Determinants of Homelessness in Metropolitan Areas

DETERMINANTS OF HOMELESSNESS INMETROPOLITAN AREAS

BARRETT A. LEEPennsylvania State University

TOWNSAND PRICE-SPRATLENOhio State University

JAMES W. KANANWestern Kentucky University

ABSTRACT: Why do rates of homelessness differ across metropolitan contexts? Only tentative

answers to this question are provided by previous investigations due to their reliance on a limited

range of settings and independent variables. In an effort to be more comprehensive, we use 1990

Census S-Night data for 335 metropolitan areas to test explanations that stress the local

housing market, economic conditions, demographic composition, the safety net, climate, and

community transience as potential determinants of homelessness. Our analysis uncovers support

for each explanation. However, median rent level has the dominant effect (þ) on metro

homelessness rates, followed by the percentage of single-person households (þ). The robustness

of these findings is demonstrated and their theoretical, methodological, and policy implications

are considered.

Observers of contemporary homelessness in the US often stress the spatial ubiquity ofthe phenomenon, claiming that homeless individuals are present in small towns and ruralsettings as well as cities (First, Rife, & Toomey, 1994; Fitchen, 1992; Vissing, 1996). Whilethis assertion is true in a literal sense, available evidence leaves no doubt about the heavyurban concentration of homeless people. Data from the 1990 Census S-Night enumerationand from the 1995–1996 National Survey of Homeless Assistance Providers and Clients(or NSHAPC) suggest that approximately nine out of every 10 homeless Americans live inmetropolitan areas, with at least 70% in central cities (Burt, Lee, & Valente, 2001; Lee &Price-Spratlen, 2002). The urban character of homelessness reflects the geographic dis-tribution of the US population as a whole, including those at-risk groups (e.g., minorities,the poor, and single-person households) from which many homeless are drawn. Large

*Direct correspondence to: Barrett A. Lee, Department of Sociology, Pennsylvania State University, 211 OswaldTower, University Park, PA 16802. E-mail: [email protected]

JOURNAL OF URBAN AFFAIRS, Volume 25, Number 3, pages 335–355.

Copyright # 2003 Urban Affairs Association

All rights of reproduction in any form reserved.

ISSN: 0735-2166.

urban places also offer more shelters and services than do smaller communities, and theyare perceived to have better employment opportunities. Because of their advantages, suchplaces contain a disproportionate share of homeless people.

We take as our point of departure the fact that the magnitude of the homelessnessproblem varies from one city or metropolis to the next. Obvious differences exist based onthe sheer number of residents. New York, for example, would be expected to have morehomeless than New Haven, and the 1990 Census confirms this (34,000 in New York and1,400 in New Haven). To gauge the relative incidence of homelessness, however, astandardized measure is needed. One that is popular among investigators indicates thenumber of homeless per 10,000 residents at a particular time. Because this rate adjusts forpopulation size, it provides estimates of point prevalence that can be meaningfullycompared across urban units. Previous research relying on some version of the rate hasdocumented substantial variation for cities and metropolitan areas (Burt, 1992; Honig &Filer, 1993; U.S. Department of Housing and Urban Development, 1984). Moreover,the patterns observed align only imperfectly with size. To complete our earlier illus-tration, New Haven’s 1990 homelessness rate (73.4 per 10,000) exceeds New York’s(42.5) by over 50%.

The key question we address is why the scale of the problem differs so much acrossurban contexts. To answer it a structural perspective must be adopted in which home-lessness rates are treated as a function of other attributes of communities. This perspectivecontrasts with individual-level strategies for explaining homelessness. A considerableamount of work has examined the reasons that homeless people give for their situation(Burt et al., 2001; Redburn & Buss, 1986; Snow & Anderson, 1993) and the pathwaysthat they follow into and out of a homeless state (Culhane & Kuhn, 1998; Dworsky &Piliavin, 2000; Shinn et al., 1998; also see Wong’s 1997 review of earlier studies). Theprimary concern of such individual-level research is with proximal determinants, i.e., thoseimmediate events or circumstances that push certain members of the poverty populationbeyond the threshold of shelter security.

On intuitive grounds, a structural approach to homelessness lacks the appeal of anapproach emphasizing the experiences of individuals. It is not too difficult for most of usto see how bad health could lead to the loss of a job, ultimately forcing a person onto thestreets, or how domestic violence victims from a poor household could be left withinsufficient resources for establishing a separate (and permanent) residence of their own.Scenarios of this kind enjoy a concreteness and familiarity that make them easy tocomprehend. By comparison, structural antecedents may seem fairly distant or abstract.We maintain, however, that they are no less important, in part because they define theextent of vulnerability in a community. A strong version of the structural position, whichwe favor, holds that the roots of homelessness are deeply embedded in the local social,economic, and institutional fabric. Our position is congruent with theoretical models ofhomelessness that blend the individual and structural perspectives (Koegel, Burnam, &Baumohl, 1996; O’Flaherty, 1996; Rossi, 1989; Wolch, Dear, & Akita, 1988). Such modelsacknowledge the role of negative events in selecting particular individuals into a homelessstate but also note that the number who are at risk of selection depends upon aspects ofthe macro-level context.

