homelessness in a scandinavian welfare state: the risk of ... · the danish national centre for...

Article

Urban Studies2016, Vol. 53(10) 2041–2063� Urban Studies Journal Limited 2015Reprints and permissions:sagepub.co.uk/journalsPermissions.navDOI: 10.1177/0042098015587818usj.sagepub.com

Homelessness in a Scandinavianwelfare state: The risk of shelter usein the Danish adult population

Lars BenjaminsenThe Danish National Centre for Social Research, Denmark

AbstractThis article analyses the risk of homelessness in the Danish adult population. The study is basedon individual, administrative micro-data for about 4.15 million Danes who were 18 years or olderon 1 January 2002. Homelessness is measured by shelter use from 2002 to 2011. Data also covercivil status, immigration background, education, employment, income, mental illness, drug andalcohol abuse, and previous imprisonment over five years prior to the measurement period.Prevalence of shelter use shows a considerable risk of homelessness amongst individuals experi-encing multidimensional social exclusion. Nonetheless, even in high-risk groups such as drug abu-sers and people with a dual diagnosis, the majority have not used shelters. A multivariate analysisshows significantly higher use of homeless shelters amongst immigrants and individuals with lowincome, unemployment, low education, mental illness, drug or alcohol abuse, or a previous impri-sonment. The highest risk of shelter use is associated with drug abuse, alcohol abuse, mental ill-ness and previous imprisonment, whereas the risk associated with low income is smaller than forthe psychosocial vulnerabilities. The results show that homelessness in Denmark is widely con-centrated amongst individuals with complex support needs, rather than associated with widerpoverty problems.

Keywordshomelessness, mental illness, poverty, risk factors, shelter use, substance abuse

Received February 2015; accepted April 2015

Introduction

Homelessness is one of the most extremeforms of social marginalisation in contempo-rary society. Even in the Scandinavian coun-tries, representing some of the world’s mostadvanced welfare systems, homelessness hasbeen shown to be a persistent social phenom-enon (Benjaminsen and Dyb, 2008;Benjaminsen and Lauritzen, 2013; Dyb andJohannessen, 2013; Socialstyrelsen, 2012). A

widespread consensus in the research litera-ture is that homelessness arises from socialmechanisms that operate on structural, sys-temic, interpersonal and individual levels,often in complex interplay (Fitzpatrick,

Corresponding author:

Lars Benjaminsen, The Danish National Centre for Social

Research, Herluf Trolles Gade 11, Copenhagen, 1052 K

Denmark.

Email: [email protected]

http://crossmark.crossref.org/dialog/?doi=10.1177%2F0042098015587818&domain=pdf&date_stamp=2015-05-22

2005; Fitzpatrick et al., 2013). Individualpsychosocial vulnerabilities interact withstructural factors such as poverty, unem-ployment and the lack of affordable housing,and systemic factors such as the lack of ade-quate social support systems. These socialmechanisms are generally embedded inbroader characteristics of welfare systems.Thus the risk factors of homelessness identi-fied in statistical models are generated bymore complex underlying social mechan-isms. Yet risk analysis is very useful in identi-fying patterns that appear on the individuallevel. Not only can risk analysis identify therelative importance of various factors suchas poverty and psychosocial vulnerabilities,but it can also identify how large a part ofrisk groups is actually affected by homeless-ness, thereby contributing to a better under-standing of the possible shortcomings inwelfare and support systems.

The risk factors of homelessness and therisk profiles of homeless people are welldescribed in the research literature. Studiesgenerally show that a broad range of riskfactors such as mental illness, substanceabuse, incarceration, institutional care inchildhood, relationship breakdown, weaksocial ties, poverty and unemployment areoverrepresented amongst homeless people(Allgood and Warren, 2003; Bearsley-Smithet al., 2008; Caton et al., 1994, 1995, 2005;Culhane and Metraux, 1999; Folsom et al.,2005; Herman et al., 1997; Kemp et al.,2006; Koegel et al., 1995; van Laere et al.,2009; Piliavin et al., 1989; Sosin and Bruni,1997; Sullivan et al., 2000; Susser et al.,1991, 1993). Quantitative studies of risk fac-tors of homelessness have generally beenbased on surveys comparing homeless peo-ple with the general population or to at-riskpopulations (such as people in poverty orindividuals with mental ill health), or com-paring subgroups of homeless people, e.g.according to the duration of their homeless-ness situation.

However, no study has used individual,administrative micro-data to analyse the riskof homelessness for the population of anentire country. This article analyses the riskof homelessness measured by the use ofhomeless shelters over a ten-year period,using individual, administrative data for theentire Danish adult population and withindividual data on key risk factors, mea-sured prior to the measurement period forshelter use.

The analysis is based on individualadministrative data from various datasources for 4.15 million people who were 18years or older on 1 January 2002. The dataare combined on individual level throughunique identifiers. Homelessness is measuredwith data from a national client registrationsystem in homeless shelters from 2002 to2011. In addition to demographic factors,socioeconomic characteristics includeincome, employment and education. Theanalysis also includes individual vulnerabil-ity factors: mental illness, drug and alcoholabuse problems, and imprisonment. Becausemicro-data are available on homelessnessand other dimensions of social exclusion forthe entire adult population, the analysisoffers unique insights into the actual occur-rence of homelessness in risk groups such asthe mentally ill, substance abusers, the poorand those excluded from the labour market.

Moreover, the analysis shows whetherhomelessness is a rare or frequent occurrencein these groups and to what extent homeless-ness is embedded in a pattern of multiplesocial exclusion. The analysis also investi-gates the relative importance of various riskfactors, including the relative importance ofpsychosocial vulnerabilities and socioeco-nomic characteristics. In the regressionmodel the large data set enables interactioneffects between key explanatory variables.This analysis provides important knowledgeon how the interplay between the main riskfactors affects the risk of homelessness.

2042 Urban Studies 53(10)

Theoretical understanding

A scholarly understanding of the relationbetween homelessness and broader forms ofhousing exclusion has been growing, with ageneral consensus in the research literaturethat not only rough sleepers (individualssleeping outdoors) and people in homelessshelters but also a broader group of peoplein various forms of short-term accommoda-tion and transitional housing are in a home-less situation, as their housing situation hasno permanency. In Europe the housing-based ETHOS (European Typology ofHomelessness and Housing Exclusion) defi-nition has been gaining widespread support(Busch-Geertsema et al., 2010). Based ontheoretical assumptions on the nature ofhousing exclusion, this definition distin-guishes amongst four main categories ofhomelessness and housing exclusion; roof-lessness, houselessness, insecure housing andinadequate housing (Edgar and Meert 2005).A revised definition (‘ETHOS light’) includescategories only for homelessness, differen-tiating amongst rough sleepers, emergencynight shelter users, shelter users, people intransitional accommodation, people await-ing institutional release without a housingsolution in place, and individuals stayingtemporarily with family or friends (Edgaret al., 2007).

Theoretical understandings of the causesand dynamics underlying the phenomenonof homelessness have been influenced by thegeneral synthesising trend in social theory(Avramov, 1999; Blid et al., 2008;Fitzpatrick, 2005, 2012; Pleace, 2000). Ashift towards a more dynamic understandingof homelessness has led to a break with pre-dominantly individualist or structuralistapproaches (Anderson & Christian, 2003;Clapham, 2003; Fitzpatrick, 2012;Fitzpatrick et al., 2000).

From a critical realist understanding of alayered reality, Fitzpatrick et al. (2013)

argues that the social mechanisms generatinghomelessness must be understood as an openand contingent interplay between structural,systemic, interpersonal and individual fac-tors, with none of these levels a priori morefundamental than others. In some caseshomelessness may be a consequence of struc-tural factors such as a lack of affordablehousing, whilst in other cases interpersonalor individual factors may be more signifi-cant. Therefore, no single ‘trigger’ of home-lessness or any one necessary or sufficientcause of homelessness can be identified(Fitzpatrick et al., 2013; Neale, 1997). Thisapproach also applies to cross-country com-parisons explaining why not only the pat-terns of homelessness but also the generativemechanisms underlying these patterns mayvary across different societies and welfaresystems (Fitzpatrick, 2012; Fitzpatrick andStephens, 2007; Fitzpatrick et al., 2013;Stephens and Fitzpatrick, 2007).

A more dynamic understanding of home-lessness has also been conceptualised withinthe ‘pathways theory’, informed by empiri-cal research showing that socially vulnerableindividuals often enter and exit homelessnessseveral times over a life course and that dif-ferent pathways into and out of homeless-ness can be identified (Anderson andTulloch, 2000; Clapham 2003; Culhaneet al., 1994, 2007; Kuhn and Culhane, 1998;MacKenzie and Chamberlain, 2003;Metraux and Culhane 1999; Shinn et al.,1998). Furthermore, homelessness has beenanalysed within the conceptualisation ofsocial exclusion, with homelessness oftenshown to be part of a pattern of multiplesocial exclusion (‘deep social exclusion’), aspeople exposed to homelessness often alsoexperience exclusion in many other lifedomains (Cornes et al., 2011; Fitzpatricket al., 2011).

