History of traumatic brain injury
Pooled across 18 studies and 9,702 homeless and marginally housed participants. A history is not a current impairment.
95% CI 46.4 to 59.7%
Stubbs et al., Lancet Public Health, 2020Homelessness tracks rents.
Rents track the rules.
Across US metros, rent and vacancy explain where homelessness concentrates. Poverty, drug use and mental illness do not. The rules that push rents up, and the cheap rooms they outlaw, never appear on the bill for managing the result.
Established4 versus 40 per 10,000
The highest poverty rate in the geography table belongs to Mississippi, which has a tenth the homelessness rate of Massachusetts, the lowest.
Poverty does not predict it. Rent does.Probable15,000 to 22,000 people
A 2% to 3% effect per 10% real minimum-wage increase, against the January 2025 count. The study and its critics.
One peer-reviewed estimate, inside a literature that disputes it.Established175,000 rooms removed
Single-room units eliminated in New York from 1955 onward, comparable to all its public housing. What replaced them is illegal to build.
The cheapest legal room, removed by code.Each number carries a tier: established means independent literatures agree, probable means one credible study in contested company.
The argument: the claim, housing, work, care, the ledger, objections, tests.
Lookups: geography table · cost ledger · commute calculator · accounting mistakes.
The argument, once
Stated in full here, and nowhere else.
Where homelessness concentrates in the United States is explained by rents and vacancy, not by poverty, drug use, mental illness or weather. Rents are highest where rules constrain housing supply and where the cheapest legal housing forms have been abolished. So the cheapest intervention against homelessness is repeal, and repeal never appears in the budget debate, because it has no line item and the people it would help are not in the room. The full ledger has two columns, each with a hidden half: the costs of homelessness that land outside public accounts, and the costs of prevention that are political rather than fiscal. Count both, then compare reforms on outcomes and rights instead of appropriations.
| Tier | Claim | Basis |
|---|---|---|
| Established | Rent and vacancy predict regional homelessness rates. Individual risk factors predict who within a metro rather than which metro. | Colburn & Aldern 2022, plus the state pattern. |
| Established | Land-use restriction raises land and housing prices in constrained metros. | Gyourko & Krimmel 2021, Glaeser & Gyourko 2018, and four supply experiments. |
| Established | The cheapest legal housing form was regulated out of existence and is hard to rebuild. | Sullivan & Burke 2013, Pew 2025, Furman Center 2018. |
| Probable | Minimum-wage increases raise homeless counts at the margin. | One peer-reviewed study against an aggregate literature that finds the channel small. |
| Probable | Psychiatric financing and capacity rules shift care costs to families, emergency rooms and jails. | Bed counts and utilization, no causal conversion. |
| Open | Whether commitment standards should change. | A rights question, stated as a tradeoff and left open. |
What working looks like. You should be able to say why which city is a rent question rather than a drug question, name the rules that would have to change for the cheapest housing form to return, and audit a local cost figure against the four quadrants.
Three rungs, drawn to the weight of their evidence
Everything above the gap is intact and the ladder reaches the same height. The question is where a person starts climbing, and what is under them when they slip.
- 01The roomSingle-room and shared-facility housing, priced or regulated out of the market.Established
- 02The jobEntry-level bargains foreclosed by wage floors and barriers to entry.Probable
- 03The treatmentFinancing rules and capacity limits, drawn solid. The commitment standard, drawn dotted, is a rights question left open.Probable Open
01 / Housing Established
Rent decides who falls off the bottom of a local market.
Supply rules and minimum standards→Fewer cheap rooms, higher entry rent→More people priced past the bottom
The mechanism. Every metro has a lowest legal rent, and where that floor sits decides how thin a person's margin can get before the market has nothing left for them. Individual trouble decides who reaches that margin. The floor decides how many people it swallows, and only the floor varies enough to explain San Jose against Detroit.
The geography evidence
If poverty, addiction or mental illness drove homelessness, the poorest and sickest states would have the most. They have the least.
| State | Homeless per 10,000 | Poverty rate | 2-bedroom fair market rent |
|---|---|---|---|
| New York | 73 | 14.0% | $2,780 |
| California | 46 | 11.8% | $2,625 |
| Massachusetts | 40 | 9.7% | $2,837 |
| Washington | 40 | 9.9% | $2,671 |
| West Virginia | 11 | 16.7% | $1,054 |
| Michigan | 10 | 13.4% | $1,378 |
| Texas | 9 | 13.4% | $1,529 |
| Mississippi | 4 | 17.8% | $1,232 |
| United States | 22 | 12.1% | 745,652 counted |
measured HUD 2025 AHAR Part 1, Appendix A (May 2026) · Census ACS Brief ACSBR-026 (Sep 2025) · HUD FY2025 Fair Market Rents, 40th-percentile gross rent for each state's principal metro.
