Readers who are interested in the topic of homelessness may also wish to read an excellent recent post by Scott Alexander, the blogger who writes "Astral Codex Ten":
Interesting overview. It would be easier to draw conclusions about local areas if charts 17-20 were shown as rates to keep them consistent with the rest of the article. San Francisco's population drop over the last few years might explain some of the decrease, and the weird sawtooth pattern for mental illness is probably a reporting change.
When you get the local region data, it would be interesting to see a comparison of different types of homelessness vs. mean low temperature in January.
If you're feeling brave, add a comparison of the political affiliation of each region as well. I've always wanted to see an unbiased resolution to the online debates of whether weather or policies attract chronically homeless people.
I have my suspicions based on four years of living in Colorado and seeing how it compares to neighboring and western states, but of course cost of housing would be a major confounder here as well.
Yeah, I'd rather have percentages rather than absolute counts. The reason for switching to absolute counts for the mental health / substance abuse plots is that, as far as I know, there are no reliable population estimates for local regions.
I'll think about the other plots. Just from looking at the maps, my impression is that:
(1) Colder states have a smaller fraction of unsheltered and/or chronic homelessness (strong effect).
(2) Colder states, all else being equal, have smaller numbers of homeless people (weak effect)
(3) Politically conservative states have smaller numbers of homeless people (medium effect)
Of course, causality is debatable, though I think the mechanism for (1) is pretty clear.
Is there a reason why the San Francisco unsheltered data seems to be bucketed in 2-year segments? That seems like it wouldn't follow the HUD requirements.
As you suggest, the sheltered numbers are different, but the unsheltered numbers are exactly the same. (Except some of the details are deleted for 2023? Hmmm.) Glancing through other places, there seem to be a few other areas where the same thing is happening:
Really enjoyed this article. I've been reading your blog for a few years and it's been such a gift to watch your interests and voice change in many ways while also maintaining a real & grounded identity. I wonder what US homelessness looks like compared to homelessness in the other countries mentioned, especially the colder ones. I also was wondering if you would be open to sharing the source code you used to come up with these visualizations?
Thank you! I *could* share the code, but I must warn that would probably be disturbing to read. It would be irresponsible to share as anything other than a cautionary tale, an example of what happens when you don't just have no organization or abstractions, but rather multiple _incompatible_ abstractions, thrown together impatiently.
In general I agree, but given that I'm also giving percentages, I thought 1/190 was more informative? So I prefer "50/10,000" to "1/200", but I think I prefer "0.5% or 1/200" to "0.5% or 5/100,000"?
It was this passage that got me, I think because it goes from 1/low hundreds to 1/high thousands :
Sheltered non-chronic homeless make up 1 out of 5900 people in Mississippi but 1 out of 210 in New York. Unsheltered chronic homeless make up in 1 in 7000 in Minnesota but 1 in 936 in California.
Re-written:
Sheltered non-chronic homeless make up 17 pcm (per 100,000) people in Mississippi but 476 pcm in New York. Unsheltered chronic homeless make up 14 pcm in Minnesota but 107 pcm in California.
___
For me, the mixed base fractions are harder to compare.
How unexpected—the problem is reducing because we are making effort to reduce it! Also, very cool to see that the city runs an independent tent count, which corroborates the numbers.
Does New York being the epicenter of the migrant crisis have any influence on the statistics? I.e. are the illegal immigrants/migrants/asylum seekers/whatever-name-won't-raise-someone's-ire temporarily housed by the City of New York counted in the statistics?
Yes, I believe the statistics count all people, with no regard for citizenship or visa status, etc. However, they don't appear to directly collect any fields related to any of this, so it wouldn't be trivial to directly check the impact of migrants. They do collect fields for other vaguely related things, e.g./age/race/sex/veteran-status, which can get detailed to the point of seeming very invasive at times. For example, there's a column called "Overall Homeless Veterans - Gender Questioning", where the total number in 2023 in the entire country was 29.
So either the survey is behind the curve on the relevant questions, or whoever gathers the data doesn't *really* want to know the answer.
> "Overall Homeless Veterans - Gender Questioning" - well that's as far from "don't ask, don't tell" as we can get... I would assume gender dysphoria/GID is a real reason for run-away teens who end up homeless, not so sure how the military influences it...
Readers who are interested in the topic of homelessness may also wish to read an excellent recent post by Scott Alexander, the blogger who writes "Astral Codex Ten":
https://www.astralcodexten.com/p/details-that-you-should-include-in
There's also a follow-up post:
https://www.astralcodexten.com/p/highlights-from-the-comments-on-mentally
Interesting overview. It would be easier to draw conclusions about local areas if charts 17-20 were shown as rates to keep them consistent with the rest of the article. San Francisco's population drop over the last few years might explain some of the decrease, and the weird sawtooth pattern for mental illness is probably a reporting change.
When you get the local region data, it would be interesting to see a comparison of different types of homelessness vs. mean low temperature in January.
If you're feeling brave, add a comparison of the political affiliation of each region as well. I've always wanted to see an unbiased resolution to the online debates of whether weather or policies attract chronically homeless people.