However, the pathways approach has alsocontributed the understanding that not allhomeless people can be characterised by

Benjaminsen 2043

deep social exclusion. Research from theUSA has shown a considerable heterogeneityamongst homeless people, and groups ofchronically, episodically and transitionallyhomeless people have been identified (Kuhnand Culhane, 1998). The chronic shelter usershave relatively few but very long shelter stays,the episodic shelter users have frequent butoften short shelter stays, and the transitionalshelter users are those with only few andshort shelter stays. This US research hasshown that whilst the chronically and episo-dically homeless usually have complex sup-port needs, and thus can be characterised asbeing in deep exclusion, fewer amongst thetransitionally homeless have complex supportneeds – and this group is more often homelessbecause of poverty and housing affordabilityproblems (Kuhn and Culhane, 1998).

The welfare system plays an importantrole in mediating the risk of homelessness,both through general social protection mea-sures, poverty reducing policies and housingpolicies, and through more specific policiessuch as the provision of specialised housingand support for groups with complex sup-port needs. According to a general hypoth-esis in homelessness research, homelessnessin countries with relatively extensive welfaresystems and lower levels of poverty tend tobe concentrated amongst individuals withcomplex support needs, whereas homeless-ness in countries with less extensive welfaresystems and a higher level of poverty tendsto affect a broader segment of poor andunemployed households and individuals(Shinn, 2007; Stephens and Fitzpatrick,2007; Toro, 2007). In the social-democraticwelfare states in Scandinavia, with their rela-tively low levels of poverty and extensivewelfare systems, we can expect homelessnessto be widely concentrated amongst peoplewith mental illness and substance abuseproblems. We can also expect these factorsto be stronger predictors of homelessnessthan poverty or unemployment.

Homelessness in Denmark

Denmark can be characterised as a typicalsocial-democratic welfare state with a rela-tively high level of income redistribution, lowlevel of income poverty and a relatively lowlevel of unemployment (Organisation forEconomic Cooperation and Development(OECD), 2014). The public housing sectorcomprises about 20% of the total housingstock, with allocation mechanisms targetedat individuals who have special support needsand are in acute need of housing (SkifterAndersen, 2010; Skifter Andersen et al.,2012).

Denmark has two main sources of home-lessness data. The first source is the nation-wide client registration system onhomelessness shelters operated under section110 of the Social Assistance Act, a data sys-tem established in 1999. Annual statisticshave shown the total number of shelter usersto be quite stable over time, at about 6000individual users each year (Ankestyrelsen,2014). The measurement of homelessness inthis article is based on data from theseshelters.

Previous research on Danish shelter usershas shown a high prevalence of mental ill-ness and addiction problems and a highmortality amongst shelter users (Nielsen etal., 2011). Moreover, previous research hasshown that the groups of chronic, episodicand transitional shelter users, identified inthe USA, can also be found in Denmark(Benjaminsen and Andrade, 2015).However, the profile of the transitional shel-ter users (i.e. with few and short shelterstays) has been shown to be different inDenmark, as even this group (along with thechronic and episodic shelter users) is widelyconcentrated amongst people with complexneeds (Benjaminsen and Andrade, 2015). Incontrast, as previously mentioned, the tran-sitional shelter users in the USA include alarger group of people who are homeless


mainly because of poverty and housingaffordability problems (Kuhn and Culhane,1998). However, these earlier Danish studiesneither include data for non-shelter usersnor estimate the risk of shelter use for thegeneral population.

The second source of data on homeless-ness in Denmark is the national point-in-time (one-week) counts that have been car-ried out biannually since 2007 (Benjaminsen,2009; Benjaminsen and Christensen, 2007;Benjaminsen and Lauritzen, 2013; Lauritzenet al., 2011). These counts include not onlyshelter users but also rough sleepers, peoplein short-term transitional housing, and peoplestaying temporarily with family or friends (tothe extent that these individuals are knownby the social services). In the last count inweek 6, 2013, 5820 individuals were found tobe homeless, with age information for 5624of them; 5480 homeless people were 18 yearsor older. This figure was equivalent to 0.12%of the adult Danish population of 4.4 millionpeople (Benjaminsen and Lauritzen, 2013:31). According to the national homelessnesscount, individuals who sleep in homeless shel-ters are the largest category amongst thehomeless. Other significant groups wererough sleepers and people staying temporarilywith family and friends.

However, research – mainly from theUSA – has shown that point-in-time countsof homelessness tend to underestimate thescale of homelessness over a longer period(Culhane et al., 1994). The pathwaysapproach suggests that repeated spells ofhomelessness during a life course is a com-mon pattern for socially vulnerable individu-als. Moreover, point-in-time countsoversample individuals with longer spells ofhomelessness and underestimate the numberof individuals experiencing homelessness ofa relatively shorter duration. Such oversam-pling may cause an over-representation ofindividuals with complex support needs, aspeople with less complex support needs are

less likely to stay in the shelter system for along time. Therefore, the analysis in thisarticle gives a new perspective on homeless-ness in Denmark, because it is based on indi-vidually linked data for the entire Danishadult population, measuring shelter use overa long period and including a wide range ofindividual risk factors measured for theentire population through administrativeregisters.

Data and measurement

The analytical understanding that structural,systemic, interpersonal and individual fac-tors interact in shaping the risk of homeless-ness for the individual also hasmethodological implications. Variable-centred risk analysis has generally been criti-cised for individualising the reasons forsocial marginalisation (France, 2008; Kelly,2001). In contrast, the critical realistapproach (Sayer, 1992, 2000) offers a morenuanced approach to the way in whichquantitative analysis can enrich our under-standing of homelessness. As previouslymentioned, according to critical realism theobserved empirical patterns are generated byunderlying social mechanisms and interact-ing factors that may operate on different lev-els, including structural and systemic factorsnot measured in the analysis of individualdata. Thus the statistical analysis identifiesthose individuals most likely to be affectedby such adverse social and economic trendsand systemic deficiencies (Fitzpatrick, 2005).Whilst the discussion section will providepossible explanations of the patterns found,a deeper understanding of these processeswould need further evidence from both qua-litative and mixed-methods studies.

Data

Denmark has extensive central databases ofhigh-quality micro-data, which can be

Benjaminsen 2045

utilised for research purposes. The empiricalanalysis is based on a combination ofadministrative data from various datasources. The main database, provided byStatistics Denmark, contains individual datafor the entire Danish population of adultswho were 18 years or older on 1 January2002. Data on homelessness have beenobtained from client registration systems onhomeless shelters. Section 110 in the SocialAssistance Act mandates that municipalitiesmust provide temporary accommodation forhomeless people. A total of 73 homelessshelters are operated under it with a total of2180 beds (Ankestyrelsen, 2014). Whenenrolling in a shelter, individuals must givetheir CPR (Central Personal Register) num-ber for registration, with the reporting ofCPR numbers to a central database beingmandatory for shelters. The CPR numberenables the linking of data on shelter use tothe main database. Almost all the Danishshelters included in the data provide emer-gency shelter. At the same time many shelterusers have longer stays, and the sheltersoften have individual rooms. Thus com-pared with similar functions in other coun-tries, the shelters widely also fulfil thefunction of providing short-term temporaryaccommodation.

The use of data on shelter use for estimat-ing homelessness means that only those indi-viduals who use shelters are categorised ashomeless, whereas individuals who sleeprough and never use a shelter or those whostay temporarily with family and friends arenot. However, data from the national countshow that even during the short span of thecount week half of all rough sleepers alsouse homeless shelters (Benjaminsen andLauritzen, 2013). Thus over a 10-year perioda considerable proportion of rough sleepersare likely to be recorded as using sheltersone or more times. However, the risk of‘false negatives’ exists, that is individuals

who did not use homeless shelters but whoindeed experienced other forms of homeless-ness situations.

Explanatory variables

Individual administrative data on demo-graphic variables, socioeconomic variablesand data from the general hospital systemand the criminal justice system have beenobtained from Statistics Denmark. Datahave also been obtained from the CentralPsychiatric Register (Mors et al., 2011) ondiagnoses of mental illness and from theRegister of Treatment for SubstanceAddiction on substance abuse. These datahave been provided by ‘Statens SerumInstitut’ (SSI), an agency under the auspicesof the Ministry of Health. All data can beindividually linked through CPR numbers.CPR numbers for all data sources includedin the study have been converted byStatistics Denmark into anonymised uniquenumbers. The anonymised data are accessedthrough the Statistics Denmark registerresearch system. Permission for the studywas granted from the Danish DataProtection Agency.

The explanatory variables are gender,age, civil status, ethnic background, residentin Copenhagen, income, employment, edu-cation, mental illness, drug addiction, alco-hol addiction and previous imprisonment.The demographic variables are measured on1 January 2002. Civil status is measured bywhether the individual is single or not. Allindividuals who are not married or cohabi-tating, and those divorced or widowed, arecategorised as ‘single’. Individuals who aremarried or cohabitating are registered as‘non-single’. This category also includesyoung adults up to 25 years still living withtheir parents. Immigrant background is mea-sured by three categories – non-immigrants,immigrants and children of immigrants.‘Resident in Copenhagen’ measures whether


the individual is recorded in the populationregister in the city of Copenhagen at theonset of the period.

