It was an early 2:00am morning in January of 2018, and the four of us were driving from Seattle to a church in Bellevue, a nearby city on the other side of Lake Washington. There, in a crowded gymnasium stocked with coffee, bagels, and bananas, we were given a map and assigned a region of the city about a square mile in size.
For the next four hours we, along with hundreds of other volunteers in the region and thousands more nationwide, would be driving around in the dark, checking off each street as we went, looking for visible signs of homelessness. A tent on the side of the road, a vehicle in a church parking lot, a sleeping bag in the entryway of a local business.
We were participating in what the Department of Housing and Urban Development (HUD) calls a “point in time count” of people experiencing homelessness, an annual effort by municipalities throughout the country to collect data on homelessness.
In Seattle, in Bellevue, and throughout the region the official point in time count is administered by a coalition of local governments called All Home. This organization is what HUD refers to as a Continuum of Care (CoC), a regional body that implements housing programs and homeless services and, crucially, is responsible for allocating grant funding from the federal government.
Continuums of Care are the core geographic unit of the nation’s homeless infrastructure, and the way that they are organized has implications for the way that data about homelessness is gathered and interpreted. All Home, like many CoCs, represents a single county. But this isn’t always the case, and there is no requirement that CoCs conform to any existing geographic boundaries. CoCs can and do represent areas as large as the state of Montana (145,552 sq mi) and as small as the city of Cambridge, Massachusetts (6.4 sq mi).
This lack of similarity between the regions underlying the CoC boundaries makes them very hard to compare—not only to one another, but also to other geographies such as those used by the US Census Bureau (census tracts, legislative districts, counties, urban areas, zip codes, etc.) Perhaps most frustrating in this regard is the lack of HUD data related to homelessness in metro areas, areas where rates of homelessness have often reached crisis levels.
Much like the general populace, people who are experiencing homelessness tend to cluster in cities and their surrounding metro areas — and undoubtedly for good reason: cities typically provide concentrated access to supportive services, subsidized housing, and employment.
But while broadly speaking a city with a large general population will typically have a large unhoused population, rates of homelessness can and do vary significantly from one city to the next.
Some of the largest metropolitan areas have rates of homelessness that are disproportionately higher than state and national averages. According to the National Alliance to End Homelessness, in 2016 just twenty-five metro areas accounted for 55% of all people in the United States who were experiencing homelessness despite only representing 42% of the nation’s overall population.
Given that homelessness is often geographically concentrated in cities and their broader metro areas, one might assume that HUD’s data on homelessness would be tabulated along Metropolitan Statistical Area (MSA) boundaries in much the same way that census information is made available for a number of different geographies. Unfortunately, this is not the case.
Compiling Data on Metro Areas
There is a wealth of data (social, economic, and political) related to metro areas that has been compiled by journalists and researchers, and I felt that it was important that quantitative information on homelessness also be available for these regions. So I set out to compile this information myself using publicly available HUD data.
Initially, I thought this might just require some clever spreadsheet manipulation to cross reference HUD counts with census data, but as I dug deeper I found that I needed to use GIS software in order to parse the data and achieve the tabulations that I was looking for. I won’t go into too much detail on that process here (you can find more comprehensive notes on the methodology that I used here) but suffice it to say that the effort was significantly more involved than I had anticipated.
In the end I was able to disaggregate HUD’s data on homelessness within Continuums of Care and geographically reapportion it to provide reasonable estimates of homeless counts in metro areas. I’ve made that data available on GitHub, and I encourage others to not only make use of those numbers but to also check and improve on my approach.
In a future post I would like to compare these numbers with other statistics relating to metro areas, but after looking over the data I felt that there were some high level takeaways that were worth mentioning here.
Total Homeless Population
There are 390 Metropolitan Statistical Areas in the United States including Puerto Rico but excluding Guam, the US Virgin Islands, and other territories. Estimated homeless counts for the top ten metro areas derived from HUD’s 2018 data are shown below. The entire dataset is available here.