I have my suspicions based on four years of living in Colorado and seeing how it compares to neighboring and western states, but of course cost of housing would be a major confounder here as well.
Yeah, I'd rather have percentages rather than absolute counts. The reason for switching to absolute counts for the mental health / substance abuse plots is that, as far as I know, there are no reliable population estimates for local regions.
I'll think about the other plots. Just from looking at the maps, my impression is that:
(1) Colder states have a smaller fraction of unsheltered and/or chronic homelessness (strong effect).
(2) Colder states, all else being equal, have smaller numbers of homeless people (weak effect)
(3) Politically conservative states have smaller numbers of homeless people (medium effect)
Of course, causality is debatable, though I think the mechanism for (1) is pretty clear.
Is there a reason why the San Francisco unsheltered data seems to be bucketed in 2-year segments? That seems like it wouldn't follow the HUD requirements.
The x-axis looks pretty wonky, but it is yearly data. The only exception is that the January 2021 data is missing (because of Covid)
Just double-checked this—the sheltered count is indeed done every year in San Francisco, but the unsheltered count only happens every other year, which is apparently acceptable to HUD. See https://www.hudexchange.info/programs/hdx/pit-hic/#2024-pit-count-and-hic-guidance-and-training
and https://www.sf.gov/data/healthy-streets-data-and-information
Many thanks for the correction! You can actually trace this back to the individual reports, e.g.
https://files.hudexchange.info/reports/published/CoC_PopSub_CoC_CA-501-2022_CA_2022.pdf
https://files.hudexchange.info/reports/published/CoC_PopSub_CoC_CA-501-2023_CA_2023.pdf
As you suggest, the sheltered numbers are different, but the unsheltered numbers are exactly the same. (Except some of the details are deleted for 2023? Hmmm.) Glancing through other places, there seem to be a few other areas where the same thing is happening:
https://dynomight.net/img/homeless/shelt_and_unshelt_coc_svg/CA-Chico,%20Paradise%20Butte%20County.svg
https://dynomight.net/img/homeless/shelt_and_unshelt_coc_svg/CA-Oakland,%20Berkeley%20Alameda%20County.svg
Really enjoyed this article. I've been reading your blog for a few years and it's been such a gift to watch your interests and voice change in many ways while also maintaining a real & grounded identity. I wonder what US homelessness looks like compared to homelessness in the other countries mentioned, especially the colder ones. I also was wondering if you would be open to sharing the source code you used to come up with these visualizations?
Thank you! I *could* share the code, but I must warn that would probably be disturbing to read. It would be irresponsible to share as anything other than a cautionary tale, an example of what happens when you don't just have no organization or abstractions, but rather multiple _incompatible_ abstractions, thrown together impatiently.
hahhhhahahhahahahhahahahahhahhahah
Small request: instead of 1/190 or 1/237 or whatever, please use _per 100,000_
In general I agree, but given that I'm also giving percentages, I thought 1/190 was more informative? So I prefer "50/10,000" to "1/200", but I think I prefer "0.5% or 1/200" to "0.5% or 5/100,000"?
It was this passage that got me, I think because it goes from 1/low hundreds to 1/high thousands :
Sheltered non-chronic homeless make up 1 out of 5900 people in Mississippi but 1 out of 210 in New York. Unsheltered chronic homeless make up in 1 in 7000 in Minnesota but 1 in 936 in California.
Re-written:
Sheltered non-chronic homeless make up 17 pcm (per 100,000) people in Mississippi but 476 pcm in New York. Unsheltered chronic homeless make up 14 pcm in Minnesota but 107 pcm in California.
___
For me, the mixed base fractions are harder to compare.
https://www.sf.gov/news/new-data-san-francisco-street-homelessness-hits-10-year-low
There is a big program to reduce unsheltered homeless. I am glad it is making progress.
They are also stepping up enforcement, so I wonder if some of the decrease is caused by the chronically unsheltered moving to other towns.
How unexpected—the problem is reducing because we are making effort to reduce it! Also, very cool to see that the city runs an independent tent count, which corroborates the numbers.
Very nice work!
Does New York being the epicenter of the migrant crisis have any influence on the statistics? I.e. are the illegal immigrants/migrants/asylum seekers/whatever-name-won't-raise-someone's-ire temporarily housed by the City of New York counted in the statistics?
Yes, I believe the statistics count all people, with no regard for citizenship or visa status, etc. However, they don't appear to directly collect any fields related to any of this, so it wouldn't be trivial to directly check the impact of migrants. They do collect fields for other vaguely related things, e.g./age/race/sex/veteran-status, which can get detailed to the point of seeming very invasive at times. For example, there's a column called "Overall Homeless Veterans - Gender Questioning", where the total number in 2023 in the entire country was 29.
So either the survey is behind the curve on the relevant questions, or whoever gathers the data doesn't *really* want to know the answer.
> "Overall Homeless Veterans - Gender Questioning" - well that's as far from "don't ask, don't tell" as we can get... I would assume gender dysphoria/GID is a real reason for run-away teens who end up homeless, not so sure how the military influences it...
Very useful.
Can you write something about the impact of mess & elite migration on productivity?
Larry Fink is interested.