Income is measured by net disposableincome after tax and interest payments, mea-sured in two categories – below and above100,000 DKK/year (approximately 13,000Euros). For a single person this income levelapproximates the OECD poverty line forDenmark. Employment is measured bybeing employed or not. Although the term‘unemployment’ usually refers to peoplewho have either lost their jobs or are seekingwork, the category ‘unemployed’ alsoincludes those who are without work butwho do not fall into the usual unemployedcategory, e.g. people on early retirement.Education is measured in two categories –compulsory level or less (9th or 10th gradeor less) and beyond compulsory level. Whilstincome and employment are measured dur-ing the 2001 calendar year, education is mea-sured on 1 January 2002.

Mental illness and substance abuse aremeasured through diagnosis registers fromthe general hospital system, the mentalhealth system and the addiction treatmentsystem. Having a mental illness, sufferingfrom drug or alcohol abuse, and havingbeen imprisoned are measured over timefrom 1997 to 2001, i.e. prior to the period inwhich homelessness is measured. A diagno-sis of mental illness covers both severe men-tal illness such as schizophrenia and bipolarcondition and also includes less severe illnesssuch as moderate depression and anxietydisorders. A diagnosis of drug abuse prob-lems covers the abuse of both hard drugs(e.g. heroin and cocaine) and cannabis. Thediagnoses of mental illness and drug or alco-hol problems are those given by medicalprofessionals in the public health systemaccording to the ICD-10 (InternationalClassification of Diseases).

The time sequence between a diagnosisfor mental illness or substance abuse, and a

recording of homelessness can in principlebe established from the data registers.However, the time of diagnosis does notyield adequate information on when an indi-vidual started to suffer from symptoms, andthe actual time order between psychosocialvulnerabilities and homelessness thereforecannot be measured in a valid way from theregisters. For example, a mental illness withescalating substance abuse may first lead tohomelessness. Then a shelter stay may be anentry point for further access to the mentalhealth or addiction treatment systems, and adiagnosis may eventually be given after psy-chiatric assessment. In this example, whilst ashelter stay is recorded in the shelter dataregisters before a diagnosis for mental illnessis recorded in the psychiatric data registers,the actual chronological order between men-tal illness and shelter use is the opposite.

To approach this issue of complex associa-tions between psychosocial vulnerabilities andhomelessness, given the data available, Iinclude two measurement periods of the psy-chosocial vulnerabilities. In addition to a mea-surement period from 1997 to 2001, prior tothe measurement period for shelter use, ameasurement of mental illness and substanceabuse over an extended time from 1997 to2011 is also included in the descriptive statis-tics. The different measurement periods affectthe share of shelter users identified as havinga mental illness or a substance abuse problem,as over a longer time span more people willbe recorded in the public health system with adiagnosis. Therefore, the longer time spanmore adequately represents the extent of psy-chosocial vulnerabilities amongst peopleaffected by homelessness.

Omitted cases

Whilst individuals who died or emigratedduring the period are included in the analy-sis, individuals who immigrate during theperiod are not. Individuals who died during

Benjaminsen 2047

the period have fewer years to be exposed tothe risk of homelessness. However, whenone analyses full population data over along period, eliminating all individuals whodie during that period would disproportio-nately diminish the study population inolder cohorts. More specifically, given ahigher mortality amongst individuals whohave been homeless (Nielsen et al., 2011),eliminating individuals who die during theperiod may incorrectly lower the risk ofhomelessness in older cohorts.

In total 4,186,953 individuals were 18years or older on 1 January 2002. However,35,672 individuals are omitted from theanalysis because of missing values on someof the variables. The final data set comprises4,151,281 individuals. The only variable forwhich individuals are not omitted from theanalysis, despite missing information, is foreducation. The reason is that about 7% ofall individuals are in the category ‘unspeci-fied education’, mainly in older cohorts, astheir education was not known when theregister data system began in 1980. I includethis category as a separate control dummyin the multivariate analysis, along with theother educational categories, as the loss ofdata resulting from omitting these individu-als would not only be too large but also notoffer any advantage to the analysis.

Statistical methods

The analysis calculates the observed bivari-ate prevalence of shelter use for each riskfactor, and the frequencies of risk character-istics for shelter users and non-shelter users.As the data set comprises the entire adultpopulation, the observed prevalence of shel-ter use provides a unique insight into howlarge a part of different risk groups actuallyexperience shelter use over a longer period.To analyse how homelessness is part of apattern of multidimensional social exclusion,I also calculate the prevalence of shelter use,depending on the cumulative number of riskcharacteristics. Then I estimate the risk ofshelter use and the relative importance ofvarious risk factors through multivariatelogistic regression models. The first modelincludes only the main effects of the expla-natory variables; the second model, the two-way interaction effects between specificvariables.

Results

Prevalence and risk profile

Table 1 shows the share of the adult popula-tion on 1 January 2002, by gender and age,who enrol in a homeless shelter at least oncefrom 2002 to 2011. Of the 4,151,281

Table 1. Shelter users (2002–2011) by age (2002) and gender.

Age by gender Shelter users (%) Shelter users (n) Total N

Men18–29 1.02 4053 396,11430–49 1.38 10,864 787,14650+ 0.37 3132 847,292Total 0.89 18,049 2,030,552Women18–29 0.27 1046 385,83030–49 0.40 3042 761,95250+ 0.09 905 972,947Total 0.24 4993 2,120,729


individuals in the analysis data who were 18years and above in 2002, 23,042 enrolled ina homeless shelter during this period – equalto 0.56%.

Men represented 78.3% of shelter users(18,049 individuals), equal to 0.89% of themale adult population in 2002. Amongstwomen, 0.23% (4993) used shelters. Thehighest use of shelters is amongst youngerand middle-aged men. Amongst men aged30–49, 1.38% used shelters over the period,and amongst women in the same age groupthe highest prevalence was 0.40%. The shareof shelter users amongst people who were18–29 years old in 2002 does not reflectyouth homelessness as such, as some of thosewho used shelters in the later part of theperiod may have been in their 30s at the timeof enrolment. For both men and women thelowest prevalence of shelter use is the agegroup of 50 years and older, at 0.37% and0.09% amongst men and women, respec-tively. This relatively lower prevalence in theolder age groups may be explained partly byhigh mortality amongst the homeless, espe-cially amongst drug abusers. However, itmay also indicate that, in Denmark, manyelderly socially vulnerable individuals areunder care in the mainstream care system.

Table 2 shows the prevalence of shelteruse for the categories of the explanatoryvariables, still by gender and age groups.Table 3 shows the share with specific charac-teristics amongst shelter users comparedwith non-shelter users within each age groupand for males and females, respectively.

For men, a higher prevalence of shelteruse appears for immigrants than for non-immigrants. Amongst immigrant males aged18–29, 2.24% used shelters over the period,compared with 0.93% of non-immigrants. Inthe same age group the prevalence of shelteruse amongst male children of immigrants, at1.28%, is closer to the prevalence for non-immigrants. Most children of immigrantswere still young at the turn of the 21st

century, as the first wave of labour immigra-tion in Denmark started in the late 1960s.Thus the age groups from 30 years andabove contain very few people who werechildren of immigrants.

For immigrant women the pattern is some-what different. Amongst women aged 30–49,the rate of shelter use for immigrants is simi-lar to that of non-immigrant women, whilstamongst younger women the share of shelterusers is higher amongst immigrants. This pat-tern indicates that younger immigrants (bothgenders) and male immigrants, regardless ofage, are more at risk of homelessness thantheir ethnic Danish counterparts, whilst thedifference levels out for middle-aged women.Nonetheless, the large majority amongst theshelter users are non-immigrants.

The risk of shelter use was substantiallyhigher for people with low income. Formales aged 30–49, 4.70% in the low-incomegroup enrolled in a homeless shelter from2002 to 2011, compared with ‘only’ 0.83%of males not in the low-income group. Forfemales aged 30–49 the corresponding fig-ures are 1.09% and 0.28%. However,amongst those aged 18–24 (both men andwomen), the difference in shelter use accord-ing to income is not as strong, as manyyoung people generally have relatively lowincomes whilst still in education. Moreover,the oldest age group shows very little differ-ence in the prevalence of shelter use accord-ing to income group, likely reflecting lowerincome differentials amongst pensioners.

The unemployed also have a higher riskof shelter use. For men aged 30–49 in 2002and with no employment in 2001, 6.56%used a homeless shelter over the 10-yearperiod, compared with only 0.61% ofemployed men in the same age group. Forwomen in the same age group 1.57% of theunemployed used shelters, compared withonly 0.15% of the employed.

A higher risk of shelter use also existsamongst individuals with low educational

Benjaminsen 2049

Tab

le2.

Pre

vale

nce

ofsh

elte

ruse

by

age

and

gender

inri

skgr

oups

(%).

