The predictors of homelessness will be discussed next week at a Policy in Practice webinar on the end of the eviction ban, featuring guest speakers from Beam and Greenwich Council. Beam does incredible work at the sharp end supporting homeless people into work, and Greenwich Council works with both Beam and Policy in Practice to combine activities to prevent homelessness, and to help people who are homeless today.
This post looks at work by local authorities who take a preventative approach to tackling homelessness. In particular, we look at how data analysis has informed council’s innovative work to identify families who are at risk of homelessness before 56 days.
Join How local authorities can prepare for the end of the eviction ban to explore how the pandemic has affected people’s pockets, and what this means for people’s housing security ahead of the lifting of the eviction ban on 31 May 2021. See further details and register here.
The costs of homelessness
Homelessness has risen over the last decade across the whole country. The consequences on the lives of everyone affected are long-lasting, and the cost to the public institutions that respond is staggering.
A report by the National Audit Office from 2017 estimated the costs of providing temporary housing to homeless families to be around £845 million. This figure is set to rise in the coming years, with research by the London School of Economics predicting that in London alone, the cost of homelessness services will increase to over £1 billion a year by 2021-22.
Preventing homelessness must begin before 56 days
Acting early and preventing at-risk families from becoming homeless is essential to counter this phenomenon. This has been formally recognised with the Homelessness Reduction Act in 2017, which shifts the focus to early intervention by:
- extending the period during which an authority should consider someone as being threatened with homelessness,
- requiring local authorities to develop an action plan for people identified as being at risk of homelessness and
- placing a duty on public services to notify a local authority if they come into contact with someone they think may be homeless or at risk of becoming homeless.
Effective prevention work, through holistic and user-centred support services, can not only spare vulnerable households the misery of eviction and homelessness but can significantly reduce public expenditure. In 2018, Crisis estimated that for every £1 invested in homelessness prevention there would be a benefit of £2.80.
Policy in Practice held an event at the House of Lords with Big Issue founder, Lord Bird, to highlight the innovative work being done by some leading local authorities on this topic. One key message from the speakers was that 56 days before homelessness is not prevention, we must act earlier.
Investment is needed, but more can be done now
Greater public investment in welfare support is key in tackling homelessness. Our research, commissioned by the LGA, explored the link between the freezing in local housing allowances and rising homelessness trends, showing that for every 1,000 households experiencing a shortfall between their LHA rate and rent, 44 households will require temporary accommodation.
We found that if LHA rates were reset to the 30th percentile of market rent, the average council would see 300 fewer households in temporary accommodation, with the gross cost of temporary accommodation reducing by between £1.4m and £3m a year. Although this measure is now in place, it has been introduced in response to the pandemic which will increase the financial pressures on those already at risk of homelessness, so we don’t believe these savings will have been realised by local authorities.
Councils need to take a different approach while recognising the need for the government to invest in this issue. We believe that more can be done with the resources councils already have available to them. Administrative data held by councils, in particular, represents a unique resource for supporting and enhancing prevention services by the council.
Local authorities should act now. Future demand for homelessness services will likely be on the rise again, given that the ban on evictions is likely to be lifted on 31 May 2021.
Using benefit data to identify predictors of homelessness
Longitudinal analysis of Housing Benefit and council tax support data can reveal clear pathways and predictors of homelessness that lead vulnerable households into homelessness.
Some councils have used monthly information on household housing, employment, income and personal circumstances to identify households who may need support from the council homelessness services. The aim is to exploit this information in order to build a predictive model that can give local authorities more informed insights into where to focus their homelessness prevention services.
Policy in Practice conducted preliminary analysis into predictors of homelessness using data from one London Borough spanning from January 2016 to September 2019.
Our first step was to identify those variables that correlate with households moving into temporary housing. We found that:
- 7.3% of households lived in temporary accommodation for at least one month
- 4.4% of households moved into temporary accommodation
- 50% of households were homeless or in temporary accommodation for less than 7 months
Behind these worrying percentages are families who are affected. From a statistical point of view, the movement into temporary accommodation is still a rare event in the context of the overall sample of analysis. This has implications for the choice of statistical approach to consider for predicting this occurrence, as most models are biased against rare events.
