Accommodation-based interventions for individuals experiencing,or at risk of experiencing,homelessness

Differences in population

Bassuk, DeCandia, Tsertsvadze, and Richard (Bassuk et al., 2014) systematically reviewed and narratively reported the findings of six studies which looked at the effectiveness of housing interventions and housing combined with additional services. The interventions included Housing First, rapid rehousing, vouchers, subsidies, emergency shelter, transitional housing, and permanent supportive housing. However, authors limited the population to American families who were experiencing homelessness and so any final conclusions on the efficacy of accommodation-based interventions on the wider population of individuals experiencing homelessness are impossible to reach.

Differences in outcomes of interest

Fitzpatrick-Lewis and colleagues (Fitzpatrick-Lewis et al., 2011) conducted a rapid systematic review on the effectiveness of interventions to improve the health and housing status of individuals experiencing homeless which located 84 relevant studies. Only those studies published between January 2004 and December 2009 were included in this review and so the current review is more up to date and broader in scope. Additionally, the primary purpose of the review was to identify literature which improved health outcomes for those experiencing homelessness and so other important outcomes were not included.

Mathew and colleagues (Mathew et al., 2018) conducted a Campbell Collaboration systematic review which looks at how various interventions impact the physical and mental health of people who are homeless alongside other social outcomes. One objective listed in the title registration form is similar to the scope of the current review. Authors assessed “What are the effects of housing models (i.e., Housing First) on the health outcomes of homeless and vulnerably housed adults compared to usual or no housing?” However, the current review has a wider scope by including additional outcomes across a wider population.

A second Campbell Collaboration systematic review (Munthe-Kaas et al., 2018) assessed the effectiveness of both housing and case management programmes for people experiencing, or at risk of experiencing homelessness. The main outcomes of interest to the authors were reduction in homelessness and housing stability. Authors searched the literature until January 2016 and uncovered 43 RCTs meeting the predetermined inclusion criteria. Authors did not include qualitative research or extract data related to the cost of the interventions, which are variables of interest to this proposed review.

Differences in analytic methods

A recent review by the What Works Centre for Wellbeing (Chambers et al., 2018) included 90 studies which included clusters of Housing First (n = 47), supported housing (n = 12), recovery housing (n = 10), housing interventions for ex-prisoners (n = 7), housing interventions for vulnerable youth (n = 3) and “other” complex interventions targeted at those with poor mental health (n = 11). Authors presented a comprehensive search strategy of both commercial and grey literature, however, due to resource constraints were unable to conduct independent screening of the potential studies and therefore risk selection bias in the review. Additionally, only studies published after 2005 were included in this review and so the current review is much broader in scope. Finally, the authors' objective was to create a conceptual pathway and evidence map between housing and wellbeing and so the results were not meta-analysed but described narratively instead.

Inclusion of qualitative studies

Finally, this review also includes qualitative data, to complement the quantitative results on effectiveness, by highlighting important implementation and process issues related to the delivery and uptake of accommodation-based services. The qualitative studies included in this element of the report are drawn from CHI's implementation and process EGM and described in more detail below.

Primary outcomes

This review primarily addresses how interventions can reduce homelessness and increase housing stability for those individuals experiencing, or at risk of experiencing, homelessness.

Secondary outcomes

Secondary outcomes include:

Access to mainstream healthcare

These outcomes reflect the domains used in the EGM (White et al., 2020), with the addition of cost.

Details of study coding categories

A data extraction tool was designed by the authors and piloted by trained research assistants using EPPI Reviewer (Appendix A). At a minimum, we extracted the following data: publication details, intervention details including setting, implementation, delivery personnel, descriptions of the outcomes of interest including instruments used to measure, design and type of trial, sample size of treatment and control groups, data required to calculate Hedge's g effect sizes, quality assessment. We extracted more detailed information on the interventions such as: duration and intensity of the programme, timing of delivery, key programme components (as described by study authors), theory of change.