In the interurban structural analyses conducted thus far, the contextual features receiv-ing the most attention include local job and housing market conditions, the representationof disadvantaged groups, and the adequacy of welfare benefits and mental health pro-grams (Bohanon, 1991; Burt, 1992; Elliott & Krivo, 1991; Honig & Filer, 1993; Tucker,1990). Because specific indicators differ by study, the results are a rather mixed bag. The

336 | JOURNAL OF URBAN AFFAIRS | Vol. 25/No. 3/2003

neglect of entire categories of predictors in some studies further complicates the picture.Greater consistency is apparent with regard to the dependent variable: most investigatorstaking a structural approach have relied on a U.S. Department of Housing and UrbanDevelopment (1984) survey of 60 metropolitan areas for their homelessness rates. How-ever, the numerator data incorporated in the rates come from informant judgments ratherthan from actual counts and must, therefore, be viewed with caution (a detailed critique ofHUD’s methodology can be found in Appelbaum, 1990). The rates suffer from temporaland coverage limitations as well. Now nearly two decades old, the data do not allowexploration of the prevalence of homelessness in the roughly 275 metro areas omitted fromthe original HUD sample.

Our objective is to provide a more comprehensive analysis of the structural determin-ants of homelessness than has been attempted to date. We improve upon previousinquiries in three ways: 1) expanding the sample of urban contexts, 2) mining a morerecent data source, and 3) broadening the range of independent variables. Our sampleconsists of all 335 US metropolitan areas recognized in the 1990 Census. Homelessnessrates are constructed for these areas with information from the Census Bureau’s S-Nightenumeration. After evaluating the credibility of these rates, we briefly demonstratetheir variation and then regress them on sets of predictors intended to capturedifferent explanations of homelessness. The article concludes with an assessment ofour findings and a discussion of their theoretical, methodological, and policy implica-tions.

EXPLANATIONS

Underlying our research is the premise that, according to Elliott and Krivo (1991),‘‘unfavorable structural conditions are necessary precursors to widespread homelessness’’at the metropolitan level (p. 14). What has yet to be determined is which conditions mattermost and in which settings. Many scholars have sidestepped this issue, giving roughlyequal weight to a variety of factors that they weave together in an interpretation ofhomelessness for the US as a whole. For analytic purposes we prefer to keep these factorsseparate. Our review of the literature identifies four major structural explanations that canbe used to account for differences in the prevalence of homelessness across metropolitanareas circa 1990. We add to the list two other explanations that emerge from street-levelinsights of advocates, service providers, and policy makers. Each explanation is summar-ized and relevant hypotheses are derived from it.

Housing Market

Proponents of the housing market explanation attribute homelessness to a decline in theavailability of affordable units (Burt, 1992; Ringheim, 1990; Wright & Lam, 1987; for acontrary argument, see Main, 1996). During recent decades, several forces have convergedto limit the housing options open to low-income renters. Price inflation throughout themarket has made the transition to home ownership more difficult and thus has increasedcompetition for rental units, pushing such units beyond the means of poor households.Other units have been lost via conversion, abandonment, demolition, and arson. Urbanrenewal and gentrification have taken a heavy toll on the stock of single-room occupancy(SRO) housing that was once plentiful in the urban core (Hoch & Slayton, 1989; Jencks,1994; Kasinitz, 1984). From a supply-side perspective, diminished profit potentialhas slowed the construction of new low-cost units as well. Together, these forces are

| Homelessness in Metropolitan Areas | 337

deemed responsible for creating a tight, expensive market in which the poor, faced withrising rent burdens and decreasing ownership opportunities, struggle to keep a roof overtheir heads.

On the assumption that access to affordable housing is not restricted to the same degreeeverywhere, we expect metropolitan homelessness rates to be affected positively by rentlevels but negatively by home ownership and vacancy rates.