Men

Wom

en

Age

1Ja

nuar

y2002

18–29

30–49

50+

18–29

30–49

50+

Tota

lN

inag

egr

oup

396,1

14

787,1

46

847,2

92

385,8

30

761,9

52

972,9

47

Vari

ab

leC

ate

go

ryPe

rcen

tsh

elte

ruse

rs

Perc

ent

shel

ter

use

rs

Perc

ent

shel

ter

use

rs

Perc

ent

shel

ter

use

rs

Perc

ent

shel

ter

use

rs

Perc

ent

shel

ter

use

rs

Imm

igra

nt

stat

us

Non-im

mig

rants

0.9

31.3

00.3

60.2

60.4

00.0

9Im

mig

rants

2.2

42.2

00.6

70.4

30.4

20.1

8C

hild

ren

ofim

mig

rants

1.2

81.7

20.9

80.3

41.0

10.1

1U

rban

ity

Not

livin

gin

Cph

1.0

61.2

90.3

50.2

80.3

80.0

9Li

ving

inC

ph

0.8

12.1

40.6

90.2

50.5

70.1

1C

ivil

stat

us

Hav

ing

apar

tner

0.5

80.6

30.1

40.1

90.2

10.0

7N

opar

tner

1.6

53.2

00.9

80.4

30.9

70.1

3In

com

eH

igh/m

iddle

inco

me

0.5

00.8

30.3

30.1

80.2

80.1

0Lo

win

com

e1.5

54.7

00.4

50.3

31.0

90.0

8Em

plo

ymen

tEm

plo

yed

0.4

70.6

10.2

40.0

70.1

50.0

8U

nem

plo

yed

2.6

36.5

60.4

90.6

31.5

70.1

0Educa

tion

Hig

hle

vel

0.0

80.3

90.2

90.0

40.1

30.1

1M

ediu

mle

vel

0.3

00.9

00.3

70.0

70.2

40.1

1C

om

puls

ory

leve

l1.8

72.9

00.4

50.5

70.8

70.1

0M

enta

lill

nes

s(9

7–01)

No

men

talill

nes

s0.8

61.1

40.3

00.1

90.2

90.0

7M

enta

lill

nes

s6.7

68.3

92.3

61.8

52.9

70.6

1M

enta

lill

nes

s(9

7–11)

No

men

talill

nes

s0.5

00.7

80.2

30.1

00.1

60.0

4M

enta

lill

nes

s6.2

07.4

51.4

61.3

42.2

60.3

7D

rug

abuse

(97–01)

No

dru

gab

use

0.8

71.2

10.3

60.2

20.3

40.0

9D

rug

abuse

16.7

523.2

811.1

713.5

918.1

64.3

1D

rug

abuse

(97–11)

No

dru

gab

use

0.4

40.9

50.3

40.1

30.2

70.0

8D

rug

abuse

17.1

324.5

113.0

113.0

017.9

04.0

0A

lcoholab

use

(97–01)

No

alco

holab

use

0.8

91.0

10.2

30.2

40.2

90.0

6A

lcoholab

use

7.0

716.8

66.4

63.0

011.7

33.9

9A

lcoholab

use

(97–11)

No

alco

holab

use

0.6

30.5

70.1

10.1

70.1

50.0

3A

lcoholab

use

9.0

615.0

74.8

24.4

110.2

92.8

5M

enta

llyill

dru

gab

use

rs(9

7–01)

Not

men

tally

illD

A0.9

61.3

10.3

60.2

40.3

70.0

9M

enta

llyill

DA

19.0

127.3

213.9

117.4

721.1

85.1

3

(con

tinue

d)


attainment. For men aged 30–49 with a max-imum of compulsory education, 2.90% usedshelters, compared with only 0.39% for menwith a high education level. For women aged30–49 the corresponding figures are 0.87%and 0.13%, respectively. However, alsoamongst men with a medium education level(of whom men with a vocational educationis the largest subgroup), there is a higher riskof shelter use – 0.90% amongst those aged30–49 – reflecting that men with a vocationaleducation have a considerably higher risk ofhomelessness than men with higher educa-tion. Amongst women this difference is notas large, as 0.24% of women aged 30–49with a medium education level used sheltersover the period, a level closer to what wefind amongst women with higher education.

Whilst the relation between a precarioussocioeconomic position and the risk ofhomelessness is evident, the highest preva-lence of shelter use is amongst people withpsychosocial vulnerabilities. As previouslymentioned, mental illness, drug and alcoholabuse problems, and previous imprisonmenthave been measured both before the periodfor which we measure shelter use, and simul-taneously with the observation of shelteruse. Whilst only about 3% of all males aged18–29 in the general population are recordedwith a mental illness from 1997 to 2001,about 9% are recorded with a mental illnessfrom 1997 to 2011. The longer measurementperiod for mental illness and substanceabuse better reflects the actual extent ofthese problems amongst people who haveexperienced homelessness. A very high num-ber of the shelter users had been recordedwith either mental illness or drug or alcoholabuse: 82.4% of all male shelter users (acrossage groups) and 87.2% of all female shelterusers measured from 1997 to 2011.

However, caution should be applied tothe causal interpretation of the sequencebetween mental illness, substance abuseproblems and homelessness. AlthoughT

ab

le2.(C

ontinued

)

Men

Wom

en

Age

1Ja

nuar

y2002

18–29

30–49

50+

18–29

30–49

50+

Tota

lN

inag

egr

oup

396,1

14

787,1

46

847,2

92

385,8

30

761,9

52

972,9

47

Vari

ab

leC

ate

go

ryPe

rcen

tsh

elte

ruse

rs

Perc

ent

shel

ter

use

rs

Perc

ent

shel

ter

use

rs

Perc

ent

shel

ter

use

rs

Perc

ent

shel

ter

use

rs

Perc

ent

shel

ter

use

rs

Men

tally

illdru

gab

use

rs(9

7–11)

Not

men

tally

illD

A0.6

61.1

40.3

50.1

70.3

20.0

9M

enta

llyill

DA

21.5

428.4

914.8

114.5

020.3

25.0

3Im

pri

sonm

ent

(97–01)

No

impri

sonm

ent

0.7

91.1

10.3

40.2

50.3

70.0

9Im

pri

sonm

ent

10.0

414.3

26.0

419.1

714.8

66.6

7Im

pri

sonm

ent

(97–11)

No

impri

sonm

ent

0.5

70.9

50.3

30.2

20.3

50.0

9Im

pri

sonm

ent

9.3

313.7

26.2

816.4

414.4

17.2

6

Benjaminsen 2051

Tab

le3.

Shel

ter

use

rsan

dnon-s

hel

ter

use

rsby

gender

and

age,

per

centa

gew

ith

each

bac

kgro

und

char

acte

rist

ic.

18–29

30–49

50+

Per

cent

of

shel

ter

use

rsPe

rce

nt

of

non-s

hel

ter

use

rs

Per

cent

of

shel

ter

use

rsPe

rce

nt

of

non-s

hel

ter

use

rs

Per

cent

ofsh

elte

ruse

rs

Per

cent

of

non-s

hel

ter

use

rs

Men

(n)

4053

392,0

61

10,8

64

776,2

82

3132

844,1

60

Non-im

mig

rants

83.8

91.2

86.3

91.4

92.3

95.8

Imm

igra

nts

14.2

6.4

13.3

8.3

7.4

4.1

Child

ren

ofim

mig

rants

2.1

1.6

0.4

0.3

0.3

0.1

Livi

ng

inC

open

hag

en12.2

15.4

16.0

10.2

12.5

6.7

Civ

ilst

atus

–N

opar

tner

66.6

41.0

67.8

28.7

73.2

27.4

Inco

me

–Lo

win

com

e75.6

49.7

48.5

13.8

41.4

34.2

Unem

plo

yed

66.2

25.3

61.3

12.2

69.8

52.8

Hig

hle

veled

uca

tion

0.4

6.0

5.3

19.1

12.1

15.6

Med

ium

leve

led

uca

tion

14.5

49.5

35.5

54.5

42.4

42.2

Com

puls

ory

leve

led

uca

tion

78.6

42.8

51.4

24.1

40.5

33.4

Men

talill

nes

s(M

I)(9

7–01)

18.6

2.7

20.2

3.1

20.2

3.1

Men

talill

nes

s(9

7–11)

55.2

8.6

48.6

8.5

46.0

11.5

Dru

gab

use

(DA

)(9

7–01)

15.6

0.8

13.0

0.6

3.6

0.1

Dru

gab

use

(97–11)

58.9

2.9

32.1

1.4

9.4

0.2

Alc

oholab

use

(AA

)(9

7–01)

14.9

2.0

28.9

2.0

38.7

2.1

Alc

oholab

use

(97–11)

41.4

4.3

61.3

4.8

71.5

5.2

MIan

d/o

rD

Aan

d/o

rA

A(9

7–11)

83.4

12.9

82.3

11.9

81.6

15.1

MIan

dD

A(9

7–11)

36.7

1.4

18.5

0.6

6.0

0.1

Impri

sonm

ent

(97–01)

24.9

2.3

21.5

1.8

8.0

0.5

Impri

sonm

ent

(97–11)

47.1

4.7

33.3

2.9

11.9

0.7

Wom

en

(n)

1046

384,7

84

3042

758,9

10

905

972,0

42

Non-im

mig

rants

85.3

90.5

90.8

91.7

91.6

95.7

Imm

igra

nts

12.7

7.9

8.5

8.0

8.3

4.2

Child

ren

ofim

mig

rants

2.0

2.0

0.8

0.3

0.1

0.1

Livi

ng

inC

open

hag

en15.7

17.2

13.2

9.2

8.8

7.7

Civ

ilst

atus

–N

opar

tner

55.2

34.9

60.2

24.6

61.3

44.9

Inco

me

–Lo

win

com

e72.0

58.5

40.3

14.7

41.1

45.6

Unem

plo

yed

82.4

35.6

69.5

17.4

72.3

66.8

Hig

hle

veled

uca

tion

1.2

9.0

8.8

26.5

15.2

12.5

Med

ium

leve

led

uca

tion

13.3

51.7

27.4

46.1

34.7

29.9

Com

puls

ory

leve

led

uca

tion

80.0

37.8

56.1

25.6

45.3

43.5 (con

tinue

d)


mental illness and substance abuse problemsincrease the risk of shelter use, homelessnessmay also cause a substance abuse problemto emerge or accelerate symptoms of mentalillness owing to the stressfulness of being ina homeless situation.

