SOCIAL SECURITY ACT (1935)

696

Predictors of Homelessness among Older Adults in New York City

Disability, Economic, Human and Social Capital and Stressful Events

MARYBETH SHINN, JAMIE GOTTLIEB, JESSICA L. WETT, & AJAY BAHL New York University, USA ARNOLD COHEN & DEBORAH BARON ELLIS The Partnership for the Homeless, USA

Abstract

We interviewed 61 housed and 79 homeless adults aged 55 and over about disability; economic, human and social capital; and stressful life events prior to becoming homeless. Over half of the homeless group had previously led conventional lives. Human capital, social capital and life events were more important than disability or economic capital in predicting homelessness. The homeless adults were younger, more likely to be male and better educated than housed adults, but had shorter job tenure and fewer social ties. Homeless adults faced multiple, cascading risks, including job loss and housing loss. Implications for prevention are discussed.

Journal of Health Psychology Copyright © 2007 SAGE Publications Los Angeles, London, New Delhi and Singapore www.sagepublications.com Vol 12(5) 696–708 DOI: 10.1177/1359105307080581

AC K N OW L E D G E M E N T S . We thank the Partnership for the Homeless for funding interview incentives, the Jacob A. Riis Settlement House for furnishing the comparison group, Marcia Liu for data entry and interviewers. We are especially grateful to respondents, both homeless and housed. This article is based in part on an undergraduate honors thesis by Jamie Gottlieb.

C O M P E T I N G I N T E R E S T S : None declared.

A D D R E S S . Correspondence should be directed to: MARYBETH SHINN, New York University, 715 Broadway, Room 201, New York, NY 10003, USA. [email: [email protected]]

Keywords

■ aging ■ disability ■ homelessness ■ life events ■ social capital

UNTIL RECENTLY, homelessness among older adults in the United States seemed to be vanquished. In 1973 a book on homeless adults in New York was entitled Old men drunk and sober (Bahr & Caplow, 1973), but by the 1980s attention shifted to ‘the new homeless’: young minority men and families (e.g. New York Commission on the Homeless, 1992). In 1990, adults aged 50–61 used shelter at less than a third the rate, and adults aged 62 and over less than a 15th the rate, of adults aged 18–39 (0.41%, 0.09% and 1.40–1.45% of the population respectively, Culhane & Metraux, 1999). But homelessness among older adults is again on the rise. This study asks why, and what to do about it.

The reduction in poverty among adults aged 65 and older in the United States from 35 percent in 1960 to 10 percent in 1995 is widely hailed as an accomplish- ment of Social Security. However, although older people continued to make progress relative to the poverty line until the mid-1990s, progress relative to median non-elderly income stagnated in the 1980s (Engelhardt & Gruber, 2004). Further, the poverty rate for people aged 65 and older is nearly twice as high in New York City as in the nation as a whole (17.7% vs 9.4% in 2004, US Census Bureau Factfinder, n.d.). Perhaps as a consequence, the age of homeless New Yorkers is creeping up again. Single adults in New York’s shelter system were an average of five years older in 2002 than in 1988. Those over 40 made up 53 percent of the total in 2002, compared to less than 30 percent in 1988 (New York City Department of Homeless Services, n.d.). By 2005, 13 percent of residents of single adult shelters were 55 and over (M. Schretzman, Associate Commissioner, NYC Department of Homeless Services, personal communication, October 2005).

It is natural that homeless adults should age along with the overall population, but why might they be aging faster? One possibility is that individuals once dubbed the ‘new homeless’ remained homeless as they grew older. Cohen suggests that ‘personal risk factors [for homelessness] may accumulate over a lifetime’ and enculturation to street or shelter may prolong homelessness, although systemic and programmatic factors also matter (2004, p. 425). He reports that older homeless men ‘commonly have long histories of homelessness’ whereas homelessness among older women is more often caused by a crisis (2004, p. 428). However, studies of shelter records indicate that relatively few people are chronically homeless; most exit from this state (Culhane, Dejowski, Ibanez, Needham, & Macchia, 1994).

Another possibility is that as incomes stagnate and housing costs rise, adverse events may lead older adults to become homeless for the first time late in life. Indeed, in a cross-national study in England, Australia and the United States, Crane et al. (2005) found that of older adults who became homeless within the last two years, two-thirds (four-fifths in the United States) had never been homeless before. However, by definition, this study did not include adults with long, continuous histories of homelessness.

The current study has two goals. First, we examine risk factors for homelessness by comparing homeless and housed but poor adults over the age of 55. Based on prior research with homeless populations of differ- ent ages, we hypothesize that five classes of factors would differentiate homeless adults from their housed counterparts. These include disability, economic, human and social capital and stressful events in the period leading up to homelessness. In order to under- stand potential causes of homelessness, we tied our assessment to the last year that study participants spent in conventional housing, that is an apartment or a house. Second, we use narrative descriptions of respondents’ lives to understand the extent to which homeless older adults always had tenuous ties to housing or led relatively conventional lives before becoming homeless in old age.

Disability

Numerous studies have found elevated levels of physical health problems, mental illness and substance abuse among homeless single adults both in the United States (Burt et al., 1999; Koegel, Burnam, & Baumohl, 1996) and in Europe (Firdion & Marpsat, 2007; Muñoz, Crespo, & Pérez-Santos, 2005; Philippot, Lecocq, Sempoux, Nachtergael, & Galand, 2007). Gelberg, Linn and Mayer-Oakes (1990) found more chronic disease and functional disability among home- less individuals over 50 than among younger homeless people. They and others have concluded that in terms of health, homeless people over 50 resemble the gen- eral population over 65 (Cohen, 1999; Gelberg et al., 1990). Substance use and mental illness accounted for 69 percent of hospitalizations among homeless adults in New York City, from 2001–3 compared with 10 percent in the general population (Kerker et al., 2005), although it is important to realize that a single person can account for multiple hospitalizations. We expected that high levels of disability would also predate homelessness.

