcrowding in Canada

A study conducted in Canada with 2001 census data utilised data re-processing to uncover patterns of overcrowding across immigrant communities (Haan, 2011). The author identified differential patterns in overcrowding between immigrant and Canadian-born households but concluded that the relationship between residential (over)crowding and hidden homelessness is nuanced and that there could be many explanations for residential (over)crowding which are not directly related to hidden homelessness. This study noted that residential (over)crowding- defined as more than one person per room in a dwelling- is not a consistent indicator of hidden homelessness or, on its own, directly related to ethnic minorities in Canada (Haan, 2011).

Case study: Housing needs from immigrant populations in Canada

Fiedler, Schuurman and Hyndman's (2006) study also re-processed data from Canada's 2001 census, but in this case, specifically explored housing needs and economic risk of homelessness for immigrant populations. This study concluded that secondary census data on its own is not a suitable data source to assess risk of homelessness for recent migrants at the time of the census in Canada. This is partly because recently arrived migrants might not yet have an income/social security number and they would therefore be excluded from the sample.

4.3 Innovative methods and data collection tools

4.3.1 Geographic Information Systems (GIS)

Geographic Information Systems are used for capturing, visualising and manipulating geographical data. In the example discussed below, GIS were used to map and compare the location of risks and resources in communities across the state of Maine, US. There, the GIS was used to map the risk factors relating to housing insecurity in communities across this state, which happens to be a in rural area for which data on homelessness and housing insecurity was sparce (Gleason, Dube, Bernier, & Martin, 2022).

The aim of this study was to plug the information gap on housing insecurity and hidden homelessness in these areas. It accomplished this by re-processing official data sources to produce a geographical map illustrating community risks of housing insecurity in relation to existing resources within a defined area (Gleason et al., 2022).

This study identified a number of community indicators associated with risks of housing insecurity which were then mapped against resources- mostly services- to better understand where needs and the support availability could lead to a higher risk of housing insecurity (Gleason et al., 2022). The data sources included were a national survey, geographic governmental data on the area, data on evictions and shelter and street counts (Gleason et al., 2022). The sources were selected based on quality and availability for the specific state.

The distribution of eight risk factors was measured as potential indicators of housing insecurity. These risk factors were poverty, unemployment, female headed households, other cohabitating households in the property, eviction, renter burden, mortgage burden and owner burden. These were then combined into two composite indicators: 'all owner burden' and 'all housing burden'. The authors highlight that these risk indicators were produced based on previous peer-reviewed studies, but there is a limitation due to the overall small quantity of research produced on homelessness in rural settings.

This study identified as potential risk factors for housing insecurity: high rates of poverty, unemployment, cost-burdened renters and cost-burdened mortgage holder; and concluded that these risk factors seem to be as prevalent in rural communities as in metropolitan areas (Gleason et al., 2022, p. 2004). However, rates of households being hosted by other households, single female parent households, cost-burdened non-mortgage holding homeowners and eviction were found to be comparably lower for urban areas. The researchers reached the conclusion that additional data sources would be needed to further explore this topic such as geocoded data to provide a better assessment of concealed households, substandard housing and moves within the community.

4.3.2 Information and Communications Technologies (ICT): WhatsApp

Information and communications technologies (ICT) are becoming more prevalent tools to gather data. In a study conducted in the city of Girona, Spain, the mobile app WhatsApp was used as a complementary data collection tool in a PiT count of people experiencing homelessness in the area (Calvo & Carbonell, 2017).

Squatting is a frequent form of homelessness in Spain and the researchers anticipated resistance from participants to identify as such because of fears of eviction or legal repercussions (Calvo & Carbonell, 2017). Thus, the researchers first developed a working relationship with the local homeless team and the network of volunteers with local knowledge. Based on their discussion with local homelessness support services, they reached the conclusion that WhatsApp was the appropriate ICT to use in the case. For this particular community, it was understood that at least one person in each household would have access to a mobile phone with internet connection or mobile data and as WhatsApp was known and used, the participants were actively reassured of the confidentiality of the data shared.

The study found that people were reluctant to reach out to the local support team in person but were instead more amenable to using WhatsApp to communicate with them. The results showed that 36.1% of the total data obtained from the count was collected through WhatsApp and this contributed an additional 55 people identified as homeless in the area which represented 19.4% of the total count (Calvo & Carbonell, 2017, p. 4). Because this is the first count conducted in this specific area it was not possible to compare the results to a baseline from previous waves.

4.4 Triangulation of multiple methods (and data sources)

As we have seen through the previous sections of this report, homelessness is not a monolith. It presents in multiple ways for different people and is in constant flux. That is why the more sensitive methods are to the complexities of homelessness, the better the chance to produce reliable estimates. Several authors have made the case for the use of combined methods and data sources as the most robust and reliable way of producing homeless counts and subsequent estimates (Agans et al., 2014; Bretherton & Pleace, 2018; Busch-Geertsema et al., 2016; Clarke, Burgess, Morris, & Udagawa, 2015; Hermans et al., 2020; Johnson et al., 2017). The rationale is that when multiple enumeration methods are employed, using a variety of data sources and data collection techniques, a larger proportion of the total homeless population can be counted, including people experiencing more concealed forms of homelessness.