For individuals recorded with a mental ill-ness from 1997 to 2011, the highest preva-lence of shelter use was found amongstmales aged 30–49 at 7.45%, compared with0.78% amongst males without a mental ill-ness observed over that period. Of all maleshelter users, 49.9% were recorded as havinghad a mental illness. The prevalence of shel-ter use at 2.26% of men aged 30–49 wasmuch lower amongst women recorded with amental illness than amongst their male coun-terparts. However, a high number of femaleshelter users, 65.1%, have been recordedwith a mental illness between 1997 and 2011.

The highest prevalence of shelter use forany of the risk groups was observed for drugabusers. Amongst males aged 30–49recorded with a drug abuse between 1997and 2011, 24.51% of those used sheltersbetween 2002 and 2011. Amongst femaledrug abusers in the same age group, 17.90%used shelters. Moreover, amongst drug abu-sers aged 18–29 the prevalence of shelter usewas also high: at 17.13% for males and13.00% for females. In addition, a relativelyhigh rate of shelter use appears amongstindividuals with a diagnosis of alcoholabuse: 15.07% for men aged 30–49 and10.29% for women in the same age group.Whilst drug abuse is more widespreadamongst the younger shelter users, alcoholabuse is more common in the older agegroups. Further analysis (not shown) indi-cates that alcohol abuse amongst youngershelter users is part of a multi-use problem,as many of the young alcohol abusers alsoabuse drugs.

Previous imprisonment is the only riskfactor in which the risk of shelter use ishigher for women than for men. AmongstT

ab

le3.(C

ontinued

)

18–29

30–49

50+

Per

cent

of

shel

ter

use

rsPe

rce

nt

of

non-s

hel

ter

use

rs

Per

cent

of

shel

ter

use

rsPe

rce

nt

of

non-s

hel

ter

use

rs

Per

cent

ofsh

elte

ruse

rs

Per

cent

of

non-s

hel

ter

use

rs

Men

talill

nes

s(M

I)(9

7–01)

32.6

4.7

31.6

4.1

32.9

5.0

Men

talill

nes

s(9

7–11)

69.2

13.9

63.4

11.0

62.7

15.5

Dru

gab

use

(DA

)(9

7–01)

19.6

0.3

14.7

0.3

4.4

0.1

Dru

gab

use

(97–11)

53.4

1.0

31.7

0.6

10.1

0.2

Alc

oholab

use

(AA

)(9

7–01)

13.5

1.2

29.3

0.9

40.9

0.9

Alc

oholab

use

(97–11)

38.9

2.3

63.8

2.2

71.8

2.3

MIan

d/o

rD

Aan

d/o

rA

A(9

7–11)

86.4

15.5

88.1

12.2

87.2

16.8

Men

tally

illdru

gab

use

rs(9

7–11)

39.5

0.6

21.2

0.3

8.6

0.2

Impri

sonm

ent

(97–01)

7.6

0.1

7.3

0.2

3.2

0.0

Impri

sonm

ent

(97–11)

18.4

0.3

12.1

0.3

4.3

0.1

Benjaminsen 2053

women aged 18–29 who were incarceratedfrom 1997 to 2001 (prior to the observationperiod for homelessness), 19.17% used shel-ters between 2001 and 2011; amongstwomen aged 30–49 who had been incarcer-ated, 14.86% used shelters. Amongst men inthe same age groups, these figures were10.04% and 14.32%, respectively. As manyas 47.1% of male shelter users aged 18–29and 33.3% of male shelter users aged 30–49have been incarcerated between 1997 and2011. For their female counterparts, the cor-responding figures are somewhat lower, at18.4% and 12.1%, respectively.

Multiple social exclusion and therisk of homelessness

Homelessness is often part of a complex pat-tern of multiple social exclusions, in whichthe individual is excluded from many dimen-sions of life. Table 4 shows, by gender andage, the prevalence of homelessness for thegeneral population according to the numberof risk characteristics. Table 5 shows theshelter users and non-shelter users groupedby the number of risk factors, also by genderand age. This analysis includes seven riskfactors: low education, unemployment, lowincome, mental illness, alcohol abuse, drugabuse and previous imprisonment. This partof the analysis, which includes the risk fac-tors simultaneously, measures all seven fac-tors prior to the observation period whenhomelessness was measured.

Whilst the youngest age group includesmany students recorded as unemployed, andthe oldest age group includes many pen-sioners who are also unemployed, the cate-gory of those aged 30–49 includes mainlyage groups active in the labour market.Although the majority of individuals in themiddle-aged group in the general populationhave none of the seven risk factors, there isa considerable group exposed to one or tworisk factors (Table 5). However, only about T

ab

le4.

Pre

vale

nce

ofsh

elte

ruse

inth

eD

anis

had

ult

popula

tion

by

the

cum

ula

tive

num

ber

ofri

skfa

ctors

,per

cent

and

Tota

lN.

0ri

skfa

ctors

1–2

risk

fact

ors

3–4

risk

fact

ors

5–7

risk

fact

ors

Gen

der

and

age

Shel

ter

use

rs,%

Tota

lN

inpopula

tion

Shel

ter

use

rs,%

Tota

lN

inpopula

tion

Shel

ter

use

rs,%

Tota

lN

inpopula

tion

Shel

ter

use

rs,%

Tota

lN

inpopula

tion

Men

18–29

0.1

5132,4

10

0.6

2198,4

17

3.3

263,0

58

23.7

82229

30–49

0.2

9481,0

04

1.5

8264,3

64

10.7

838,3

36

33.2

73442

50+

0.,1

4265,4

70

0.3

2448,4

32

0.9

0132,5

32

14.6

9858

Tota

l0.2

2878,8

84

0.7

5911,2

13

3.1

7233,9

26

27.5

86529

Wom

en18–29

0.0

2107,0

43

0.1

6207,7

45

0.7

970,2

48

18.1

4794

30–49

0.0

8443,0

17

0.4

2281,8

27

3.4

535,9

93

24.4

81115

50+

0.0

3195,2

32

0.0

8532,2

55

0.1

6244,6

84

5.8

0776

Tota

l0.0

6745,2

92

0.1

91,0

21,8

27

0.6

2350,9

25

17.2

12685


5% in the middle-aged group in the generalpopulation have three risk factors or more.In this age group, the risk of shelter useincreases sharply when the number of riskfactors exceeds three or more. Amongst menaged 30–49 only 1.58% of those with one ortwo risk factors have used shelters, com-pared with 10.78% amongst those with threeor four risk factors, and 33.27% of thosewith five or more risk factors have used shel-ters (Table 4). The same pattern is observedfor women, albeit with a lower prevalence ofshelter use in all groups.

Almost two-thirds of young shelter users(both genders) and about half of middle-aged shelter users have three risk character-istics or more (Table 5). The length of themeasurement period for the risk factors

obviously affects the shares with risk charac-teristics in Table 5. If mental illness, sub-stance abuse and incarceration are measuredfrom 1997 to 2011 (not shown), even fewerof the shelter users have none of the risk fac-tors (e.g. 3.64% and 2.40% amongst maleand female shelter users aged 30–49, respec-tively). The analysis thus shows that home-lessness in Denmark is largely part of apattern of multiple social exclusions.

Multivariate analysis of the risk ofshelter use for the Danish adultpopulation

A multivariate logistic regression model ofthe risk of homelessness over the 10-yearperiod has been estimated for the entire adult

Table 5. Shelter users and non-shelter users, percentage with cumulative number of risk factors.

Gender Age group by shelteruse or non-shelter use

0 riskfactors

1–2 riskfactors

3–4 riskfactors

5–7 riskfactors

Men 18–29Shelter users 4.98 30.32 51.62 13.08Non-shelter users 33.72 50.30 15.55 0.43Total 33.43 50.09 15.92 0.5630–49Shelter users 12.92 38.48 38.05 10.54Non-shelter users 61.78 33.52 4.41 0.30Total 61.11 33.59 4.87 0.4450+Shelter users 11.49 46.26 38.22 4.02Non-shelter users 31.41 52.95 15.56 0.09Total 31.33 52.93 15.64 0.10

Women 18–29Shelter users 2.39 30.98 52.87 13.77Non-shelter users 27.81 53.91 18.11 0.17Total 27.74 53.84 18.21 0.2130–49Shelter users 11.21 38.99 40.83 8.97Non-shelter users 58.33 36.98 4.58 0.11Total 58.14 36.99 4.72 0.1550+Shelter users 6.85 45.08 43.09 4.97Non-shelter users 20.08 54.71 25.13 0.08Total 20.07 54.71 25.15 0.08

Benjaminsen 2055

population 18 years or older in 2002, sepa-rately for men and women (Table 6). Wheninteraction effects are included in the model,

a large number of interactions become signif-icant owing to the large number of individu-als in the data set. Therefore, the interaction

Table 6. Logistic regression model, shelter use (OR) (2002–2011) by risk factors (1997–2001), separatefor men and women.