SHINN ET AL.: PREDICTORS OF HOMELESSNESS AMONG OLDER ADULTS

697

Economic capital

Homeless individuals live in deep poverty, with aver- age incomes in a national survey of only $367 per month in 1996 (Burt et al., 1999). We hypothesized that poverty would predate homelessness. In particu- lar we expected that low levels of income (in the job participants held for the longest period) and high levels of economic stressors (in the last year in conventional housing) would predict homelessness. In the United States, a home is the primary economic asset for middle-class adults, but New York is a city of renters, and poor adults are unlikely to own homes. Thus we examined title to housing, defined as owning a home or having one’s name on the lease rather than doubling up with others. We also examined housing quality. The protective effect of a housing subsidy— another form of economic capital—could not be exam- ined, because comparison group members all had subsidized housing.

Human capital

Human capital refers not to economic assets but to the ability to earn them. We considered educational attain- ment and work history as measures of human capital, and predicted that low levels of both would predict homelessness. Caton et al. (2000) found low educa- tional levels to be a risk factor for homelessness among men in New York.

Social capital

Social capital refers to the social and organizational ties on which individuals can draw for assistance. Many studies have found that homeless individuals and families lack social supports, or wear out their welcome with relatives and friends before becoming homeless, although findings are not uniform (Shinn, Knickman, & Weitzman, 1991). We assessed the extent to which children, or other relatives and friends, would serve as housing resources, and also examined respondents’ participation in community-based orga- nizations. We hypothesized that social capital would be negatively related to homelessness. A number of studies have found that disruptive childhood experi- ences, such as abuse or being in foster care, are asso- ciated with homelessness (Herman, Susser, Struening, & Link, 1997; Shinn et al., 1991). We assessed such experiences as negative indicators of social capital but posed no hypothesis, because it was unclear whether

childhood disruptions would have enduring conse- quences for older adults.

Stressful life events

Crane et al. (2005) describe events or transitions that may serve as ‘triggers’ for homelessness in older adults such as widowhood, marital breakdown, stop- ping work, evictions and onset or increased severity of mental illness. (We included the last under disability.) We hypothesized that stressful life events would be associated with homelessness. However, life transi- tions may also be common for older adults who remain securely housed. Here, as for all measures, our strategy was comparative. We asked not simply about the levels of disability, capital and stressful events among homeless adults during their last period of stable housing, but about the relative levels in homeless and housed but poor adults, and the extent to which each factor predicted homelessness in the context of the others.

Method

Participants Participants were 79 homeless and 61 housed adults aged 55 and older. Homeless adults were recruited from Peter’s Place, the only drop-in center in New York City dedicated to serving adults 55 and older. Drop-in centers are low-demand settings that provide food, social, medical and housing services to home- less individuals coming off the street. They are open 24 hours a day, seven days a week. Some participants go to informal night shelters in churches and return to the drop-in center during the day; others remain on chairs in the drop-in center at night or return intermit- tently to the street. Peter’s Place often serves older adults wary of the city’s mixed-age shelter system. Housed respondents were recruited from a settlement house serving a public housing project in New York City, ensuring that all were low income. Based on directors’ estimates of attendance at the two agencies during the interview period, response rates were approximately 82 percent for the homeless adults and 68 percent for the housed adults.

Procedure Interviewers (undergraduate and graduate psychology students who received extensive training) visited the drop-in center and settlement house repeatedly over a period of several months, becoming a familiar

JOURNAL OF HEALTH PSYCHOLOGY 12(5)

698

presence. Respondents, who were paid $20, could sign up for interviews, or were solicited informally by interviewers. After giving informed consent, partici- pants were interviewed in English or Spanish in private spaces for about an hour.

Two measures were used to assure respondents’ competence to provide data. First, we included a mea- sure of cognitive competence (Chestnut Health Systems, n.d.), however, individuals with failing scores were primarily non-native speakers of English and, in the interviewers’ judgments, difficulties had to do with language rather than memory. Interviewers also rated the coherence and consistency of the interview. Three interviews with homeless respondents, one rated as having ‘serious problems of coherence or consistency’, one terminated by the interviewer when the respondent seemed confused and one broken off by an agitated respondent were excluded from analysis.

Measures Most measures were tied to a ‘target year’, that is the most recent 12-month period in which the respondent had lived continuously in conventional housing with- out a move, in order to understand how events and conditions in this year may have precipitated home- lessness. The interviewer obtained a housing history to identify the last residence that qualified, and asked several questions about its location, who else lived there and when and why the respondent left in order to fix the location in the respondent’s mind. For 59 of 61 comparison respondents, but none of the homeless respondents, this target year was the 12 months imme- diately preceding the interview.

Measures of disabilities included physical disabil- ity, mental disability and substance use in the target year. Physical disability included reports that health problems affected ability to carry out any of five tasks (e.g. engage in moderate physical activity such as car- rying groceries or climbing stairs), or hospitalization for a medical problem. Mental disability included reports that a ‘mental or nervous problem’ affected ‘your ability to do the things you had to do’ or hospi- talization ‘for a nervous problem’. A substance prob- lem included reports of using marijuana or other drugs weekly, having any of four other symptoms of abuse of alcohol or drugs from the GAIN–Short Screener (Chestnut Health Systems, n.d., e.g. ‘did you try to hide that you were using alcohol, marijuana or other drugs’), or staying overnight in a detox facility.