Alongside extracting data on programme components, descriptive information for each of the studies was extracted and coded to allow for sensitivity and subgroup analysis. This included information regarding:

Setting in which the intervention is delivered

Quantitative data were extracted at immediate post-test to allow for calculation of effect sizes (such as mean change scores and standard error or pre- and post- means and SDs or binary 2 × 2 tables). Data were then extracted for the intervention and control groups on the relevant outcomes measured, in order to assess the intervention effects.

Statistical procedures and conventions

Most outcomes reported were based on continuous variables and so the main effect size metric that was used for the purposes of the meta-analyses was the standardised mean difference, with its 95% confidence interval. Within this, Hedges' g was used to correct for any small sample bias. Where other effect sizes were reported, such as Cohen's d or risk ratios (for dichotomous outcomes) these were converted to Hedges' g for the purposes of the meta-analysis using formulae provided in the Cochrane Handbook (Higgins et al., 2019).

Most outcomes were calculated using the David-Wilson Calculator (Wilson, 2019), utilising formulae to find the effect size of several continuous data, including means and SDs. Hozo's Formula (Hozo et al., 2005) was also used to help calculate effect sizes when Interquartile range and Median data were provided.

Treatment of qualitative research

The qualitative research that was included in this review is based upon existing evidence collated through the second implementation and process EGM constructed by White et al. (2018). The EGM includes 292 qualitative process evaluations on the implementation issues associated with interventions designed to target homelessness. These are not the same studies that are included in the effectiveness EGM or included in the meta-analyses reported below. These qualitative reports were downloaded from EPPI reviewer on 10th May 2019 and screened for relevance to the current review.

The EGM categorises included studies into broad categories of barriers and facilitators to the implementation of interventions. These categories were developed by the original authors of the EGM using an iterative process and were initially based on the implementation science framework (Aarons et al., 2011). The categories were independently piloted against a small number of process evaluations and agreement was reached by researchers in the Campbell Collaboration, Campbell UK and Ireland, and Heriot-Watt University. The five broad categories are contextual factors, policy makers/funders, programme managers/implementing agency, staff/case workers, and recipients. The review team recognise that in the majority of accommodation-based interventions, more than one of the agreed categories could act as a factor that impacts positively or negatively on the effectiveness of the intervention, or both in some cases. This potential overlap reflects the complexity of the implementation of the interventions and the multifaceted evaluation tools needed within this review. For this reason, the review team decided to focus on factors that influence intervention effectiveness in order to formulate a coherent Synthesis Framework.

We included process evaluations and other relevant qualitative studies that provide data that enables a deeper understanding of why the accommodation-based programmes included in the quantitative synthesis do (or do not) work as intended, for whom and under what circumstances. We conducted a “Best Fit” Framework synthesis in order to have a highly structured approach to organising and analysing data, which can prove difficult to do with qualitative data. This method is largely informed by background material and team discussions to extract and synthesise findings. This is particularly useful given the mixed methods approach, as the quantitative and qualitative data can work in tandem to give the clearest results possible.

Descriptive account of reported accommodation interventions

As presented in Table 2, interventions varied considerably between studies, with some evaluating Housing First interventions (e.g., Brown et al., 2016; Goering et al., 2011) and others evaluating accommodation with specific services like case management (e.g., Sosin et al., 1996) and enhanced care (Milby et al., 1996). The most common aspect of the interventions was providing accommodation alongside some other form of additional service such as case management (e.g., Sosin et al., 1996), continuum of care (e.g., Gulcur et al., 2003), and other services delivered through a supportive housing approach (e.g., Lipton et al., 2000).

All the interventions aimed to improve outcomes for those individuals experiencing homelessness by focusing primarily on providing some form of housing. Although the interventions shared the same basis, the theories of change varied due to the other, additional, services that may or may not have been offered to participants. Some focused more on addressing adherence to medical care services (e.g., Appel et al., 2012; Buchanan et al., 2006; Buchanan et al., 2009) while others focused primarily on improving housing stability outcomes (Cheng et al., 2007; Lim et al., 2017; Srebnik et al., 2013).

See Table 3 for frequencies of effect sizes and number of studies which measured each of the five outcomes of interest. Table 4 demonstrates the diversity of outcomes covered by these accommodation-based approaches and provides detail on how primary study authors described the outcomes measured in the included studies. Appendix B contains a table which provides additional details on the included studies including the geographical spread of the intervention (Scattered vs. congregate), and the years in which the intervention was delivered.