Read the two ends. Mississippi has the highest poverty rate here and a tenth the homelessness rate of Massachusetts, the lowest. West Virginia leads the country in overdose deaths at 48.9 per 100,000 in 2024, double California's 21.7, at a quarter of California's homelessness rate. The variable moving with the outcome is rent, which is the state-level version of the cross-metro pattern in Colburn and Aldern.
The boundary. Rent explains which city, and individual risk explains which person inside it. Untreated psychosis in Jackson is a crisis whatever the rent is.
What the rules do to the price
Evidence: the regulatory wedge in land and housing prices
| Measure | Finding | What it supports |
|---|---|---|
| The zoning tax on land | estimated Median regulatory wedge per quarter-acre lot, 2018 dollars, within 30 miles of the center: San Francisco $409,706, Los Angeles $198,769, Seattle $174,850, New York $152,417, against Detroit $10,089 and Dallas $2,217. Gyourko & Krimmel, JUE 126, 2021, Table 1 | From large vacant-lot sales, which are scarce and unlike infill parcels. In the coastal metros the wedge runs two to four times median household income. |
| Price against cost | estimated A median San Francisco home costs about $281,690 to build and deliver at a normal margin against a median price of $800,000. Manhattan permitted 13,000 units in 1960 alone against 21,000 across the 1990s. Glaeser & Gyourko, JEP 32(1), 2018 | Where price sits far above building cost, the binding input is permission. |
| The standard strictness measure | measured The Wharton regulation index, 2,472 communities surveyed in 2018, standardized to mean zero. San Francisco and New York are the only metros more than one standard deviation above it; Houston, Columbus, San Antonio and Pittsburgh sit within a tenth. Gyourko, Hartley & Krimmel, JUE 124, 2021 | Built from survey answers, with no price data in it, so its agreement with the zoning tax is not mechanical. |
The abolished form
The cheapest legal housing in America was a rented room: an SRO hotel, a rooming house, a lodging house. Code, zoning and urban renewal removed it. New York eliminated roughly 175,000 SRO units from 1955 onward, comparable to all of its public housing. Sullivan & Burke, 17 CUNY L. Rev. 113 (2013), an estimate by two legal-services attorneys that later citations trace back to. Chicago lost 32,000 rooms, 80% of its stock, from 1973 to 1984. Pew, July 2025.
The test that matters now is whether the unit can be built again. New York's Housing Maintenance Code prohibits new rooming units outside a narrow nonprofit channel, and its zoning density factor allowed a sample lot with 60,000 square feet of floor area only 88 apartments, averaging 681 square feet. Furman Center, Feb 2018. Seattle allowed units as small as 90 square feet until 2013, then set a 220-square-foot floor, and the format stopped. Most large cities still mandate unit sizes, restrict shared kitchens and baths, or require parking. Washington State reversed course in 2024, requiring cities to allow single-room housing wherever six-unit multifamily is allowed.
The published model
What the outlawed unit looks like
Studios, private facilities
$1,820 per month, 400 sq ft each
Rooms, shared kitchen and bath
$1,040 per month, 160 sq ft each
The Furman Center modeled both formats in one 67,100-square-foot New York envelope, asking what rent each needs to support construction at a competitive return. At $200 per square foot of land: $1,820 − $1,040 = $780 a month, or $9,360 a year, and 251 − 126 = 125 additional places to live. On free land the shared room needs $840 against $1,480. Furman Center, 2018, Tables 2 and 3.
What the figure is. The difference between two building formats, in one city, in 2018 dollars. The regulatory share is the part of the gap that exists because the smaller unit is illegal, which the model does not isolate. The room buys less space and a shared bathroom, and that is the trade the tenant is forbidden to make.
