Prevalence and Risk of Homelessness Among US Veterans

ORIGINAL RESEARCH

Jamison Fargo, PhD, MS; Stephen Metraux, PhD; Thomas Byrne, MSW; Ellen Munley; Ann Elizabeth Montgomery, PhD; Harlan Jones; George Sheldon, PhD; Vincent Kane, MSW; Dennis Culhane, PhD

Suggested citation for this article: Fargo J,
Metraux S, Byrne T, Munley E, Montgomery AE, Jones H, et al. Prevalence and
risk of homelessness among US veterans. Prev Chronic Dis 2012;9:110112. DOI:
http://dx.doi.org/10.5888/pcd9.110112.

PEER REVIEWED

Abstract

Introduction

Understanding the prevalence of and risk for homelessness among veterans is
prerequisite to preventing and ending homelessness among this population.
Homeless veterans are at higher risk for chronic disease; understanding the
dynamics of homelessness among veterans can contribute to our understanding of
their health needs.

Methods

We obtained data on demographic characteristics and veteran status for 130,554
homeless people from 7 jurisdictions that provide homelessness services, and for
the population living in poverty and the general population from the American
Community Survey for those same jurisdictions. We calculated prevalence of
veterans in the homeless, poverty, and general populations, and risk ratios (RR)
for veteran status in these populations. Risk for homelessness, as a function of
demographic characteristics and veteran status, was estimated by using multivariate
regression models.

Results

Veterans were overrepresented in the homeless population, compared with both the
general and poverty populations, among both men (RR, 1.3 and 2.1, respectively)
and women (RR, 2.1 and 3.0, respectively). Veteran status and black race
significantly increased the risk for homelessness for both men and women. Men in
the 45- to 54-year-old age group and women in the 18- to 29-year-old age group
were at higher risk compared with other ages.

Conclusion

Our findings confirm previous research associating veteran status with higher
risk for homelessness and imply that there will be specific health needs among
the aging homeless population. This study is a basis for understanding variation
in rates of, and risks for, homelessness in general population groups, and
inclusion of health data from US Department of Veterans Affairs records can
extend these results to identifying links between homelessness and health risks.

Top of Page

Introduction

Veterans are overrepresented among the homeless in the United States and are
at greater risk than nonveterans of becoming homeless (1-10). Homelessness is
associated with chronic health conditions, either causing or preceding such
conditions, becoming a consequence of such conditions, or complicating the
treatment and care of such conditions (11-14). Furthermore, among the 136,000
homeless veterans in 2009, 53% had a chronic health condition (15).
Understanding the epidemiology of homelessness and the specific factors
associated with increased risk of becoming homeless is prerequisite to both
reducing homelessness and more effectively addressing the health needs of this
population.

The objective of this study was to provide a more detailed assessment of risk
for homelessness among veterans than has been previously reported, in comparison
with the nonveteran population and after controlling for various demographic
characteristics. Specifically, we sought to answer 2 research questions: 1) Is
veteran status associated with an increased risk of homelessness? and 2) Does
risk of homelessness among veterans vary as a function of age, race, and sex?

Top of Page

Methods

Study design

Homeless Management Information Systems (HMIS) and American Community Survey
(ACS) data from 7 jurisdictions provided a basis for estimating the prevalence of
veterans in the homeless, poverty, and overall populations; calculating risk
ratios for veteran status in the homeless population compared with veteran
status in the poverty and overall populations; determining if veteran status is
associated with an increased risk of homelessness; and identifying whether risk
of homelessness among veterans varies as a function of age, race, or sex.

Data collection

Data for this study came from the 2008 HMIS and the 2006-2008 ACS. Service
providers use HMIS to record data on client characteristics and use of services
in homeless populations across a local area known as a continuum of care (CoC).
A CoC is a planning entity established by the US Department of Housing and Urban
Development (HUD) for a geographic unit, which can range in size from a large
city to multiple rural counties. In a CoC, stakeholders and service providers
coordinate resources and provide services (eg, shelter, housing, food) to
address homelessness (16). The more than 400 CoCs throughout the United States
are each mandated by HUD to maintain an HMIS that collects data on the local
service-using homeless population. The data fields collected include identifying
information, veteran status, demographics, the presence of disabling conditions,
and dates of program entry and exit.

