Homelessness Feasibility study - Causes of Homelessness and Rough Sleeping - GOV.UK
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
MARCH 2019 | Rough sleeping feasibility Study
Homelessness
Homelessness
Causes of Homelessness and Rough Sleeping
Feasibility study
Page 1 of 84Homelessness | Feasibility Study
Homelessness
Causes of Homelessness and Rough Sleeping
Feasibility study
Page 2 of 84Homelessness | Feasibility Study
Table of Contents
Non-technical summary .......................................................... 4
1. Scope ................................................................................. 8
1.1 Overview and aims ................................................................................................................ 8
1.2 Purposes of the model suite .................................................................................................. 9
1.3 Modelling options ................................................................................................................ 13
2. Collections of data ............................................................ 17
2.1 Overview .............................................................................................................................. 17
2.2 Evidence gaps and areas for improvement ......................................................................... 18
3. Key choices for designing a homelessness model suite .... 31
3.1 Model development ............................................................................................................. 31
3.2 Ease of use .......................................................................................................................... 33
3.3 Flexibility .............................................................................................................................. 34
3.4 Granularity of outputs .......................................................................................................... 35
3.5 Transparency ....................................................................................................................... 36
4. Considerations of model specific issues ............................ 38
4.1 Time-series models.............................................................................................................. 38
4.2 Simulation models ............................................................................................................... 42
5. Conclusions and recommendations .................................. 54
References ........................................................................... 57
Appendix: Data overview ...................................................... 58
A.1 Existing data sources .......................................................................................................... 58
A.2 Upcoming homelessness data collections .......................................................................... 80
Page 3 of 84Homelessness | Feasibility Study
Non-technical o appraise the impact of suggested
policy changes.
summary Key choices for the development
of a suite of models on
homelessness
The feasibility study explores a set of
available options for a suite of models The key message from assessing the
on homelessness and rough sleeping characteristics of different classes of
in England. This is part of a three models on homelessness is that each
series project that includes a rapid model class has specific aspects that
evidence assessment of render it more suitable for certain
homelessness and rough sleeping purposes than others. Based on this
causes in the UK and abroad as well finding, we recommend the
as a review of existing models on development of a suite of different
homelessness. Key findings from models to address each distinct
these strands of research inform drive objective rather than a single, multi-
our recommendations for developing purpose model. The suggested suite
models to estimate future trends in should include the following:
homelessness and rough sleeping and
o time-series models for accurate
appraise government policies.
short-term forecasts
Specifically, the models could be used o simple, ad-hoc simulation models
by analysts in the Ministry of Housing, for appraisal of specific policies
Community and Local Government
(MHCLG) and the Department for o complex simulation models for
Work and Pensions (DWP) to address medium to long term projections of
the following objectives: homelessness types conditional on
a broad set of predictive factors
o generate accurate short-term that are shown in the literature to
forecasts of various types of influence homelessness.
homelessness including statutory
homelessness,1 single people Time series models
homelessness and rough sleeping
The optimal solution for predicting
o project medium to long term trends levels of homelessness and rough
in various types of homelessness sleeping in the short-term is the
development of time series models
that are empirically shown to generate
11 putting this report into context, we use the term
Under
UnderthetheHomelessness
HomelessnessReduction
Reduction Act
Act2017,
2017, the definition of statutory homelessness
‘statutory homelessness’ to referhas been
to the former
the definition
recently of statutory
extended homelessness
to include has people
all homeless been (includingofficial
singledefinition
homeless and those in hidden
(i.e. homeless households in
recently extended
homelessness) to include
who turn to all homeless
Local people
Authorities for homelessness and
priority rough
needs sleeping
that apply toservices. For ease of
LAs for temporary
(including
referencesingle
and tohomeless
avoid any and those in hidden
confusion when putting this report into context, we use the term ‘statutory
accommodation), which is still universally used in
homelessness)
homelessness’who turn to the
to refer Local Authorities
former officialfor
definition (i.e. homeless households
the literature that are and
on homelessness accepted
rough as
sleeping
homeless andand
homelessness in priority need by services.
rough sleeping LAs, whichForis still universally used in the literature on homelessness and
in England.
roughofsleeping
ease referenceinandEngland.
to avoid any confusion when
Page 4 of 84Homelessness | Feasibility Study
accurate predictions in the near future. trends. However, the development of
The models are simple in that they such a model is a long-term process
arrive at short-term forecasts based that requires high levels of expertise
on historical trends and are not and substantial investment in
dependent on factors that are shown resources.
to predict or cause homelessness and
rough sleeping. While there are In the short-term, we suggest the
versions of time series models that development of simple, ad hoc
include a set of predictive factors and simulation models to provide timely
can be used to evaluate the impact of evidence-based assessments of future
policy changes, they are not an policies. These simpler versions of
optimal method for policy appraisal as economics-based simulation models
they won’t correctly identify the can be used to help quantify the net
relationships from predicting factors to effects from introducing new policies
homelessness. without having to consider baseline
trends in homelessness (in the
Simulation models absence of the policy) and the factors
that drive them.