Economic Conditions

Another popular explanation of homelessness stresses the peculiar economic conditionsin the US that have resulted from globalization, deindustrialization, restructuring, tech-nological change, and similar trends (Blau, 1992; Hardin, 1996; Timmer, Eitzen, & Talley,1994). In metropolitan areas hit hardest by these trends, stable, decent-paying jobs(especially of the sort once common in manufacturing) have diminished in number, andlocal unemployment tends to be relatively high even when overall economic prosperity(such as occurred during the 1990s) manages to keep the national unemployment rate incheck. Those residents with modest levels of human capital who are still able to find workoften wind up in unskilled, temporary, or part-time positions in the service sector (Law &Wolch, 1991). The combination of unemployment and fragile service employment holdsdown earnings, putting more people in jeopardy of deep, chronic poverty. After they enterthat state, the range of problems experienced, from domestic violence and substance abuseto social isolation, is thought to further heighten their vulnerability to homelessness(Wright, Rubin, & Devine, 1998). Consistent with this reasoning, the extent of unemploy-ment, low-wage employment, and extreme poverty should each have a positive impact onthe homelessness rate across metro areas.

Demographic Composition

Several aspects of demographic composition have been used in different ways toaccount for homelessness. Given the overrepresentation of African Americans amonghomeless persons (Burt et al., 2001; Culhane & Metraux, 1999), a metro area’s blackpopulation can be considered an at-risk group. Compared with whites, blacks are morelikely to encounter housing discrimination, to suffer skill and locational mismatches in thejob market, and to have weakened family supports, each of which increases their suscep-tibility to homelessness (Culhane, Chang-Moo, & Wachter, 1996; Hopper & Milburn 1996;Rosenheck, Bassuk, & Salomon, 1999; Roth, Toomey, & First, 1992). A similar logic ofdisadvantage applies to single-person households. These individuals lack the protectionoffered by a second wage earner when crises arise, and factors that may originally selectcertain people into single status–youthfulness or old age, pre-existing personal problems–can threaten their shelter security (Burt, 1992; Jencks, 1994).

Because of their rapidly growing numbers, single-person households could also influ-ence the level of homelessness by increasing competition for affordable housing. So mighttwo other demographic groups–baby boomers and Hispanics. Although cohort size hasundoubtedly depressed some boomers’well-being along economic and other lines, produ-cing what Wright and his colleagues (1998) term ‘‘yuffies’’ (young urban failures), thematuration of the baby boom generation is believed to have intensified pressures on allsegments of the housing market. In parallel fashion, the presence of a large Hispanicpopulation may signify escalating demand associated with post-1965 immigration streams.However, based on their proportional representation, Hispanics do not seem particularly


vulnerable to homelessness as a group (Burt et al., 2001; Culhane & Metraux, 1999).According to Baker (1996), they are more inclined than African Americans to usedoubling-up and other non-traditional housing arrangements to stay off the streets (alsosee Rosenheck et al., 1999).

With respect to demographic composition, the greater the proportion of African Ameri-cans, one-person households, and baby boomers in a metropolitan area, the higher thehomelessness rate is predicted to be. Our prediction regarding Hispanics remains indeter-minate–a positive, negative, or null effect is plausible.

Safety Net

Many experts view the magnitude of homelessness in 1990 as a function of the size ofthe social safety net, which contracted (or continued to contract) during the Reagan era ofthe 1980s (Blau, 1992; Koegel, Burnam, & Baumohl, 1996). Across a wide range ofprograms, eligibility criteria became increasingly restrictive and benefits shrank. However,the changes must be judged incremental when compared to the dramatic reforms set inmotion by the Personal Responsibility and Work Opportunity Act (PRWORA) of 1996(Lichter & Crowley, 2002; Weaver, 2000). Aid to Families with Dependent Children(AFDC), a federal entitlement administered by the states, was heavily impacted prior toits PRWORA-induced abolition. Local public assistance programs lost ground during the1980s as well, reducing the purchasing power of the many single individuals who failed toqualify for federal forms of income support (Greenberg & Baumohl, 1996). These cut-backs put an added burden on Social Security as the primary strand in the safety netresponsible for keeping people from slipping into poverty and for lifting them out (Burt,1992). Central to our research is the fact that by 1990 the proportion of persons coveredby Social Security and public assistance varied substantially from place to place, as did themagnitude of AFDC payments. We predict negative relationships in both cases: the moreextensive the coverage and the higher the payments, the lower the metropolitan home-lessness rate.

Another aspect of the safety net explanation concerns the care of the mentally ill in thewake of the deinstitutionalization movement. The total population of patients in statemental hospitals declined from 560,000 in 1955 to just over 100,000 by 1989 (Federal TaskForce on Homelessness and Severe Mental Illness, 1992). Scholars have attributed thistrend to changing theories about treatment, expanded patients’ rights, advances in drugtherapy, and budgetary constraints (Blau, 1992; Jencks, 1994; National Institute ofMental Health, 1989). The effect of deinstitutionalization on the level of homelessnesscannot be determined with any precision given the methodological difficulties involved. Amore appropriate issue, in our opinion, has to do with the availability of community-based residential care and outpatient services for mentally ill persons whose counterpartsin the past were hospitalized. Once again, there is marked geographic variation in theextent to which local jurisdictions are able to provide these alternative types of treatmentto members of at-risk groups. In particular, the amount that a state government invests inmental health programs can be hypothesized to negatively affect the homelessness rates ofthe metro areas within its borders.