Measures of economic capital included income for the final year at the longest job the respondent had held, and economic stressors, title to housing and

building problems during the target year. Income at the longest job was divided by the poverty threshold for the relevant year, to adjust for inflation. Economic stressors were assessed by an eight-item scale (based on Pearlin & Schooler, 1978) with high scores indicat- ing high levels of stressors (Cronbach’s alpha = .90). Items asked about inability to afford necessities (e.g. ‘the kind of food you should have’) and difficulties with finances. Because the items used different response scales, they were standardized before averag- ing, and the average was again standardized to make units meaningful. Title to housing assessed whether the respondent owned a residence or was named on a rental lease. A count of four serious building problems (e.g. lack of heat for a week or more in winter, rats; Shinn et al., 1998) indexed housing quality.

Measures of human capital included receipt of a high school (or equivalency) diploma and length of the longest job the respondent had ever held. Measures of social capital included a count of six dis- ruptive events in youth such as living in foster care, or being physically abused, reported by the respondent before age 18 (Shinn et al., 1991), and three adult measures: child housing resource indicated that the respondent had at least one child who would allow the respondent to stay with him or her. Relative/friend housing resource indicated that the respondent had a friend or a relative who would allow this. Organiza- tional ties were scored on a three-point scale where 0 indicated no organizational affiliations in the target year; 1 indicated attendance at a place of worship, community or senior center, or other club or regular meeting (excluding the agencies where we sampled respondents) and 2 indicated that someone would ask about a respondent who missed a meeting or did not go for a long time.

Stressful life events was a count of the number out of 11 events the respondent experienced during the target year. Events were related to housing (eviction, being told to leave), employment (job loss), relation- ships (divorce, ceasing to live with a partner, spouse or family member’s death or illness) and criminal victim- ization or involvement (self or family member arrested or jailed).

After the interview, the interviewer wrote a ‘thumbnail sketch’ of the respondent’s life and, for homeless respondents, the circumstances that led to housing loss. The interviewer also rated the coher- ence and consistency of the interview, and the extent to which homeless respondents had a ‘con- ventional life’ in terms of housing and employment prior to becoming homeless. To assure consistency

SHINN ET AL.: PREDICTORS OF HOMELESSNESS AMONG OLDER ADULTS

699

in the latter ratings, two experienced interviewers re-read the full set of thumbnail sketches for homeless respondents, and rated respondents again, focusing on whether the respondent had a stable lifestyle for a decade or more before becoming homeless. Agreement, corrected for chance (kappa) between the two sets of ratings was .75; disagreements were resolved by consensus.

Missing data Thirty-five respondents (25%) were unable to recall their income for the final year at their longest job, and we doubted the accuracy of additional reports. Thus we use this variable descriptively, but exclude it from regression analyses. No other variable was missing data for more than 3 percent of cases, and missing data were scattered. We used the Expectation Maximization method, SPSS version 14.0, to impute missing values for regression analyses (excluding income from the data used for imputation).

Results

Description of sample Table 1 shows the demographic characteristics of homeless and housed respondents as of the time of the interview. Housed respondents were approximately seven years older (and 13 years older during the ‘tar- get year’ in which both groups were in conventional housing, see Table 2). They were also much more likely to be female and Black, and marginally less likely to be foreign born. Few respondents in either group were currently married. Housed respondents were more likely to be widowed, and homeless respon- dents more likely never to have married.

Surprisingly, the homeless respondents were sub- stantially better educated than the housed comparison group. Just under half of the housed group had com- pleted high school, and only 13 percent had any post- high school education, whereas three-quarters of the homeless group had completed high school and

JOURNAL OF HEALTH PSYCHOLOGY 12(5)

700

Table 1. Descriptive characteristics of homeless and housed groups

Homeless group Housed group Test of difference (N = 79) (N = 61) (t or χ²)

Age, years, mean (SD) 63.6 (7.6) 70.5 (7.4) 6.88*** Female, % 19 77 49.74*** Race/ethnicity, % 22.26***

Blacka 41 75 Latino 13 13 White 33 8 Other 14 3

Foreign birth % 30 17 3.57t

Marital status, % 30.19*** Married 4 13 Separated 17 16 Divorced 23 15 Widowed 19 49 Single (never married) 37 7

Education, % 18.55** 8th grade or below 12 21 9th to 11th grade 14 30 Completed high school 31 36 Some college 27 8 College graduate 10 2 Post-graduate 6 3

Income/poverty line 4.1 (2.5) 2.6 (1.9) 3.51** (at end of longest job)b

tp < .10; *p < .05; **p < .01; ***p < .001 a Mostly African-American, but also African and Caribbean b Excluded from regression analyses due to 25 percent missing data

43 percent had some higher education. Reported jobs (including teacher, engineer, army officer and many business posts) were consistent with these educations. The homeless group also reported higher incomes at their longest job.

Prediction of homelessness We predicted homelessness from age (as of the target year), gender and measures of disability; economic, human and social capital; and stressful life events. Table 2 shows univariate relationships between variables in each domain and homelessness. Taken one at a time, only physical disability, building problems and disrup- tive events in youth failed to predict homelessness at p < .05.