Economics-based simulation models
project outcomes of interest Data inputs and model elements
conditional on a set of predictive
factors. They are based on a solid The rapid evidence review of the
theoretical framework that allows for causes of homelessness and rough
modelling homelessness and rough sleeping revealed that homelessness
sleeping as the outcome of complex is a complex phenomenon that
relationships between a broad set of emerges as a result of intricate
predicting factors. In theory, the interactions between a broad set of
models can produce short-term policy, economic and personal factors.
predictions – however, their outputs Policy analysts can choose the set of
depend on estimated future trends predictive factors that should be
and potential relationships between a included in the models based on the
broad set of determinants that are model’s objectives.
likely to materialise in the medium to For example, time series models can
long term. Therefore, these type of generate forecasts simply by using
models are better placed for historical series of data for the variable
appraising policies and estimating of interest. Simple ad hoc models can
long-term trends rather than include a set of variables that are
producing predictions in the short- relevant to the policy in question while
term. more complex simulation models
In the long run, it is important that usually integrate a number of modules
MHCLG and DWP can use complex to model the links between outcomes
simulation models to conduct a of interest and a broad set of
comprehensive analysis of the explanatory factors.
mechanisms driving future
homelessness and rough sleeping
Page 5 of 84Homelessness | Feasibility Study
The suggested models can be models on homelessness in England.
developed using existing sources of Moreover, the literature suggests that
data on homelessness and predictive experiences of homelessness differ
factors – i.e. administrative sources of across vulnerable groups. Therefore, it
data on homelessness and rough is important to understand the impact
sleeping collected by LAs, data from of national policies on different
surveys that are either centred around segments of the population (e.g. low-
homelessness or include relevant income households, victims of
information, and official statistics for domestic violence, immigrants, people
predictive factors. However, better with mental health and drug abuse
data can result in more reliable problems).
outputs using the same methodology.
For example, more granular and It is important that the suite of models
precise estimates of different types of can produce highly granular outputs
homelessness can be achieved if the across different levels of geographic
following data improvements are disaggregation (e.g. regions, local
realised: authorities), types of homelessness
that are driven by different underlying
o covering different types of factors (e.g. sofa surfers, concealed
homelessness (e.g. sofa-surfing, homelessness) and population groups.
overcrowding), The entire set of components of the
model suite should be developed
o linking data from various sources,
using detailed data that allow for
and
disaggregation at the geographical
o improving consistency and data level and across different population
sharing across LAs. segments.
As reliability of outputs depends on Moreover, the suggested suite of
quality of available data, improvements models should be easy to use and
in data on homelessness are maintain by in-house analysts. While
considered to be of equal priority to the development of some elements of
development of robust models. the model suite could be externally
commissioned (e.g. time series model
Key modelling choices and complex simulation model), the
departments should be able to
Patterns of homelessness and rough
operate, revise and update the entire
sleeping vary from place to place
set of components of the model suite
across England and are likely driven by
using their own resources and
interactions between a range of
expertise.
different factors specific to each area.
In this context, it is likely that national Documentation, including guidance for
policies around homelessness and model applications as well as front
rough sleeping have different impacts ends that will allow users to easily
throughout England. Therefore, operate the models, should be
regional variation is a critical aspect to provided along with the core models.
consider in developing a suite of The departments should also invest
Page 6 of 84Homelessness | Feasibility Study
time and resources to train in-house
analysts to revise and update the
models – for example, using updated
data or different assumptions about
key model parameters. It should be
noted here that developing an easily-
accessible front end and detailed
guide can often be as difficult and time
consuming as developing a model’s
core ‘engine’.
In the case of simulation models,
implementing a full modular structure
is important to ensure that even a
complex model can be accessible to
in-house analysts. A large model
should build in separate components
explicitly considering and planning for
future adjustments in the development
stage. It is important that the separate
modules are built in a consistent way
that allows different teams of analysts
to revise the model or add new
modules without having to change the
core model structure.
Finally, the development of a suite of
models that produces highly granular
outputs and considers the impact of
broad sets of determinants is a
demanding and long-term process.
Therefore, it is important that the
departments develop or retain the
expertise to design and use ad hoc
simulation models that consider a
limited set of links between predictive
factors and outcomes of interest to
conduct ex ante evaluation of potential
policy changes within limited time
frames.
Page 7 of 84Homelessness | Feasibility Study
of models to predict homelessness
1. Scope trends in the future and appraise
planned changes in broad policy
areas. Available evidence regarding
data inputs, modelling options and
1.1 Overview and explanatory factors will guide our
recommendations about the
aims development of methodologies
This feasibility study seeks to explore suitable for addressing distinct national
options for the development of a policy questions in the most effective
model, or a suite of models, that could way.
be used to assess the impacts of
Instead of focusing on a single and
Government intervention on levels of
complex model that can potentially
homelessness.
address all different objectives, we
This study is informed by three strands propose the development of a
of research that were conducted as a composite model suite that comprises
part of the wider feasibility project:2 various components. In our
recommendations, we take into
o a rapid evidence assessment of the account an array of issues related to
factors that cause various types of data inputs, outputs, resources,
homelessness in the UK and methodological considerations,
overseas, modelling choices and types of
o a review and assessment of the policies that the models should
suitability of existing methodologies consider.
that have been applied to
This section outlines the purposes that
accommodate different policy
the Ministry of Housing, Communities
purposes related to homelessness,
and Local Government (MHCLG) and
and
the Department for Work and
o an overview of existing data and Pensions (DWP) seek to
recommendations regarding accommodate by conducting
potential areas of improvement for empirical research. Moreover, we
data that feeds into homelessness identify and discuss a range of options
models. for adapting approaches to address
The findings of these exercises provide these objectives. Section 2 maps
an evidence base for identifying existing sources of administrative and
available options for developing a suite survey data on homelessness as well
other types of data that can potentially
22 is presented in a separate report under the title
Findingsfrom
Findings fromthethe rapid
rapid evidence
evidence assessment
assessment on on causes of homelessness are presented in a separate
“Review of Homelessness Models”. The overview of
report of
causes under the title “Rapid
homelessness EvidenceinAssessment”.