Climate

The role played by climate in attracting or repelling homeless individuals has been thesource of much speculation but little systematic research. As the ethnographic work of


Snow and Anderson (1993) suggests, some of the subsistence strategies that the homelessemploy–sleeping outdoors, panhandling, engaging in scavenging and casual labor–areclearly weather-dependent. This may help account for the large numbers of homelessmigrants reported in Los Angeles, Las Vegas, Jacksonville, and other sunny locations(Stefl, 1987; Timmer & Knottnerus, 1989). In a similar vein, local government officialsacross the Sunbelt often contend that the warm climate draws outsiders to their cities,unfairly increasing their service burdens. But because the homeless are so fully exposed tothe elements, we suspect that too much heat can be just as much of a deterrent as toomuch cold. Piecing together our own climate explanation, we predict that homelessnessrates will be higher in more temperate metropolitan areas that avoid the extremes ofsummer and winter weather. The rates should also be negatively influenced by precipita-tion levels, given the difficulty of surviving on the streets during lengthy periods of rainor snow.

Transience

Places vary not only in climate but also in the transience of their populations. Howunsettled an area is depends on the overall pace of its growth, net migration, intraurbanmovement, and the like. While such demographic dynamics are themselves driven by avariety of forces–the vitality of the local economy, new housing construction activity, thearea’s popularity as a destination for immigrants (Wolch & Dear, 1993)–they all impactthe residential mobility rate. Places with high rates of mobility tend to be competitive, thusincreasing the vulnerability of less successful people. Transience can also be fueled bytemporary ebbs and flows of tourists, migrant laborers, students, and persons just passingthrough, some of whom become stranded or experience periods of homelessness (Morrow-Jones & van Vliet, 1989; Stefl, 1987; Timmer & Knottnerus, 1989). All else equal, theaccessibility of a community via affordable means of transportation should influence thevolume of these temporary streams. We, therefore, predict that the higher the mobilityrate of a metropolis and the better the transportation access to it, the greater will be itshomelessness rate.

Although much of the work devoted to the six explanations, and especially to the firstfour, has focused on the rise of the post-1980 new homelessness at the national level, theexplanations apply at the local level as well. This point is confirmed by the metropolitanstructural inquiries conducted thus far. As noted earlier, most of these studies utilize the1984 HUD data on 60 metro areas for their homelessness rates and the 1980 Census formost of their independent variables. Despite differences in operational details, they gen-erally find support for multiple explanations. Elliott and Krivo (1991), for example, showthat homelessness is affected negatively by the availability of affordable rental units(consistent with what we refer to as the housing market explanation) and investments inmental health care (in line with the safety net perspective) but positively by the representa-tion of African Americans (demographic composition). Other predictors emerging assignificant in the HUD-based studies include the housing vacancy rate (�), rent control(þ), unemployment (þ), job growth (�), average household size (þ), AFDC benefit levels(�), state mental hospital caseload (�), mean temperature (þ), and community popula-tion change (þ) (Appelbaum, Dolny, Dreier, & Gilderbloom, 1991; Bohanon, 1991; Honig& Filer, 1993; Redburn & Buss, 1986; Tucker, 1990).

Only a couple of structural investigations do not rely on the HUD data. Culhane et al.(1996) have conducted a rare intraurban study, identifying characteristics of Philadelphia


and New York neighborhoods (census tracts) that are associated with an overrepresentationof families admitted to homeless shelters. More relevant for our purposes is Burt’s (1992)analysis of homelessness rates for 147 central cities with populations of 100,000 or more.These rates, pertaining to 1989, have the advantage of being more recent. (Burt alsoreports rates for 1981, 1983, and 1986.) However, their numerators treat shelter beds asa surrogate for the actual number of homeless people. Regressing the shelter bed rates(beds per 10,000 residents) on a wide range of potential antecedents, Burt finds that therates increase with the local unemployment level and with the percentage of single-personhouseholds in a community. Contrary to expectations, her measure of the intensity of theaffordable housing shortage has a negative impact on homelessness, and the size of themaximum public assistance benefit exerts a positive effect. The sign for the housingshortage measure is surprising in light of the broad support that closely related predictorshave received in the literature. In our opinion, though, the deficiencies noted with existingstudies render premature any definitive conclusions about the key structural determinantsof homelessness in metropolitan areas.