A convenience sample of 11 urban CoCs from geographic regions throughout the
country initially provided HMIS data for this study. These HMIS datasets
consisted of unduplicated, de-identified, individual records for adults who used
emergency shelter or transitional housing within their CoC during 2008. HMIS
data from 7 of these 11 jurisdictions were usable and sufficiently complete
(<10% missing): New York, New York; San Jose/Santa Clara County, California;
Columbus/Franklin County, Ohio; Denver, Colorado (Denver, Adams, Arapahoe,
Boulder, Broomfield, Douglas, and Jefferson counties); Tampa/Hillsborough
County, Florida; Phoenix/Maricopa County, Arizona; and Lansing/Ingham County,
Michigan.

We estimated data missing because of nonresponse to an item in the dataset
(17) (ie,
missing 1 or more data elements) from these CoCs by using single imputation
techniques and SOLAS version 3.2 (Statistical Solutions, Saugus, Massachusetts). Some users of homeless services were not included in HMIS
data because of the service providers’ lack of participation; this unit nonresponse was addressed by applying a variation of the extrapolation
procedures used in the Annual Homeless Assessment Report (AHAR) to estimate
additional homeless people (veterans and nonveterans) who used homeless services
but were not recorded doing so (18). Extrapolation procedures produce reliable
estimates when more than half of providers in a CoC participate in HMIS (ie,
<50% data missing because of unit nonresponse); all CoCs included in this study
were well above this threshold. Extrapolation increased our homeless sample by
20,964 people (2,455 veterans, 18,509 nonveterans), a 16% increase over the
original sample.

To compute rates of homelessness, we used ACS data to estimate the total
veteran and nonveteran populations in each CoC. The ACS is an annual survey
administered by the US Census Bureau that collects social, economic, and
demographic information from samples of housing units in all counties in the
United States (19). We selected 3-year estimates (2006-2008) for this study
because they are based on a larger sample size than the 1-year estimates and
offer better precision, especially in examining smaller populations such as
veterans, and smaller geographic areas. When CoC boundaries varied from the
geographic areas for which ACS estimates are publicly available, the US Census
Bureau provided customized ACS estimates for these CoCs. For each geographic
area, we aggregated ACS data by age, sex, race, veteran status, and poverty
status.

This study received approval as an exempt study from institutional review
boards at the University of Pennsylvania and the US Department of Veterans
Affairs (VA).

Outcomes

Homelessness status was our outcome of interest. Data collected through HMIS
for the homeless population included age (18-29, 30-44, 45-54, 55-64, >65 y),
race (black, nonblack), sex, and veteran status (veteran, nonveteran). ACS
variables included in this study were age, race, sex, and veteran status in
categories consistent with HMIS data. In addition, ACS data were stratified by
poverty status, that is, whether household income was below the federal poverty
threshold. All people in the HMIS database were considered as living in poverty
on the basis of their homeless status. ACS, which collects data from group
quarters in addition to private housing units, included both homeless and housed
members of the population but did not differentiate the population on this
basis. Veteran status is defined as having served in the US military and is
based on self-report in both HMIS and ACS data.

Data analysis

Two phases of analysis used pooled data from the 7 CoCs. All analyses were
weighted by CoC size, were conducted separately for men and women as well as for
the total population and for the population living in poverty (from the ACS),
and were conducted using the R language and environment for statistical
computing, version 2.13 (R Foundation for Statistical Computing, Vienna,
Austria) (20).

In the first phase, we estimated the prevalence of veterans in the homeless,
poverty, and overall populations and calculated corresponding risk ratios (RR). This
process provided a simple measure of whether veterans were overrepresented in the
homeless population. We computed prevalence and risk ratios for each age, race,
and sex subgroup. Risk ratios for men and women were age- and race-adjusted.

In the second phase, we conducted binomial generalized estimating equation
(GEE) analyses in which homeless status was the outcome, and age, race, and
veteran status were potential predictors. Because we were modeling frequencies,
the outcome was a ratio of homeless (from HMIS data) to total (general or
poverty population from ACS data) people for each subpopulation, as defined by
the frequency within each subgroup, weighted by that same frequency
(21). GEE
modeling adjusted for dependence because of clustering within individual CoCs.
The phase 2 analysis consisted of main-effects-only multivariate models. Three
interaction effects were selected a priori and tested but were later
discarded because they were found to be nonsignificant: veteran status by 1)
age, 2) race, and 3) age by race.

