Japan’s 2,591 Homeless — And the 100,000 Who Don’t Count

Once a year, before dawn, employees of Japanese local governments walk through urban parks, along riverbanks, and past station buildings, counting bodies. They record people sleeping on cardboard, in tents, under bridges, and in doorways. The 2025 census found 2,591 such people [1] — 2,346 men, 163 women, and 82 of unidentified gender. Osaka led with 763, Tokyo followed with 565, and Kanagawa contributed 366. The total represented an 8.1% decline from 2024, continuing a fall that has reduced visible homelessness in Japan by more than 90% from a 2003 peak of 25,296.

The decline is real. The streets of Tokyo, Osaka, and Yokohama in 2026 contain fewer tents, fewer cardboard encampments, and fewer rough sleepers than at any point in the past quarter-century. ✓ Established Compared to the visible homeless populations of San Francisco, Los Angeles, Paris, or Berlin, Japanese cities present an extraordinary tableau of order and apparent affluence. International commentators routinely cite the 2,591 figure as evidence that Japan has, through some combination of culture, welfare, and policy design, solved a problem that defeats wealthier states.

This appearance is the genuine policy achievement at the centre of every English-language explainer on Japanese homelessness. Visiting Western journalists since the early 2010s have produced a steady stream of reports framing Japan as a low-homelessness success story — variously crediting Confucian family obligation, post-war welfare design, a strong shame culture, or some unspecified combination of the three. The framing reflects a real observation: anyone landing at Haneda and travelling into central Tokyo will see a public order that San Francisco or Seattle long ago lost. ◈ Strong Evidence But every framing assumes that the ratio of visible homelessness to total housing precarity is roughly comparable across countries — and that assumption is precisely what the Japanese case violates.

The trouble starts the moment one asks what was actually measured. ◈ Strong Evidence The Ministry survey records only people staying overnight in spaces it designates as public outdoor — parks (25.5%), roadsides (24.1%), riversides (21.6%), station buildings (5.8%), and miscellaneous outdoor locations [1]. Anyone sleeping under a roof — however precarious, however temporary, however close to destitution — is by definition not homeless. The OECD’s 2024 country note flagged this as a major incomparability with European, North American, and Oceanian definitions [5].

The consequence is statistical: Japan is using a 1990s definition designed for the visible tent populations of Sumida-gawa and Shinjuku — populations that have either died, been moved indoors via welfare, or migrated into one of three other forms of housing precarity that the survey does not see. The country that pioneered global rankings of urban liveability has, in this single domain, opted for a metric that no other rich democracy considers complete. ✓ Established The 2,591 number is real. So is everything it leaves out [2][5].

The net café refugee — ネットカフェ難民 — is a category invented in the mid-2000s to describe a phenomenon Japanese welfare statistics still refuse to name. A reclining-seat booth in a 24-hour manga kissa rents for ¥1,500 to ¥2,500 per night, often with free showers, vending machines, blankets, and a power outlet. For someone working irregular shifts, unable to pass landlord guarantor checks, or fleeing a former address, it is the cheapest viable form of indoor sleeping in central Tokyo [3]. By the legal definition, it is not homelessness. By the lived definition, it is.

The numbers are contested but consistent in direction. The Tokyo Metropolitan Government’s 2018 survey found 15,000 nightly net café users in Tokyo alone, of whom an estimated 4,000 were effectively homeless — meaning they had no other residence [4]. Tokyo Challenge Net, the housing-assistance NGO commissioned by the metropolitan government, has extrapolated national estimates of 100,000 to 300,000 across Japan ⚖ Contested. The high end is contested; the low end is not.

Beyond net cafés sits a second layer of invisibility: the capsule hotel long-stayer. Originally designed for the salaryman who missed the last train, capsule hotels in Shinjuku, Shimbashi, and Umeda increasingly host weekly and monthly residents — people whose entire wardrobe fits in a coin locker and whose monthly cost is two-thirds of a one-room apartment with deposits and guarantor fees. ◈ Strong Evidence A 2024 OECD review noted that no jurisdiction in Japan counts these residents, even though equivalent populations in EU countries are categorised under ETHOS Light typology as ‘people in accommodation for the homeless’ [5].