Four places that put a rung back
| Place and change | What happened | What it does not prove |
|---|---|---|
| Auckland, 2016 Unitary Plan | estimated Most of the city upzoned at once. 21,808 additional dwellings permitted in five years, 4.11% of the region's stock. Greenaway-McGrevy & Phillips, JUE 136, 2023. Six years on, rents run about 28% below the synthetic counterfactual. Greenaway-McGrevy & So, Economic Inquiry, 2026 | Two papers and two methods, each resting on a constructed comparison. Critiques of the construction result were rejected in Donovan & Maltman, 2025. |
| Minneapolis, 2040 plan and the end of parking mandates | measured From 2017 to 2022 the city added 12% to its housing stock while rents rose 1%, against 4% and 14% in the rest of Minnesota, a gap worth about $1,700 a year per renter. Hennepin County homelessness fell 12% while rising 14% statewide. Pew, Jan 2024 | Compared trends rather than an experiment, by an organization that argues for these reforms. A 2025 study credits weaker demand instead, a disagreement about mechanism, not direction. |
| Houston, minimum lot size 1998 and 2013 | measured The by-right minimum lot fell from 5,000 to 3,500 square feet inside the Loop in 1998 and citywide in 2013. Roughly 80,000 houses have since been built on lots the old rule forbade, and land values show no increase, with estimates running to 9% lower. Hamilton, Cityscape 26(3), 2024. The regional count went from 8,538 in 2011 to 3,325 in 2025. Coalition for the Homeless | A joint treatment: cheap supply and coordinated housing arrived together and cannot be separated here. The count's geography also changed. |
| Tokyo, uniform national zoning rules | measured Tokyo's private rent index sits about 5% below its 1994 peak in nominal terms and 15% below in real terms. Statistics Bureau CPI, 2020 base. The city started 2.77 million dwellings from 2006 to 2025 while gaining 1.6 million residents. MLIT | Japan sets thirteen zone categories nationally while municipalities map them, so this is different zoning rather than none. Population decline pushes the same way. |
Four instruments, one result: the constraint was permission, and removing it moved quantities within a few years.
The free-market response: let people offer and choose small rooms, shared facilities and small lots, and make each remaining restriction justify itself against the alternative it removes rather than an imaginary better apartment. Keep enforceable protection against fraud and hazard. Success is measured by the entry price of the cheapest legal unit.
02 / Work Probable
A higher legal wage guarantees a wage, not a job.
Wage floors and entry barriers→Some bargains stop being offered→Income and experience forgone at the margin
The mechanism. When an employer's expected value from a hire falls below the required cost, that hire stops being offered. The worker who loses is the one closest to the line: least experienced, least healthy, longest out of work. Licensing costs close the self-employment route the same way. This is a claim about a tail, and it needs tail evidence.
The lead estimate. A 10% year-over-year real increase in a local minimum wage is associated with a 2% to 3% relative increase in homeless counts across 203 localities from 2006 to 2019. Localities in the top third of real increases from 2012 to 2015 ran about 21% higher in 2016 to 2019 than those whose real minimum fell. Hill, Southern Economic Journal 92(4), 2026. Against the January 2025 count of 745,652, that is roughly 15,000 to 22,000 people, if the estimate holds and generalizes.
Evidence: the case against this mechanism, at full strength
| Finding | Evidence | What it settles |
|---|---|---|
| Aggregate effects are small | estimated Across 138 state minimum-wage changes, 1979 to 2016, employment of affected workers changed +2.8% (s.e. 2.9) while their wages rose about 6.8%. The minimums studied ran 37% to 59% of the local median wage. Cengiz, Dube, Lindner & Zipperer, QJE 134(3), 2019 | Counts missing jobs below the new minimum against new jobs above it, and the strongest evidence that Hill's channel is small. |
| The international review agrees | estimated Effects stay muted up to about 59% of the median wage, and to 81% in low-wage counties. Dube for HM Treasury, Nov 2019 | A commissioned review rather than a peer-reviewed paper, setting the range in which that holds. |
| The tail behaves differently | estimated A $1 real increase cut employment of young adults with cognitive disabilities by 0.9 to 1.0 points, 1.6 points without college, with nothing significant for ambulatory or sensory disability or the nondisabled. Chiswick, Corman, Dave & Reichman, SEJ, June 2026 | Direct evidence for the concentration claim, in the population the bottom rung describes. |
| Evidence pointing the other way | estimated Among 6 million people released from prison, a $0.50 increase cut the probability of returning within a year by 2.8%. Agan & Makowsky, Journal of Human Resources 58(5), 2023 | For one group with very weak attachment the higher wage pulled people into legal work. |
| Licensing and entry barriers | gap No quantified link from licensing to homelessness inflow was located. | Needs entry-barrier measures matched against inflow, not employment. |
The honest position. Hill's channel is the same effect the bunching literature finds near zero in aggregate. His argument is that it lands on workers too few to move an aggregate estimate, and the disability evidence supports that reading. One study of homeless counts is not a settled parameter, and if the wage effect vanished tomorrow the geography of homelessness would look the same.