are presented a The review and assessment of existing classes of
existing data, evidence gaps and recommendations
models report
separate used to predict
under the and measure
title “A homelessness is presented in a separate report under the title
Rapid Evidence for potential areas for improvements in collections
“Review of About
Assessment models ofCauses
the homelessness”. The overview of existing data, evidence gaps and recommendations
of Homelessness”. of data on homelessness is presented in section 2
for review
The potential
andareas for improvements
assessment in collections
of existing classes of of data on homelessness is presented in section 2 of
of the current report.
the current
models used report.
to predict and measure homelessness
Page 8 of 84Homelessness | Feasibility Study
feed into homelessness models, We will discuss the applicability of
summarises upcoming data different classes of models in
collections and highlights gaps in the addressing these three distinct
evidence to make suggestions about objectives for various types of
potential improvements in data homelessness. Theoretically, a
collection. Section 3 discusses a set of methodological approach applies
key modelling choices that apply to equally well to all outcomes related to
the entire collection of models that homelessness since these outcomes
should be developed while section 4 are measured by variables of the same
highlights modelling issues that are type (e.g. continuous variables for
specific to the particular components population counts, probabilities for
of the model suite. Finally, section 5 estimation of homelessness risks, etc.)
sets out recommendations and For example, the same time series
considerations for developing a suite model (e.g. Autoreggressive Integrated
of models to inform policies on Moving Average – ARIMA – model)
homelessness. can handle different series of data
inputs to forecast the entire range of
homelessness types (for example,
1.2 Purposes of the inclusing single homeless, and sofa
model suite surfers) and rough sleeping. What
might change across different types of
The model suite is meant to be used homelessness are the assumptions
by MHCLG and DWP to address the about model determinants in the
following set of objectives related to sense that each homelessness type is
various types of homelessness and likely to be driven by different mixtures
rough sleeping: of causal and predicting factors.
o short-term forecasting, Our objective is not just to recommend
o projections of medium to longer- ways for producing forecasts and
term trends, and projections for broad categories of
homelessness such as statutory
o appraisal of hypothetical policy
homelessness, single homeless and
scenarios designed to influence
people who sleep rough. Instead, we
levels of homelessness.
will identify a range of options for
The discussion is centred around the generating outputs at different levels of
development of an empirical approach disaggregation (e.g. new
to forecasting and projecting future homelessness levels among former
levels of different types of care leavers, homelessness among
homelessness types under baseline black and minority ethnic (BME)
assumptions or alternative policy groups, returns to rough sleeping
scenarios that are likely to affect among people with complex needs).
homelessness (for example, policies
affecting the supply of housing or
levels of welfare support for housing
costs).
Page 9 of 84Homelessness | Feasibility Study
Box 1. Types of homelessness – a discussion about
definitions
According to the Office for National Statistics (ONS) and MHCLG, a
household or an individual is considered homeless and can apply for
homelessness support when they:
“no longer have a legal right to occupy their accommodation or if it would
no longer be reasonable to continue to live there, for example if living there
would lead to violence against them”.
Moreover, the official MHCLG definition for people who sleep rough is the
following:
People sleeping, about to bed down (sitting on/in or standing next to their
bedding) or actually bedded down in the open air (such as on the streets,
in tents, doorways, parks, bus shelters or encampments). People in
buildings or other places not designed for habitation (such as stairwells,
barns, sheds, car parks, cars, derelict boats, stations, or “bashes”).
The rough sleeping definition does not include people in hostels or
shelters, people in campsites or other sites used for recreational purposes
or organised protest, squatters or travellers.
Prior to the recent Homelessness Reduction Act 2017, the official
definition of statutory homelessness comprised three criteria:
o being eligible for assistance,
o being unintentionally homeless – ‘intentionally homeless’ are
considered the households that left a home that could have stayed in,
and
o falling within a specified priority need group – households with
dependent children or a pregnant woman; individuals who are
vulnerable as a result of mental illness or physical disbaility ; individuals
aged 16-17 years old; individuals aged 17-19 who were previously in
care; vulnerable individuals as a result of previously being in care, ΗΜ
forces or under custody; vulnerable individuals who had to flee their
home as a result of violence or threat of violence.
Page 10 of 84Homelessness | Feasibility Study
The Housing Act 1996 –provides that where an applicant meets the
above three criteria, then local authorities (LAs) have a statutory duty to
provide them with a settled home, and where this is not possible straight
away, they are under a duty to provide suitable temporary
accommodation until settled accommodation can be found. Counts of
households and people that were in temporary accommodation
following accepted homelessness applications were reported at the end
of each quarter. National statistics on statutory homelessness were
derived from these counts reported by LAs. While rejected applications
for homelessness (either because households were found not to be in
priority need or because they were considered to be intentionally
homeless) were also reported, no other information on these groups that
are considered to be non-statutory homeless was reported.
According to MHCLG, there are three sub-groups in the non-statutory
homelessness category:
o single homeless,
o people who sleep rough – people bedded down in the open air, and
o hidden homeless – people who are homelessness but are not visible
in official statistics (sofa-surfing).
The new Homelessness Reduction Act 2017, which came into force in
April 2018, leads to important changes in the delivery of homelessness
services. Under the new Act, LAs are required to offer two new duties
(prevention and relief) to all applicants that are eligible even if they are
intentionally homeless or do not fall into any priority needs category.