Table 2 also shows the adjusted odds ratios and 95% confidence intervals for a parsimonious multivariate

model, arrived at by backwards elimination: non- significant predictors were removed, one at a time, from a full model with all the predictors in the table, until only variables that were related to homelessness at p < .05 remained. To check whether the order of removal mattered, each excluded variable was added back to this parsimonious model individually; none of the excluded variables was related to homelessness at p < .10.1 The adjusted odds ratio is the amount by which the odds of homelessness are multiplied for each unit increase in the predictor variable, controlling for other variables in the model. For dichotomous vari- ables (such as gender or high school education) it is simply the amount by which the odds are multiplied for women, or for high school graduates. For variables measured in years, such as age or length of longest job, it is the amount by which the odds are multiplied for

SHINN ET AL.: PREDICTORS OF HOMELESSNESS AMONG OLDER ADULTS

701

Table 2. Predictors of homelessness along with odds ratios for logistic regression model predicting homelessness from life history and conditions in last year in conventional housing (target year)

Adjusted odds ratio Homeless group Housed group (95% confidence

Predictor: mean (with SD) or proportion (N = 79) (N = 61) interval)b

Demographics Age during target year 57.6 (10.3) 70.5 (7.4)*** 0.84 (0.74, 0.94) Female .19 .77*** 0.01 (0.00, 0.12)

Disability in target year Physical disability .42 .57t

Mental disability .24 .08* Substance problem .34 .05***

Economic capital in target year Economic stressors (Z-score) 0.2 (1.1) -0.3 (0.7)** Housing title .69 .89** # Building problems (out of 4) 0.6 (0.9) 0.6 (0.9)

Human capital High school or GED .74 .49** 38.52 (2.29, 648.68) Length of longest job (years) 10.8 (8.2) 17.1 (10.3)*** 0.81 (0.71, 0.93)

Social capital # Disruptive events in youth 0.8 (1.1) 0.6 (1.1) Child housing resource .33 .85*** 0.08 (0.01, 0.59) Relative/friend housing resource .25 .79*** 0.03 (0.00, 0.28) Organizational ties, mean (0–2 scale) 1.0 (0.9) 1.7 (0.6)***

Stressful life events in target year # of events 1.0 (1.1) 0.4 (0.6)*** Apartment or job lossa .48 .02*** 31.02 (1.99, 483.68)

tp < .10; *p < .05; **p < .01; ***p < .001 in univariate analyses predicting homelessness a Apartment or job loss in the target year was substituted, post hoc, for full index of stressful life events. The substitution did not change the significance of other predictors b Odds ratios are from the parsimonious model including all variables where odds ratios are given. No other variable reached significance at p < .10 in the context of this basic set

each year—a five-year increase in age multiplied the odds of homelessness by (.84)5 or .42. If the confidence interval includes 1, the variable is not significant (mul- tiplying by 1 yields no change). Because of the rela- tively small sample size, only rather substantial effects reached statistical significance.

In terms of demographic variables, homeless indi- viduals were younger and more likely to be male than housed individuals, and both these variables remained significant in the context of all other variables.

No form of disability was a significant predictor of homelessness in the context of other variables. The adjusted odds ratios when each variable was added to the parsimonious model were 1.18 for physical dis- ability and 1.08 for mental disability, suggesting that these forms of disability were not important to home- lessness. However, the confidence intervals were quite broad: (0.13, 10.54) for physical disability and (0.12, 9.38) for mental disability, so that neither large increases in the odds of homelessness nor large decreases could be ruled out. In the case of substance problems the adjusted odds ratio was substantial (9.13), suggesting that substance abuse could well be important to homelessness in this age group, but the broad confidence interval (0.58, 145.05), made it impossible to specify this association. (Both sub- stance abuse and homelessness were correlated with younger age and male gender.) No indicator of eco- nomic capital contributed to the final model, although economic stressors and title to housing were signifi- cant taking one variable at a time. Of these, title to housing seemed more likely to be important, with an adjusted odds ratio (when added to the parsimonious model) of 7.98, but a very broad confidence interval (0.54, 118.05). Both groups experienced relatively high levels of economic stressors in the target year, e.g. 38 percent of homeless and 31 percent of housed adults reported not having enough money to afford the kind of food they should have at least once in a while. Although we could not examine the association of housing subsidies with homelessness, because the comparison group was recruited from subsidized public housing, only 24 percent of homeless respon- dents had received a housing subsidy in their last year in stable housing.

The two indicators of human capital had opposite relationships to homelessness. As already noted, educational attainment was positively associated with homelessness, but length of the longest job was negatively associated. Housed individuals had worked over six years longer, on average, at the longest job they had held, but the homeless group

also had substantial work histories, with tenure at the longest job averaging almost 11 years.

Two measures of social capital were also signifi- cant predictors of homelessness controlling for other variables. Housed respondents were much more likely than homeless respondents to report having a child or another relative or friend who would house them. Nevertheless, a third of homeless respondents said a child would allow them to stay, and a quarter said someone else would do so. Why, then, were they homeless? Respondents commonly reported contentious relationships with someone in the house- hold, not wanting to impose or wanting to remain independent (14–19 respondents each). Other network members lived far away, had not been in touch with the respondent for years, lived in crowded circum- stances or were in the military or institutional settings (5–9 respondents each). (Respondents could offer dif- ferent reasons for different network members.)

Although homeless individuals reported over twice as many stressful life events in their last year in con- ventional housing as did housed respondents, the overall index of life events did not contribute to the prediction of homelessness, controlling for other vari- ables. A post hoc examination of specific life events showed that homeless respondents were far more likely than housed respondents to report events relat- ing to loss of housing or jobs during their last year in conventional housing: 25 percent had been evicted, 8 percent had been asked to leave by someone they were staying with and 22 percent had lost a job. Altogether, 48 percent of homeless but only 2 percent of housed respondents had experienced one or more of these events in their last year in conventional hous- ing. The average number of other events reported by the two groups (0.43 for homeless, 0.42 for housed) was virtually identical. An indicator that the respon- dent had lost an apartment or a job in the year before becoming homeless, when substituted for the life events index in the logistic regression analysis, was highly significant, increasing the odds of homeless- ness by a factor of 31 (with broad confidence bounds). No other variable changed in significance as a result. The odds ratios and confidence intervals in Table 2 are from the equation using the indicator of apartment or job loss.