DATA

Our own effort to advance the structural line of research relies heavily on the muchpublicized S-Night operation undertaken during the 1990 Census (for a fuller methodo-logical treatment, see Taeuber & Siegel, 1991). On S-Night, Census Bureau employeesattempted a nationwide count of the homeless population as part of a larger initiative toimprove the coverage of underrepresented groups. The design of the count was two-pronged: 1) homeless people in shelters were enumerated on March 20 between 6 p.m.and midnight; 2) those in street locations were enumerated on March 21 between 2 a.m.and 4 a.m. Prior to S-Night, census staff compiled a master list of shelter and street sitesfrom administrative records and through contacts with local experts (government officials,service providers, advocacy groups, etc.). Afterwards a mop-up phase was conducted toinclude any shelters that had been missed. Taken together, the shelter and street compon-ents suggest a 1990 national homeless population of roughly 230,000, with over 93% ofthat total located in metropolitan areas.

Unlike previous multi-site studies, the S-Night data are based on direct observationrather than on informant opinion and are available for all census-defined metro areas, notjust a small subset. Nevertheless, S-Night has been heavily criticized. The furor beganbeforehand when advocates, fearing the consequences of a low count, proposed a boycottof the operation. Though the boycott failed to materialize, subsequent allegations aboutdeficiencies in coverage have limited the use of S-Night results for both programmatic andscholarly purposes. Of greatest concern to us are criticisms that have some empirical basis.Independent evaluations in Chicago (Eden, 1992), Los Angeles (Cousineau & Ward,1992), New Orleans (Devine & Wright, 1992), New York (Hopper, 1992a, 1992b), andPhoenix (Stark, 1992), performed under contract to the Census Bureau, indicate that theshelter counts were fairly complete, thanks in part to a high level of local cooperation. Ofthe 1,300 municipalities throughout the US with populations of 50,000 or more, all but 19provided census staff with lists of shelters and other sites prior to S-Night (U.S. GeneralAccounting Office, 1991).

The most serious problems occurred with the street portion of the enumeration. Obser-vers were positioned in a sample of predesignated street sites in the five evaluation cities tomonitor enumerator performance. At some sites enumerators violated procedural


guidelines by arriving late or leaving early, remaining in their cars (particularly if the sitelooked dangerous), not interviewing Hispanics (due to a lack of Spanish languagefluency), or being selective in which persons they approached. At other sites, they failedto appear altogether, either because their workload caused them to run out of time orbecause the site was hard to find. Such problems led Martin (1992) to conclude that the‘‘street enumeration was not carried out in a comparable, standardized way in the districtoffices in the assessment’’ (p. 436). The fact that site lists were prepared well in advance ofS-Night and were often based on homeless occupancy during the day rather than theevening further complicated enumerators’ jobs, yielding numerous dead ends on S-Night.Moreover, the existence of ‘‘uncountables’’–homeless individuals who spent S-Night inlocations beyond the intended reach of the enumeration–increased worries about a grossunderestimate of the size of the street population (Wright & Devine, 1995).

Because of the difficulties involved in any attempt to count homeless people, theS-Night data cannot be considered flawless, but neither should they be dismissed out ofhand. In a separate paper (Lee & Price-Spratlen, 2002), we offer an extensive justificationfor taking these data seriously. Our argument rests on the following points:

1. By S-Night in 1990, the number of people on the streets had decreased dramaticallyin response to the expansion of shelter capacity, making them easier to count (Burt,1992; Jencks, 1994).

2. The winter weather on S-Night may have assisted census enumerators by pushing ahigher percentage of homeless persons than normal to indoor shelters (Taeuber &Siegel, 1991).

3. In spite of their shortcomings, the enumerators responsible for the street countvisited a majority of the sites assigned to them and produced tallies significantlycorrelated with those of the independent observers (Martin, 1992).

4. Certain aspects of the S-Night operation, such as the automatic classification ofeveryone observed in a predesignated street site as homeless and the potentialdouble-counting of those who changed locations while the enumerators were in thefield, may have served to artificially inflate rather than deflate the street estimates.

Several types of empirical comparisons enhance the credibility of our argument. Forinstance, when juxtaposed with estimates from rigorous local studies in Chicago (Rossi,1989), Los Angeles (Koegel, Burnam, & Morton, 1996), Nashville (Lee, 1991), Washington,DC (Robinson, 1985), and elsewhere, the S-Night figures usually seem reasonable,and they deviate on the high side as well as the low. In a preliminary analysis,we attempted a similar comparison on a grander scale. Central-city homelessnessrates (shelter beds per 10,000 residents) for 1989 were obtained from Burt (1992)and paired with corresponding 1990 Census rates that use homeless persons instead ofbeds in the numerator. The resulting correlation across the 147 cities in Burt’s sample equals.80.