03 / Care Probable
Closing an institution does not remove the need for care.
Financing and capacity limits→Unmet need for sustained care→Families, emergency rooms, jails, streets
The mechanism. For some people with severe mental illness, a cheap room and a job leave the problem intact. When treatment is unavailable and community support cannot substitute, the need moves to relatives and crisis systems, the most expensive places to meet it. That is displacement, not saving.
The economic half: who pays for the bed
Financing follows the building
Since 1965 Medicaid has excluded federal payment for adults aged 21 to 64 in an institution for mental diseases, defined as a facility of more than 16 beds primarily treating mental illness. Social Security Act § 1905(a)(30)(B) and § 1905(i), per CRS, Jan 2025. A payment rule keyed to building size shapes what gets built.
The exception is a waiver
As of January 2025, 36 states held approved Section 1115 demonstrations for substance-use care and 15 plus the District of Columbia for serious mental illness. CMS. Each is time-limited, so capacity planning runs on a renewal cycle.
Capacity is a separate limit
State hospitals held 36,150 beds, 10.8 per 100,000 people, in the 2023 census, 52% of them occupied through the criminal courts. Treatment Advocacy Center, Jan 2024. It counts state hospitals alone, and comes from an organization that argues for more beds.
Evidence: where the displaced care shows up
| Where it lands | Evidence | Boundary |
|---|---|---|
| The emergency department | measured 310 visits per 100 people per year among people experiencing homelessness in 2020 to 2021, against 40 per 100 for everyone else, up from 141 a decade earlier. CDC/NCHS, MMWR 72(42), Oct 2023 | A federal survey with a matched comparison group. |
| The jail | estimated 14.5% of male and 31.0% of female jail entrants met criteria for current serious mental illness on structured interview. Steadman et al., Psychiatric Services 60(6), 2009 | The best instrument in this literature, and five jails in two states. A federal survey using a different measure put distress at 26%. |
| The family | gap No national estimate of hours or income lost to caregiving for an adult relative with untreated severe mental illness. | The largest displacement channel has no series behind it. |
The rights half: commitment is a different argument
O'Connor v. Donaldson, 422 U.S. 563 (1975) holds that a state cannot confine, without more, a nondangerous person capable of surviving safely in freedom alone or with willing help. The case on this page is a case for voluntary exchange and against coercion. Expanding involuntary commitment is a rights cost that has to be argued on its own terms, against the liberty it takes. Nothing in the housing or financing argument decides it.
The reform that follows from the economics: pay for treatment according to need rather than the number of beds in the building, and count the emergency rooms, jails and families absorbing the difference.
The bottom rung has to hold people with impairments.
The people most dependent on a cheap room, an attainable job or sustained treatment are also the most likely to be managing brain injury, psychiatric symptoms or cognitive difficulty.
Psychosis
Pooled across 31 studies in 13 countries. The same review puts schizophrenia at 10.3%, a subset rather than an addition.
95% CI 13.7 to 31.3%
Ayano, Tesfaw & Shumet, BMC Psychiatry, 2019Cognitive impairment
Summary estimate across 24 studies and 2,969 adults, covering several acquired and developmental conditions.
Methods and samples vary substantially
Depp et al., J Nerv Ment Dis, 2015Separate estimates, overlapping people. Different samples and definitions, so these cannot be added into a combined rate, and none establishes what came first. HUD's sheltered-population estimate puts disability at 57.5% among adults with known status in 2022, on a broader definition. HUD, 2022 AHAR Part 2
The implication this page will not dodge. If about half of people experiencing homelessness carry a brain-injury history and about a quarter show measurable cognitive impairment, then a cheap room and a job are the complete fix for some and not for others. Rent still governs where homelessness concentrates, because it decides who falls off the bottom of a local market. The reform package still has to carry care capacity, because repeal alone would leave the most impaired share where it found them. This is the strongest reason on the page to treat care as a mechanism at all, and why the third rung is drawn thinner than the first rather than left out.
Two columns, each with a hidden half
Budget debate compares one quadrant against one other.