Network meta-analysis

Using the available data from 21 studies which contained 59 measures of housing stability, it was possible to conduct a network meta-analysis to estimate the relative effect of different categories of intervention (described by the typology), on housing stability. These head-to-head comparisons are shown in Figure 3 and in Table 5. When the numbers in the table are negative, it means that the intervention in the row had worse outcomes than the intervention in the column. The first number denotes the point estimates, while the number in brackets present the confidence intervals. The confidence intervals can be understood as “good” and “bad” scenarios that are also reasonably in line with the data.

OPEN IN VIEWER

Figure 3

Network diagram of comparisons with the (no intervention) control group on housing stability outcomes

Two categories of interventions (Basic (Conditional) and Housing Only (Conditional)) did not have sufficient numbers of studies for head-to-head comparisons and so these are not included in Table 5.

Based on the point estimates, some important trends are noteworthy, even if only indicative as the confidence intervals for most comparisons remain wide suggesting that there is still substantial uncertainty around the plausible “good” and “bad” scenarios.

First, in the row categorised as Basic/Unconditional support, which describes interventions that offer only a bed (alongside some very basic sustenance such as an evening meal) suggests that all other available categories performed better for outcomes related to housing stability, even in the no intervention groups. Second, in the column categorised as high/unconditional support, which describes interventions than offer accommodation alongside assertive and individualised support, results suggest that this type of intervention provided better outcomes on housing stability than other available categories of intervention. Finally, we can see that all interventions performed better than no intervention, except for the group of interventions categorised as basic/unconditional.

Two of these comparisons of intervention categories were statistically significant (shaded in light grey):

1.

Interventions classified as high support and unconditional resulted in greater housing stability than interventions classified as basic and unconditional. The effect size describing this difference is 1.10

We have the most information about comparisons with the no intervention control group. The forest plot in Figure 4 shows these estimated effects, comparing each intervention with the no-treatment control.

OPEN IN VIEWER

Figure 4

Forest plot of comparisons with the (no intervention) control group (housing stability)

Forest plots, which are the graphical representation of a meta-analysis, present effect sizes in a way that provides simple interpretation of effectiveness. From Figure 4, one can quickly assess that five studies included in this analysis were not statistically significant as all have CI's which cross the line of no effect. In this review, the effect size metric used was the standardised mean difference, represented by square boxes. The size of the boxes is representative of the weight of the study, larger boxes mean more weight, while smaller mean less. The combined SMD of all the primary studies is represented by the black diamond at the bottom of the graph.

The only statistically significant comparison is with interventions that are categorised as high support and unconditional. As previously reported, this effect size is 0.62 (95% CI [0.19, 1.06]), in favour of the intervention.

Moderator analyses

Earlier, in Table 4, we see that 21 studies looked at outcomes related to housing stability. These outcomes include measures of periods of time spent homeless, stable housing, participants housed, time spent in specific residential setting, days in institution and days homeless. Table 3 reports the number of housing stability effect sizes per comparison of intervention categories. In order to conduct a moderator analysis, it is necessary to have a sufficient number of effect sizes within each comparison. Due to missing outcome data in the primary research, there is an increased risk of presenting incorrect results which appear simply through chance (false positive conclusion) (Hedges & Pigott, 2009). Thus, moderator analyses were attempted for only two comparisons with sufficient data (Table 6):

1.

Moderate support and unconditional interventions versus no intervention (10 effect sizes)

Network meta-analysis

There were 65 measures of treatment effect across 20 studies, therefore, it was possible to conduct a network meta-analysis to estimate the relative effect of different categories of intervention (described by the typology), on health outcomes. These head-to-head comparisons are shown in Figure 5 and in Table 7. As in the previous analysis (housing stability outcomes) some obvious trends emerge when considering the point estimates. However, as above, the confidence intervals which denote “good” and “bad” scenarios, that are also consistent with the data, remain wide. First, Basic/Unconditional interventions again performed worse for health outcomes than all other interventions including no intervention groups. Also, noteworthy that all interventions performed better than no intervention, except for the interventions categorised as basic/unconditional. Two comparisons were statistically significant (shaded in light grey):

1.