Public perceptions also align as expected with the S-Night data. Appending these datato individual records from a national survey of attitudes toward the homeless (Link et al.,1995), we find significant positive correlations between the homelessness rate in therespondents’ area of residence and their judgment of the magnitude of the problem locally(r¼ .20), the frequency with which they see homeless people during a typical week(r¼ .33), and the frequency with which they see homeless people in their neighborhood(r¼ .20). Finally, Garfinkel and Piliavin (1994) have employed multivariate regressiontechniques to examine differences in street-to-shelter ratios from a variety of studiesconducted throughout the 1980 to 1990 period. They find no evidence that S-Night


estimates systematically understate the extent of street homelessness relative to estimatesfrom other sources.

These comparisons suggest that the data from S-Night are free of consistent bias andmay rank-order metropolitan areas in a fairly accurate manner. However, we shouldemphasize the nature of that ranking. The census information ideally provides a snapshotor, in more technical terms, an estimate of the point prevalence of literal homelessness: thenumber of persons at a particular time who lack a permanent residence of their own. Thecrucial issue is that any type of point estimate will understate the impact that frequententries into and exits from the population have on the prevalence of homelessness over alonger period. Thus low point estimates are not necessarily inconsistent with much higherperiod prevalences. If the work of Culhane, Dejowski, Ibanez, Needham, and Macchia(1994) and Metraux et al. (2001) on the annual incidence of homelessness in New York,Philadelphia, Washington, DC, Columbus, and other places can be taken as a guide, aone-day count may translate into two-and-a-half to 10 times that many homeless duringthe course of a year. The numbers become more striking when calculated on a five-year orlifetime basis (Link et al., 1995).

Readers should keep in mind that the S-Night data tap only the most accessiblemembers of the literal homeless group, best described as visibly homeless. People whowere staying with friends or relatives or who were otherwise hidden from view wentundetected. People homeless for brief durations, though a major portion of the targetpopulation over an extended period, were also at lower risk of being enumerated by cross-sectional census procedures. The 1990 S-Night count thus disproportionately representsthose who experience homelessness on a long-term basis, face the gravest problems, andconsume the most resources.

Our decision to use 1990 data was motivated by Census Bureau restrictions placed onthe 2000 data. During the most recent census, a service-based enumeration similar toS-Night was conducted over a three-day period. From March 27 to March 29, 2000,homeless people, or more precisely, people without conventional housing, were countedat emergency and transitional shelters, at soup kitchens and mobile food programs, and atpredesignated outdoor sites (Smith & Smith, 2001). Because of coverage and data qualityconcerns, Bureau officials elected not to combine the components of the service-basedenumeration in a single report or file. Instead, only the shelter results are available andonly for geographic units with at least 100 persons in shelters. Persons counted at soupkitchens or outdoor sites are contained in the ‘‘other non-institutional group quarters’’category and cannot be broken out separately. What these restrictions mean is that the2000 Census falls short of its 1990 predecessor in both the number of metropolitan areasand number of segments of the homeless population discernible in tabulations released tothe public.

Measures

We construct our principal dependent variable, the 1990 metro area homelessness rate(number of homeless people per 10,000 general population), from S-Night data. Theregular version of that variable is described in the top row of Table 1; some adjustmentsare introduced later. The remainder of the table defines the multiple measures used tocapture each of the six explanations of homelessness. Housing market influences areexamined with metro-specific data on median rent level (the effect of which we previouslyhypothesized to be positive) and home ownership (�) and vacancy (�) rates. Economic


conditions are tapped by the unemployment rate (þ), the percentage of persons in low-wage service and unskilled jobs (þ), and the incidence of extreme poverty (þ). Potentialdemographic determinants of homelessness include the representation of blacks (þ) andHispanics (þ or -), the size of the local baby boom cohort (þ), and the percentage of one-person households (þ). All of these measures are drawn from the 1990 Census.

The variables reflecting the three other explanations of homelessness come from variedsources.With regard to the safety net explanation, the percentage of households receiving publicassistance income (�) is basedon the 1990Census, but thepercentage of persons receivingSocialSecurity benefits (�) was obtained from a county-city data compendium (Slater & Hall, 1992)and themeanAFDCpayment (�) and per capitamental health expenditure (�) measures fromspecial HHS reports (U.S. Department of Health and Human Services, 1990, 1992). The SlaterandHall compendiumalsoprovides information for our two climate variables, the July-Januarytemperature range (�) and the average annual precipitation (�) in the central city of eachmetropolitan area. Finally, to test the transience explanation we use a conventional censusmeasure of 1985 to 1990 geographic mobility (þ), complemented by a transportation accessindex (þ) that sums the number of interstate freeways, US highway routes, and railroad linesserving each metropolis (RandMcNally, 1990a, 1990b).