A ledger with one column is not a ledger. Set the costs of homelessness against the costs of preventing it, then split each column by what lands in public accounts and what lands outside. The quadrants are the argument, and the figures inside them are examples that do not add to a total.
| Column A / the costs of homelessness | Column B / the costs of preventing it | |
|---|---|---|
| Seen appears in public program accounts | Emergency care, jail, shelter, outreach, cleanup, sanitation. Public budget Sanitation | Housing capital and operations, treatment, case management, vouchers. Cost per person |
| Unseen borne outside those accounts | Lost public space, hardened and removed amenities, avoidance commutes, protective spending, damage that never becomes a claim, victimization of people outside. Public space Hardening | Political costs: repealing a rule that has a constituency, restoring treatment capacity, enforcing against people with nowhere to go. Each has an owner and an electoral price. The housing rules |
The asymmetry that decides the debate. Column A's unseen costs are diffuse and unowned. Column B's are concentrated and owned: the neighbor who wants the parking minimum, the board that wants the lot size, the union that wants the wage floor. A cost with an owner gets defended at the hearing. A cost spread across everyone who no longer sits down downtown gets nothing, because nobody is assigned to notice it. This explains attention and measures nothing.
The sign inversion
One category runs backwards through the accounts. Remove a bench, close a restroom or replace a plaza with boulders, and the budget records a saving while the people who used it record a loss no document holds. A ledger built from budget files reads the removal of a commons as prudence. Washington State spent close to $700,000 on boulders at former encampment sites, a rare case where hardening was priced at all, and the removals it stands in for are priced nowhere. The hardening register holds the rest.
Bastiat's question applies to both columns: what would those resources have produced elsewhere, and what would exist had the rule never been written. Bastiat, What Is Seen and What Is Not Seen. Broken windows is a separate empirical claim about disorder and policing, and opportunity-cost logic does not settle it. Wilson & Kelling.
Who pays when the option disappears
The person excludedHousing, earnings, autonomy, safety.
Family and friendsCaregiving, crisis management, lost work time.
Neighbors and businessesDamage, protective spending, lost use of public space.
Taxpayers and service usersEmergency care, shelter, enforcement, and whatever those displace.
Paying twice means the lost opportunity and then the coping cost. Counting one loss twice is the error the accounting mistakes name: rent is a transfer, lost wages and lost output overlap, and property values may already capitalize the losses above.
Eight objections, answered
Each gets an answer, and the part the answer leaves standing.
01
It is drugs and fentanyl, not rent.
For it: street homelessness and open drug use appear together in every photograph.
Answer: compare places. West Virginia leads the country in overdose deaths at 48.9 per 100,000 in 2024 against California's 21.7, with a quarter of California's homelessness rate. CDC/NCHS. Kentucky, Tennessee and Ohio also beat California on overdose deaths and stay far below it on homelessness. Addiction is distributed like despair, and homelessness like rent.
Residual: drugs still explain much about who becomes homeless inside a city.
02
SROs vanished because the land got valuable, not because of the code.
For it: converting a rooming house on expensive land needs no ordinance to explain it.
Answer: both happened, and the test that matters is prospective. Can a developer build that unit today in a city that wants it? New York bans new rooming units outside a narrow nonprofit channel, and Seattle's 220-square-foot floor ended the format there in 2013. Pew finds SROs were profitable when allowed, which is why prohibiting them was necessary.
Residual: some conversion was market-driven, part of the stock was dangerous, and urban renewal was itself a public program.
03
Minimum-wage employment effects are near zero in the best studies.
For it: the bunching estimator covers 138 state changes and rules out own-wage elasticities past −0.45.
Answer: correct about the aggregate, and the work section states it as the strongest evidence against this page's second mechanism. The narrower claim is concentration: a $1 real increase cut employment of young adults with cognitive disabilities by about a point while moving nothing measurable for the nondisabled.
Residual: testable rather than settled, and it runs the other way for people leaving prison, where higher minimums reduced one-year returns to custody.
04
Housing First has failed. Look at the West Coast.
For it: California and Washington spent heavily under that policy and their counts rose.
Answer: separate the intervention from the constraint. Denver's randomized trial kept 77% housed at three years and cut 38 jail days each. Urban Institute, 2021. Finland cut long-term homelessness about 70% from 2008 to 2024. ARA and Varke. No program can house people into units that do not exist.
Residual: Finland's series reversed in 2024 and 2025, which its municipalities blame on a shortage of small affordable homes. Denver's offsets covered half the program cost.
05
Deregulation just builds luxury units.
For it: what gets built after an upzoning rents far above what anyone at risk can pay.
Answer: people at the bottom move into what its tenants left. Tracing 52,000 residents of new buildings back six rounds of moves, a building housing 100 people leads 45 to 70 to move out of below-median-income neighborhoods, mostly within three years. Mast, JUE 133, 2023. The unit is expensive and the chain it starts is not.