In this context, the new official definition for statutory homelessness has
been broadened to include the entire range of single people and
households that apply to the LAs for homelessness support (even if they
are not eligible for temporary accommodation). Therefore, the new
national statistics need to integrate figures on what previously was
considered non-statutory homelessness in addition to rough sleeping.
Developing a broader definition is critical for guiding collection of data
that cover the entire range of homelessness types including statutory
homelessness, rough sleeping, sofa surfing and concealed
homelessness (‘over-crowding’).
Bramley (2017) suggests the following two alternative definitions that are
broader in the sense that they integrate forms of non-statutory
homelessness that fall out of official statistics:
Page 11 of 84Homelessness | Feasibility Study
o Core homelessness that includes the most acute forms of
homelessness (rough sleeping, sleeping in tents and cars, unlicensed
and insecure squatting, unsuitable, non-residential accommodation,
hostel residents, users of night/winter shelters, domestic violence
victims in refuge, unsuitable temporary accommodation, sofa surfing),
and
o Wider homelessness that refers to people who are at risk of
homelessness or stay in some form of temporary accommodation
(staying with friends and relatives due to inability to find proper
accommodation, eviction/under notice to quit, asked to leave by
parents/relatives, intermediate accommodation and receiving support,
in other temporary accommodation, discharged from prison, hospital
or other state institution without permanent housing).
Finally, efforts have been made to establish a harmonised official
definition of homelessness across the UK. The Government Statistical
Service (GSS) Harmonisation Team, which is part of ONS, has been
recently commissioned by MHCLG to map the definitions of
homelessness that are used in the UK and investigate options for
developing a harmonised homelessness definition.
It was found that different homelessness definitions reflect differences in
homelessness policies and priorities in delivery of prevention and support
services across the UK countries. Moreover, information regarding the
comparability between different definitions appears to be limited.
The GSS harmonisation team has further explored a set of
homelessness definitions that are used across government bodies (e.g.
MHCLG for national statistics on homelessness, DWP for those in need
of benefits and the Ministry of Justice (MoJ) for assessing
accommodation of ex-offenders) and non-government organisations (for
example, core and wider homelessness definitions used for CRISIS
projections of future trends in homelessness).
Variations in homelessness legislation and operational differences when
applying the definitions to produce homelessness statistics were also
examined across UK countries.
Findings from this research revealed that introducing a harmonised
definition would require changes in legislation and data collections
across the devolved nations that are not straightforward to implement.
Therefore, the GSS harmonisation team recommended that a
conceptual framework for homelessness should be created in order to
map different definitions and data collections in the UK and improve
comparability of existing statistics.
Page 12 of 84Homelessness | Feasibility Study
model that can address all purposes.
1.3 Modelling options Moreover, the development of a
Accounting for the complexity of the complex comprehensive model
phenomenon under analysis and the requires time and resources while
theory underpinning homelessness as smaller models can be designed in the
well as distinguishing between the short term to address immediate
distinct purposes of short-term policy objectives. The design of small
forecasting, long-term projections and ad hoc models can also be seen as a
policy appraisal are key considerations critical step to the long-run process of
when developing models around developing a robust complex
homelessness. simulation model that can be used for
the entire set of policy purposes
For instance, the complexity of associated with homelessness and
relationships between different factors rough sleeping.
– such as the interconnections
between the housing and labour For example, the CLG-Affordability
markets across English areas – that model – a complex simulation model
influence homelessness levels is an that estimates housing affordability in
important element that should be England as the outcome of a number
considered when projecting long-term of interconnected determinants
homelessness trends. However, (NHPAU, 2009) – consists of a set of
including assumptions about such simpler simulation models on house
relationships to estimate trends in the prices as well as housing demand and
short-term is likely to result in supply that can be used separately.
decreased forecasting accuracy. These models have been also utilised
in the development of the components
A key choice that needs to be made is of the Sub-Regional Housing Market
between developing a single, large- Model (SRHMM) developed by
scale and complex model that Bramley and Watkins (2016).
integrates multiple features or a suite
of simpler models that are used to In the review and assessment of
accommodate distinct purposes. A classes of models that are used to
number of issues, including the measure and predict homelessness,
applicability of the available we identified a set of methodologies
methodologies as well as the costs that can be applied to address policy
associated with each option, should purposes around homelessness. The
be considered in informing the choice key take-away from the model review
of the optimal strategy. and assessment is that there is merit
to applying different models for
Specifically, the costs of developing different purposes.
and using a large-scale, complex
model that integrates various features As shown in figure 1, which
to model all possible links and summarises the main findings of the
interdependencies between related model review and assessment, each
factors and homelessness types might class has particular statistical
exceed the benefits of having a single properties that makes it more suitable
Page 13 of 84Homelessness | Feasibility Study
for some purposes than for others. For Projections of trends under baseline
example, time-series models are assumptions and evaluation of
simple trend-based methods that changes in homelessness levels under
generate accurate forecasts of alternative policy scenarios are
outcomes of interest in the short-term separate exercises.
based on the underlying assumption
that patterns that existed in the past Though these objectives can be
will continue into the future. While they covered within the same overall
can be applied to estimate medium to framework (for example, a simulation
longer-term trends, they lack the model can do both), different versions
theoretical framework that is needed of models that fall within this class can
to account for relationships between be used to address these two distinct
explanatory factors and outcomes that objectives.
play out in the long-term. Essentially, simple ad hoc models can
3
Based on our findings and the above be used to quantify the impact of
discussion, we recommend the specific policies compared to a
development of a flexible suite of baseline ‘do nothing’ scenario. The
models that will comprise a set of estimation of additional effects from
methodologies applied to address launching a new policy does not
different objectives instead of a necessarily require considering the
complex, large-scale model. baseline levels of homelessness and
Specifically, we suggest that models rough sleeping. SRHMM (Bramley and
from the following two broad classes Watkins, 2016) is an example of a
should be applied to accommodate comprehensive simulation model that
MHCLG and DWP policy objectives: projects housing needs, including
homelessness, under composite
o time-series models for short-term policy and economic scenarios.
forecasting,4 and
o economics-based simulation
models for medium to long-term
projections and policy appraisal.