TABLE 1

Description of Variables

Explanation/Variable Description Mean

Standard

Deviation

Homelessness Rate Homeless persons per 10,000 residents 7.9 7.0

Housing Market:

Rent Level Median contract rent ($) for renter-occupied units 350.9 107.8

Home ownership % of all occupied housing units with owner occupant 64.9 6.6

Vacancy rate % of all housing units vacant 8.6 4.4

Economic Conditions:

Unemployment rate % of civilian labor force unemployed 6.3 1.8

Low-wage jobs % of employed persons 16þ in service and unskilled jobs 19.9 3.3

Extreme poverty % of all persons with incomes <50% of poverty level 5.7 2.6

Demographic Composition:

Blacks % black in total population 9.6 9.4

Hispanics % Hispanic in total population 7.3 12.8

Baby boomers % of persons 25–44 in total population 32.1 2.5

1-person households % of all occupied housing units with single occupant 24.0 2.7

Safety Net:

PA recipients % of all households receiving public assistance income 7.1 2.6

SS recipientsb % of all persons receiving Social Security benefits 15.7 5.2

AFDC paymentc Mean monthly AFDC payment ($) per recipient, for state 122.7 51.3

MH expendituresd Per capita mental health expenditures ($) for state, 1986 75.1 32.0

Climate:

Temperature rangeb July average maxium minus January average minimum (F�) 61.4 11.0

Precipitationb Average annual precipitation (inches) 36.8 13.2

Transience:

Mobility rate % of persons 5þ living in different house than in 1985 48.0 7.7

Transport accesse N of interstates, US routes, and railroads serving area 8.1 4.6

Note. Source is 1990 U.S. Census unless otherwise noted.aSlater and Hall (1992). bU.S. Department of Health and Human Services (1992).cU.S. Department of Health and Human Services (1990). dRand McNally (1990a, 1990b).


The means and standard deviations in Table 1 were calculated in unweighted fashionacross the 335 metropolitan areas (including both stand-alone areas and primary areasdecomposed from larger consolidated units) that are available in 1990 Census data filesand that constitute the cases in our study. Nearly 193 million people, 77.5% of the totalUS population, resided in these metro areas as of 1990. We focus on such areas ratherthan central cities for two reasons. First, many of our independent variables are mostappropriately conceptualized as operating within metro-wide housing and labor markets(e.g., rent level and unemployment and mobility rates), as referring to phenomena not wellcontained by political jurisdictions (demographic composition), or as measuring relativelyconstant aspects of the larger setting (climate, transportation access). Second, bothNSHAPC and S-Night data show that roughly one fifth of all metropolitan homelessnationally are located outside central cities, especially in the more densely settled suburbs(Burt et al., 2001). Hence our choice of cases yields a more complete picture of home-lessness and its determinants than would be possible with city-specific data.

For ease of interpretation, descriptive statistics on all variables are presented in rawform in Table 1. Distributional properties of several of the variables lead us to employlogged versions of the homelessness rate and 10 of the predictors in later regressionanalyses. Though the predictors fail to satisfy the assumption of statistical independenceunderlying regression, they do not display the excessive degree of collinearity that plaguesmuch aggregate-level research in which metro areas or other spatial units are featured. Themean zero-order correlation among the 18 predictors barely reaches .22 in absolute value,and the mean correlation among the predictors associated with each explanation exceeds.30 only for the measures of economic conditions (mean r¼ .45). Across all 153 correl-ations involving the independent variables, the shared variance tops 50% just once: thepercentage of households on public assistance is strongly and positively related to theunemployment rate (r¼ .76) at the metropolitan level.

RESULTS

Rates for Metro Areas

Our first objective is to establish that the metropolitan homelessness rates based on the1990 Census actually vary to a non-trivial degree. Toward that end, Table 2 breaks therates down by regional location and metro population size. We define a population of onemillion or more as ‘‘large,’’ 250,000 through 999,999 as ‘‘medium,’’ and less than 250,000as ‘‘small.’’ On average, the rates are highest for western metro areas, followed by those inthe Northeast. This rank order holds for each size class with one exception: small mid-western metropolises exhibit a slightly higher mean than their small northeastern counter-parts. The ranges are also greater in the West and the Northeast. Differences between SanFrancisco and Provo (a gap of 42.7 homeless per 10,000) and Orange County, NY andSharon, PA (a gap of 50.6) far exceed the extremes found in the South or the Midwest.