Variable (reference group) Category Model 1, Main effects Model 2, Interaction model

OR SE P OR SE P

MenAge (18–29) 30–49 1.95 0.04 0.000 1.93 0.04 0.000

50+ 0.43 0.01 0.000 0.43 0.01 0.000Immigrant status(Non-immigrant)

Immigrant 1.89 0.05 0.000 1.90 0.05 0.000Children ofimmigrants

1.03 0.10 0.765 1.02 0.09 0.822

Urbanity (not in Cph) Living in Cph 0.96 0.02 0.089 0.96 0.02 0.093Marital status (couple) Single 2.68 0.05 0.000 2.61 0.05 0.000Income (high income) Low income 1.66 0.03 0.000 1.65 0.03 0.000Employment (yes) Unemployed 2.18 0.04 0.000 2.11 0.04 0.000Education (high level) Compulsory level 2.35 0.08 0.000 2.30 0.08 0.000

Medium level 1.38 0.05 0.000 1.37 0.05 0.000Unspecified 1.82 0.08 0.000 1.80 0.08 0.000

Mental illness (MI) (none) Mental illness 2.10 0.05 0.000 2.55 0.07 0.000Drug abuse (DA) (none) Drug abuse 3.08 0.10 0.000 5.87 0.24 0.000Alcohol abuse (AA) (none) Alcohol abuse 5.91 0.12 0.000 7.55 0.18 0.000Imprisonment (none) Imprisonment 3.98 0.09 0.000 3.76 0.09 0.000MI+DA (none) Both MI and DA – – – 0.64 0.04 0.000MI+AA (none) Both MI and AA – – – 0.71 0.03 0.000DA+AA (none) Both DA and AA – – – 0.30 0.02 0.000Constant 0.001 \0.001 0.000 0.001 \0.001 0.000WomenAge (18–29) 30–49 2.07 0.08 0.000 2.06 0.08 0.000

50+ 0.27 0.01 0.000 0.29 0.01 0.000Immigrant status(non-immigrant)

Immigrant 1.31 0.07 0.000 1.36 0.07 0.000Children ofimmigrant

1.52 0.25 0.009 1.56 0.25 0.006

Urbanity (not in Cph) Living in Cph 0.91 0.04 0.030 0.89 0.04 0.008Marital status (couple) Single 1.92 0.06 0.000 1.81 0.06 0.000Income (high income) Low income 1.33 0.05 0.000 1.30 0.04 0.000Employment (yes) Unemployed 2.82 0.11 0.000 2.61 0.10 0.000Education (high level) Compulsory level 2.00 0.11 0.000 1.89 0.11 0.000

Medium level 1.14 0.07 0.023 1.13 0.07 0.032Unspecified 1.53 0.12 0.000 1.48 0.12 0.000

Mental illness (none) Mental illness 3.01 0.11 0.000 4.28 0.18 0.000Drug abuse (none) Drug abuse 6.61 0.36 0.000 21.05 1.57 0.000Alcohol abuse (none) Alcohol abuse 10.35 0.41 0.000 20.46 0.98 0.000Imprisonment (none) Imprisonment 6.80 0.52 0.000 4.94 0.38 0.000MI+DA (none) Both MI and DA – – – 0.50 0.05 0.000MI+AA (none) Both MI and AA – – – 0.41 0.03 0.000DA+AA (none) Both DA and AA – – – 0.15 0.02 0.000Constant \0.001 \0.001 0.000 0.000 0.000 0.000


model includes only specific two-way inter-actions between the three variables: mentalillness, drug abuse and alcohol abuse. Thesevariables represent the most intrinsic aspectsof homelessness in the advanced welfarestate – those individuals who fall through thesocial safety net because of highly complexsupport needs. The analysis with interactioneffects thus offers an insight into how a dualdiagnosis of mental illness and substanceabuse, and the combination of alcohol anddrug abuse, affects the parameter estimatesof shelter use. Incarceration has not beenincluded amongst the interaction variablesbecause of a very high multicollinearity withdrug abuse.

The table shows both a model (1) withonly main effects and a model (2) includingthe interaction effects. Mental illness, sub-stance abuse and incarceration have beenmeasured from 1997 to 2001, prior to themeasurement period for homelessness.

The analysis shows that for both men andwomen almost all risk factors in both mod-els have a significant impact on the risk ofhomelessness: age, gender, civil status, immi-grant background, low income, unemploy-ment, low education, drug and alcoholabuse, mental illness and previous imprison-ment. Only the variable for being a residentin Copenhagen at the beginning of theperiod is insignificant for men, as well as thecoefficient for being a child of immigrants.

For those not belonging to any of the riskgroups identified in the model, the risk ofbecoming homeless is extremely small, asillustrated by the very low odds ratio for theconstant. All variables in the model contrib-ute significantly to explaining the risk ofhomelessness. The coefficients for most riskfactors are relatively large and, given thepopulation size, they are estimated veryprecisely.

The effects for the demographic variablesshow that both men and women share ahigher risk of homelessness in the age group

30–49 years. Being an immigrant is associ-ated with a higher risk of shelter use forboth men and women. When I control forother risk factors, I find no enlarged risk ofshelter use amongst male children of immi-grants compared with non-immigrant males,whereas an increased risk of shelter useappears for female children of immigrantscompared with non-immigrant females.

There is a higher risk of shelter use forboth single men and women, with the oddsratio being highest for single men.Surprisingly, for both men and women, nohigher risk of shelter use appears for individ-uals who were residents of Copenhagen atthe beginning of the period. For both menand women the effect is actually slightly neg-ative (OR\1), although significant only forwomen. This result indicates that a higheroverall risk of homelessness found in thebivariate analysis (Tables 2 and 3) for indi-viduals living in the capital (except for theyoungest age group) is attributable to ahigher prevalence of other risk factors, espe-cially those related to substance abuse andmental illness, amongst people living in thecity.

The socioeconomic risk factors of lowincome, unemployment and low educationalattainment are all associated with a higherrisk of homelessness for both men andwomen. The largest coefficients for thesocioeconomic variables are found for hav-ing no employment, with an odds ratio of2.3 for men and 2.8 for women (model 1).Thus even though I control for vulnerabilityfactors such as mental illness and substanceuse, the lack of employment remains animportant risk factor for homelessness,along with income poverty and lack of edu-cation. In contrast, the coefficient for lowincome is considerably smaller than forunemployment, with an odds ratio of 1.7 formen and 1.3 for women (model 1).

Substance abuse problems stronglyincrease the risk of homelessness, with odds

Benjaminsen 2057

ratios for the main effects (model 1) for menat 5.9 for alcohol abuse and 3.1 for drugabuse, and at 10.4 and 6.6, respectively, forwomen. The relative risk of homelessnessassociated with mental illness also remainssubstantial, with an odds ratio of 2.1 formen and 3.0 for women, but considerablylower than the risk attached to drug andalcohol abuse. The interaction effects (model2) give valuable insights into how the combi-nation of these main risk factors affects therisk of homelessness. For drug abuse, alco-hol abuse and mental illness, comorbiditybetween any two of these variables moder-ates the combined main effects, so that thecumulative effect of each of these variablesis not fully reflected in the total effect on therisk of shelter use. This pattern is found forboth men and women. Nonetheless, giventhe overall cumulative structure of the riskfactors in the model, the highest overall riskof homelessness is still found for those indi-viduals with comorbidity between the majorrisk factors of drug and alcohol abuse andmental illness, that is people with a dualdiagnosis (mental illness and substanceabuse) or combined use of alcohol anddrugs.

Moreover, previous imprisonment is alsostrongly associated with a risk of shelter use,with an odds ratio of 4.0 for men and 6.8 forwomen. As a strong multicollinearity existsbetween drug abuse and incarceration, asmentioned earlier, further interaction effectswith incarceration are thus problematic forinclusion in the model.

Conclusion

This article has presented the first statisticalanalysis of the risk of homelessness based onmicro-data for the entire adult population ofa country, Denmark. The analysis has beenbased on robust administrative data fromvarious data sources, including data from anationwide client registration system on

homeless shelters and data from the publichealth system. The independent variableshave included not only demographic andsocioeconomic factors but also mental ill-ness, substance abuse and previousincarceration.

The findings show that in a typicalScandinavian welfare state such asDenmark, homelessness is widely concen-trated amongst people with severe psychoso-cial problems and complex support needsand that the homelessness problem inDenmark is not widely associated with moregeneral poverty problems. The results alsoshow that homelessness is part of a patternof multiple social exclusion. For individualsfacing exclusion in many domains, homeless-ness is a common occurrence. For peopleexposed to five or more of the risk factors inthe analysis, about one in four have usedshelters over the observation period of tenyears. For drug users in particular, home-lessness is a common experience: about onein four drug abusers have used shelters overthe ten-year period, whereas for the mentallyill without a dual diagnosis the prevalence ofshelter use is considerably lower than fordrug abusers.