The third layer is a phenomenon Japanese welfare workers call ‘sofa-surfing within obligation’ — adult children, divorced spouses, and elderly relatives who are formally housed because they live with kin, but whose presence is unwelcome, conditional, or violent. Surveys by Moyai Support Centre, Tokyo’s leading anti-poverty NGO, suggest that this group constitutes the largest single category of women experiencing housing instability [11]. They will never appear in any homeless count, Japanese or otherwise, because the unit of measurement is the dwelling, not the person’s right to remain in it.

A fourth layer sits behind even these — the population of long-stay weekly hotel residents, particularly in Osaka, Kawasaki, and the outskirts of Nagoya, where former mid-tier business hotels have been converted into de facto single-room-occupancy buildings. Rent runs at ¥35,000–¥50,000 per month for a 9 m² room, payable in advance, with neither tenancy rights nor deposit-refund protections. ◈ Strong Evidence Welfare workers describe these residents as a growing category: former salaried workers in their 40s and 50s, often after a divorce, debt settlement, or company collapse, who cannot return to the rental market because they failed standard landlord and guarantor checks. They have a roof and a key. They have no lease, no recourse, and no listing in any homelessness survey.

The cumulative effect is that the 2,591 figure functions as a category, not a measurement. ✓ Established It captures the population most visible to passers-by and most embarrassing to municipal governments — and converts the rest of the housing-precarious population into a statistical absence [2][5]. The disappearance is not denial; it is definition.

The three districts are remnants of Japan’s post-war reconstruction. From the 1950s through the bubble years, they served as labour-market reservoirs: morning hiring halls dispatched men to construction sites, ports, and factories on day contracts, paid in cash at evening. Workers slept in doya — single-room cubicles of three or four tatami mats — and ate at common cafeterias. ✓ Established Kamagasaki at its peak in 1990 housed roughly 30,000 men. Today the population is 19,000 to 25,000 within twenty hectares, predominantly elderly single men [8].

The bubble’s collapse, the Lehman shock, and four decades of construction-sector mechanisation extinguished the day-labour economy. The men who built Japan’s expressways did not retire to provincial bungalows; they aged in the doya. Where the morning hiring hall once moved tens of thousands of workers, the daily ritual is now welfare administration. In San’ya, around 3,800 men live in the 145 remaining doya buildings, with more than 90% on seikatsu hogo [8]. The streets are calm not because the men are housed in the conventional sense, but because the state pays the doya owner directly and the men are obliged not to die in public.

The doya districts function, in policy effect, as the indoor end-state of the 2002 Special Measures Law on Homeless Self-Support. The law was written for an employable population: it funded shelters, vocational training, and job-search assistance. Within five years, around 40 such shelters had opened nationwide. But the population they were built to serve aged out of the labour market before the programs could reabsorb them [5]. The result was that men flowed from the streets into shelters, from shelters into doya, and from doya into welfare — a pipeline that produces declining outdoor counts without addressing structural poverty.

This is what makes Kamagasaki and San’ya statistical exceptions rather than urban renewal stories. ◈ Strong Evidence The men live indoors, but indoors is a 1.6-metre-by-1.8-metre cubicle, often with shared bathrooms, no kitchens, and rules against guests. [8] The buildings are termed ‘simple lodging facilities’ in legal code — neither apartments nor shelters. By European or American standards, these would be categorised as ‘people in non-conventional dwellings’ under ETHOS, and counted as homeless. By Japanese categorisation, they are housed.

Kotobuki in Yokohama and other regional doya-gai follow similar trajectories. The geography reveals the underlying mechanism: Japan has not so much solved homelessness as confined it to administrative zones where it does not register as such. The men are no longer on the streets, which is a humanitarian gain in absolute terms — they sleep on a mattress, with a locking door, in heated rooms. ✓ Established It is also a categorical maneuver. The 2,591 number is what is visible after this maneuver is complete.

To apply for seikatsu hogo in Japan, an eligible person must present themselves at the welfare counter of their local ward office, document their assets and income, and submit to an institution called fuyo shokai — the family support inquiry. Welfare staff contact the applicant’s parents, adult children, and in some cases siblings, asking whether the relatives can support the applicant before public funds are deployed [6]. The inquiry is procedural in theory and devastating in practice. Many applicants withdraw rather than have their estrangement, addiction, debt, or business failure broadcast to relatives.