Residual: filtering is slow, which answers nobody losing housing this month.
06
These costs are already in property values, so you are double counting.
For it: a house near an encampment already sells for less because buyers priced the lost use of the park.
Answer: correct, and the ledger says so at row level. The Seattle price effect capitalizes what the public-space register would otherwise value again, and the accounting mistakes name the overlap. The point is direction and ownership rather than a sum.
Residual: capitalization covers only nearby owners and renters, pricing nothing for the visitor, the bus rider, or the person in the encampment.
07
You have no total, so you have no argument.
For it: policy runs on comparisons, and a comparison needs two numbers.
Answer: the argument is about which column the comparison omits. Every sourced row here is a lower bound on a category the budget treats as zero, and that zero is a number chosen without evidence. A total assembled from different cities and years would be worse than an honest gap.
Residual: this cannot tell a council which program to fund. It can tell them what their comparison left out.
08
Camping bans work.
For it: Grants Pass v. Johnson (2024) lifted the Eighth Amendment constraint, and a cleared block does improve.
Answer: the measured effect is local and short. Across 303 Denver displacements, crime within a quarter mile fell 9.3% at seven days and 3.9% at twenty-one, with no consistent change further out and crimes against persons clustering nearby. Padmanabhan et al., 2024. Enforcement without a destination moves the cost, and the ledger shows where.
Residual: the public-order gain is real for the people on that block, and enforcement paired with shelter capacity is a package no study here prices.
Restore the options. Then measure the result.
The change, the outcome to measure, and the cost to count against it.
Legalize the cheap room.
Change: minimum unit and lot sizes, parking minimums, occupancy limits, and the rules that make single-room housing unbuildable.
Measure: permits and completions at the bottom of the market, entry rents rather than average rents, and inflow into homelessness.
Count against it: construction resources, fire and habitability risk in the units built, effects on other tenants.
Mechanism and evidenceWiden the entry to paid work.
Change: licensing scope, day-labor and self-employment barriers, and the wage floor as it applies to the least employable.
Measure: hiring, hours and earnings for workers with disability, records or long gaps, and their housing retention.
Count against it: lower pay for workers who keep their jobs, and unsafe arrangements. The aggregate literature finds these effects small, so measure the tail.
Mechanism and evidencePay for treatment where the patient is.
Change: the Medicaid exclusion for large psychiatric institutions, waiver renewals, and the licensing limits on residential capacity.
Measure: bed-days available, waiting time from crisis to placement, continuity after discharge, repeat emergency and jail contacts.
Count against it: treatment cost, institutional harm, and the risk that capacity built for voluntary care becomes capacity used for confinement.
Mechanism and evidenceArgue commitment separately.
Change: the commitment standard itself, which is a question about liberty rather than supply.
Measure: where a state changes it, outcomes a year later for the people committed, including mortality and housing.
Count against it: coercion, lost liberty, due-process burden.
The tradeoff statedThe cheapest intervention is the one that was prohibited. Judge a restriction against the housing, work and care a person could obtain without it, and reform against the cost it leaves behind.
How this page was built
Every figure comes from the primary series named beside it. Where only an advocacy source exists, the row says so. Where nothing defensible exists, it says gap, because a gap is information and an invented number is not. Volatile inputs move annually: the count each December, poverty each September, rents each October. Court decisions move without warning, so dates sit inline and a stale number stays visible as one.
Related: Housing Comparison: From Basic to Billionaire puts the room at the bottom of this argument beside everything above it. Red State vs Blue State Economies holds the state cost-of-living series. Economic Systems Compared covers the price-signal argument underneath.
Follow any part of the bill.
The registers hold the sourced rows, boundaries and calculators behind the argument above. Each one opens on its own.