Our preliminary findings are based on an ‘Ordinary Least Square (OLS) regression model’, employed in order to estimate the likelihood of being homeless based on a set of variables related to demographic, housing and employment circumstances.
Figure 1 below shows the five predictors of homelessness by plotting the coefficients of the five variables with the strongest correlation with movement into temporary housing. Correlation coefficients are used to measure how strong a relationship is between two variables.
The predictors of homelessness in order of importance are:
- Previously living in temporary accommodation
- End of a relationship
- Start of a relationship
- Children under three years old
- In full-time work
Having previously lived in temporary housing represents the biggest predictor of whether someone may become homeless. Around 4% of all households observed moving into temporary accommodation (TA) were previously housed by the council in a temporary home. This finding highlights the importance of a holistic approach to homelessness services.
Providing short-term accommodation alone, without tackling some of the underlying issues that make a household vulnerable, can simply mean kicking the can down the line.
Factors related to personal circumstances, such as ending a relationship and starting a new one are also correlated with becoming homeless. The presence of children between the age of 0 and 3 is positively correlated with the possibility of requiring temporary accommodation. This may be due to the barrier to employment which childcare represents for households with children of preschool age, and is likely to grow as a result of the benefit cap and two-child limit.
On the contrary and in line with our expectations, being in full-time employment shows a negative correlation with the likelihood of being homeless.
Figure 1: The correlation between household circumstances and the risk of homelessness
Using data analysis for good
The findings of the analysis above show the potential that this type of data holds in providing useful and actionable insight in contrasting homelessness at the local level. However, the coefficients shown, based on data from only one council, leaves a lot that is left unexplained. There will be other important factors not present in the data might also be associated with households becoming homeless, such as domestic violence, drug or alcohol abuse.
A recent project involving Policy in Practice and backed by the LGA and NHS Digital, has looked at linking data across adult services, children’s services, public health, the NHS, Police and Fire and Rescue Services and will provide useful information for improving our model. In addition, we are working with local authorities to widen access to and use cases for Universal Credit data, this will provide invaluable information to improve the accuracy of these models.
The next step for Policy in Practice is to move from unveiling correlations to making useful predictions. From a methodological point of view, the choice of the statistical model should be centred around three guiding criteria.
- Accuracy. Models will be developed on large datasets and tested on specific local samples. Predictions should be geared towards rare events and should tolerate false positives (i.e households identified at risk who will not end up homeless) over false negatives (i.e. households requiring temporary accommodation but not identified as at risk)
- Interpretability. The outcome of the models should be easily interpreted by service providers as well as providing clear explanations on the main factors behind predictions
- Security. Storing and processing of data should follow strict data security protocols that carry out the analysis in aggregate and safeguards the confidentiality of all information collected
We believe that our initial research, made possible thanks to the support of the Trust for London and a number of London boroughs, can deliver significant positive outcomes for thousands of vulnerable households, as well as for local authorities.
Over the coming months we will be working closely with councils to understand how our predictive analytics can help them improve their homelessness prevention services.
Find out more
Join our free webinar, How local authorities can prepare for the end of the eviction ban, on Wednesday 28 April from 10.30 to 11.45.
We will be joined by guest speakers from Beam and Greenwich Council to explore how the pandemic has affected people’s pockets and what this means for people’s housing security. We will look at analysis on the predictors of homelessness and best practices of leading local authorities who are tackling homelessness, and we will identify effective solutions for others to adopt. See further details and register here.
In the meantime, to find out more about Policy in Practice’s data-driven approach to homelessness reduction, please email email@example.com.
Giovanni Tonutti is a PhD student in data science at the Scuola Normale Superiore in Pisa, Italy. He has worked with Policy in Practice since 2015.