Is the US in the midst of a homelessness crisis? Many people think so, but that’s largely based on based on anecdotes. What does the data say?
At a glance, this doesn’t look very crisisy. Since 2015, things have gone up by less than three percent.
Still, I think there is a crisis, we just have to work harder to see it. We need to look at different locations, rates of change, different types of homelessness, and mental health and substance abuse issues. Let’s do that.
Homelessness in general
Homelessness is much higher in some places.
The rate of homelessness varies hugely between different states. Here’s the percentage of the population in each state that was homeless in 2020.
In Mississippi it is 1 in 2500 (0.04%), while in New York State it is 1 in 210 (0.47%). That’s a huge difference.
Now, when we talk about a crisis, there’s an implication that things are getting worse. (You don’t hear much about the everyone you love will die and be forgotten “crisis”…) We already saw that things are pretty stable at the national level. How are things changing in individual states?
Homelessness is increasing in some places and decreasing in others.
Here’s the change between 2015 and 2020, again as a percentage of each state’s population.
The general pattern is increases close to California and decreases close to Florida. The exceptions are the random huge decreases in Nevada and North Dakota (maybe because of the end of the fracking boom?), and the confusing mess in the Northeast.
This is a relatively simple story so far, but I don’t think it fully captures what’s going on. We have to go deeper.
Types of homelessness
There are different types of homelessness.
Some people run out of money, get evicted, and stay at a shelter for a few weeks before moving in with family and eventually getting back on their feet. Other people have mental health issues and stay on the street for years. When we talk about a “homelessness crisis”, we should try to distinguish these different situations.
Let’s back up. Where is all this data coming from? Well, every year during the last 10 days of January, the US tries to count every homeless person. This is done by hundreds of local government organizations and nonprofits that cover almost the entire country. They must participate in the counting or they won’t get any funding from the US Department of Housing and Urban Development (HUD), which collects all the data.
We’ll look at two attributes that people get during this count. First, people are classified as sheltered or unsheltered. Someone is sheltered if they are staying in an emergency shelter, transitional housing program, or a safe haven. Someone is unsheltered if they’re staying in a vehicle, an abandoned building, or on the street. Second, people are chronically homeless if they’ve been so for at least a year, and otherwise non-chronically homeless.
Aside: Technically—because nothing can be simple—the definition of “chronically homeless” is much more complex. You also qualify if you’ve been homeless for a total of a year in the last three years, plus that happened in at least four episodes, plus those episodes were separated by at least a week. Yes, that means that someone homeless their entire life except the first week of each year is “non-chronic”, and yes that is crazy, but probably it never happens and anyway these categorizations are done by random people around the country who may not much care about the legalistic definition anyway. Oh, and also, this changed in 2016.
So, what types of homelessness does the US have?
Most homeless are non-chronic, and most non-chronic homeless are sheltered. But the chronic homeless are usually unsheltered.
Here are the fractions of people that fell into each of the four possible groups in 2020.
Here’s how I think about this:
- Most homeless are non-chronic (around 80%).
- The non-chronic homeless are mostly sheltered (around 2/3).
- The chronic homeless are mostly unsheltered (around 2/3).
That’s the overall mix. But we have to worry about two things. First, is the mix changing over time? And second, how does the mix vary in different places?
Unsheltered and chronic homelessness is increasing.
Unfortunately, the mix is changing, and for the worse. The “best” type of homelessness (sheltered and non-chronic) is decreasing, while the other types are increasing.
Since it’s a bit hard to see, here’s the number of people in each group in 2020 divided by the number in 2015:
While sheltered and non-chronic homelessness remains the most common, it actually decreased by around 12.5% since 2015, whereas all the other types have increased by around 25%.
OK, but how do things look in different places?
Different types in different places
Chronic and unsheltered homelessness is much more common in some places than others.
Let’s compare New York and California:
Homelessness overall is similarly high in both places—0.47% in New York and 0.41% in California. But California has much more unsheltered and chronic homelessness.
Aside: Did you know that there is a constitutional right to shelter in New York? This is a result of a 1979 New York State Supreme Court decision. Two other places have a weaker version of this. Massachusetts has a mandate from a 1983 law but it only applies to families, not individuals. The District of Columbia guarantees shelter to families year-round, and to individuals when the temperature is below 32° F or above 95 °F (below 0° C or above 35 °C).
Anyway, we can picture the situation for all states by visualizing the four types of homelessness on a map. Here they all are, each as a percentage of the state’s population. (Click for a .pdf if you want to look closer.)
For sheltered non-chronic homelessness, the West Coast isn’t exceptional. Instead, the standouts are New York at 0.42% and—press your face against the screen—the District of Columbia at 0.49%. These are followed by Massachusetts at 0.23% and Alaska at 0.20%.