Interventions classified as moderate support and conditional resulted in better health outcomes than no intervention. The effect size describing this difference is 0.36 (95% CI [0.03, 0.69]).

OPEN IN VIEWER

Figure 5

Network diagram of comparisons with the (no intervention) control group across health outcomes

The forest plot in Figure 6 shows the estimated effects associated with comparing each intervention with the no-treatment control. The only statistically significant comparisons are those already identified in Table 7, that is, interventions classified as either moderate support and conditional or high support and unconditional, resulted in better health outcomes for participants compared to a no intervention control.

OPEN IN VIEWER

Figure 6

Forest plot of comparisons with the (no intervention) control group (health outcomes)

Moderate support and unconditional interventions versus no intervention

There were 16 effect sizes that compared the interventions categorised as moderate support and unconditional with no intervention control. These 16 effect sizes were from three studies and so no subgroup analysis was possible.

High support and unconditional interventions versus no intervention

There were 20 effect sizes that compared the interventions classified as high support and unconditional with no intervention control. These 20 effect sizes were from seven studies. The same subgroup analyses described above were also conducted for the health outcomes.

Study Design: For these 20 effect sizes, 11 were from nonrandomised studies and nine were from randomised controlled studies. There was no difference in effect size between non-RCT and RCT studies.

Study Quality: Similarly, when using ROB as the moderator variable, there was no difference in the mean effect size between the 10 studies rated as moderate ROB and the 10 studies rated as high.

Age: Twelve of the 20 studies included an estimate of the age of the participants, coded as an integer. Age was not significantly related to effect size magnitude in these 20 studies.

Geographical spread of housing: In this moderator analysis we have two effect sizes which are scattered site housing, and nine that are congregate. Geographical spread was not significantly related to effect size magnitude in these studies.

Background and aims

The second element of the current review involved synthesising qualitative data extracted from process evaluations included in CHI's implementation and process evaluation EGM. The purpose of this synthesis was to complement the quantitative evidence reported above and provide a better understanding of what factors influence programme effectiveness. It focused on the following question:

What implementation and process factors influence intervention effectiveness?

The typology used to construct the original EGM (White et al., 2018) was developed using a grounded theory approach piloted on 25 papers initially. This iterative process was combined with expert knowledge, ensuring that the broad concepts identified would adequately capture all papers included in the map. From the piloted typology, categories were created to include all process evaluations found during the searching period. The team in Heriot-Watt coded each process evaluation under five main analytical categories of factors or levels of influence, namely: contextual factors, policy makers/funders, programme administrators/managers/implementing agencies, staff/case workers and recipients of the programme. Using a best fit synthesis framework, it is these five analytical categories that have been used to synthesise and organise the data analysis reported in the following section.

In this way, the EGM provided an initial framework around which to synthesise the data; a framework that, for the most part, fits better than anything else. This decision also ensured that the EGM structure could be used to inform the synthesis process but also provided the team with a degree of flexibility. This flexibility became an essential component to the review as the map captures barriers and facilitators to the process of implementing interventions whereas this review focuses on the process factors that impact upon the effectiveness of an intervention.

We included process evaluations and other relevant qualitative studies that provided data to enable a deeper understanding of why accommodation programmes, in general, do (or do not) work as intended, for whom and under what circumstances. Studies were selected on the basis of providing insight into implementing accommodation programmes with a diverse range of populations and geographical locations. Studies that provided most data were selected first and additional studies added until we reached saturation. There was no overlap between the studies in the effectiveness analysis and the qualitative papers.

Framework synthesis

Framework synthesis is an approach that originates from a process of analysing primary research data to address policy concerns. The background theoretical and empirical literature help create an understanding of the issue into an initial conceptual framework, which develops iteratively as new data are incorporated and themes are derived from the data. This process was carried out in collaboration with researchers and academics in Heriot Watt University and the Campbell Collaboration (White et al., 2018). This synthesis method presents an opportunity to use a “scaffold against which findings from the different components of an assessment may be brought together and organised” (Carroll et al., 2011, p. 1). Its flexibility captures new understanding as data are incorporated into the framework.