Top of Page

Results

Phase 1 results

An estimated 130,554 adults received homelessness services in the 7 CoCs in
this study; 10,726 of these adults (8.2%) reported veteran status
(Table 1).
This rate was higher than the veteran rate among the ACS poverty (n = 63,655,
3.34%) and ACS general (n = 1,023,515, 6.96%) populations. Compared with nonveterans, veterans in each population (HMIS, ACS poverty, ACS general) were
disproportionately male and in the older age categories.

Veterans were overrepresented in the homeless population for both sexes
(Table 2). For men, 13.6% of the homeless adults were veterans,
whereas for women
1.8% of homeless adults were veterans. These rates yielded age- and
race-adjusted RRs of 2.1 (men) and 3.0 (women) compared with the population
living in poverty, and 1.3 (men) and 2.1 (women) compared with the general
population. RRs for demographic subgroups were generally consistent with the
overall RRs.

The age- and race-adjusted RRs for homelessness among both men and women were
higher for veterans than for nonveterans in both the poverty (RR, 2.2 for men
and 3.0 for women) and general populations (RR, 1.4 for men and 2.3 for women)
(Table 3). Rates of homelessness were consistently higher in veteran populations
than in nonveteran populations, and among both veterans and nonveterans, black
adults, especially in the younger age groups, had higher rates of homelessness.

Phase 2 results

Veteran status, older age, and black race were significantly and
independently associated with risk of homelessness among both men and women.
Similarly, the patterns of results found in the general population were
consistent with those found in the poverty population; however, in the latter,
veteran status was associated with a greater risk for homelessness.

For the veteran indicator, male veterans were almost 50% as likely (adjusted
odds ratio [AOR], 1.47; 95% confidence interval [CI], 1.19-1.81) and female
veterans were almost twice as likely (AOR, 1.97; 95% CI, 1.25-3.12) to be
homeless as nonveterans in the general population. Among the population in
poverty, male veterans were more than twice as likely (AOR, 2.20; 95% CI,
1.96-2.48) and female veterans were more than 3 times as likely (AOR, 3.33; 95%
CI, 2.17-5.13) to be homeless as nonveterans.

Among the control variables, increased age was significantly associated with
homelessness, but its effect differed between sexes. Among men, risk for
homelessness generally increased as a function of age up to the 45- to
54-year-old age range, but declined thereafter. This was so among both veterans
and nonveterans and in both the general and poverty populations. Men in the 45-
to 54-year-old age group appeared to be at the highest risk of homelessness,
nearly twice as likely (AOR, 1.85; 95% CI, 1.18-1.93) in the general population
and 3 times as likely (AOR, 2.65; 95% CI, 1.41-4.97) in the poverty population
as their 18- to 29-year-old counterparts to be homeless. Male veterans in the 45- to
54-year-old age group made up 41% of the homeless veterans. Risk for
homelessness among women declined with age at an increasing rate in both the
general and poverty populations, so that older women were at the lowest risk for
homelessness, compared with the youngest group.

Finally, black race was also a significant predictor of homelessness among
all subgroups. In the general population, the risk associated with black race
increased more than 5-fold for both men and women (AOR, 5.49; 95% CI, 4.25-7.09
for men and AOR, 5.45; 95% CI, 4.23-7.01 for women). This risk was lower in the
poverty population but remained high; the AOR for men was 2.18 (95% CI,
1.95-2.45) and for women was 3.32 (95% CI, 2.16-5.11).

Top of Page

Discussion

The findings in this report support those of earlier studies that showed
veterans to be overrepresented in the homeless population and reach beyond by
showing veteran status to be associated with increased risk for homelessness
after controlling for race, sex, and age. The magnitude of this association
became greater after controlling for poverty; veteran status was associated with
more than a 2-fold increase for men and a 3-fold increase for women in the odds
of becoming homeless.