Lawyers and welfare workers describe a parallel front-line practice called mizugiwa sakusen — literally ‘the water’s-edge tactic’, the policing of the border before applicants reach it. The practice involves discouraging applications at the counter: telling claimants they should look for work first, suggesting they consult relatives, asserting that benefits will be small or denied, demanding documentation not legally required [6]. The Ministry denies the practice is policy. The 22.9% take-up rate is, regardless, structurally indistinguishable from a system designed to deflect eligible applicants.

The cumulative numbers are stark. As of December 2025, about 1.64 million households — roughly 2.9% of all Japanese households — received seikatsu hogo. Of these, more than 50% were headed by elderly recipients and more than 90% were single-person households [6]. The program reaches the institutionally embedded elderly, increasingly through Kamagasaki and San’ya, but disproportionately fails to reach the working-poor, single mothers under 65, and those with intact but estranged family networks. The dependency ratio is high precisely because the entry barrier is high.

The system also encodes an asymmetry foreign residents cannot escape. In July 2014, Japan’s Supreme Court ruled that non-citizens — including special permanent residents born in Japan, taxed for decades, and contributing to the social security system their entire lives — have no statutory right to seikatsu hogo [7]. ✓ Established They may be granted benefits at administrative discretion, but they cannot appeal a denial as a violation of legal entitlement. Welfare for foreigners thus exists in a precarious, gift-of-the-state status that municipal governments have tightened repeatedly since the ruling.

Matthew Penney, writing in the Asia-Pacific Journal, argues that this combination of stigma, family-inquiry, and discretionary gatekeeping is not cultural accident but state policy: a fiscally efficient method of generating low official poverty figures while leaving an undercount of distressed households unreached [11]. The take-up rate is a price signal — a measure of how much friction the state introduces between eligibility and entitlement. ◈ Strong Evidence At 22.9%, the friction is doing exactly what it was designed to do.

Japan’s welfare system was built on the assumption of the male breadwinner household: a married salaryman with a homemaker spouse and dependent children, eligible for pension contributions through formal employment and entitled to spousal benefits if widowed. The system functions reasonably well for this household, which is now a minority of all Japanese households. The fastest-growing categories — single mothers, lifelong single women, divorced women, and elderly widows — sit awkwardly across the model’s seams. ✓ Established Approximately 50% of children in single-mother households in Japan live in poverty [10], the worst single-parent rate in the OECD.

For elderly women the gap is widening. Roughly 25% of elderly Japanese women live below the relative poverty line, against approximately 10% of elderly men. ◈ Strong Evidence For elderly women who never married or who divorced, the figure rises to around 50% [10]. The mechanism is mechanical: the basic Japanese state pension, paid in full only to those with 40 years of contributions, amounts to about ¥777,800 per year — roughly ¥65,000 per month, well below the official poverty line for a single-person urban household. Women’s incomplete contribution records, due to part-time work, child-rearing breaks, and care obligations, deliver smaller annuities than men’s.

Where do these women go? Almost never to the street. Cultural and bureaucratic mechanisms route female poverty into less visible forms: continued cohabitation with adult children, even where the relationship is strained; sub-market rental in aging UR public housing; informal cash work; and, increasingly, late-life net café and capsule hotel use. The same architecture that diverts male precarity into San’ya or Kamagasaki diverts female precarity into co-residence and silent withdrawal from social life. ◈ Strong Evidence The category that emerges in the data is kodokushi — lonely deaths — of which Japan recorded 76,020 in 2024, with 76.4% aged 65 or older and many discovered weeks or months after death [10].

Foreign residents face a separate set of exclusions. Following the 2014 Supreme Court ruling, foreign nationals who lose employment or fall ill must rely on administrative goodwill rather than legal entitlement to receive seikatsu hogo [7]. The 2025 crackdown on foreign residents with unpaid pension and health insurance contributions tightened the discretionary margin further; municipal governments have used the new compliance framework to deny benefits to long-term residents who had received them previously. Working-age foreign residents who lose income, including special permanent residents and visa holders contributing to the system for decades, are pushed disproportionately into the same indoor invisibility — capsule hotels, net cafés, and dependence on co-ethnic networks.

The combined effect is that the official homeless census — almost entirely male, almost entirely elderly, almost entirely Japanese — fails to capture the demographics actually most exposed to housing precarity in 2026. ◈ Strong Evidence The single mother working two part-time jobs, the divorced 68-year-old in a fading UR danchi, the Brazilian-Japanese factory worker whose plant closed in Toyota — none of them appear in the 2,591. The census measures what visible homelessness looked like in 1995 [11][5]. It does not measure what poverty looks like now.