Framework synthesis comprises five methodological stages:

1.

Familiarisation

These stages are often overlapping and may be revisited throughout the process.

The first is the familiarisation stage in which a reviewer becomes familiar with current issues and ideas about the topic, by drawing iteratively on a variety of sources (Booth & Carroll, 2015). This leads to the second stage: framework selection where an initial framework is chosen, which might be a conceptual or policy framework, logic model, causal chain or established theory that might explain the issue (Bruton et al., 2020). During the third indexing stage, studies are searched for, screened and data extracted using the initial conceptual framework. Much of this work was carried out in the development of the Implementation issues EGM (White et al., 2018). Here, studies are sorted to determine their relevance to the review questions and to identify their main characteristics. During this stage, Campbell UK and Ireland screened the process evaluations for relevance to the review. During the fourth charting stage, the main characteristics of each study are analysed by grouping characteristics into categories and deriving themes directly from those data (Brunton et al., 2020). At this stage, a process of purposive sampling (Booth et al., 2016) was completed by Campbell UK and Ireland due to the available team expertise and resources. This purposive sample endeavoured to include process evaluations spanning geography, targeted populations and types of intervention in order to exhibit an accurate representation of accommodation programmes available. The selected process evaluations presented the most “rich” and “thick” data (Booth et al., 2016) from the studies included in the map. At this stage, Campbell UK and Ireland synthesised much of the available data from the selected studies against the original agreed framework embedded in the EGM. During the final stage of mapping and interpretation stage, the derived themes are considered in light of the original research questions (Brunton et al., 2020) and in this case, policy implications. This stage has been completed in collaboration with content experts who could consider these themes in light of the available empirical and theoretical literature.

Included papers

On 10th May 2019, 246 process evaluations were downloaded from the implementation and process EGM. Title and abstract screening of these evaluations for inclusion in this review was undertaken independently by the review team and 135 papers were identified as relevant to accommodation programmes for individuals experiencing or at risk of experiencing homelessness. Papers that considered a wide variety of factors from legislation and housing markets to perceptions held by services users were initially viewed for full text screening. From the 135 papers related to accommodation that were reconciled, 10 papers were selected for synthesis using purposive sampling to create a manageable and rich data set (Austin et al., 2014; Booth et al., 2016; Burt, 2009; Busch-Geertsema, 2013; Greenwood et al., 2005; HRDC, 2003; Keast et al., 2008; Lawlor & Bowen, 2017; Pleace & Bretherton, 2013; Turner Research & Strategy, 2015; Sewel, 2016). These studies are presented in Tables 9 and 10 below.

Three process evaluations focus on using the Housing First programme to tackle homelessness. Another evaluation focuses on veterans accessing accommodation after discharging from the armed forces. One of the selected studies concentrates on people with mental health issues accessing appropriate accommodation. Three evaluations focus on young people as a target group, one of which focuses on ensuring care experienced young people move into secure and stable accommodation after “growing out” of care services; two others target a more general population. One of the selected studies are based on interventions conducted in the UK, two in Ireland, one in Australia, one across Europe and the remaining five were carried out in North America; three in the United States and two in Canada. All evaluations took place between 2003 and 2017.

The following analysis takes each of the five main analytical categories of factors or levels of influence (described above and reflected in the EGM) in turn, namely: contextual factors, policy makers/funders, programme administrators/managers/implementing agencies, staff/case workers and recipients of the programme.

Quality appraisal of included studies

The quality appraisal of the selected process evaluations was carried out using the tool developed by White and Keenan (2018) in collaboration with CHI. This tool assesses the quality of each of the 10 process evaluations by asking a series of questions regarding methodology, data analysis and usefulness of findings. This section aims to provide a synopsis of the quality of the process evaluations used in this synthesis.

