Housing First: does unconditional housing fail?

The retention paradox at the core of Housing First

Systematic reviews and long-term evaluations consistently show housing stability rates above 80–90% for chronically homeless individuals placed through the model — figures that, for a population long considered the hardest to house, represent a structural break with the historical pattern of street re-entry. And yet the model remains politically embattled. The reason is not that Housing First fails at housing. It is that high retention does not, by itself, solve the upstream flow of new entrants into homelessness, and the policy apparatus on both sides of the Atlantic has spent a decade arguing about which denominator matters.

This is the analytical fault line that runs through every serious critique of the model. Housing First optimizes for one variable — whether a housed individual remains housed. It does not, on its own, optimize for whether the visibly homeless population on a given street, in a given city, declines. Treating those two questions as identical has produced a policy discourse riven by what looks like contradiction but is actually a category error.

Housing First does not fail at housing. It fails, when deployed in isolation, at outpacing the inflow of new entrants into the system.

How the model actually operates

The architecture of Housing First was codified in New York in 1992 by Sam Tsemberis, who inverted the dominant sequencing logic of homeless services. The prior paradigm — what critics and proponents both now refer to as "Treatment First" — required sobriety, psychiatric stability, and program compliance as gatekeeping conditions before a client could access permanent housing. Chronically homeless individuals, by definition the population with the deepest co-occurring disorders, were structurally locked out of the very resource most likely to stabilize them. Tsemberis's restructuring was a procedural inversion, not a therapeutic one. Clients are offered permanent, scatter-site housing immediately, with no treatment precondition, paired with voluntary supportive services that scale to need rather than to compliance.

Three design choices define the operating logic. No treatment gatekeeping removes the abstinence-or-housing binary that had defined intake for decades. Scattered-site placement — dispersed across ordinary rental units rather than concentrated in single facilities — sidesteps the institutional stigma that historically accompanied congregate shelters. Harm reduction orientation allows active substance use to coexist with housing tenure, decoupling shelter from sobriety. The US federal government codified these principles across its homelessness portfolio in 2013, embedding them as binding policy rather than experimental model. Finland adopted the framework nationally in 2008 and built an entire homelessness strategy around it.

The conceptual move from Treatment First to Housing First is best understood as a substitution of acceptance criteria, not an abandonment of clinical ambition. Under the old gatekeeping architecture, treatment failure produced housing failure. Under Housing First, housing tenure functions as a stable platform from which clinical engagement becomes possible — though, as the empirical record shows, that engagement cannot be assumed.

The retention record: what the data does and does not show

Housing First's housing stability figures are the strongest evidence in its favor. Long-term follow-ups and systematic reviews place one-year and multi-year retention rates for previously chronically homeless participants in the 80–90%+ range, with some implementations reporting figures at the top of that band. For a population where pre-intervention street-readmission was the norm, this is the most relevant outcome metric and the one that the model was designed to produce. Critics who dispute the model on retention grounds are arguing against the empirical record.

Finland's national implementation provides the cleanest population-scale test. After adopting Housing First as the structural backbone of its homelessness strategy in 2008, the country has reported roughly a 30% reduction in overall homelessness and a reduction of more than 35% in long-term homelessness. Those are unusually large movements in a policy domain where most interventions produce marginal shifts. The Finnish government has paired those reductions with per-person public expenditure savings estimated at EUR 9,600 to 15,000 annually, drawn from reduced emergency medical care, social services, and criminal justice contact. The cost arithmetic is favorable even before the moral case is counted.

What the data does not show — and what the model's defenders are sometimes reluctant to concede clearly enough — is that Housing First, when measured against secondary clinical outcomes, does not consistently outperform standard treatment models. Housing tenure does not reliably translate into recovery from severe substance use disorders, nor does it produce predictable improvements in mental health functioning when specialized clinical infrastructure is not layered on top. A stable door, even an open one, is not a clinic. The model retains people. It does not, by mechanism, treat them. Conflating those two claims has been the most common analytical error in the model's defense, and it has handed critics an unnecessary line of attack.

The structural objection: why retention does not equal system-level reduction

Population-level homelessness is a flow problem, not a stock problem. The number of visibly homeless individuals on a street at any given moment is determined by two variables: inflow (people entering homelessness over a period) and outflow (people exiting it). Housing First, in its pure form, addresses only the second variable. It places people into housing at a defined rate. If inflow continues to outpace that rate — driven by eviction, wage stagnation, mental illness, addiction, migration, or housing-market dysfunction — the street count will plateau or grow even as the model's own housing program performs well on every internal metric.

This is the analytically serious version of the Housing First vs. Treatment First debate, and it has less to do with therapeutic philosophy than with policy architecture. Critics operating from this objection — and the most rigorous of them are not anti-Housing First but pro-supply-side — argue that a target of roughly ten new permanent supportive housing (PSH) beds per single visible street-homeless individual is a reasonable estimate of what is required to bend the curve downward in cities where the affordable housing supply itself is shrinking. Below that ratio, Housing First becomes a high-quality triage ward inside a hemorrhage.