Walk through Ikebukuro East Park, Shinjuku Chuo Park, or Osaka’s Tennoji station forecourt and a curious feature recurs: the public benches are unusable. They slope at five-degree angles that prevent sleeping. They are subdivided by metal armrests into two or three discrete seats too short for an adult body. They are made of metal tubing that is uncomfortably hot in summer and painful in winter. They are, in the technical language of urban design, ‘hostile’ — deliberately engineered to deter behaviour rather than enable use [9].

The practice predates the 2002 Special Measures Law. In the early 1990s, as the bubble economy collapsed and visible homelessness in Tokyo rose, several wards began retrofitting parks, station forecourts, and underpasses with what officials called ‘behaviour-correcting’ design. By the early 2000s, the practice had spread to private developers and JR East stations. ✓ Established The Japan Times documented the practice in detail in 2020, recording on-record statements from the Toshima parks office acknowledging the explicit intent to prevent homeless people from sleeping in the area [9].

The Toshima statement is unusual only in its candor. Most ward governments use euphemism — ‘revitalisation,’ ‘safety improvements,’ ‘alignment with the new urban vision’ — for what is the same intervention. ◈ Strong Evidence The hostile-design literature documents the practice across Shibuya (round metal poles at station entrances), Shinjuku (the Mosaic Road covered benches with central armrests), and Yokohama (sloped concrete platforms in waiting areas) [9]. The 2024 Shinjuku ‘bad benches’ campaign — installed to deter both rough sleepers and late-night street drinkers — confirmed that the policy is still active and expanding.

The effect is multi-layered. Most directly, the redesign makes the public spaces useless to people experiencing homelessness — they cannot rest, they cannot sleep, they cannot pause. The visible city becomes uninhabitable to the bodies that would otherwise inhabit it. Secondarily, the redesign makes the public spaces less usable by everyone else — elderly people who cannot stand for long, disabled people who require flat resting surfaces, parents with sleeping infants. The hostile-design strategy is collective punishment in the service of statistical management. ◈ Strong Evidence It works: people who would once have slept in the park now sleep in a net café, do not appear in the homeless count, and do appear, eventually, in the kodokushi statistics.

The architectural strategy connects directly to the statistical strategy. ✓ Established When a park bench cannot be slept on, the person who would have slept on it is rerouted to a manga kissa or a covered station passage. When the manga kissa is also surveyed, the route shifts again — to a 24-hour McDonald’s, to a karaoke booth, to a family member’s spare floor space. Each redirection moves the body across an administrative boundary that determines whether it is counted as homeless. The Japan model is an integrated system: the bench, the welfare counter, the doya, the capsule hotel, the fuyo shokai inquiry, all working in concert to convert visible homelessness into invisible housing precarity.

This is what makes the question of ‘how Japan solved homelessness’ a category error. Japan did not solve homelessness. Japan rendered it invisible through a coordinated combination of physical design, statistical definition, and bureaucratic friction. ◈ Strong Evidence The outcome — a city with no tent encampments — is real and, in the narrowest sense, humane. The disappearance is real. So is the question of where the people went [2][5][11].

The United States produces the highest absolute homeless count among rich democracies: 771,480 people in the January 2024 point-in-time enumeration, a record since data collection began in 2007 and an 18% jump from 2023 [12]. The U.S. methodology counts sheltered and unsheltered populations in a single overnight enumeration coordinated by HUD-funded Continuums of Care. The shelter system itself is substantial — roughly 60% of the count is sheltered — and the unsheltered population is highly visible in cities like San Francisco, Los Angeles, Portland, and Phoenix. The U.S. number is also widely understood to undercount the actual flow population over a year, which is approximately 4–5 times larger.

Germany’s 2024 estimate by the Federal Working Group for Homeless Assistance (BAG-W) puts the homeless population at 1,029,000, including approximately 440,000 Ukrainian refugees and asylum-seekers housed in temporary accommodation. ◈ Strong Evidence The figure represents a 10.9% rise from 2023 and a 70% rise since 2022, reflecting both refugee inflows and a deepening domestic housing crisis [13]. Germany counts anyone without secure independent housing, including those in shelters and refugee facilities — a far broader definition than the U.S. or Japan. Excluding refugees, the underlying domestic homeless population is approximately 600,000.