The ledger: who bears each category
The ledger
| Category | Who bears it | Status | Figure or gap | Source and boundary |
|---|---|---|---|---|
| Public budget | Taxpayers | estimated | NYC cohort $40,451 per person-year, 1999 dollars. Denver: $6,876 of emergency costs avoided per participant-year. | Culhane et al., 2002 Urban Institute, 2021 A baseline and a treatment effect differ. |
| Property crime | Residents, merchants, insurers, people living outside | estimated gap | Boulder: crime about 3× as concentrated near reported encampments. No dollar total. | Reinhard & Stafford, 2023 Spatial association only. |
| Public-space exclusion | Park users, transit riders, neighbors | estimated gap | Seattle sales about 2.0% lower within 0.75 miles of encampments. | Kallberg & Shimizu, Aug 2026 A capitalized effect, not an annual access bill. |
| Business and tax base | Owners, workers, landlords | measured gap | San Francisco office vacancy 34.2%, Q4 2024. Attributable share unidentified. | Cushman & Wakefield, Q4 2024, p. 4 Remote work and financing moved at once. |
| Sanitation and health | Agencies, patients, neighbors | measured | San Diego hepatitis A: 592 cases, 407 hospitalizations, 20 deaths, and about $12.5 million of county response through April 2018. | County after-action report, May 2018 Onset dates span Nov 2016 to Oct 2018, and it overlaps the public-budget row. |
| Victimization | People living outside | estimated gap | 38% experienced physical or sexual violence during the current episode. | UCSF, June 2023 Episode prevalence, not an annual rate. |
| Private security | Commercial owners, tenants, customers | measured gap | Downtown Denver district: $1,056,568 on security in 2024, inside $8,488,133 of spending. | Denver BID, 2024, p. 10 All-cause district spending, so a ceiling rather than an estimate. |
| Hardening and lost amenity | Everyone using public space | measured gap | Washington State: nearly $700,000 in defensive boulders. No inventory of removed seating or restrooms exists. | KUOW, Jan 2024 Removals enter budgets as savings, so this row carries the wrong sign. |
| Avoidance and sorting | Households who move or reroute | estimated gap | Roughly one resident lost per additional reported crime. | Cullen & Levitt, 1999 Crime is a different exposure. Register |
Read across, not down. Cleanup already sits inside public spending, and lost access inside property prices.
The public-budget number
The public-budget number
| Evidence | Finding | Boundary |
|---|---|---|
| NYC supportive-housing cohort | estimated $40,451 per person-year before placement, 1999 dollars; housing at $17,277 against $16,282 of offsets, a net $995 per unit-year. Culhane et al., 2002 | Service records at average per-diems, dominated by health care, for a selected severe-mental-illness cohort. The unit figure adjusts for occupancy and cannot be subtracted from the per-person figure. |
| Denver randomized trial | estimated 34% fewer police contacts, 40% fewer arrests, 38 fewer jail days, and $6,876 per participant-year of avoided emergency costs against a program cost of $12,078 to $15,484. Urban Institute, 2021 | Random assignment among heavy users of justice and emergency systems, so a result for that group rather than a profile for everyone. |
Public-space exclusion
Public-space exclusion
| Measure | Evidence | Boundary |
|---|---|---|
| Seattle home prices | estimated About 2.0% lower within 0.75 miles, an average effect of $17,445. Kallberg & Shimizu, International Review of Finance, Aug 2026 | From 2017 to 2020 transactions, bundling safety, nuisance and access into one number. A Los Angeles working paper finds 1.41% within 0.2 miles. |
| Transit satisfaction | measured BART rider satisfaction 73% in 2024 against 67% in 2022. BART survey, Feb 2025 | Counterevidence to a simple deterioration story, though surveys miss whoever stopped riding. |
The hardening of public space
The hardening of public space
A removed bench leaves no invoice. It leaves an absence.
| Item | Figure | What it shows |
|---|---|---|
| Defensive landscaping, Washington State | measured Nearly $700,000 in boulders at former encampment sites, $643,000 at one interchange. KUOW, Jan 2024 | One agency, three sites, and a rare case where hardening carried a price. |
| Corporate restroom access | measured Open to non-customers from 2018, purchase required from 27 January 2025. NPR, Jan 2025 | Much of the usable urban restroom network sits on private property. Access to it changed nationwide by memo, with no entry in any municipal account. |
| Seating, hours, fountains, outlets | gap No inventory, no cost series, no removal dates. | No number exists. |
Crime and violence, both directions
Crime and violence, both directions
| Study | Finding | Accounting implication |
|---|---|---|
| Boulder, 2021 to 2022 | estimated About 3× crime concentration near resident reports, 7× near police-identified encampments. Reinhard & Stafford, GeoJournal, 2023 | Location quotients rather than individual risk ratios, and encampments form where conditions were already bad. |
| Denver displacement, 2019 to 2023 | estimated Crime within a quarter mile fell 9.3% at 7 days and 3.9% at 21. Journal of Urban Health, 2024 | Retrospective, with nothing consistent further out. Objection 8. |
| Violence against homeless people | estimated 38% experienced physical or sexual violence during the current episode. UCSF California study, June 2023 | Self-reported episode prevalence, so it is not an annual rate. |
Avoidance and the commute
Avoidance and the commute
| Element | Evidence | Boundary |
|---|---|---|
| The sorting mechanism | estimated Roughly one fewer city resident per additional reported crime, through out-migration, with movers staying inside the metro. Cullen & Levitt, REStat, 1999 | Crime rather than homelessness, on an older panel. It shows households answering street conditions by moving out while staying in commuting range, and prices none of it. |
| The inputs for pricing it | measured Mean one-way commute 27.2 minutes in 2024. ACS 2024. $11,577 a year to own and run a new vehicle at 15,000 miles, about 77 cents a mile. AAA, Sep 2025. USDOT values local personal travel time at 50% of hourly income. USDOT guidance | The commute series is all-cause, and the AAA figure is full ownership across 45 models, so the per-mile rate overstates a marginal mile. |
| The homelessness share | gap No study identifying it was located. | Needs exposure at the origin, moves with stated reasons, and a matched metro. |
Private-cost worksheet
Price your own avoidance commute
estimated This prices a commute. It does not attribute one. If you would have moved anyway, the whole figure belongs to those reasons.