But unsheltered chronic homelessness is very different. It’s almost nonexistent in some places—it’s 0.10% in California and 0.08% in Hawaii, but 0.0013% in Wisconsin. In Wisconsin, that is 80 people total. In California, that’s 42,195.
However, let me remind you—I can’t emphasize this enough—this survey is done in late January, when it’s cold in New York and stupidly cold in Wisconsin. You have to imagine that things would look different if things were done in Summer, though it’s hard to quantify. Still, the difference might be significant in that almost all homeless people in cold places have at least some occasional contact with social services.
OK. The types of homelessness are different in different places, and they are changing over time. But how are they changing in each place? We still need to go deeper.
Unsheltered and chronic homelessness is getting worse in some places, particularly the West coast.
How are the different types of homelessness changing over time in each state? Let’s again contrast New York and California:
The situation in New York is reasonably stable since 2015, albeit at very high levels. Meanwhile, California has a small decline in sheltered non-chronic homelessness, but big increases in the other groups.
Other places are different. For example, in Florida, everything is decreasing at once.
What does the rest of the country look like? Remember, that for the country as a whole, we saw above that sheltered non-chronic homelessness is decreasing, and the other categories are increasing.
But what about individual states? Since it’s not convenient to look at 50 different charts, I made four maps to show the change in each type of homelessness between 2015 and 2020, again as a percentage of each state’s population. In each of these, grey is a decrease, white is no change, and a non-grey color is an increase. (Click for a .pdf to zoom.)
How should we think about this? Here’s my best attempt at a summary:
- Sheltered non-chronic homelessness decreased almost everywhere.
- All homelessness declined in Florida and places near Florida.
- Chronic and unsheltered homelessness increased a lot in California and places near California.
There are some exceptions. For one thing, despite being close to California, Nevada and some of the Montana-esque states saw big decreases in certain categories. For another, the Northeast is weird and defies any attempt to summarize. Sheltered non-chronic homelessness decreased a lot in Massachusetts and Vermont but barely changed in New York or New Hampshire. Every other category is a random mishmash with no pattern. I tell you, on my worst days it’s almost like reality is just completely indifferent to our desires to understand it with tidy little narratives.
Drugs and mental health
And what about meth? Theories abound that mental health and substance abuse are a huge part of the homelessness crisis.
Well, the yearly homelessness survey collects data on if the homeless are “severely mentally ill” or suffer from “chronic substance abuse”. I can’t figure out exactly how these are defined. It’s implied that it’s done by literally asking people, but I think it varies from place to place.
So, I did the sensible thing. I downloaded the .pdf files for each of the 6082 different reports, wrote a script to convert each .pdf to plain text, wrote a parser for that text, compensated for 8 billion inconsistencies in how the reports were laid out, damn you HUD, damn you to hell, extracted the data for each of the above categories, and made plots.
Nationally, there is only a small uptick in mental illness and substance abuse.
Blue is people with mental health issues, while orange is substance abuse. Some people will be in both categories, but we don’t know how many. (My guess is a lot because they’re strongly correlated.) Grey is all people, including those with no issues.
Like the other national data, this doesn’t scream crisis. If anything, it’s a bit reassuring. We already knew that unsheltered homelessness was increasing, but here we see that mental health and substance abuse aren’t increasing at the same rate.
Next, of course, we want to know how this looks in different places, and how it’s changing.
I made plots for each of the 437 regions and states (you can see them below). This data is quite noisy, which makes it a challenge to visualize. For example, here’s Los Angeles.
To get more reliable numbers, I combined sheltered and unsheltered homelessness, took the average of mental health and substance abuse, and then applied a smoothing function. I used the smoothed values below, which are hopefully less polluted by noise.
There are significant increases in mental illness and substance abuse in certain states.
How many homeless people are there in each state with severe mental illness of substance abuse problems? Here are the numbers in 2020.
Again, this is high if you’re close to California and in certain parts of the Northeast. This is pretty similar to the map of homelessness overall we started with, though you’ll see a few places that stand out more here, e.g. Nevada.
OK, and how are things changing?
Again, we have a crisis mostly in the West. California was already really high in 2015 and has gotten even higher. The real standout is Washington state, where things more than doubled. And sure enough, the plot for Seattle is stark.
Things just exploded between 2016 and 2017 and then went up from there. You might remember that 2016-2017 is exactly when meth metabolites in sewage in Seattle also exploded. Of course, that could just be a coincidence, or might be noise in either of these datasets. Still, I think we can see why people in Seattle might feel there’s a crisis. Maybe anecdotal knowledge ain’t so bad.
P.S. If you want to see data for your local state or city it is below.