A Better Way to Ask Where People Live: Testing Housing Questions in Fulton County, Georgia

People experiencing homelessness and those living in congregate settings — group homes, shelters, jails and the like — face higher risks of illness and death than people in their own homes. Knowing someone's housing status helps target health promotion and outbreak response. But no question set measuring it had been validated against an independent data source. In June 2023, the Fulton County Board of Health in Georgia asked CDC for help building one for case interviews and surveillance.

Building the questions

A November 2022 search of the Department of Housing and Urban Development (HUD), the Veterans Administration, the Centers for Medicare and Medicaid Services, Google Scholar and PubMed, plus advice from experts, turned up 21 existing tools. None matched the federal definitions of homelessness in enough detail, and none met public health's needs: telling sheltered from unsheltered homelessness and congregate from non-congregate living.

So a new set was written, over several rounds, by CDC experts, federal partners, local health officials, service providers and clinician-researchers, and reviewed by people with lived experience of homelessness on a national partner's consumer panel. It asks, for any period of interest:

1. whether the person stayed in one place or several;
2. where — described in their own words, which the interviewer matches to a set list of places;
3. multiple-choice follow-ups, such as whether the arrangement is short- or long-term.

It was designed to take under 2 minutes, and its list includes places not common locally, such as "beach" or "boat," so it can be used anywhere.

The field test

From August 2 to September 1, 2023, interviewers asked about the previous 2 weeks at food pantries and the county's tuberculosis, vaccination and sexual health clinics. Respondents got a $10 gift card. Interpretation was available, including a CDC phone line covering 171 languages.

• 481 people took part, 26 (5.4%) in a language other than English — Spanish, Creole, Korean or Mandarin.
• Interviewers judged 99.2% of those assessed somewhat or completely engaged, and nobody declined the housing questions (some did decline personal or demographic ones).
• Paper (461) and electronic (20) versions worked equally well. Interviewers were unsure at first with the list of housing types, and grew confident once they knew it.

What people reported:

• 331 (68.8%) housed; 139 (28.9%) homeless; 11 (2.3%) unclear, unsure how long their arrangement would last.
• Of the 139: 56 sheltered, 58 unsheltered, 25 both.
• 75 (15.6%) had lived in a congregate setting.

Did it work?

• Consistency: among people staying at the same place, 20 of 21 pairs of answers (95.2%) gave the same classification.
• Against an outside record: 26 respondents could be matched to the county's Homeless Management Information System (HMIS), the fullest client-level record of people receiving HUD-funded housing and support services. For 24 of the 26, the question set agreed — 18 homeless, 6 housed. The other two were listed as housed in HMIS but described at least one homeless situation in the previous 2 weeks. Everyone HMIS recorded as unhoused, the questions also found unhoused.

Limits

• Results may not carry over beyond Fulton County.
• Only about 5% of respondents could be matched to HMIS, perhaps partly from errors in names and birth dates.
• The matched answers covered 10 of the 14 housing types.
• The consistency check covered only people living with others.
• Interviewers need training, which may make the questions hard to fit into existing workflows — a free-text interface that picks from the list automatically, for example with a large language model, could help.

Why it matters

This is the first housing question set shown to be externally valid, for a small group of respondents, and it could help agencies screen for homelessness and focus help on groups at high risk, especially in emergencies — and in health care, where housing affects individual and community health. It should be validated in other places. Because systems increasingly share data, CDC is working with partners on a data standard for housing status, for consideration by bodies such as those behind the U.S. Core Data for Interoperability.

Sources

Based on Bratcher A, Waddell CJ, Kava CM, et al., "Field Testing and Validation of a New Question Set to Measure Housing Status — Fulton County, Georgia, August–September 2023," MMWR Morbidity and Mortality Weekly Report, volume 74, Centers for Disease Control and Prevention; a work of the United States government in the public domain. The report gives the match with the housing database as 92%, 92.2% and about 90%, so only the count is given here.