Conventional lives and qualitative analyses We coded 42 or 53 percent of the 79 homeless respondents as having conventional lives prior to becoming homeless. This designation did not mean

JOURNAL OF HEALTH PSYCHOLOGY 12(5)

702

that individuals had no problems, simply that they managed to keep conventional housing and jobs for extended periods of time prior to becoming home- less late in life.

Table 3 shows differences between the homeless respondents coded as having more conventional and less conventional lives. The two groups were nearly the same age, but those with less conventional lives were 15 years younger on average, when they first became homeless, were four times as likely to have had multiple bouts of homelessness and, based on the thumbnail sketches, often had tenuous ties to housing throughout adulthood. The less conventional group was twice as likely to have had a substance problem in the target year and twice as likely to have experi- enced disruptive events in childhood. The more con- ventional group was more likely to have graduated from high school, and had held their longest job for almost twice as long, on average. They were also more likely to report organizational ties and that a child would allow them to stay. Although the groups did not differ on overall stressful life events or the combined index of apartment and job loss, the less conventional group was more likely to have lost a job and the more conventional group more likely to have lost housing during the target year. The groups did not differ on any other variables in Tables 1 and 2.

In summary, the less conventional group fit the pro- file of individuals with long histories of homelessness or housing instability and accumulated risk; the more conventional group did not. Why, then, did the latter group become homeless in old age? Summaries of the thumbnail sketches for five more conventional respondents give qualitative answers.

José (not his real name), age 77, graduated from college in Cuba, and came to the United States, where he owned a furniture business and raised eight chil- dren, before retiring and selling the business at age 72. Although he thought that he could live on his pension, he was soon unable to afford the rent for his long-time apartment, and was evicted. José is close to his chil- dren, but says that they are enjoying their lives and he does not want to be a burden to them, or a ‘pain in the neck’.

Bill, age 74, graduated from high school and lived in the same apartment for 50 years until age 68, when he developed a crippling physical illness which pre- vented him from working or living alone. He thus lost the construction job he had held for 15 years and was unable to afford his rent. He stayed with a nephew for two years and a sister for one year; but became home- less after ‘using up’ these social resources.

James, age 68, came to New York from a southern state in the 1960s. He had some college education,

SHINN ET AL.: PREDICTORS OF HOMELESSNESS AMONG OLDER ADULTS

703

Table 3. Differences between homeless individuals with more and less conventional lives

Less conventional More conventional Test of difference Characteristic: mean (SD) or proportion (N = 37) (N = 42) (t or χ²)

History of homelessness Age at interview 62.5 (6.6) 64.7 (8.4) 1.30 Age first homeless 44.2 (15.0) 59.4 (10.2) 5.06*** Multiple bouts of homelessness .70 .17 22.62***

Disability in target year Substance problem .49 .21 6.56*

Human capital High School or GED .62 .85 5.59* Length of longest job (years) 7.4 (6.1) 13.7 (8.8) 3.66***

Social capital # Disruptive events in youth 1.2 (1.3) 0.5 (0.8) 2.49* Child housing resource .19 .45 6.36* Organizational ties, mean (0–2 scale) 0.8 (0.9) 1.2 (0.8) 2.16*

Stressful life events in target year Evicted or asked to leave .19 .45 6.36* Job loss .35 .10 7.89**

*p < .05; ** p< .01; *** p< .001 The groups did not differ on gender, race, foreign birth, marital status, income relative to poverty line, mental or physical health, economic stressors, title to housing, building problems, relative/friend housing resource or the full index of stressful events

and owned a grocery store, from which he retired at age 67. A year later, he lost his apartment in a fire. The City placed him in a single-room occupancy hotel, which he left because it was dirty and infested with roaches. Although he could stay with either of his two children, he wants to get back on his feet in a place of his own.

Bob, age 58, is a Vietnam veteran with some college education who was diagnosed with generalized anxi- ety disorder, bipolar disorder and post-traumatic stress disorder (PTSD), but nevertheless earned $70,000 a year in a management position for a bank. He became homeless at age 56 when the girlfriend with whom he had lived for seven years left. Two months later he left his job and lacked money to pay the rent. He had problems with both gambling and alcohol abuse, and reported stays in detox and in a hospital for both med- ical and nervous problems in that year.

Susan, age 86, became homeless at age 74 when she was evicted from the apartment where she had lived for 29 years for hoarding. She never married and had no children. She had some college education and had served in the military, done fundraising for a social service agency and worked as assistant public- ity director for a large arts organization, among other jobs. After an injury restricted her ability to work, she began to manage a thrift shop. She brought so many items home that it created a fire hazard. Susan has three elderly siblings who are in nursing homes or living with children and unable to help her.

These case studies suggest that the quantitative measures were sometimes too specific to capture respondents’ situations. Neither James, whose apart- ment burned down, nor Bob, who could not afford the rent after his girlfriend left, reported being evicted or asked to leave by someone they were stay- ing with. Other ‘conventional’ respondents reported losing their housing in ways our stressful event inventory did not capture (e.g. a flood, a death of the primary tenant, a shooting, a move that did not work out). No case-study respondent reported losing a job in the target year, although Bill and Susan lost jobs earlier, due to illness and injury, and Bob left work for unspecified reasons that may have been related to mental disability.