A program measured only by retention is a program that has been asked to optimize for the wrong denominator.

The friction here is structural, not operational. In high-cost housing markets — much of coastal North America, parts of the UK and Ireland, Western European capitals — the rental stock capable of absorbing scatter-site PSH placement has tightened considerably since the model's diffusion in the 2010s. Landlords willing to accept housing-voucher tenants are scarcer, acquisition costs are higher, and the per-unit subsidy required to place someone has grown. The same model that functions at acceptable cost in Helsinki, where the social housing stock is deep and the rental market is regulated, can require two to three times the public expenditure per placement in markets without that underlying foundation. The model's critics have often elided this difference, treating cost data from one jurisdiction as if it applied universally. It does not.

The Finland counter-example, read carefully

Finland's outcomes are real. They are also not a portable template. Three features of the Finnish housing system preceded the 2008 Housing First launch and made it possible. The country already operated an extensive social rental sector. Eviction rates for non-payment were and remain structurally lower than in most comparable economies. Municipalities had the legal and financial capacity to acquire and dedicate housing units to the program at scale. In other words, Finland did not deploy Housing First into a constrained supply environment. It deployed it into a housing system in which supply had already been treated as a public good for decades.

This sequencing matters because it separates the model's intrinsic performance from the conditions required for that performance to translate into population-level reductions. Critics who point out that Housing First has not produced comparable headline figures in cities that adopted it without parallel investment in affordable supply are not attacking the model. They are pointing to the absence of an enabling infrastructure. The Cicero Institute and OECD-aligned analysts who have raised this objection — and the more measured ones have done so without disputing Housing First's individual-level effectiveness — are arguably making the same point the model's original architects made: housing first requires housing to exist.

A useful frame is to treat Housing First as one subsystem inside a larger architecture. The model governs how a person is housed once a unit exists. It does not, and was never designed to, generate those units. In jurisdictions where unit generation has stalled, the model performs well on its own metrics and the surrounding homelessness count still moves in the wrong direction. The failure then gets attributed to Housing First, when the actual cause is the gap between intake and construction.

What the critique is actually demanding

Read carefully, the published criticisms of Housing First — from think tanks, from frontline service directors, from mayors in cities where street counts have risen despite high retention — rarely call for the model's abolition. They call for two things, both of which are consistent with the model's own logic.

First, supply-side integration. Permanent supportive housing functions as a destination, not as an origin. If the pipeline leading into that destination narrows because of overall rental market dysfunction, the model's effectiveness at the individual level becomes irrelevant to the political question that drives funding decisions. The structural demand is for concurrent capital investment in affordable housing — acquisition funds, social rental expansion, voucher adequacy indexed to local rents, anti-displacement measures tied to PSH placement zones. The aim is to prevent the inflow variable from swamping the outflow variable the model so reliably generates.

Second, layered clinical infrastructure. The model's data on secondary clinical outcomes is not an indictment of Housing First; it is a description of what Housing First does and does not include. Voluntary, in-home psychiatric support, medication-assisted treatment access, addiction medicine consultation, and peer-led engagement are all compatible with the model. Many implementations include them. Some — particularly those that have been forced to scale rapidly with constrained budgets — have not. The criticism here is not philosophical. It is operational: a housing placement without a real clinical on-ramp produces retention without recovery, which is sufficient to satisfy one metric and insufficient to satisfy the political coalition that funds the program.

Position: the model is not failing, but the deployment often is

The defensible reading of the evidence is that Housing First is a high-performing component of a homelessness system, not a complete homelessness system. It reliably stabilizes housing for the population it was built to serve. It produces public expenditure savings that, in well-implemented contexts, exceed its costs within a measurable timeframe. It has demonstrated population-level effects where it has been paired with the housing supply and clinical infrastructure that the underlying logic of the model implicitly requires.

It does not solve addiction. It does not solve serious mental illness. It does not solve the structural causes of inflow into homelessness. No model does, because those are different problems with different policy instruments. Treating Housing First as a comprehensive answer to street homelessness is the analytical mistake that has produced both inflated expectations from its advocates and disappointed conclusions from its critics. The evidence supports its retention as a core operating protocol for chronically homeless individuals and a structural priority for federal and municipal homelessness funding. It does not support the claim that the model, standing alone, will produce the visible street-count reductions that political constituencies demand.

The policy question is therefore not whether Housing First works. The retention record answers that. The policy question is sequencing — how aggressively to expand the affordable housing supply alongside the model, how to fund the clinical infrastructure the model assumes but does not itself provide, and how to measure success against the population-level denominator rather than the individual-level one. Cities that treat Housing First as the whole strategy will continue to defend an indefensible position. Cities that treat it as the operational backbone of a broader supply-and-support architecture will continue to produce the kind of outcome data that made the model controversial in the first place.