The quality of these process evaluations varies across sectors, where they were published and by whom. None of the process evaluations are linked to an effectiveness study in this review; this creates issues in assessment as they may not follow the pattern of a RCT or quasi-experimental study. However, in the context of assessing how these evaluations effect the implementation of access programmes, all provide relevance in recommendations for future accommodation programmes.

Six process evaluations presented clear research questions that the programme sought to explore, while two others presented a series of aims that they wanted to achieve during the programme; two process evaluations did not present any research questions or hypothesis; this may be a result of the succinct nature of the reports and their intended audiences.

Only four of the selected evaluations discussed a recognised qualitative research methodology, such as phenomenology or the use of case studies. However, most of the implementation studies did describe some data collection methods such as semi structured interviews, survey data and focus groups with study participants, staff and stakeholders. The methods reflect the researchers desire to collect and collate rich data from service users and staff implementers about factors that influenced the accommodation programme. It is this data that can facilitate in the development of future programmes targeting homeless populations for implementers as it uncovers what works and why.

The process of recruitment was discussed fully in only one of the evaluations, presenting full eligibility criteria. Others discussed this partly or not at all in some cases. Although all evaluations were focused on factors influencing the accessibility of accommodation, some evaluations were clear that their intake and referrals included users who were not homeless (but may have been previously). Although these users were generally separated in any results presented, this does reflect the wide scope of service users that many of these organisations need to accommodate. Another issue is that only four of the studies discussed ethical considerations in any detail, however this is expected as many of the evaluations are not based on trials. The others do not provide sufficient detail regarding ethical considerations or do not report these at all.

None of the interventions have a control group to compare outcomes against. Some evaluations discuss the ethical issues of a control group with no access to accommodation and the implications of this.

In three of the evaluations, a data analysis approach was fully described and with an approach that seemed systematic and sufficiently rigorous, therefore presenting a lower ROB. The other seven evaluations did not describe a rigorous and systematic analysis.

However, all of the selected evaluations present a clear list of recommendations that were based upon the evidence collected and collated in their separate programmes. These recommendations presented valuable insights into what worked, what did not and why for managers, staff and service users implementing and availing of the accommodation programme. It is these insights that are presented as implementation factors within levels of influence in this report and will be useful to implementers of homelessness programmes in the future.

Contextual factors

The framework that this synthesis was initially aligned with describes contextual factors as those involving housing and labour markets, however, these themes essentially point towards one issue: access to sufficient and suitable housing. Within this theme, four key topics emerge: social welfare, supply, prejudice, and conditionality.

Policy makers and funders

Policy makers and funders are key stakeholders in tackling homelessness. In this section, three key themes are explored: collaborative approaches, community engagement, and sustainable funding. Each theme can influence how well an intervention is implemented. This is discussed using examples in different jurisdictions and targeting various subgroups of homeless populations.

Community engagement

Wider public and community engagement has also been identified as a key component of successful programme implementation. In their 2014 study, Austin et al. (2014) reports that while leaders in the Veterans Affairs facility were interested in the Housing First programme, they had little understanding of the challenging pragmatics associated with setting up a Housing First intervention. However, without the publicly visible commitment of facility leaders to tackling veteran homelessness the programme would have faced many more issues, particularly in the interface between Veteran Affairs and other agencies both at government and community levels. Human Resources Development Canada (HRDC, 2003) go further, reporting that one of the key success factors was that through their Supporting Communities Partnership Initiative model, devolving control of funding to a community level, with appropriate accountability safeguards in place, enables communities to mobilise together to address homelessness in their local area. This in turn can result in a significant increase in partnerships, planning and decision making. Paired with the flexible nature of the terms and conditions of the funding and requirements, this can create a culture of investment, therefore increasing community capacity. One way to mitigate this issue is illustrated in the HRDC evaluation where it was reported that because there were specific officers assigned to each of the project components, they worked closely together with the majority of staff and communities, building solid relationships and creating a culture of trust. This approach can increase buy in from managers and additional collaboration between staff and agencies. However, it should be noted that devolving funding to communities can cause problems in terms of intervention fidelity. Approaches such as this should be given careful consideration before they are embedded into a programme.