estimated About $9,146 per year: $3,833 of time and $5,313 of driving.
Roughly 192 hours in transit, or about 8.0 full days a year, and about $182,927 over twenty undiscounted years.
Roughly 192 hours in transit, or about 8.0 full days a year, and about $182,927 over twenty undiscounted years.
Cost per person, compared
Cost per person, compared
| Approach | Public cost or outcome | Comparison limit |
|---|---|---|
| Street homelessness, usual services | estimated $40,451 per person-year, NYC cohort, 1999 dollars. Culhane et al., 2002 | A selected cohort in one city a quarter century ago, with no credible external range beside it. |
| Housing First, Denver trial | estimated $5,202 to $8,608 additional public cost per participant-year after offsets. Urban Institute, 2021 | $12,078 to $15,484 minus $6,876, for heavy users of jail and emergency services. |
Sensitivity worksheet
What would an omitted cost change?
For one cohort and year, let B be the usual-service public cost, E the net external cost per person-year, and A the alternative's full social cost. Then street total = B + E, and the alternative's advantage is B + E − A. The baseline holds Denver's measured 2021 cost of $12,078. The external fields start blank because no defensible general value exists, and the worksheet stays silent until all four are filled.
Enter all four values to see a range. Everything it produces is arithmetic on your assumptions.
Include in A: recurring services, annualized capital, remaining emergency use and the intervention's own burdens. Include in E only harms B does not already contain.
Ten ways to misread this ledger
Ten ways to misread this ledger
| Tempting claim | What the evidence permits | What would close the gap |
|---|---|---|
| "The full cost is a multiple of the budget." | External costs are omitted. The multiplier remains unestablished. | Same-place, same-year costs, overlap removed. |
| "Public space is the largest cost." | A reasonable hypothesis. The evidence here ranks nothing. | Comparable welfare estimates across categories. |
| "Nearby crime was committed by homeless people." | Proximity measures place. Identity needs other data. | Offense-level data, exposure denominators, a comparison area. |
| "A capital price is an annual cost." | A building lasts beyond the budget year; operations recur. | Annualization with useful life and financing stated. |
| "Add property losses, security and lost visits." | Sale prices already capitalize much of what you would add. | Pick one valuation route, or reconcile the overlap. |
| "Vacancy proves encampments emptied downtown." | Remote work, retail competition and financing moved at once. | Matched locations, pre-trends and encampment timing. |
| "An unmeasured category costs zero." | Missing evidence licenses no figure in either direction. | A disclosed sensitivity range and a measurement plan. |
| "Living far out is a preference, so the commute is free." | Preference and avoidance look identical in a commute time. | Matched movers with stated reasons. |
| "Hostile architecture is a homelessness cost." | Defensive design also targets skateboarding, vandalism and vehicle attack. | Per-installation rationale and dates against comparable places. |
| "Deregulation would have prevented it, so the rules caused it." | Rent predicts where homelessness concentrates and rules raise land prices. Neither counts episodes caused by one rule. | Policy variation with credible timing, measured on inflow. |
How to read the evidence tags
Measured is a recorded count, expenditure or market observation. Estimated is a modeled association, survey estimate or stated arithmetic. Gap means no defensible value was found, which differs from nobody having looked. The tiers in the argument rate claims; tags rate figures.