The case studies also show that it is typically the confluence of multiple risk factors or a cascade of events that make someone homeless, rather than just one. Susan was coded as having both a physical and a mental health problem in her last year in con- ventional housing, no child, friend or relative hous- ing resources and a relatively short period of five

years in her longest job (she held many jobs over the years). None the less, she remained housed until age 74, living in her last apartment for 29 years.

Based on the qualitative findings that homeless respondents experienced multiple risk factors, we did a final post hoc analysis. For each respondent we counted the number of 12 risk factors: physical dis- ability; mental disability; substance problem; eco- nomic stressors above the sample mean; lack of title to housing; longest job of less than 10 years; any dis- ruptive event in youth; lack of child housing resource; lack of relative/friend housing resource; lack of organizational ties; job loss in target year; and housing loss in target year. On average, homeless individuals had three more risk factors than did housed individuals (homeless M = 4.97, SD = 1.97; housed M = 2.05, SD = 1.51, t(138) = 9.60, p < .001). Among housed respondents, 67 percent had 0–2 risk factors and none had more than six. Among homeless respondents, only 10 percent had 0–2 risk factors, and 25 percent had 7–10. The only homeless respon- dent with no identifiable risk factors described giving everything up and taking to the street after his wife died, but his example shows the value of a compara- tive approach. As Table 1 shows, widowhood was far more common in the housed sample.

Discussion

Conventional lives Perhaps the most interesting finding to emerge from the study is that over half of the homeless respon- dents lived relatively conventional lives, typically involving long periods of employment and residen- tial stability before becoming homeless at an aver- age age of 59. Multiple events shifted people who were unsupported by family and society from these conventional lives into homelessness. Just under half of the homeless respondents had longer histo- ries of instability, more in line with earlier findings (e.g. Cohen, 2004). The dividing line between these groups is a fuzzy one—the slide into homelessness was often slow, with lives looking less conventional as time went on.

Predictors of homelessness The quantitative analyses isolated factors that differen- tiated the entire group of homeless adults from poor adults who remained housed. Key predictors were male gender, younger age, higher levels of education, shorter tenure in the longest job held, loss of an

JOURNAL OF HEALTH PSYCHOLOGY 12(5)

704

apartment or job while in conventional housing and lack of children or other ties who would provide hous- ing. Because confidence intervals were often broad, the study should not be interpreted as providing evi- dence against the contributions of other factors to homelessness. In particular, substance problems and title to housing may play important roles. The sam- pling design, in which all housed respondents were drawn from public housing, meant that the role of housing subsidies in protecting against homelessness could not be examined, despite their scarcity in the homeless group and importance to other populations in the same city (Shinn et al., 1998).

Demographic differences between groups were unsurprising. Studies of single homeless adults (e.g. Burt et al., 1999) typically find many more men than women whereas differential mortality leads to larger numbers of women than men among older adults generally. Very old adults may not be able to survive on the street (Gelberg et al., 1990). Adults who are too old to readily gain employment if they lose jobs but who are too young to be eligible for social secu- rity benefits may be at special risk as Okamoto (2007) also found in Japan. As in national studies (e.g. Burt et al., 1999), there were relatively more Black respondents in the homeless group than in New York City, but this was even truer of the comparison group.

More surprisingly, health and disability did not play a statistically significant role in predicting homelessness, although confidence intervals were broad, so that the data cannot rule out important asso- ciations, especially for substance abuse. Health may have deteriorated after individuals became homeless, consistent with other literature (e.g. Burt et al., 1999; Cohen, 1999; Firdion & Marpsat, 2007; Gelberg et al., 1990; Muñoz et al., 2005; Philippot et al., 2007). Also, the qualitative data suggest that disability sometimes precipitated other, more proximal causes of homelessness, and substance abuse was more common among the homeless adults with less con- ventional lives. Crane et al. (2005) also found that newly homeless older adults reported housing and relationship problems as more direct antecedents of homelessness than physical or mental health or sub- stance problems, which were sometimes ‘predispos- ing or contributory’.

Sample biases may have affected reported health. Adults with physical disabilities or dementia may be more likely to be in institutional settings, and those with substance problems or paranoia may be less likely to come into a drop-in center than to stay on the street. Interviewers and staff believed that poten-

tial homeless respondents who were not interviewed had more cognitive problems and mental illness than those who were, and the three homeless respondents whose interviews were not usable were more agitated or less coherent than those whose interviews were analyzed. On the other hand, Gelberg et al. (1990) found older homeless adults less likely than younger ones to have psychotic symptoms, and no more likely to have memory loss.

Economic capital also seemed relatively unimpor- tant. No predictors were significant, although having title to housing could not be ruled out as a protective factor. Levels of economic stressors were high for both groups, but may not have threatened homeless- ness for the comparison group whose public housing rents were tied to income.

The high levels of educational-level attainment among homeless respondents, and their rates of par- ticipation in college and post-graduate education, were surprising. Nor did the higher education levels of homeless adults with conventional than with unconventional lives protect them from housing loss. The other indicator of human capital, tenure in the longest job a respondent had held, favored the housed group, as expected. Even so, homeless respondents averaged 10.8 years and those with conventional lives averaged 13.7 years in their longest jobs. We did not assess the total number of years that respondents worked, but it is clear that many had a series of responsible and often well-paid jobs commensurate with their educational levels.