The United Kingdom operates a hybrid system. The official rough sleeping snapshot — taken on a single night in October or November — found 4,667 people in autumn 2024 and 4,793 in autumn 2025, the highest figure since the count began and 171% above the 2010 baseline [14]. But the UK additionally categorises around 325,000 households as statutorily homeless under the 1996 Housing Act, owing them rehousing duty. The dual count produces a high visible-precarity number and a far higher institutional-precarity number — closer to the German methodology than the U.S. or Japanese one.

Finland sits at the opposite end of the methodological spectrum. The country adopted a national Housing First policy in 2008, providing immediate independent tenancies to people experiencing homelessness without preconditions of sobriety, employment, or treatment compliance. Long-term homelessness fell 68% between 2008 and 2022, and the 2024 total population experiencing homelessness was approximately 3,806 — about 0.06% of the country’s 5.5 million population [15]. ✓ Established Finland is the only OECD country to have approached statistical homelessness elimination, and it did so via legal entitlement to housing, not via redefinition of who counts.

The four comparator models — Japan’s exclusion-by-definition, the U.S.’s sheltered-plus-unsheltered, Germany’s broad-precarity, and Finland’s legal-entitlement — produce dramatically different numbers because they measure dramatically different things. ✓ Established The temptation to compare them is irresistible and largely useless. What can be compared is policy framework. ◈ Strong Evidence On that dimension, Japan looks less like a high-performing outlier and more like a country with a narrow camera angle.

Three propositions emerge from the evidence. First, Japan’s official homeless count of 2,591 is internally accurate but internationally misleading. ✓ Established It measures outdoor rough sleeping under a 1990s definition, and on those terms the decline is real. The country has fewer tents under bridges than at any time in the past quarter-century, and the streets of central Tokyo and Osaka are, by any measure, safer and more orderly than those of comparable Western capitals [1].

Second, the gap between the official count and the actual precarious population is large, structural, and partly intentional. ◈ Strong Evidence Net café refugees number in the tens of thousands in Tokyo alone, with national estimates ranging from 100,000 to 300,000. The doya districts absorb another 20,000–40,000 welfare-dependent former workers who would be counted as homeless under any G7 definition except Japan’s [4][8]. Welfare reaches 22.9% of eligible households — the lowest take-up rate in the developed world — and 50% of elderly women who never married or who divorced are in poverty. The hidden population is at least an order of magnitude larger than the visible one.

Third, the visible-versus-hidden split is not accident but design. ◈ Strong Evidence The Japanese model integrates physical hostile design (sloped benches, divided seats), statistical definition (outdoor-only counts), bureaucratic friction (fuyo shokai, mizugiwa sakusen), and informal absorption (doya, family co-residence) into a coherent system whose output is a low official poverty number and a high informal poverty reality. The system is fiscally efficient — Japan’s welfare spending as a share of GDP is below the OECD average — and politically resilient, because the precarious population it produces is dispersed, indoor, and unseen [11].

The implications are not abstract. As Japan’s population ages and the doya cohort dies off, the absorptive capacity that defined the post-1990s containment model is contracting. The aging-out of San’ya and Kamagasaki removes a structural buffer; the rise in female and foreign precarity introduces populations the existing system was not built to handle. ◈ Strong Evidence The 2024 kodokushi figure of 76,020 lonely deaths — 76.4% over age 65, many discovered weeks or months after death — is the leading indicator of where the precarity has gone. Lonely death in a one-room apartment is invisible to a homelessness survey. It is not invisible to a coroner.

The policy question is whether Japan continues to manage precarity through definition and friction, or moves toward an entitlement-based model on the Finnish pattern. The two paths produce different numbers but more importantly different lives. ✓ Established Finland’s Housing First framework, having reduced long-term homelessness by 68% across 14 years with cost savings of approximately €21,000 per rehoused person, demonstrates that legal entitlement to housing is operationally feasible at OECD-country resource levels [15]. The Japanese alternative — discretionary, stigmatised, low-take-up — is also operationally feasible, but produces the kodokushi end-state in slow motion.

The 2,591 number is a story Japan tells about itself, and the world has largely chosen to believe it. ◈ Strong Evidence The story is not false. It is selective. The cost of the selection is borne by the people the count does not see: the women in the manga kissa booth, the foreign worker turned away at the welfare counter, the elderly widow in the UR danchi who has not spoken to anyone for a week. The visible city is calm. The invisible one is where the policy decisions are settled — and where, increasingly, they are paid for [2][11][5].