Q&A: After developing a better way to count homelessness, UW researchers discuss how more accurate data can help providers and people

America’s homeless services system relies on a massive amount of data, and at first glance, that data is exacting. Federal reports describe the country’s unhoused population in granular detail, listing precisely how many people are experiencing homelessness in each city along with detailed demographic data. Want to know how many people ages 55-64 slept outside in Spokane last year? A spreadsheet confidently provides the answer: just one. 

That data influences decisions at every level of government, from how the U.S. Department of Housing and Urban Development (HUD) distributes $3 billion in funding to how local service providers target their outreach efforts. It’s also not that accurate. As a result, communities across the country — including King County — don’t really know exactly how many of their residents are unhoused and have a limited window into people’s circumstances and needs.  

So, a team of University of Washington researchers designed a better way to count. Led by Zack Almquist, a UW associate professor of sociology, and Amy Hagopian, professor emeritus of health systems and population health, researchers developed a method that taps into people’s social networks to generate a more representative sample, which they use to estimate the total unhoused population. Along the way, agency staff and volunteers gather information on people’s demographics, resources and needs.  

The researchers launched this method in partnership with King County in 2022 and repeated the process in 2024, publishing their findings Sept. 4 in the American Journal of Epidemiology. UW News sat down with Almquist and Hagopian to discuss their new approach and how it could help close the gaps in our understanding of homelessness in America. 

Statistics on homelessness and the demographics of unhoused populations are often quite specific. The federal government reported that 653,104 people experienced homelessness on a single night in January 2023, for example. How do we get these statistics, and how reliable are they?  

Amy Hagopian: I’m always a little amused at numbers that create a false specificity; for example, an airline says my flight will arrive in Chicago at 11:33 a.m. Everyone knows that number isn’t true, except sometimes by accident, and yet we entertain the airline by pretending to believe the number. After all, there are no consequences for being wrong! 

The national count is an amalgamation of counts reported by each community’s “continuum of care” jurisdiction, designed by the U.S. Department of Housing and Urban Development. Most jurisdictions are still attempting a single-night head count of people found by volunteers who move about in the dark with flashlights and clipboards — a highly problematic approach King County has abandoned in favor of our sampling method. When these numbers come in, HUD just adds them up, and of course the number won’t be round. We all know it’s way below the actual number, because a middle-of-the-night census isn’t going to find everyone. 

Zack Almquist: There is a common fiction that when we do a census it is exact, because government reports often do not provide a margin of error. I think if you asked many experts, they would say they know the reality is a range, not a single number. In fact, not providing a range provides a level of confidence that we really don’t have, regardless of how we get there.

One nice thing about using a statistical estimate is that people are trained to expect a margin of error or confidence interval. We can say, plus or minus 5%, or 100-200 people. In other words, by moving into a space where we expect to see a range, we can be more honest, and ideally be more prepared to handle the real situation. 

Why does it matter how accurate this data is? 

AH: America has the worst homelessness problem in the world created by an economic system – as opposed to war and other disasters – largely because we make no attempt to recognize the human right to housing as established by the United Nations. One reason to count by jurisdiction is to learn where the hot spots are, and which areas have managed to lower their counts, and why.  

ZA: This is also an equity and respect issue for the people who are experiencing homelessness. We owe it to our community members to do our best to capture the real state of the problem in our area and to best represent their race, ethnicity, gender, disability status, and causal or associated factors like eviction. We cannot hope to adequately engage a problem if we can’t accurately quantify it.

Your team developed a new method to estimate the unhoused population. How does your method work, and how does it differ from the traditional PIT count?  

ZA: Our method takes the approach that there is no reliable way for us to obtain a census of people living unsheltered in our community, and that we need to move from a biased counting exercise to an approach that leverages modern statistical methods to obtain a best estimate of the population given our current resources. Modern sampling methods can improve how we count people. Sampling is the process of selecting a small group from a larger population to study and make conclusions about the entire population. 

We leveraged a sampling strategy that comes out of public health literature and is endorsed by the National Institutes of Health and World Health Organization. First, we collect a roster and bed count from shelters. The HUD-mandated Point-in-Time count was always split between the roster or bed count and an unsheltered count; the latter was historically counted in King County by a visual census. So, the total number of people experiencing homelessness is the number of people in emergency shelters on a given night plus the number of people living outside on a given night. Through some ratios and algebra, we can estimate the total number of people if we know who slept in an emergency shelter and know from historical measures the relative proportion of people who slept outside.  

Our sampling strategy of leveraging people’s social networks and peer referral allows us to estimate the proportion of people who slept outside to those who slept in an emergency shelter on a given night. Further, this allows us to better find and count people who would be hard to find in the traditional visual census — people living in the woods or hiding — and also provides a clear method for the margin of error of our estimate of the number of people experiencing unsheltered homelessness. 

Your count creates a more reliable estimate of the unhoused population, but that’s not all. What other information can you collect with this method, and how might it be useful?  

AH: When other jurisdictions do their midnight census counts, they are just counting bodies seen. There is no opportunity to collect demographic or life history or health status data unless they shake people awake and interview them in the moment, which few people do. Instead, they conduct a post-count interview process in places like food banks. Our approach provides the opportunity to count people during daylight hours while also learning something about their life course and circumstances. This provides King County with some valuable information about the causes of homelessness. Once we move towards a quarterly count, we can also learn about the “churn” — the number of people moving into and out of homelessness and what the drivers are for those changes in circumstance. 

ZA: I think this point can’t be emphasized enough, as running a post-count survey is almost always conducted as a spatial convenience sample that surveys both those using emergency shelters and those who slept outside. It’s unlikely to include the same people who were in the one-night body count.  

What have you heard from people who’ve participated in your method? How do participants’ experiences differ from the old Point-In-Time count? 

AH: We conducted a couple of focus groups recently with people experiencing homelessness in Seattle. We asked them about their impressions of the recent methods change in how we count. We found people appreciated the motivations behind the change, and the more respectful approach we are now using.  

ZA: I just want to second what Amy said, and to point out that people really appreciate being directly engaged with and having a chance to be paid for their time and effort. 

How else could this method be used? Are there potential applications outside of homelessness and housing services?  

AH: I have helped conduct mortality counts in war zones, and some of the lessons learned from those experiences were helpful here. For example, in Iraq conducted a door-to-door survey to ask adult household members to tell us about the alive or dead status of their siblings. This allowed us to calculate a total war-related mortality rate for the country, as our sample was selected proportionate to size of the governorate sampled.  

ZA: I think the basic ideas used here could end up influencing health and demography measurement efforts for several hard-to-estimate populations. For example, international migration can often be split between those we can count with high fidelity, like registered immigrants, and unregistered immigrants. Combining new sampling methods with administrative data to count hard-to-reach populations could be employed for a number of problems in industry, health and public policy. I hope to see these ideas picked up broadly. 

AH: We are grateful to the UW’s Population Health Initiative for the opportunity to develop these methods, and to our partners at King County Regional Homelessness Authority for being willing to try something new with us. 

For more information or to contact Hagopian and/or Almquist, contact Alden Woods at acwoods@uw.edu.