Social ties, especially ties to children who would allow the respondent to stay with them, were an important protective factor. Organizational ties (which were not explicitly tied to housing) were less important. The number of homeless individuals who declined opportunities to stay with children or relatives may suggest that they overestimated these resources. In some cases, such as Bill’s, respon- dents had stayed with others and had worn out their welcome. (Note that 11 percent of the housed sample did not have title to housing, but were dou- bled up with others who may have protected them from homelessness.)

It is also interesting that disruptive experiences in childhood, which have been robust predictors of homelessness in younger samples (e.g. Herman et al., 1997), were not important after controlling for other variables here. Such disruptive childhood experiences were relatively high among homeless adults with unconventional lives, suggesting that they may play an indirect role, by setting processes

SHINN ET AL.: PREDICTORS OF HOMELESSNESS AMONG OLDER ADULTS

705

in motion that lead to more proximal predictors of homelessness.

Stressful life events during the last year in conven- tional housing did not predict homelessness, but events that indicated loss of resources (eviction, being asked to leave, job loss) did. This post hoc selection of events that best differentiated the groups should be replicated. It is also possible that the low levels of such events in the comparison group is an artifact. Housed individuals had been living in their present apartment in public housing for a median of 26 years, and were largely retired (only 21 percent of the comparison group, compared to 53 percent of the homeless group had been employed since 2000, χ²(1, n = 140) = 15.2, p < .001). Thus they were unlikely to have lost jobs or housing in the past year. Nevertheless, it is plausible that events that affected access to housing resources and income would be particularly important to homelessness (see similar findings by Okamoto, 2007).

Limitations The study has several limitations. Ideally, we would compare a random sample of homeless individuals over the age of 55 with a random sample of poor adults of the same age. Because Peter’s Place is the only drop-in center for older adults in New York City, the homeless sample is not a bad one, but may still differ from samples drawn from institutions or the street. The comparison sample is more limited because all had access to subsidized housing and were recruited at a settlement house, so they were unlikely to be socially isolated.

Focusing the interview on the last year in stable housing (the target year) was both a strength and lim- itation of the study. Collecting information on respon- dents’ circumstances prior to homelessness (or the most recent instance of homelessness, for respondents with multiple bouts) justified considering these cir- cumstances as predictors rather than consequences of homelessness, and may account for differences between this study and others with respect to health. However, retrospective recall of past events may mag- nify biases inherent in self-report data and the longer time lag for homeless than for housed respondents may have led to differential recall in the two groups. Future research might have housed respondents recall a period three years in the past (the median time lag for homeless respondents). Focusing on a year in stable housing also minimized reporting of events incom- patible with such housing, such as imprisonment. In

addition to focusing on the last period in stable hous- ing, future research might inquire about earlier events.

Implications for prevention Despite these limitations, this study provides useful guidance for preventing homelessness among older adults, and challenges some assumptions that might be drawn from considering only the circumstances of people who are currently homeless. The analysis of health and disability suggests that efforts to provide more health services, however valuable on other grounds, may do little to prevent homelessness. Rather, the analysis of stressful life events suggests that efforts to prevent homelessness late in life should target those who lose jobs or housing for any reason. Cohen (1999) found that 80 percent of older homeless men wanted to be employed and 56 percent had been continually looking for work. Over half of the home- less respondents in our sample had recent work histo- ries. Age discrimination in employment is illegal in the United States, although laws are not always enforced. Older workers can face difficulties finding new work if they are laid off, or if illness or injury requires a period of unemployment or a change in activity. Providing jobs for adults aged 50–64, who do not qualify for entitlements available to older adults, could prevent some homelessness (Cohen, 1999).

Older adults and those with disabilities should be helped to apply for available income supports. However, housing costs put even adults who work full time at risk of homelessness, and place unsubsidized housing out of reach for people receiving disability benefits. The fair market rent for a one-bedroom apartment in New York City during the study was $1003 per month—more than the entire after-tax income of a full-time minimum wage worker (National Low Income Housing Coalition, 2005), and much more than such a person would receive in retire- ment. Supplemental Security Income (SSI) for dis- abled individuals was only $666 per month.

Thus, rent subsidies are important supports for older adults, but in our study, only one-quarter of respondents on the verge of homelessness received one. Both general subsidies (e.g. Section 8) and those targeted to older adults (e.g. Section 202) should be expanded. New York’s Senior Citizens Rent Increase Exemption Program, which exempts low-income senior citizens from increases in rent by giving landlords reductions in property taxes, is an entitlement, but should be publicized more broadly and extended to subsidize base rents (not

JOURNAL OF HEALTH PSYCHOLOGY 12(5)

706

just increases). New York City currently offers legal help to prevent eviction, but many older adults are unaware of their rights and do not access these programs. Additional inexpensive housing options, such as clean, safe single-room occupancy hotels for seniors, could also reduce homelessness.

Efforts might also focus on adults who lack family, especially children who would take them in. Social policy cannot change social ties, but it can provide in-home services to allow older people to remain independent, legal and other forms of advo- cacy with housing providers and access to benefits that might substitute for social resources.

Note

1. To check the robustness of this model, we added race (Black vs other) as a control; no variable changed significance at p<.05, but confidence intervals were broader. Race itself was not a sig- nificant predictor.

References

Bahr, H. M., & Caplow, T. (1973). Old men drunk and sober. New York: New York University Press.

Burt, M., Aron, L. Y., Douglas, T., Valente, J., Lee, E., & Iwen, B. (1999). Homelessness: Programs and the people they serve: Findings of the National Survey of Homeless Assistance Providers and Clients. Washington, DC: The Urban Institute.

Caton, C. L., Hasin, D., Shrout, P. E., Opler, L. A., Hirshfield, S., Dominguez, B., & Felix, A. (2000). Risk factors for homelessness among indigent urban adults with no history of psychotic illness: A case-control study. American Journal of Public Health, 90, 258–263.