An argument in favour of using multiple methods is that triangulating multiple estimates can counteract the individual shortcomings of each method used to produce them (Berry, 2007; Busch-Geertsema et al., 2016). This is particularly relevant to large geographic areas where there is also a benefit to using multiple methods to capture the different experiences of homelessness (sleeping rough, sleeping in places unsuitable for habitation, squatting, sofa surfing, etc.) as seen in the previously discussed Greater Los Angeles, US count from 2009 and 2011 (see section 4.1.2) (Agans et al., 2014).

This approach often uses a variety of primary (mostly surveys and structured interviews) and secondary research methods (re-processing of routine data collection sources). Using a mix of primary and secondary methods has the advantage to make the most of already existing data sources and reducing costs, while also designing collection tools that cover the aspects of homelessness that might not be already present in routine/administrative data sources. This was noted as having the potential to be methodologically robust and cost-effective and have the best chances of capturing people experiencing hidden homelessness (Agans et al., 2014; Bretherton & Pleace, 2018; Busch-Geertsema et al., 2016; Clarke et al., 2015).

The studies reviewed for this report indicate that there are some methods that are more likely to identify and count certain populations when experiencing homelessness. For example, Bretherton and Pleace (2018) made the case for using multiple data points specifically to count women experiencing homelessness as their trajectories are often not picked up by traditional data collection methods like overnight snapshot counts. In Busch-Geertsema et al. (2016) the authors make the same point with regards to young people and Robinson (2002) states that the use of a combination of enumeration methods and multiplier techniques in rural areas can provide a reliable estimation of the hidden homeless population there.

Case study: Research on the experiences of Black and minority ethnic people accessing homelessness support services in Scotland

The Homelessness Task Force commissioned a report to better understand the experiences of homelessness by Black and Minority Ethnic people (BME) in Scotland (Netto et al., 2004). The aim of this study was to map services supporting people experiencing homelessness across Scotland, examine services use and the experiences of minority ethnic people with those services. In addition, it looked to identify good practices. The study involved a range of qualitative and quantitative methods.

Qualitative methods included:

• in-depth interviews and group interviews with BME people who were currently experiencing homelessness, who had been homeless in the past or who were at risk of becoming homeless
• in-depth interviews were also conducted with diverse mainstream and BME agencies which provided homelessness services.
• focus groups supplemented by interviews with some local authorities

Quantitative methods included:

• a postal questionnaire which was extensively circulated to all identified agencies providing homelessness services in Scotland
• analysis of local authority homelessness data monitoring

A mapping exercise of homelessness services was conducted by approaching the 32 local authorities to provide information on the services within their area. The definition of homelessness support services considered was broad and a wide range of services were included. Recruitment of minority ethnic people with experience of homelessness was done through these previously identified homeless support services using information leaflets. It was also done through community interpreters who were likely to know minority ethnic people previously or currently affected by homelessness or at risk of being homeless and had English as a second language.

The use of qualitative data supported the understanding of each community's perspectives on what it means to be homeless, as these usually differ across communities. It also explored service use, best practices and the experiences of minority ethnic people of support services.

This study presented several recommendations for the understanding, better identification of homelessness among minority ethnic communities and provision of culturally appropriate services (Netto et al., 2004). The use of mixed methods and the breath of topics covered makes this study makes a valuable contribution to the understanding of the particularities of homelessness for minority ethnic people in Scotland and remains a key text in the topic. More on the hidden homelessness of Minority Ethnic people is discussed in section 5.1.4 of this report.

Case study: Denmark's biennial homelessness mapping

Another example of using multiple methods and data sources is published on the 2017 Denmark's biennial homelessness mapping report. Every two years, a mix of surveys and administrative data collection tools are used to count the homeless population across the country (Benjaminsen et al., 2020). Statistical data is pulled from the HMIS used in homeless shelters and it is processed alongside a week-long survey with all community services homeless people are likely to have contact with.

Administrative records offer longitudinal data on who use homelessness support services and survey results provide coverage of those people experiencing homelessness who might be missed from regular homelessness data collection because they do not use services related with homelessness (Bretherton & Pleace, 2018).

Case study: Belgium's approach to homelessness counts

Belgium was described as utilising a combination of data sources to calculate homelessness figures that includes: a national periodic count conducted by each city, use of administrative and social security system databases, specific data on evictions and waiting lists for social housing and the national statistics on housing quality and availability (Hermans et al., 2020). City counts are also conducted through services and collect data on nationality, country of birth and place of stay.

The paper outlining Belgium's approach by Hermans, et. al. (2020) raises two main points about the reliance in support services for the count of people who are migrants and are homeless. First, the mapping and selection of services phase has a significant influence on the results and, therefore, it vital to dedicate resources to get this aspect right. Second, is that legal status has an impact on eligibility for most services which also affects the population reached, leading to undercounts of migrants.