3 figure 1, from
machine learningand
techniques arethe
anset of
3 Our recommendations
Our recommendationsfor forsuitable
suitablemodels
models are based on findings
are reviewing assessing
model classes that have been used to predict and measurealternative
types of option for generating
homelessness. For a accurate
detailed short-
based on findings from reviewing and assessing the term
discussion
set of modelabout
classesthethat
characteristics
have been used of existing models, see
to predict theforecasts.
“Review of However,
model ofthe reliability of machine
homelessness”
report. learning outputs relies on the amount and level of
and measure types of homelessness. For a detailed detail of data on homelessness. Therefore, applying
4
discussion aboutinthe
As discussed thecharacteristics
model reviewofand existing
shown in figure 1, such
machine learning
models techniques
to English are ana alternative
data (facing number of
models, see
option for the “Review
generating of Homelessness
accurate short-termModel”
forecasts. However, the reliability
limitations that areof machineinlearning
discussed sectionoutputs
2 of this
report.
relies on the amount and level of detail of data on homelessness. Therefore,
report) is applying
likely to be such models to
suboptimal.
4English data (facing a number of limitations that are discussed in section 2 of this report) is likely to be
As discussed in the model review and shown in
suboptimal.
Page 14 of 84Homelessness | Feasibility Study
Box 2. Other policy objectives and methods to address
them
The following policy purposes can be also addressed by applying
empirical models:
o identifying homelessness risks for households and individuals and
single people,
o measuring homeless groups that are not straightforward to capture,
and
o evaluating existing policy interventions that address homelessness.
The models used to accommodate these purposes include:
o homelessness risk models,
o non-standard sampling models such as the capture-recapture method,
and
o models developed to quantify intervention (treatment) effects for
participants in the period following the intervention.
The focus of this feasibility study is not to recommend ways to explicitly
address these additional objectives. However, the above methods can
potentially complement the main models developed to predict
homelessness levels and appraise policies. For example, outputs from
homelessness risk models can be integrated into larger and more
complex policy models that simulate homelessness outcomes under
different scenarios.
Alternatively, these methods can be used as stand-alone policy tools
developed outside the main models. It may be worthwhile for MHCLG to
sponsor a project that brings together expertise from LAs that use
homelessness risk models to develop a common approach to identifying
households and single people that are in priority need for homelessness
prevention services. Such an approach can also be adopted to assess
differential impacts of the implementation of central policies in different UK
regions. For example, policies around private rent prices (e.g. housing
benefits) and housing supply (e.g. investment in council housing) are
expected to exert significant impact on homelessness risks in London
Boroughs where high private rents and shortages in supply of council
housing are important drivers of homelessness. On the other hand, such
policies are not expected to have a similar impact in Northern England
where access to social housing does not appear to be a major issue
(Fitzpatrick et al. 2018).
Page 15 of 84Homelessness | Feasibility Study
Capture-recapture methods, which use a set of sampling techniques to
estimate the size of populations that are elusive and thus not
straightforward to measure,1 can be implemented to guide new data
collection that can improve the outputs of models. Alternatively, these
methods can be applied to existing data to produce reliable counts of
populations that are not easy to measure such as sofa surfers and
households in concealed homelessness, improving area-based counts
of homelessness groups.
Finally, ad hoc models that identify treatment effects can be used to
assess the effectiveness of previous or existing interventions. For
example, a similar strategy was adopted to measure the impacts of
realised changes in Local Housing Allowance (LHA) on a number of
outcomes, including LHA entitlements, contractual rents and types of
properties claimants live in. (Beatty et al., 2014). A difference-in-
differences model was applied to administrative data on housing benefits
claims from the Single Housing Benefit Extract (SHBE) to compare
trends in outcomes (for example, rents and types of properties) for
groups who moved into the new LHA system to groups with similar
characteristics that have not rolled onto the new system yet.2
Notes
1
For a more detailed discussion about capture-recapture methods see the “Review of models
of homelessness” report.
2
For a comprehensive outline of the model developed to measure the impact of LHA reforms,
see the report by Brewer et al. (2014).