For male veterans, those in the 45- to 54-year-old age group made up 41% of
the homeless veterans and also had the highest risk for becoming homeless. This
finding is consistent with other research (2) that identified a cohort effect in
this age group of veterans. This cohort, whose key characteristic was service
during the initial years of the All Volunteer Force, instituted in 1973, has
continuously been the veteran age group at highest risk for homelessness as
these veterans have aged over the last 2 decades. Similarly, members of the
general population who are now aged 45 to 54 have continuously been at highest
risk for homelessness (21-23).

Veterans make up a discrete subgroup in this general age cohort, in terms of
both the increased risk for homelessness associated with their veteran status
and their access to health care and homeless services through the VA. The
susceptibility of homeless people to chronic disease and disability increases as
they age, and the veterans among them will increasingly turn to the VA for
health care. Given their lack of housing and heightened susceptibility to
chronic health problems, homeless veterans will likely contribute
disproportionately to the increased demand for long-term care through the VA
(24). But beyond that, the changing health and need for housing support services
of an aging homeless population are poorly understood. As the VA responds to an
aging veteran population through increased reliance on community-based care to
treat chronic illness (25), those with the most tenuous ties to the community
will be the ones who present the most pressing challenges.

Among women, particularly black women, the youngest age groups were at
highest risk for homelessness. This finding is consistent with media accounts
that women who served in more recent conflicts such as those in Iraq and
Afghanistan are more likely than older female veterans to be homeless
(26). This
finding is also consistent with other research indicating that among women in
general, the period of highest vulnerability for homelessness is during the time
period when they are heading families with young children (27). Because younger
cohorts are most at risk, female veterans stand to benefit more from existing
homelessness-prevention efforts tied to reentering civilian life, which focus on
housing needs, than from efforts that combine housing with health care services.

Veterans who are living in poverty are more vulnerable to homelessness, an
effect that is magnified by black race. For example, for the youngest age group
living in poverty, more than
50% of black male veterans and more than 30% of black female veterans were
homeless (compared with only 7% for nonblack males and 12% for nonblack females),
according to HMIS data. These alarmingly high rates suggest that
homelessness-prevention activities—including tenant/landlord mediation or
short-term rent and utility payments—among veterans may be particularly
effective because they can target a finite poverty population and can further
refine this effort by focusing on black veterans. Our findings highlight the
usefulness of these data for such targeting, but future investigations of risk
factors must go beyond the simple focus on race and poverty status. The addition
of health-related data to the datasets used here could make specific links
between health conditions and risk for homelessness. The VA is currently
building a registry of veterans using homelessness services that can be linked
to VA health care records, which promises such assessments of health-related
risks for homelessness and for which this study could be a prototype.

Although the 7 CoCs included in our study represented approximately 10% of
the US homeless population, they are a convenience sample of urban
jurisdictions, which limits our study’s comparability to other studies. This
difference likely contributed to the divergence in a key finding between this
study and the Veteran Supplement to the Annual Homelessness Assessment Report
(15). Whereas this study demonstrated that male veterans were overrepresented
among the homeless population (RR, 1.3), the Vet-AHAR found them to be
underrepresented (RR, 0.7). This disparity is explained in part by the
differences in geographic areas, as the Vet-AHAR was a nationally representative
estimate. Further explanation for this difference in findings is the Vet-AHAR’s
inability to adjust its risk assessments by age and race.

Another limitation of our study is that the veteran status was based on
self-report and likely included people who reported veteran status but may have
been ineligible for VA services. Conversely, we may have included people
eligible for VA services who did not acknowledge veteran status. The HMIS data
are also limited in their universally available data fields, and a more
comprehensive range of data fields would go further toward understanding and
eliminating homelessness.

In conclusion, this study offers evidence that supports and expands on prior
findings that veterans, particularly older veterans, are vulnerable to
homelessness. As more and richer data on veteran homelessness, and homelessness
in general, become available through HMIS and other administrative sources,
future research should be able to increasingly relate health data to the
demographic characteristics included in this study.

Top of Page

Acknowledgments

This study was conducted by the US Department of Veterans Affairs National
Center on Homelessness among Veterans. We acknowledge the generous collaboration
of the following Continuum of Care organizations: New York City; San Jose,
California; Columbus, Ohio; Denver, Colorado; Tampa, Florida; Phoenix, Arizona;
Lansing, Michigan; and Fall River/Attleboro, Massachusetts.