In another example, Lawlor and Bowen (2017) acknowledged that a key finding of the evaluation was the clear partnership between Limerick County Council, the agency Tusla and Focus Ireland in delivering the programme which garnered operational activity (allocations meetings every 6 weeks) and an organisational commitment to communication and problem solving. At the time of writing, the report suggested that this partnership approach was a unique working arrangement in the county; Lawlor and Bowen (2017) proposes that greater success could be achieved if this were adopted across the island of Ireland. However, Lawlor and Bowen (2017) suggests that this model requires interagency trust and understanding which should be established through joint staff meetings to encourage transparency. As Keast et al. (2008) observe, while a strong, coherent and meaningful policy framework, formulated on cross sector dialogue can facilitate greater cohesion of resources and effort (potentially resulting in more sustained tenancy and better relationships across organisations), it does not solve the problem of an under resourced service system.

Sustainable funding

The amount of funding available, the time constraints often placed on the funding (in terms of applying for it and spending it) as well as competing services and needs, all impact on the support and delivery of accommodation-based programmes. For example, Busch-Geertsema (2013) reported that in their Housing First project, overall funding was severally reduced across Europe by around 45% in comparison to the previous year. This had implications on rent subsidies, where lower rents had to be renegotiated with landlords and higher subsidies procured from elsewhere. The number of service users had to be decreased from 60 to 50 and staff capacity had to be reduced from six to four. Additionally, Busch-Geertsema (2013) suggested that given the lack of any sustainable welfare provision to cover the costs of living and housing (except for those who could receive an old age or disability pension), it was essential for participants (in Budapest) to find a job, however most of them had only a very low level of education and no formal qualifications.

Sustainable funding is a key and ongoing issue for those implementing accommodation interventions. A commitment from government and other stakeholders to provide this sustained funding over several years is essential having a continued service. In relation to funding, Burt (2009) recommends that where possible, resources should be streamlined into one funding package from different government departments. This is evident in Keast et al. (2008) where three government departments (the Department of Housing, Department of Communities and Department of Justice & Attorney-General) were involved in providing funding for the Responding to Homelessness Strategy. If this fully integrated approach to funding were implemented, it could create capacity to provide both sufficient accommodation and services for individuals experiencing homelessness. However, HRDC (2003) point to the fact that “urgent” funding needs are often in conflict with a planned, consultative approach, particularly for communities at an earlier stage in dealing with homelessness. Keast agrees, reporting that programmes driven by immediacy compounded with complex participant needs and housing shortages create an unsustainable model, particularly for funders to consider.

Programme administrators, managers and implementing agencies

People and agencies implementing an intervention hold much of the responsibility in ensuring successful outcomes for service users. The managers often lead on projects, set the tone for the culture that ultimately shapes the programme pathway and take responsibility for prioritising key targets and the resources needed for the intervention to work well. They also draw the map on how service users are referred and access the service in a timely and well managed fashion. The section below takes examples from previous interventions on some factors that have influenced why some aspects of interventions have worked well while others have not.

Staff and case workers

Staff and case workers are often the key implementers of accommodation interventions and therefore play an essential role in ensuring that the intervention works for participants; their commitment to the programme is important to gain in order for it to be implemented correctly. They often have the most contact with participants and are well placed to engage with them and other implementing agencies. It is important for implementing agencies and managers to engage with staff and workers before a new programme is rolled out in order to have the best chance of achieving their buy in. It is of equal importance that staff are given the time and space to actively engage with service users and to receive quality training that will enable skills and attributes that many will have already developed.

Recipients of the programme

A focussed consideration of the recipients of accommodation-based interventions is central to understanding effectiveness of interventions. Recipients may have particular responsibilities to ensuring success, such as a commitment to engage with the processes of an intervention or adherence to a conditionality advised by implementers. Recipients also have the right to expect good communication from project staff. They might also have particular needs that could be addressed prior to commencement of the programme, such as trust-building, peer support, flexibility around conditions, or additional and individualised support for illness.