Page 16 of 84Homelessness | Feasibility Study
2. Collections population groups) to produce
granular short-term forecasts.
of data Previous collections of administrative
data on homelessness (collected using
P1E forms for people in temporary
accommodation), which are
2.1 Overview aggregated at the local authority level,
included a limited set of background
There are three types of data which
information. The Homelessness Case
can be used to project homelessness
Level Information Classification (H-
and rough sleeping in the future and
CLIC) system for data collection which
evaluate the effects of policies aimed
will replaces the P1E forms, collects
at tackling homelessness and
household case level data providing
supporting people in need: 5
more detailed information on the
o administrative data on causes and impacts of homelessness,
homelessness and rough sleeping long-term outcomes for homeless
collected by LAs and reported by households and what works best for
MHCLG at frequent time intervals, preventing homelessness. Moreover,
administrative data on rough sleeping
o data at the household and/or collected using the Rough Sleeping
individual level from large scale Evaluation Questionnaire (RSEQ)
household surveys which include include information on individual socio-
information on homelessness or economic characteristics that have
surveys that were designed to been shown to be associated with
explicitly cover homelessness and homelessness (e.g. financial strain,
rough sleeping experiences, and use of other public services, mental
o administrative data (for example, health problems, etc.) Therefore,
official statistics) on homelessness forthcoming collections of
determinants – e.g. housing and administrative data can be used to
unemployment benefits, housing estimate time series models that
supply, private rents, demographic include limited sets of explanatory
trends, health indicators, key variables in addition to historical values
economic variables, etc. of the variables of interest (multivariate
models).
Time series models can be applied to
series of administrative data on H-CLIC and RSEQ data can be also
homelessness and rough sleeping that used to measure the effects of
are reported frequently (e.g. every predictive factors on different types of
quarter). These models can handle homelessness at the first stage of
large series of data inputs (for simulation models. Survey data and
example, across LAs and for particular other sources of statistics on key
55 be found in the appendix.
AAdetailed
detailedoverview
overviewof of existing
existing data
data sources
sources cancan be found in the appendix.
Page 17 of 84Homelessness | Feasibility Study
determinants can also useful for extrapolation of missing predictors
estimating homelessness projections using other observable characteristics)
using simulation models. Individual might lead to decreased output
data drawn from surveys can feed into accuracy. For instance, if data is not
components of simulation models to available on a predictor that is highly
quantify behavioural responses to correlated with homelessness such as
changes in important predictive income, we would have to use an
factors. Detailed individual data from observable proxy such as socio-
surveys are necessary for developing economic status or educational
models that can produce granular achievement to approximate individual
outputs – e.g. micro-simulation income. We would then quantify the
components that produce link from income to homelessness
distributional outcomes or generate based on this approximation, which
projections for subsamples with would result in reliability losses in our
specific characteristics. Data on other homelessness estimates conditional
explanatory variables are also used in on income.
simulation models to arrive at
homelessness projections conditional In the case of homelessness and
on future trends in determinants.6 rough sleeping in England, existing
sources of data on outcomes of
In principle, selecting suitable interest and their determinants are
methodologies to project outcomes of adequate for applying models to
interest and evaluate policies does not predict future levels of homelessness
depend on data availability and quality under composite policy and economic
in the sense that there is no merit in scenarios. However, improving the
developing different methodologies for quality of existing data or collecting
different data. Analysts select the new detailed data on homelessness
methods they will use from a set of and rough sleeping will certainly
existing options and rely on available influence the model outputs – more
data to accommodate policy detailed data lead to more reliable
objectives. When data is imperfect or outputs under the same empirical
not available, they make assumptions design.
to address the limitations imposed by
lack of data or data of low quality.
For example, when detailed data on 2.2 Evidence gaps
other life domains of homeless people and areas for
are not available, assumptions are
used to compensate for missing improvement
knowledge about personal
In this section, we highlight potential
characteristics that might influence
areas for improvement in data
paths in and out of homelessness.
collection based on gaps that we have
Such assumptions (including the
66 predictive factors.
Seebox
See boxA1
A1ininthe
theappendix
appendixforforforecasts
forecasts
of of predictive factors.
Page 18 of 84Homelessness | Feasibility Study
identified in existing data sources on contribution of broad policy areas to
homelessness and rough sleeping in reduction and preventions of
England. New collections of data and homelessness.
enhancements to already existing
systems for gathering information will Covering all homelessness types
result in more reliable estimates of
future trends in homelessness under The new Homelessness Reduction Act
alternative policy and economic 2017 required local authorities (LA) to
scenarios. More detailed evidence of meet two new duties (relief and
homelessness experiences at the prevention) to all those affected,
individual level will contribute to a regardless of priority need or
better understanding of the causes intentionality.7
and impacts of homelessness as well Following this major change in policy,
as what works best for preventing and it is important that LAs gather
reducing homelessness. information about types of
We also consider the importance of homelessness in addition to the former
suggested enhancements in data as definition of statutory homelessness –
part of developing a comprehensive for example, sofa surfing, squatting
evidence base that will result in more and living in hostels and other types of
reliable estimates of homelessness short-term or emergency
and rough sleeping. We categorise accommodation.
them in two groups: The development of comprehensive
o top priority – data that are definitions of various homelessness
necessary for generating robust types is central to the design of a
projections of various systematic recording of homelessness
homelessness types, and types that covers the entire range of
homelessness experiences in England
o further priority – data that can add – for instance, single people
depth but are not central to homelessness, rough sleeping and
achieving the aims of a suite of sofa surfing. A common and
models around homelessness. comprehensive description of what
homelessness is and which groups of
people are owed support by public
2.2.1 Top priority services in England will guide the
collection of consistent data on
In this section, we discuss homelessness outcomes of interest.
recommendations for improving data
on homelessness that are critical to Examples of collecting data on various
conducting a robust empirical analysis homelessness types are the additional
of homelessness trends, pathways in modules and questions included in the
and out of homelessness and the Rough Sleeping Questionnaire as well
7
7 For a detailed discussion about the Act see here: http://www.legislation.gov.uk/ukpga/2017/13/cont
For a detailed discussion about the Act see here:
ents/enacted
http://www.legislation.gov.uk/ukpga/2017/13/contents/enacted
Page 19 of 84Homelessness | Feasibility Study
as well as in the H-CLIC form for data array of paths to the incidence of
collection. The Rough Sleeping social problems such as
Questionnaire includes questions that homelessness.