Top of Page

Author Information

Corresponding Author: Jamison Fargo, PhD, MS, National Center on
Homelessness Among Veterans and Utah State University Department of Psychology,
2810 Old Main Hill St, Logan, UT 84322-2810. Telephone: 435-881-8797. E-mail:
jamison.fargo@usu.edu.

Author Affiliations: Stephen Metraux, National Center on Homelessness Among
Veterans and University of the Sciences, Philadelphia, Pennsylvania; Thomas Byrne, Ann
Elizabeth Montgomery, Harlan Jones, Dennis Culhane, National Center on
Homelessness Among Veterans and University of Pennsylvania, Philadelphia,
Pennsylvania; Ellen Munley, National Center on Homelessness Among Veterans and
The City University of New York, Philadelphia, Pennsylvania; George Sheldon, US
Department of Veterans Affairs, Washington, DC; Vincent Kane, National Center on
Homelessness Among Veterans, Philadelphia, Pennsylvania.

Top of Page

References

1. Rosenheck R, Bassuk E, Salomon A. Special populations of homeless
   veterans. In: Fosburg LB, Dennis DL, editors. Practical lessons: the 1998
   Symposium on Homelessness Research. Washington (DC): US Department of
   Housing and Urban Development; 1998.
2. Rosenheck R, Frisman L, Chung AM.
   The proportion of veterans among
   homeless men. Am J Public Health 1994;84(3):466-69.
3. Gamache G, Rosenheck R, Tessler R.
   The proportion of veterans among
   homeless men: a decade later. Soc Psychiatry and Psychiatr Epidemiol 2001;36:481-5.
4. Tessler R, Rosenheck RA, Gamache G.
   Comparison of homeless veterans with
   other homeless men in a large clinical outreach program. Psychiat Quart
   2002;73(2):109-19.
5. Robertson M. Homeless veterans: an emerging problem? In Bingham RD, Green
   RE, White SB, editors. The Homeless in Contemporary Society. Beverly
   Hills (CA): Sage; 1987.
6. Richardson C, Waldrop J. Veterans: 2000. Census 2000 Brief. Washington
   (DC): United States Census Bureau; 2003.
7. VetPop2007. Washington (DC): National Center for Veterans’ Analysis and
   Statistics; 2008. http://www.va.gov/vetdata/Veteran_Population.asp.
   Accessed November 29, 2011.
8. Office of Policy and Planning. Women veterans: past, present & future.
   Washington (DC): United States Department of Veterans Affairs; 2007.
9. Gamache G, Rosenheck R, Tessler R.
   Overrepresentation of women veterans
   among homeless women. Am J Public Health 2003;93(7):1132-6.
10. Washington DL, Yano EM, McGuire J, Hines V, Lee M, Gelberg L.
    Risk factors
    for homelessness among women veterans. J Health Care Poor Underserved 2010;21(1):82-91.
11. Wolitski RJ, Kidder DP, Fenton FA.
    HIV, homelessness, and public health:
    critical issues and a call for increased action. AIDS Behav 2007;11(6 Supp1):167-71.
12. Zerger S. A preliminary review of literature: chronic medical illness and
    homelessness. Nashville (TN): National Health Care for the Homeless Council;
    2002.
13. Babatsikou FP. Homelessness: a high-risk group for the public health.
    Health Sci J 2010;4:66-7.
14. Institute of Medicine. Homelessness, Health and Human Needs. Washington
    (DC): National Academy Press; 1988.
15. The 2009 annual homeless assessment report to Congress, veteran
    supplement. Washington (DC): United States Department of Housing and Urban
    Development and Department of Veterans Affairs; 2011.
16. McGah J, Sokol B, Spellman B, Sullivan N. HMIS Project Management Topics
    and Tools. Boston (MA): McCormick Graduate School of Policy Studies,
    University of Massachusetts, Boston; 2004.
17. Graham JW.
    Missing data analysis: making it work in the real world.
    Annu Rev Psychol 2009;60:549-76.
18. The 2009 annual homeless assessment report to Congress. Washington (DC):
    US Department of Housing and Urban Development; 2010.
19. Mather M, Rivers KL, Jacobsen LA. The American Community Survey.
    Population bulletin (of the Population Reference Bureau) 2005;60(3):3-20.
20. Venables WN, Ripley BD. Modern applied statistics with S. New York:
    Springer. p.190.
21. Hahn JA, Kushel MB, Bangsberg DR, Riley E, Moss AR.
    BRIEF REPORT: The aging of the
    homeless population: fourteen-year trends in San Francisco. J Gen Intern Med
    2006;21(7):775–8.
22. Sermons MW, Henry M. Demographics of homelessness series: the rising
    elderly population. Washington (DC): National Alliance to End Homelessness;
    2010.
23. Culhane, DP, Metraux S, Bainbridge J. The age structure of contemporary
    homelessness: risk period or cohort effect? University of Pennsylvania
    School of Social Policy and Practice Working Paper; 2010.
24. Kinosian B, Stallard E, Wieland D.
    Projected use of long-term-care
    services by enrolled veterans. Gerontologist 2007;47(3):356-64.
25. Malphurs FL, Striano JA.
    Gaze into the long-term crystal ball: the
    Veterans Health Administration and aging. J Gerontol A Biol Sci Med 2001;56(11):M666-73.
26. Eckholm E. Surge seen in number of homeless veterans. New York Times,
    November 8 2007:A22.
27. Culhane DP, Metraux S. Assessing relative risk for homeless shelter usage
    in New York City and Philadelphia. Population ResPolicy Rev
    1999;18(3):219-36.