Thus a considerable part of homelessnessin Denmark is the housing problem of sub-stance abusers and part of their generalexclusion in society, likely reflecting theirparticular vulnerability to evictions, barriersto housing access, and less extensive avail-able support. In contrast, the mentally ill areoften in supported accommodation undersocial-psychiatric services or receive floatingsupport in their own home. However, despitethe high prevalence of shelter use amongstindividuals suffering from multiple socialexclusion, the majority – even in high-riskgroups such as people with drug abuse prob-lems or a dual diagnosis – have not usedshelters over the period.

In the multivariate model, low income,unemployment and low education remain


significant factors. This result shows thatsocioeconomic conditions play a role at theindividual level, adding to the risk of home-lessness – even in a Scandinavian welfarestate, with its overall relatively low levels ofpoverty. However, that the statistical coeffi-cients for the socioeconomic variables aremuch lower than for substance abuse andmental illness is also evident. Moreover,amongst the socioeconomic explanatoryvariables, the magnitude of the coefficientfor unemployment is substantially largerthan the coefficient for low income on therisk of shelter use. These results indicate thatthe extensive redistribution of income in theScandinavian welfare state model substan-tially modifies the risk of homelessness dueto poverty. This pattern also indicates thatexclusion from the labour market for mar-ginalised groups not only has financial con-sequences for the individual but alsorepresents an exclusion from daily meaning-ful activities and social contacts.

The findings are in line with results of theDanish national counts of homelessness,which have shown that about 80% of indi-viduals recorded as homeless in these point-in-time surveys have either a mental illness,a substance abuse problem, or both(Benjaminsen and Lauritzen, 2013). Thesefindings correspond with the prevailinghypothesis in homelessness theory thathomelessness in countries with low levels ofpoverty and extensive welfare systems ismore highly concentrated amongst peoplewith complex needs, whereas homelessnessin more unequal societies with less extensivewelfare systems more widely affects broadersegments of poor households (Stephens andFitzpatrick, 2007).

As mentioned earlier, analysis from theUSA on similar shelter data has shown thatdespite a considerable part of homeless peo-ple having complex support needs in the US,there is also a significant part withoutrecords of mental illness or substance abuse

problems. Those individuals are more likelyto be homeless because of more general pov-erty and housing affordability problems(Kuhn and Culhane, 1998). Although thegroup with a transitional pattern of shelteruse (i.e. few and short shelter stays) has alsobeen identified in recent Danish research, inDenmark even transitional shelter use isconcentrated mainly amongst people withmental illness and substance abuse problems(Benjaminsen and Andrade, 2015). The anal-ysis in this article further corroborates theseearlier findings, as it has demonstrated therisk pattern of shelter use for the populationas a whole and has shown that the risk ofshelter use in the Danish adult population iswidely concentrated amongst people withcomplex needs.

Whilst the overall extent of homelessnessin Denmark is relatively small, the resultsindicate that the otherwise strong socialsafety net in Denmark is apparently nottight enough for the most socially vulnerablegroups. In terms of policy implications, theresults indicate a need for developing poli-cies and interventions for strengthening theprovision of housing and social support forthe most marginalised groups, particularlysubstance abusers. As many substance abu-sers have a criminal record and as a previousprison sentence is also a significant risk fac-tor for shelter use, rehousing efforts shouldalso incorporate a focus on people leavingthe prison system without a housingsolution.

The research literature generally points to‘Housing First’– immediate permanent hous-ing with intensive social support followingevidence-based methods such as AssertiveCommunity Treatment (ACT) or IntensiveCase Management (ICM) – as an effectiveintervention for homeless individuals withmental illness and addiction problems(Coldwell and Bendner, 2007; Nelson et al.,2007; Tsemberis et al., 2004). Nonethelessthere is an ongoing debate as to whether this

Benjaminsen 2059

method also applies to individuals with themost severe addiction problems (Kerteszand Weiner, 2009; Kertesz et al., 2009;Pleace, 2011).

A turn towards Housing First-orientedpolicies and practices only occurred rela-tively late in Denmark: the first nationalhomelessness strategy, based on the HousingFirst approach, was adopted in 2008, andthe strategy programme from 2009 to 2013emphasised providing access to permanenthousing in combination with intensive socialsupport following the ACT, ICM and CTImethods (Ministry of Internal and SocialAffairs, 2009). In accordance with the resultsfrom international research, the evaluationresearch on the Danish strategy programmegenerally shows that Housing First and theseevidence-based floating support methods arehighly effective in rehousing homeless peo-ple. However, the evaluation research alsoshows that these types of interventions stillcover only a minor part of homeless peoplein Denmark (Benjaminsen, 2013; Rambølland SFI, 2013). The results of the analysis inthis article thus point to a need to extendsuch intensive interventions to cover a widerpart of individuals with severe addictionproblems and dual diagnoses – individualswith a high risk of falling through the wel-fare safety net and becoming homeless.Moreover, the evaluation of the Danish pro-gramme also points to considerable struc-tural barriers to implementing HousingFirst-based policies, especially given anincreasing lack of affordable housing inDenmark’s larger cities.

Thus one should be cautious about inter-preting the empirical findings in this articlefrom an individualist perspective. Whilst theanalysis indicates that homelessness inDenmark is not widely associated with moregeneral poverty problems, the statisticalanalysis did not measure possible structuraland systemic factors such as barriers ofaccess to affordable housing for vulnerable

groups, nor did it address possible shortcom-ings in the social support system for thesegroups. Rather, the analysis has identifiedthose individuals most likely to be affectedby such adverse structural trends and sys-temic deficiencies and who are in need ofmore intensive and targeted interventions.

Funding

This research was financed by a grant from theDanish Council for Independent Research, SocialSciences.

References

Allgood S and Warren RS Jr (2003) The duration

of homelessness: Evidence from a national sur-

vey. Journal of Housing Economics 12(4):

273–290.Anderson I and Christian J (2003) Causes of

homelessness in the UK: A dynamic analysis.

Journal of Community & Applied Social Psy-

chology 13(2): 105–118.Anderson I and Tulloch D (2000) Pathways

Through Homelessness: A Review of the

Research Evidence. Edinburgh: Scottish

Homes.Ankestyrelsen (2014) Brugere af botilbud efter ser-

vicelovens § 110. Arsstatistik 2013 [Users of §

110 Accommodation under the Social Assis-

tance Law. Yearly Statistics 2013]. Copenha-

gen: The Social Appeals Board.Avramov D (1999) Coping with Homelessness:

Issues to be Tackled and Best Practices in Eur-

ope. Aldershot: Ashgate.Bearsley-Smith CA, Bond LM, Littlefield L, et al.

(2008) The psychosocial profile of adolescent

risk of homelessness. European Child & Adoles-

cent Psychiatry 17(4): 226–234.Benjaminsen L (2009) Hjemløshed i Danmark

2009. National kortlægning [Homelessness in

Denmark 2009. National Mapping]. Copenha-

gen: SFI (09:25).Benjaminsen L (2013) Policy review up-date:

Results from the Housing First based Danish

homelessness strategy. European Journal of

Homelessness 7(2): 109–131.Benjaminsen L and Andrade SB (2015) Testing a

typology of homelessness across welfare


regimes. Shelter use in Denmark and the

USA. Housing Studies 30(6): 858–876.Benjaminsen L and Christensen I (2007) Hjem-

løshed i Danmark 2007. National kortlægning

[Homelessness in Denmark 2007. National

mapping]. Copenhagen: SFI (07:22).Benjaminsen L and Dyb E (2008) The effective-

ness of homelessness policies. Variations

among the Scandinavian countries. European

Journal of Homelessness 2(1): 45–67.Benjaminsen L and Lauritzen HH (2013) Hjem-

løshed i Danmark 2013. National kortlægning

[Homelessness in Denmark 2013. National

Mapping]. Copenhagen: SFI (13:21).Blid M, Gerdner A and Bergmark A (2008) Pre-

diction of homelessness and housing provi-

sions in Swedish municipalities. European

Journal of Housing Policy 8(4): 399–421.Busch-Geertsema V, Edgar W, O’Sullivan E, et

al. (2010) Homelessness and Homeless Policies

in Europe: Lessons from Research. European

Consensus Conference on Homelessness. Brus-

sels: FEANTSA.Caton CLM, Dominguez B, Schanzer B, et al.

(2005) Risk factors for long-term homeless-

ness: Findings from a longitudinal study of

first-time homeless single adults. American

Journal of Public Health 95(10): 1753–1759.Caton CLM, Shrout PE, Dominguez B, et al.

(1995) Risk factors for homelessness among

women with schizophrenia. American Journal

of Public Health 85(8): 1153–1156.Caton CLM, Shrout PE, Eagle PF, et al. (1994)

Risk factors for homelessness among schizo-

phrenic men: A case-control study. American

Journal of Public Health 84(2): 265–270.Clapham D (2003) Pathways approaches to

homelessness research. Journal of Community

and Applied Social Psychology 13(2): 119–127.Coldwell C and Bendner W (2007) The effective-

ness of assertive community treatment for

homeless populations with severe mental ill-

ness: A meta-analysis. American Journal of

Psychiatry 164(3): 393–399.Cornes M, Joly L, Manthorpe J, et al. (2011)

Working together to address multiple exclu-

sion homelessness. Social Policy and Society

10(4): 513–522.Culhane D and Metraux S (1999) Assessing

relative risk for homeless shelter usage in

New York City and Philadelphia.