capture past experiences of sofa
surfing in addition to rough sleeping. It Poor linkage of data in the English
could be administered to all local context is a major limitation to a
authorities in England and become comprehensive analysis of
part of official statistics. Moreover, homelessness that could contribute to
collecting data at regular time intervals a better understanding of the problem,
– for example, in annual or bi-annual its causes at the personal, economic
waves – as well as adding a and policy level and what policies are
longitudinal element to data collection needed to tackle it.
would improve statistics on rough Administrative data covering a number
sleeping and contribute to a better of areas including welfare benefits,
understanding of individual health and use of public services can
experiences. The H-CLIC form is be linked to other administrative data
completed by all local authorities and on homelessness and rough sleeping.
includes a question about last settled For example, the Single Housing
accommodation and type of Benefit Extract (SHBE) dataset,
accommodation at the time of the collected from LA records, is the key
application.8 administrative source of monthly data
on housing benefits claim. This
Data linking contains data on household type and
demographic characteristics, amount
Using datasets that comprise linked of monthly rent, share of the rent that
administrative data from distinct is covered by Local Housing
sources that cover large numbers of Allowance and type of
areas (e.g. benefits, health, institutional accommodation. Linking such benefit
history) is an important tool for data to data on people who are either
research that aims to understand homeless or at risk of homelessness
complex social issues and inform would allow analysts to identify the
policy. It allows for capturing links contribution of housing benefits to
between a broad set of predictors and homelessness prevention.
outcomes of interest and mapping the
88 LAhomelessness
homelessnessservices
services applicants Service accommodation; no fixedatadobe;
LA applicants areare asked about the
asked type of their accommodation the time of the
caravan/houseboat. In the cases where the
application.
about the typeThey canaccommodation
of their choose between: at owner-occupier;
the time shared ownership; private rented sector; council
applicants report that their current accommodation
oftenant; registeredThey
the application. provider tenant; between:
can choose Armed Forces accommodation; tied accommodation, looked after
is not their last settled home, they are asked about
children replacement;
owner-occupier; sharedliving with family;
ownership; privateliving with friends; social rented supported housing (or hostel);
rented
refuge;council
rough tenant;
sleeping; homeless on departure their accommodation when they were last settled in
sector; registered provider tenant; from institution
order
(custody/hospital);
to capture routes
temporary
into homelessness. The
accommodation; student
Armed Forces accommodation; tiedaccommodation; National Asylum Support Service accommodation; no fixed
adobe; caravan/houseboat. applicants can choose between owner/occupier;
accommodation, looked after In the cases
children where the applicants report that their current accommodation is not
replacement; shared ownership;
their with
living last family;
settledliving
home, they
with are asked
friends; socialabout
rentedtheir accommodation when theyprivate
were rented sector;
last settled lodging
in order to
capture routes into homelessness. The applicants can (not with
choose family/friends);
between council tenant;
owner/occupier; shared registered
ownership;
supported housing (or hostel); refuge; rough Provider tenant; living with family tenant;
or friends; looked
private rented
sleeping; sector;
homeless lodging (not
on departure fromwith family/friends); council
institution tenant; registered Provider living with
family or friends; looked after after children placement; social rented
children placement; social rented or supported housing; tied accommodation; or supported
(custody/hospital); temporary accommodation; housing; tied accommodation; Armed Forces
Armed accommodation;
student Forces accommodation.
National Asylum Support accommodation.
Page 20 of 84Homelessness | Feasibility Study
For example, the Public Health o data on groups who are vulnerable
Outcomes Framework sets out because of physical and mental
desirable health outcomes at the health issues drawn by Mortality
national and subnational level and Statistics, Mental Health Minimum
measures health indicators across LAs dataset, Hospital Episode Statistics
in England. The dataset also includes and the Health Improvement
two indicators on homelessness that Network.
potentially allow for modelling links
between physical and mental health There are various considerations
outcomes and homelessness at the concerning issues related to technical
LA level. However, the indicators only and legal aspects of the data linking
capture statutory homelessness at the process. An important issue is
LA level, hindering the assessment of anonymisation of data and security of
links between health outcomes and information. Explicit guidelines and
other types of homelessness – such protocols should be put in place to
as sofa surfing and rough sleeping. ensure that it is not possible for
Given that mental health appears to be analysts who use the dataset to link
a major determinant of rough sleeping, data to people. For example, the
there is merit in expanding the number of attributes included in the
accommodation response category in compilation of administrative data is an
the Public Health Outcomes issue to consider – a wide variety of
Framework to capture other types of attributes could lead to the
homelessness and link the identification of specific service users
observations to official statistics on in small LAs, where limited numbers of
rough sleeping or administrative data people experience homelessness.
collected using the Rough Sleeping Despite the variety of issues that need
Questionnaire. to be considered, linking existing
Other sources of data that contain sources of data could be a more
information on accommodation types, straightforward and less costly – in
including homelessness, that could be both resources and time – alternative
linked to homelessness data, such as to expanding existing data sources or
H-CLIC, are the following: designing new collections to capture
additional information about people
o data on care leavers aged 17-21 who are either homeless or at high risk
years old drawn by Children of homelessness.