Top of Page

Table 1. Demographic and Geographic Characteristics of Homeless People in Selected US Metropolitan Areas^a

Abbreviations: HMIS, Homeless Management Information System; ACS, American Community Survey; CoC, Continuum of Care.
^a Source: CoC data are collected for geographic units established by the US Department of Housing and Urban Development to track resource use for homeless populations.
^b People within a CoC who used homelessness services, according to HMIS 2008.
^c Adults identified by the ACS 2006-2008 whose incomes fell below the federal poverty threshold.
^d ACS, 2006-2008.

Table 2. Prevalence and Risk of Veteran Status in Homeless, Poverty, and Overall Populations in 7 US Metropolitan Areas^a

Abbreviations: M, male; F, female; RR, risk ratio.
^a Continuum of Care (CoC) data are collected for geographic units established by the US Department of Housing and Urban Development to track resource use for homeless populations. The 7 CoC metropolitan areas included in this analysis are Columbus, Ohio; Denver, Colorado; Lansing, Michigan; New York City; Phoenix, Arizona; San Jose, California; and Tampa, Florida.
^b People within a CoC who used homelessness services, according to Homeless Management Information System 2008.
^c Adults whose incomes fell below the federal poverty threshold, according to the American Community Survey (ACS) 2006-2008.
^d Prevalence of veterans in homeless population divided by prevalence of veterans in poverty population.
^e ACS 2006-2008.
^f Prevalence of veterans in homeless population divided by prevalence of adults in general population.
^g Risk ratios are age-adjusted.
^h Risk ratios are both age- and race-adjusted.

Table 3. Prevalence and Risk of Homelessness^a Among Veterans and Nonveterans in Poverty and General Populations in 7 US Metropolitan Areas^b

Abbreviations: M, male; F, female; RR, risk ratio.
^a People within a Continuum of Care (CoC) who used homelessness services, according to Homeless Management Information System 2008.
^b CoC data are collected for geographic units established by the US Department of Housing and Urban Development to track resource use for homeless populations. The 7 CoC metropolitan areas included in this analysis are Columbus, Ohio; Denver, Colorado; Lansing, Michigan; New York, New York; Phoenix, Arizona; San Jose, California; and Tampa, Florida.
^c People whose incomes fell below the federal poverty threshold, according to the American Community Survey (ACS) 2006-2008.
^d Prevalence of homelessness among veterans divided by prevalence of homelessness among nonveterans in poverty population.
^e ACS 2006-2008.
^f Prevalence of homelessness among veterans divided by prevalence of homelessness among nonveterans in the general population.
^g Risk ratios are age-adjusted.
^h Risk ratios are both age- and race-adjusted.