Today NRS published experimental statistics on homeless deaths in Scotland for the first time, in response to user demand.
Similar information was produced by the Office for National Statistics (ONS) for England and Wales in 2018, and there has been considerable demand since then from a wide range of stakeholders for comparable Scottish figures. We have worked closely with ONS in the development of the methodology used for the Scottish figures.
We hope that these statistics will provide useful information to policy makers in central and local government and add to the evidence base on homelessness within Scotland and the factors which impact upon it.
The methodology used to produce these figures is still being developed, and we will encourage feedback from users to help us develop and improve the statistics further.
Julie Ramsay, the NRS statistician responsible for publishing the Scottish statistics, explains how she and her colleagues approached the task.
Methodology
Identifying whether a person was homeless when they died is not straightforward using the information recorded at death registration. There is no specific question on the death certificate asking if a person was homeless at the time of death.
To compensate for this lack of information we searched death registration records for any evidence that the deceased may have been homeless at the time of their death.
We used a number of search strategies to identify these deaths and then applied a statistical method called capture-recapture modelling to estimate the likely number of additional deaths which were not identified via these searches.
Search Strategies
We searched death registration records to see whether the terms “no fixed abode”, “shelter”, “homeless”, “rough sleeper” or “no address known” were recorded either as the place or residence, place of death or in the registrar notes.
We also got information from local authorities on addresses used to provide temporary accommodation for homeless people, then identified deaths where these appeared either as the usual residence or the place of death.
Death registration records also contain an institution code where the person died in a hospital, care home, hostel or other type of institution, so we were able to identify deaths which had occurred in a homeless hostel or shelter.
We searched for deaths where the place of death was recorded as being in a hospital and there was no information provided for the place of residence. This was an attempt to identify homeless people who may have been found unwell on the street and taken to hospital where they subsequently died.
As it is possible for people (particularly elderly people) to spend a long time in hospital prior to their death, we excluded any deaths where the deceased had been in hospital for more than a year.
Quality Assurance
All records identified by one or more of these searches were checked individually to prevent errors. For example, the search for the term “shelter” identified some records where the deceased lived in a sheltered housing complex. These records were excluded.
For searches one and two, where we found a match with one of the addresses being used as temporary accommodation by a local authority, we asked them to confirm that the address was being used as temporary homeless accommodation at the date of death.
A lower age cut-off of 15 and an upper age cut-off of 75 was applied to the data. This decision was driven by evidence that deaths of elderly people who had spent time in an institutional setting for some time prior to their death, and for whom no prior place of residence was recorded, could have erroneously been included.
How we calculated the estimates
The estimation was carried out using the Rcapture package in the R programming language. The calculations estimate the likely size of an unknown closed population based on multiple captures (searches).
The model looks at how many deaths were identified in each of the five searches and the degree of overlap between searches (it is possible for a death to appear in more than one search). It then provides an estimate of the likely number of deaths which were not picked up in these searches to give an estimate of the total number of homeless deaths.
The model we have used is a robust but conservative model, so the figures produced should be taken as the lowest probable estimates and it is likely that the true number may be higher.
Definitions of homeless
The identification of homeless people for the purposes of these statistics is not based on an existing definition of homelessness but is based on our ability to identify such individuals in the death registration records.
The records we have identified are mainly those people using emergency accommodation such as homeless shelters, hostels and temporary homeless accommodation at the time of their death.
In some instances we have been able to identify those who are rough sleeping, but we expect there are more which we have not been able to identify from the death registration records.
Comparisons with ONS
These statistics are broadly comparable with those published by ONS for England and Wales. The same methodology and statistical model have been applied.
Although the death registration systems in Scotland, England and Wales are similar, there were some cases where it was not possible to mirror the five searches which ONS carried out on their death registration data.
For example, ONS used information received from coroners for one of their searches, but due to the different system in Scotland, this information wasn’t available. Despite these differences, it is felt that the statistics are broadly comparable.
The ONS report includes estimates of homeless deaths caused by drug poisoning. This is based on the wider definition of drug deaths and is not the same definition of drug-related deaths that we have used in this report.
Improving our methodology
These statistics are experimental and the methodology is still under development. We recognise that there may be better methods to estimate homeless deaths or better sources of information to feed into our current methodology. We are publishing them at an early stage to involve users and stakeholders in assessing their suitability and quality.
We welcome any feedback from users on ways in which the methodology or data sources may be developed to improve the quality of these statistics in future years.
Further information
If you have any questions about the publication of these figures, please contact our Statistics Customer Service helpdesk.
Open Book 