Chestnut Health Systems. (n.d.). GAIN–Short Screener (GAIN-SS). www.chestnut.org/li/gain (accessed 15 February 2005).

Cohen, C. I. (1999). Aging and homelessness. The Gerontol- ogist, 39, 5–14.

Cohen, C. I. (2004). Older homeless persons. In Encyclopedia of homelessness (pp. 425–431). Thousand Oaks, CA: Sage Publications.

Crane, M., Byrne, K., Fu, R., Lipmann, B., Mirabelli, F., Rota-Bartelink, A. et al. (2005). The causes of home- lessness in later life: Findings from a 3-nation study. Journals of Gerontology, 60B, S152–S159.

Culhane, D. P., Dejowski, E. F., Ibanez, J., Needham, E., & Macchia, I. (1994). Public shelter admission rates in Philadelphia and New York City: The implications of turnover for sheltered population counts. Housing Policy Debate, 5, 107–140.

Culhane, D. P., & Metraux, S. (1999). One-year rates of public shelter utilization by race/ethnicity, age, sex and

poverty status for New York City (1990 and 1995) and Philadelphia (1995). Population Research and Policy Review, 18, 219–236.

Engelhardt, G. V., & Gruber, J. (2004). Social security and the evolution of elderly poverty. National Bureau of Economic Research Working Paper 10466. Available at http:// www.nber.org/papers/w10466 (accessed 17 July 2006).

Firdion, J.-M., & Marpsat, M. (2007). A research program on homelessness in France. Journal of Social Issues 63(3), 567–587.

Gelberg, L., Linn, L. S., & Mayer-Oakes, S. A. (1990). Differences in health status between older and younger homeless adults. Journal of the American Geriatrics Society, 38, 1220–1229.

Herman, D. B., Susser, E. S., Struening, E. L., & Link, B. L. (1997). Adverse childhood experiences: Are they risk factors for adult homelessness? American Journal of Public Health, 87, 249–255.

Kerker, B., Bainbridg, E. J., Li, W., Kennedy, J., Bennani, Y., Agerton, T. et al. (2005). The health of homeless adults in New York City: A report from the New York City Depart- ments of Health and Mental Hygiene and Homeless Services. Available at http://www.nyc.gov/html/doh/down- loads/pdf/epi/epi-homeless-200512.pdf (accessed 7 July 2006).

Koegel, P., Burnam, M. A., & Baumohl, J. (1996). The causes of homelessness. In J. Baumohl (Ed.), Homelessness in America (pp. 24–33). Phoenix, AZ: Oryx Press.

Muñoz, M., Crespo, M., & Pérez-Santos, E. (2005). Homelessness effects on men’s and women’s health. International Journal of Mental Health, 34, 47–61.

National Low Income Housing Coalition. (2005). Out of reach 2005. http://www.nlihc.org/oor2005/ (accessed 3 July 2006).

New York City Commission on the Homeless. (1992). The way home: A new direction in social policy. New York: New York City Commission on the Homeless.

New York City Department of Homeless Services. (n.d.). Emerging trends in client demographics. Available at http://www.nyc.gov/html/dhs/downloads/pdf/demo- graphic.pdf (accessed 3 July 2006).

Okamoto,Y. (2007). A comparative study of homelessness in the United Kingdom and Japan. Journal of Social Issues 63(3), 525–542.

Pearlin, L. I., & Schooler, C. (1978). The structure of cop- ing. Journal of Health and Social Behavior, 19, 2–21.

Philippot, P., Lecocq, C., Sempoux, F., Nachtergael, H., & Galand, B. (2007). Psychological research on homeless- ness in Western Europe: A review from 1970 to 2001. Journal of Social Issues 63(3), 483–503.

Shinn, M., Knickman, J. R., & Weitzman, B. C. (1991). Social relations and vulnerability to becoming homeless among poor families. American Psychologist, 46, 1180–1187.

Shinn, M., Weitzman, B. C., Stojanovic, D., Knickman, J. R., Jimenez, L., Duchon, L. et al. (1998). Predictors of homelessness among families in New York City: From

SHINN ET AL.: PREDICTORS OF HOMELESSNESS AMONG OLDER ADULTS

707

shelter request to housing stability. American Journal of Public Health, 88, 1651–1657.

US Census Bureau Factfinder. (n.d.). American commu- nity survey data for 2004. Available at http://factfinder.

census.gov/servlet/SAFFPeople?_submenuId=people_9 (accessed 17 July 2006).

JOURNAL OF HEALTH PSYCHOLOGY 12(5)

708

Author biographies

MARYBETH SHINN is Professor of Applied Psychology and Public Policy in the Steinhardt School of Culture, Education and Human Development and Wagner School of Public Service at New York University.

JAMIE GOTTLIEB is a first-year Distinguished Public Interest Scholar at Seton Hall School of Law. She graduated with honors from New York University in 2006 with a BA in Psychology.

JESSICA WETT received her BA in Psychology from New York University in 2006. She plans a career in social work.

AJAY BAHL is an aspiring psychiatrist who will graduate from NYU with a Psychology major in spring 2007. He hopes to pursue a Masters in Bioethics before attending medical school.

ARNOLD S. COHEN is President and CEO of the Partnership for the Homeless, which offers services, research and education to end homelessness. He worked previously as a public interest attorney.

DEBORAH BARON ELLIS, LCSW is Director of Older Adult Services at The Partnership for the Homeless. She is also director of Peter’s Place, a multi-service center for homeless older adults.