Using Private Information to Predict Homelessness Entries: Evidence and Prospects

Abstract

Do people at risk of homelessness have private information—information that social service agencies cannot credibly obtain—that helps predict whether they will become homeless? This article asserts that the answer to this question is yes: homeless people and people at risk of homelessness know important things about their future. Data from Journeys Home (JH), a pathbreaking longitudinal study of people experiencing homelessness and people at risk of homelessness in Australia, are used in this article. In many cases, the private information that participants have predicts entries better than the public information that agencies can obtain. Ways in which this private information can be used to improve service delivery are suggested.

Keywords:

• Homelessness
• homelessness entries
• homelessness prevention
• private information
• longitudinal studies

Acknowledgments

We are grateful to Martha Burt, Pierre-Andre Chiappori, Abraham Chigavazira, Dennis Culhane, Beth Shinn, the participants at the 2016 Workshop on Homelessness and Housing Insecurity at the Melbourne Institute, and members of the National Alliance to End Homelessness (United States) research council for helpful comments and questions. We also thank two anonymous referees for major improvements in the manuscript.

Notes

1. A large body of literature on mechanism design gives many examples, which range from optimal taxes (Mirrlees, 1971) to the assignment of residents to hospitals (Roth, 1990) to the awarding of electromagnetic spectrum licenses (Fox & Bajari, 2013).

2. In 2007–2011, about 225,000 families with children lived in poverty in the city (U.S. Bureau of the Census, American Community Survey, 2011) and about 1,000 families per month entered shelters (Goodman et al., 2016), so in a 6-month period about 6,000 families would enter shelters. Since some families may enter twice, since there is turnover among poor families, and since some families not previously resident in New York may enter shelters, the estimate of 2.7% is probably high.

3. For evidence of this see the information brochures all participants received at: https://melbourneinstitute.com/journeys_home/participate/publications.html (Melbourne Institute, 2011).

4. Correlations in misreporting were also looked at, but they seemed to make little difference to the results.

5. Corresponding regression results are presented in Appendix B.

6. Remember also that the entire JH sample is disadvantaged.