Of the 552,830 people nationally who were experiencing homelessness in 2018 an estimated 507,536 (or 91.8%) were residing in America’s metro areas. Those same metro areas accounted for 86.9% of the general population.
The cities that top the list when comparing metro areas differ from those that rank the highest when comparing Continuums of Care. For example, when we compare CoCs, Boston CoC (MA-500) counted 6,188 homeless in 2018 and ranked 14th among other CoCs. But Boston CoC is relatively small. At 48 square miles, it represents the City of Boston rather than the much larger metropolitan area. For comparison, San Diego City and County CoC (CA-601) ranks 4th among CoCs with 8,576 counted homeless—but the region is 4,207 square miles, about 87 times the size of Boston CoC.
This “apples to oranges” comparison of CoCs demonstrates why reallocating the HUD data can help give us a better picture of homelessness in America’s metros. We see that among MSAs, the Boston-Cambridge-Newton Metro Area (with a homeless population of 12,998) ranks 5th while the San Diego-Carlsbad Metro Area (which has the same boundary and homeless count as the CoC) ranks 8th.
San Diego and Boston have two of the largest homeless populations nationally, and they are among a number of other major US metros with disproportionately high rates of homelessness. In 2018 just 20 metros accounted for an estimated 54.7% of all homeless counted nationally while representing only 35.2% of the country’s total population. The map below shows the 20 metro areas with the largest estimated homeless counts.
Rate of Homelessness
Highly populous metros are likely to have large homeless populations, but I also wanted to compare rates of homelessness per capita across metro areas. In addition to tabulating the total homeless population, I also used the HUD counts and the census figures to generate a “representation ratio” for each MSA.
The representation ratio is calculated by dividing a metro area’s share of the nation’s total homeless count by its share of the nation’s general population. A value of 1 indicates that the rate of homelessness is equal to the national rate of homelessness. A value of 2 indicates that the rate of homelessness is double the national rate of homelessness, ie, that there are twice as many homeless people per capita residing within the MSA’s boundaries. Estimated homeless representation ratios for the top ten metro areas derived from HUD’s 2018 data are shown below. The entire dataset is available here.
Here again the metros at the high end of the tabulation are different. The Chicago-Naperville-Elgin Metro Area, which ranks 9th in terms of overall homeless count, falls to 205th when ranked on representation ratio. At 0.53 it has a rate of homelessness that is about half of the country’s average.
The Santa Cruz-Watsonville Metro Area, on the other hand, which ranks 36th in terms of total homeless counted, sits at number one when comparing representation ratios. Santa Cruz-Watsonville has five times the national rate of homelessness. That equates to about 85 people in 10,000 who were experiencing homelessness in 2018 (compared to 17 people in 10,000 nationally).
While one might expect that the majority of metro areas would have a rate of homelessness greater than the national average (due, perhaps, to the clustering effect) in fact the opposite is true. Of the 390 metro areas included in this analysis, only 80 have a homeless representation ratio greater than one.
When I first compiled the ratios, one of the things that immediately jumped out at me was how the metros with disproportionately high rates of homelessness were largely concentrated in west coast states. With the exception of New York-Newark-Jersey City Metro Area and Springfield Metro Area, the top 25 metros by rate of homelessness are predominantly located in California, Oregon, Hawaii, Washington, and Alaska.
It should be stated that the representation ratio is by no means perfect given the methodology for how homeless populations are apportioned. Particularly affected are MSAs that fall within larger CoCs, such as the five MSAs in Oregon that fall within the Oregon Balance of State CoC (OR-505). Currently they are all assumed to have the same average ratio as the CoC that they reside in, but it is likely that their actual representation ratios would be higher or lower.
For more on the challenges associated with geographically reapportioning population counts, I encourage you to check out the more in depth explanation that I posted on GitHub along with the accompanying data.
While comparing homelessness in metros is inherently difficult and imperfect due to the way that HUD administers their counts, I think it is important nevertheless that the data on homelessness be presented (and analyzed) within these geographic boundaries — and my hope is that my efforts here will allow others to do that and to expand on this work.