Population Research and Policy Review 18:

219–236.

Culhane D, Dejowki EF, Ibanez J, et al. (1994)

Public shelter admission rates in Philadelphia

and New York City: The implications of turn-

over for sheltered population count. Housing

Policy Debate 5(2): 107–140.Culhane D, Metraux S, Park JM, et al. (2007)

Testing a typology of family homelessness

based on patterns of public shelter utilization

in four U.S. jurisdictions. Implications for pol-

icy and program planning. Housing Policy

Debate 18(1): 1–28.Dyb E and Johannessen K (2013) Bostedsløse i

Norge 2012 – En kartlegging [Homelessness in

Norway 2012 – A Mapping]. Oslo: NIBR

(Norwegian Institue of Urban and Regional

Research), Report 2013:5.Edgar B and Meert H (2005) Fourth Review of

Statistics on Homelessness in Europe. The

ETHOS Definition of Homelessness. Brussels:

FEANTSA.Edgar W, Harrison M, Watson P, et al. (2007)

Measurement of Homelessness at European

Union Level. Brussels: European Commission,

DG Employment, Social Affairs and Equal

Opportunities.

Fitzpatrick S (2005) Explaining homelessness. A

critical realist perspective. Housing, Theory

and Society 22(1): 1–17.Fitzpatrick S (2012) Homelessness. In: Clapham

D, Clark W and Gibb K (eds) The SAGE

Handbook of Housing Studies. London: Sage,

pp. 359–378.Fitzpatrick S and Stephens M (2007) An Interna-

tional Review of Homelessness and Social

Housing Policy. London: Department for

Communities and Local Government.Fitzpatrick S, Bramley G and Johnsen S (2013)

Pathways into multiple exclusion homelessness

in seven UK cities. Urban Studies 50(1):

148–168.Fitzpatrick S, Johnsen S and White M (2011)

Multiple exclusion homelessness in the UK:

Key patterns and intersections. Social Policy

& Society 10(4): 501–512.Fitzpatrick S, Kemp P and Klinker S (2000) Sin-

gle Homelessness: An Overview of Research in

Britain. Bristol: Policy Press.

Benjaminsen 2061

Folsom DP, Hawthorne W, Lindamer L, et al.

(2005) Prevalence and risk factors for home-

lessness and utilization of mental health ser-

vices among 10,340 patients with serious

mental illness in a large public mental health

system. American Journal of Psychiatry 162(2):

370–376.France A (2008) Risk factor analysis and the youth

question. Journal of Youth Studies 11(1): 1–15.Herman DB, Susser ES, Struening E, et al. (1997)

Adverse childhood experiences: Are they risk

factors for adult homelessness? American Jour-

nal of Public Health 87(2): 249–255.Kelly P (2001) Youth at risk: Processes of indivi-

dualisation and responsibilisation in the risk

society. Discourse: Studies in the Cultural Poli-

tics of Education 22(1): 23–33.Kemp PA, Neale J and Robertson M (2006)

Homelessness among problem drug users: Pre-

valence, risk factors and trigger events. Health

and Social Care in the Community 14(4):

319–328.Kertesz SG and Weiner SJ (2009) Housing the

chronically homeless: High hopes, complex

realities. Journal of the American Medical

Association 301(17): 1822–1824.Kertesz SG, Crouch K, Milby JB, et al. (2009)

Housing first for homeless persons with active

addiction: Are we overreaching? The Milbank

Quarterly 87(2): 495–534.Koegel P, Melamid E and Burnham A (1995)

Childhood risk factors for homelessness

among homeless adults. American Journal of

Public Health 85(12): 1642–1649.Kuhn R and Culhane D (1998) Applying cluster

analysis to test a typology of homelessness by

pattern of shelter utilization: Results from the

analysis of administrative data. American Jour-

nal of Community Psychology 26(2): 207–232.Lauritzen HH, Boje-Kovacs B and Benjaminsen

L (2011)Hjemløshed I Danmark 2011. National

kortlægning [Homelessness in Denmark 2011.

National Mapping]. Copenhagen: SFI (11:45).MacKenzie D and Chamberlain C (2003) Home-

less Careers: Pathways In and Out of Home-

lessness. Melbourne: Swinburne and RMIT

Universities.Metraux S and Culhane D (1999) Family

dynamics, housing and recurring homelessness

among women in New York City homeless

shelters. Journal of Family Issues 20(3):

371–396.Ministry of Internal and Social Affairs (2009)

The Government’s Homelessness Strategy – A

Strategy to Reduce Homelessness in Denmark

2009–2012. Copenhagen: Indenrigs- og Social-

ministeriet and Ramboll Management Con-

sulting. Available at: www.feantsa.org/code/

en/pg.asp?Page=1169 (accessed 17 January

2013).Mors O, Perto GP and Mortensen PB (2011) The

Danish Psychiatric Central Research Register.

Scandinavian Journal of Public Health 39(S7):

54–57.Neale J (1997) Homelessness and theory reconsid-

ered. Housing Studies 12(1): 47–61.Nelson G, Aubry T and Lafrance A (2007) A

review of the literature on the effectiveness of

housing and support, assertive community

treatment, and intensive case management

interventions for persons with mental illness

who have been homeless. American Journal of

Orthopsychiatry 77(3): 350–361.Nielsen SF, Hjorthøj CR, Erlangsen A, et al.

(2011) Psychiatric disorders and mortality

among people in homeless shelters in Den-

mark: A nationwide register-based cohort

study. Lancet 377: 2205–2214.Organisation for Economic Cooperation and

Development (OECD) (2014) Society at a

Glance. Highlights DENMARK. OECD Social

Indicators. Available at: http://www.oecd.org/

denmark/OECD-SocietyAtaGlance2014-High

lights-Denmark.pdf (accessed 12 October

2014).Piliavin I, Westerfelt H and Elliott E (1989) Esti-

mating mental illness among the homeless:

The effects of choice-based sampling. Social

Problems 36(5): 525–541.Pleace N (2000) The new consensus, the old con-

sensus and the provision of services for people

sleeping rough. Housing Studies 15(4):

581–594.Pleace N (2011) The ambiguities, limits and risks of

Housing First from a European perspective. Eur-

opean Journal of Homelessness 5(2): 113–127.Rambøll and SFI (2013) Hjemløsestrategien.

Afsluttende Rapport [The Homelessness Strat-

egy: Final Report]. Copenhagen: Rambøll and

SFI.


www.feantsa.org/code/en/pg.asp?Page=1169

www.feantsa.org/code/en/pg.asp?Page=1169

http://www.oecd.org/denmark/OECD-SocietyAtaGlance2014-Highlights-Denmark.pdf



Sayer A (1992) Method in Social Science: A Rea-

list Approach. London: Routledge.Sayer A (2000) Realism and Social Science. Lon-

don: Sage.Shinn M (2007) International homelessness: Pol-

icy, socio-cultural, and individual perspectives.

Journal of Social Issues 63(3): 657–677.Shinn M, Weitzman BC, Stojanovic D, et al.

(1998) Predictors of homelessness among fam-

ilies in New York City: From shelter request

to housing stability. American Journal of Pub-

lic Health 88(10): 1651–1657.Skifter Andersen H (2010) Spatial assimilation in

Denmark? Why do immigrants move to and

from multi-ethnic neighbourhoods? Housing

Studies 25(3): 281–300.Skifter Andersen H, Turner LM and Søholt S

(2012) The special importance of housing pol-

icy for ethnic minorities: Evidence from a com-

parison of four Nordic countries, International

Journal of Housing Policy 13(1): 20–44.Socialstyrelsen (2012) Hemloshet och utestangning

fran bostadsmarknaden – Omfattning och kar-

aktar [Homelessness and Exclusion from the

Housing Market – Extent and Profile]. Stock-

holm: Socialstyrelsen (The National Board of

Health and Welfare).Sosin MR and Bruni M (1997) Homelessness and

vulnerability among adults with and without

alcohol problems. Substance Use & Misuse

32(7&8): 939–968.Stephens M and Fitzpatrick S (2007) Welfare

regimes, housing systems and homelessness.How are they linked? European Journal of

Homelessness 1(1): 201–212.Sullivan G, Burnam A and Koegel P (2000) Path-

ways to homelessness among the mentally ill.Social Psychiatry and Psychiatric Epidemiology

35: 444–450.Susser ES, Lin SP and Conover SA (1991) Risk

factors for homelessness in patients admitted

to a state mental hospital. American Journal of

Psychiatry 148(12): 1659–1664.Susser ES, Moore R and Link B (1993) Risk fac-

tors for homelessness. Epidemiologic Reviews

15(2): 546–556.Toro PA (2007) Toward an international under-

standing of homelessness. Journal of Social

Issues 63(3): 461–481.Tsemberis S, Gulcur L and Nakae M (2004)

Housing first, consumer choice and harmreduction for homeless individuals with a dualdiagnosis. American Journal of Public Health

94(4): 651–656.Van Laere IR, de Wit MA and Klazinga NS

(2009) Pathways into homelessness: Recentlyhomeless adults problems and service usebefore and after becoming homeless inAmsterdam. BMC Public Health 9: 3.

Benjaminsen 2063

homelessness in a scandinavian welfare state: the risk of ... · the danish national centre for...

Documents