Looked After in England,
o data on prisoners drawn by
Accommodation Status of
Prisoners and Police Records,
o data on groups of drug treatment
services drawn by National Drug
Treatment Monitoring System, and
Page 21 of 84Homelessness | Feasibility Study
Box 3. Steps toward data linking in England: the Rough
Sleeping Evaluation Questionnaire (RSEQ)
The Rough Sleeping Evaluation Questionnaire (RSEQ) was introduced as part of
the recent MHCLG initiative to tackle the most severe form of homelessness – i.e.
rough sleeping. The new instrument for data collection contributes to existing
approaches by collecting detailed data on individuals’ past and current
experiences of rough sleeping and capturing a wider set of factors that are
related to such experiences, including support needs, feelings and attitudes and
health indicators.
In addition to this contribution, the new method goes beyond prior approaches
to data collection by proposing a scheme for data linking across administrative
datasets. Personal details of service users interviewed with the RSEQ – such as
names, date of birth, and national insurance number (if known) – are linked to:
o administrative data on receipt of welfare benefits (DWP),
o criminal justice system records (MoJ),
o administrative data on statutory homelessness applications collected by LAs
(MHCLG),
o health care services use (NHS Digital),1 and
o alcohol and drug treatment use (PHE).
The output of this process is a comprehensive dataset that includes detailed
information about a broad set of areas – history of rough sleeping, statutory
homelessness applications, support needs, contact with the criminal justice
system, receipt of welfare benefits, healthcare use and participation in substance
use treatment – but excludes service users’ personal details.
Assembling such detailed lists of administrative data for users of homelessness
prevention and treatment services is important for understanding the needs of
people who sleep rough or are homeless and assessing wider costs of
homelessness that potentially exceed the costs of delivery of homelessness
services alone.
Notes
1
Linking RSEQ data to information on public health service use is likely to be
challenging. Evidence from the Homelessness Link survey on the health
outcomes of homeless people shows that while 90% of the 2,500 surveyed
homeless and rough sleeping individuals are registered with a GP, the rough
sleeping groups use GP services the least. See here for more information:
https://www.homeless.org.uk/sites/default/files/site-
attachments/The%20unhealthy%20state%20of%20homelessness%20FINAL.pdf
Page 22 of 84Homelessness | Feasibility Study
Data from people in households that have been assessed as homeless
by Scottish local authorities (LAs) were linked to a number of health
datasets covering the following areas: accident and emergency
attendance; alcohol-related admissions; drug misuse-related
admissions; emergency admissions related to injury and poisoning;
psychiatric admissions; and non-attendance at outpatient appointments.
LAs in Scotland do not normally reveal personal information of applicants
when sharing data with government departments. For the purpose of
this project, all LAs were asked to submit personal identifiable
information for people who were considered homeless or at risk of
homelessness to the National Records of Scotland (NRS) Indexing
Service.
A dataset including personal information about the applicants – such as
homelessness application number, name, gender, date of birth,
postcode and local authority code – was created particularly for the
purpose of data linking. Using the application numbers, this new dataset
could be linked back to the homelessness datasets assembled by
Scottish LAs.
In order to match homelessness with health data, a ‘separation of
function’ approach was adopted to ensure that no single organisation or
individual had access to the entire range of datasets required for this
project. A third party (the NRS Indexing Service) matched the
homelessness dataset that was created for the purpose of this project
with the Research Indexing Spine (RIS) – a population compiled by NRS
that uses information drawn from general practitioner (GP) registries at a
single point in time (snapshot). The NRS Indexing Service performed the
matching only using personal identifiers across the datasets – access to
the rest of the data was restricted. Each matched individual was then
assigned the Community Health Index (CHI) number that tracks
individual usage of health care services.
When matching was completed, the matched results were combined
with the rest of the data and the personal identifiers were removed.
Analysts accessed this secondary dataset in a separate and secure
environment.
Page 23 of 84Homelessness | Feasibility Study
Box 4. Best practices in data linkage: the Scottish
example
The Scottish government has recently adopted a strategy to promote
better use of existing administrative data to understand important social
and economic issues and evaluate policies.
Data linking is central to this new approach, which draws on a thorough
Data Linkage Framework established in 2012 to promote collaboration
and best-practice sharing among key public sector organisations that
collect and handle registry data.
A set of guiding principles has been developed to “support the legal,
ethical and efficient use of data for linkage purposes within a controlled
and secure environment”.1 The principles set out important priorities
and considerations related to acting in the public interest, transparency,
privacy (consent, anonymisation and security of individual data), data
access and consequences when these principles are disregarded.
Efforts have been made to ensure that linked administrative data are
anonymised and secure, personal information is protected, and
individuals cannot be identified in the datasets. Several anonymisation
methods are applied, including complete anonymisation, which
excludes all identifiers of personal information from the datasets, and
pseudonymisation, where identifying fields (such as names) are
replaced with artificial identifiers (such as unique serial numbers).
Moreover, safe havens were launched as a way to ensure privacy –
these are secure environments where researchers have access only to
the anonymised segments of secondary datasets relevant to their
research.
Homelessness data linking
One example relevant to analysing homelessness is linking data on
homelessness to national-level health datasets. Homelessness data
were linked with individual health indicators in order to quantify the use
of health services by homeless groups in Scotland (Waugh et al., 2018).
Notes
1
For more information about guiding principles for data linkage in Scotland see here:
https://www.gov.scot/Topics/Statistics/datalinkageframework/GuidingPrinciples
Page 24 of 84You can also read