Housing

The primary outcome of interest was housing stability. Network meta-analysis enabled us to examine head-to-head (direct and indirect) comparisons of different categories of intervention, based on the typology developed by the review team. Note that most of these comparisons are indirect, considering the relative impacts of different interventions versus control conditions to gauge their relative effectiveness. Results showed that interventions offering the highest levels of support alongside unconditional accommodation (High/Unconditional) were more effective in improving housing stability compared to basic support alongside unconditional housing (Basic/Unconditional) (ES = 1.10, 95% CI [0.39, 1.82]), and also in comparison to a no-intervention control group (ES = 0.62, 95%CI [0.19, 1.06]). These results are comparable to a meta-analysis on Housing First interventions (Baxter et al., 2019) where participants in the Housing First group spend more days stably housed than the control groups (SMD = 1.24; 95% CI 0.86 to 1.62). There were no other significant differences in effectiveness between the typology categories.

Comparing studies in the different categories of the typology also shows some important trends but two caveats are necessary. First, these results should only be considered indicative in nature given the intrinsic limitations of the indirect comparisons being made. Second, given the indirect comparison being made, the resulting confidence intervals are wide, which represent the uncertainty of “good” and “bad” scenarios that are also consistent with the data. With these caveats in place, there are very important trends that were observed.

First, Basic/Unconditional interventions might harm people: the housing stability outcomes are worse than for all other interventions, including control groups who are not receiving an intervention. The relative effect size is negative and favours control; or, in other words, comparing groups receiving Basic/Unconditional interventions against groups who are receiving no intervention at all (control condition), the Basic/Unconditional group have worse housing stability outcomes.

While many studies have demonstrated the existence of more effective interventions than basic/unconditional interventions (Fitzpatrick-Lewis et al., 2011; Lako et al., 2018), few have quantified any harm relative to doing nothing. This review provides quantitative evidence congruent with qualitative works which have described the harmful effects of hostels (Johnsen et al., 2018; Watts & Blenkinsopp, 2021; Watts et al., 2018). Notable damage to the individual includes the risk of harm from others (Watts & Blenkinsopp, 2021) and the coercion experienced through lack of alternative offerings aside from acceptance of “highly institutionalised, restrictive and intimidating large-scale hostels or shelters” (Watts et al., 2018; pg.14).

Second, interventions described as High/Unconditional support achieve better outcomes on housing stability than other available categories of intervention, including those classified as High/Conditional which have similar supports and services in place, but with conditions attached.

Third, in most cases any intervention at all (with the exception of Basic/Unconditional) will perform better than no intervention.

Health

A second network meta-analysis was possible on health outcomes. Again, there were only two comparisons that were statistically significant, and both were against a no-intervention control. In this analysis interventions categorised as offering Moderate/Conditional (ES = 0.36, 95% CI [0.03, 0.69]) and High/Unconditional (ES = 0.22, 95% CI [0.01, 0.43]) were effective in improving health outcomes compared to no intervention. These effects were smaller than those observed for housing stability. In a meta-analysis on Housing First programmes, health effects were also much smaller than housing stability effects: Baxter et al. (2019) present effect sizes on three health outcomes including emergency room visits (incidence rate ratio (IRR) = 0.63; 95% CI, 0.48–0.82), number of hospitalisations (IRR = 0.76; 95% CI, 0.70–0.83) and time spent hospitalised (standardised mean difference (SMD) = −0.14; 95% CI, −0.41–0.14). They found no clear differences between the Housing First group and the control group on outcomes related to mental health, quality of life and substance use. Thus, these results are consistent with other related evidence syntheses on the topic. Examination of the relative effectiveness of each category of the typology compared to a no intervention control showed very similar effect sizes for those categories at the more intensive end of the typology (0.22–0.27) and smaller effects at the lower, less intensive end of the typology (−0.1–0.1).

Head-to-head comparisons of studies which tested health outcomes also showed some important trends, with the same caveats described above regarding the indirect comparisons and the width of the confidence intervals. As with housing stability outcomes, Basic/Unconditional interventions might harm people: their health outcomes were worse than for all other interventions, including control groups who are not receiving any intervention. This trend across both outcomes may have important implications for practice and policy. Additionally, in most cases any intervention at all (with the exception of Basic/Unconditional) performed better than no intervention for health outcomes.