RECAP: Administrative Data – Ways to Inform Equitable Homelessness Prevention

What is Administrative Data? Administrative data is defined by Dr. Fowler as information that tracks service delivery and outcomes, and signals how a system is responding to housing emergencies. Administrative data can be used to evaluate and improve services, uncover biases, and engage stakeholders with decision-making and address inequities in how services are delivered. 

Housing is a scarce resource. Dr. Fowler says that because of this, service delivery poses an ethical and moral question for providers. What informs this resource allocation (i.e. whom to service first) includes policies, experience, information, and the service providers’ own preferences and biases. Dr. Fowler says administrative data can be an additional lens to inform this resource allocation.

Big data technology such as machine learning can harness administrative data to understand who is at greatest risk, what works best for who, and who will be harmed most without services. These are complex, dynamic, nonlinear systems, and big data technology is designed to account for this complexity. The potential of big data technology posed the question for Dr. Fowler whether we can automate homelessness prevention. Can we predict who will do best in each type of intervention, from supportive housing, to rapid rehousing? Dr. Fowler showed an example of how he used Counterfactual machine learning to estimate how people in need of housing services in St. Louis, Washington would fare if they were given a different intervention. The model predicted nearly all would do best if they received a prevention intervention, and that it would reduce homelessness by 16%. 

Equity has to be defined locally when it comes to housing allocation. Dr. Fowler posed the question: is equity a matter of access to services, or access to good service? For example, he modelled a scenario of optimizing housing allocations to a 4% reduction in homelessness. In this scenario, while there were more accessing services, there were also some who received interventions that were making their homelessness worse. Dr. Fowler concludes that there is no such thing as a global equity definition; it has to be defined locally. 

How can community inform data-driven decision making? Dr. Fowler emphasizes that when it comes to leveraging data, technology needs people – both are required to leverage large administrative datasets to make ethical data-driven decisions to address homelessness. When Dr. Fowler’s team simulated potential COVID-19 policy responses in St. Louis, community leaders were first asked to locally define what equitable housing allocation would look like to them. The dynamic data modelling was based on this community-informed equity framework, and the results showed the best opportunity was to funnel resources into rental assistance as a preventative measure. This was implemented in St. Louis as an impactful systems-wide approach over the COVID-19 pandemic.

About Dr. Patrick Fowler: Dr. Fowler’s research aims to prevent homelessness and its deleterious effects on child, family, and community well-being. Trained in child clinical-community psychology, Fowler uses innovative methods that rigorously investigate policies and programs intended to promote housing and family stability. Recent work integrates community and data sciences to improve fair and efficient social service delivery. Fowler currently directs the Public Health Sciences Ph.D. Program and co-directs the Division of Computational and Data Sciences at Washington University in St. Louis. 

This blog is part of the Insights from the MtS Research Gathering 2022 blog series. In this series, we will share the key takeaways from this research gathering. To check out the other blogs in this series, click here