By Emily Keller

A hand hovers over the touchpad of a lap, a notepad and pen next to it. Photo, University of Washington

Photo, University of Washington

The University of Washington (UW) eScience Institute has received a five-year, $754,601 grant from the Bill & Melinda Gates Foundation to advance data analytics at organizations serving the homeless in the Puget Sound region, with a focus on educational outcomes.

Through this grant, researchers at the UW will collaborate with foundation grantees that focus on improvements to educational attainment in our region, to incorporate data science approaches and develop actionable insights from their data while protecting personally identifiable information.

Partnerships with foundation grantees will include the following activities:

  • Collaborative development of open source software to implement data science methods using automation and capacity building to utilize the software and methods developed
  • Training in basic data science methods, use of data outcomes and interpretation of results
  • Developing an understanding of unintended bias introduced through input data and sometimes preserved through algorithms and analyses, and creating tools and methods that apply algorithmic fairness constraints

Three principal investigators will lead the project. Bill Howe, associate professor in the Information School and former associate director of the eScience Institute, conducts applied research in data-intensive urban science, data management systems and methods, and curation, analytics, and visualization. Ariel Rokem, a data scientist at the eScience Institute, is a neuroscientist who develops open source software for analysis of biomedical imaging data, and tools for reproducible and open research. Bryna Hazelton, a research scientist at the UW Department of Physics and the eScience Institute, is an astrophysicist with expertise in advanced statistical measurements and workflows for processing petabyte-scale data.

A full-time data scientist will be hired to oversee the work of fellows and students at the eScience Institute, in collaboration with the Information School’s Urbanalytics group, directed by Howe. Urbanalytics applies interdisciplinary research and analysis to fine-grained data on urban social issues, with a focus on responsible data management. The eScience Institute supports researchers in the development and application of advanced computational methods and tools to solve real-world problems across disciplines.

The grant continues the ongoing collaboration between the eScience Institute and the foundation that began with a ten-week summer project, “Predictors of Permanent Housing for Homeless Families,” which was part of the inaugural Data Science for Social Good program at the eScience Institute in 2015. Hazelton and Rokem served as data science mentors working with fellows, students and the foundation. Through a collaboration with Building Changes, the foundation, and King, Pierce and Snohomish counties, the team analyzed service enrollment data of homeless families to identify factors that predict success in finding permanent housing, and to examine the ways families transition between programs and episodes of homelessness.

The summer project utilized client-level data collected by publicly funded service organizations as part of the federally mandated Homelessness Management Information System (HMIS). Data points include demographics, income, disability and veteran status, housing history and the provision of services. Through a follow-up grant awarded to Hazelton and Rokem at the beginning of this year, eScience researchers are currently continuing this work to include information about other services in King County and the City of Seattle that support individuals and families who are at risk of becoming homeless, primarily from the Seattle and King County Public Housing Authorities.

Ultimately, the work conducted under the new grant could enable a powerful combination of HMIS and public housing authority data, with information on educational outcomes through sources such as the Washington State Education Research & Data Center, to examine the impacts of housing instability on educational attainment. The project aims to use newer data science methods to go beyond population-level evaluations and uncover subtle dependencies between multiple factors, while also quantifying the uncertainty in inferences based on these data. These methods could show how particular groups benefit from specific interventions and predict the success of varying approaches based on individual backgrounds, characteristics and circumstances.

The interest of the eScience Institute in reproducible and open science dovetails with the foundation’s established open access policy. In support of open access principles, the research products developed under the grant will be publicly available for free; the software developed will be hosted publicly on GitHub; and papers or other products will be made freely available to anyone through the University of Washington Libraries.

These projects undertaken by the eScience Institute are a part of a growing focus on homelessness through research and teaching across UW disciplines and campuses, as detailed in a new report by Urban@UW’s Homelessness Research Initiative.