San Francisco is not the only familiar place among the metro areas with relatively highrates. In the large size class, New York, Chicago, and Washington, DC (well known fortheir homelessness problems) rank at the top of their respective regions. Perhaps moreinteresting, however, are the areas that constitute surprises. Orange County, NY, has thehighest rate of all, a fact due to the presence of Camp LaGuardia, a farm-style shelterfacility to which New York City’s homeless residents have been shipped regularly since theDepression (Hernandez, 1994). Table 2 reveals further that Rapid City, SD, and SanAngelo, TX, surpass Chicago and Washington in the point prevalence of homelessness,


and Redding, CA, approaches its better-known neighbor, San Francisco, 200 miles to thesouth. In addition to small metro areas having high rates, large areas can have low ones.Cleveland andRochester,NY, offer themost obvious illustrations of this possibility in the table.

One conclusion we draw from these descriptive results is that homelessness should notbe regarded as a simple function of metropolitan population size. We realize such a viewchallenges popular thinking about the concentration of the homeless in the biggest cities.But while the means in Table 2 underscore the desirability of considering metro size, giventhe tendency for the homelessness rate to increase with size, the magnitude of the relation-ship must be judged moderate when estimated with ungrouped data (r¼ .27). Put anotherway, metro size leaves unexplained the vast majority of variance in the homelessness rate,raising the question of which factors might account for it.

Understanding Intermetropolitan Variation

We use ordinary least squares regression to assess the relative importance of the sixexplanations summarized earlier. In the full equation of Table 3 the logged 1990 home-lessness rate is regressed on the housing market, economic conditions, demographiccomposition, safety net, climate, and transience measures. We also enter the inverse ofmetro area population size in the equation. This term, labeled ‘‘1/population’’ in the table,acts as a control for size though it serves another, equally critical purpose: Firebaugh andGibbs (1985) argue that it offers one solution to the ratio variables problem. Specifically,it corrects for the kind of dependency that arises when variables on both sides of anequation have the same denominator. Most of the variables in our analysis are rates orpercentages with population size (or a size surrogate) in common.

TABLE 2

Homelessness Rate by Regional Location and Population Size of Metro Area

Region/Size Mean

Standard

Deviation Highest Rate/City Lowest Rate/City

Number of

Metro Areas

Northeast:

Large 12.7 12.5 42.5 New York, NY 3.7 Rochester, NY 9

Medium 11.5 10.0 50.8 Orange Co., NY 2.5 Pawtucket, RI 29

Small 6.2 4.2 15.9 Brockton, MA .2 Sharon, PA 30

Midwest:

Large 6.3 2.5 12.0 Chicago, IL 4.0 Cleveland, OH 10

Medium 5.0 2.9 12.0 Rockford, IL 1.4 Lake County, IL 26

Small 6.8 5.0 21.3 Rapid City, SD .9 Eau Claire, WI 49

South:

Large 8.8 3.0 17.5 Washington, DC 5.5 Norfolk, VAa 14

Medium 6.0 3.0 15.1 Ft. Pierce, FL 2.2 McAllen, TXb 44

Small 4.9 4.8 24.5 San Angelo, TX .3 Laredo, TX 65

West:

Large 17.0 11.6 45.5 San Francisco, CA 8.0 Anaheim, CAc 13

Medium 12.3 6.1 25.5 Reno, NV 2.8 Provo, UT 21

Small 12.4 9.7 41.7 Redding, CA 3.1 Olympia, WA 25

Note. Regions are defined by US Census categories. Size: Large metro area>1,000,000 population in 1990;medium¼ 250,000–999,000; small<250,000.aNorfolk includes Virginia Beach and Newport News. bMcAllen includes Edinburg and Mission.cAnaheim includes Santa Ana.


The regression results for the full equation show that two predictors have significantimpacts in the hypothesized direction. Consistent with the housing market explanation,high rents boost homelessness, presumably by putting more residents in a competitive andhence precarious situation. In a similar vein, the greater the representation of single-person households in a metropolis, the greater the homelessness rate of that metropolis.As noted previously, we see this demographic variable as tapping both the size of the localat-risk population and the intensity of housing demand. One measure of economic con-ditions deserves mention as well: a concentration of workers in low-wage jobs affectshomelessness in a positive, nearly significant manner.

For the sake of parsimony, the full equation has been reduced in backwards stepwisefashion to its most influential predictors, which are presented in the right half of Table 3.The largest standardized coefficients again belong to rent level (B¼ .47) and the propor-tion of one-person households (B¼ .28), which represent the housing market and demo-graphic composition models respectively. However, each of the four remainingexplanations also places a predictor in the reduced equation. In metro areas with manylow-wage jobs (an economic measure) and extensive geographic mobility (an aspect oftransience), the 1990 homelessness rate tends to be greater; in areas of heavy precipitation(a climate indicator), it tends to be lower. All of these effects are in the expected direction.The most confusing result is for the public assistance variable, which takes a positive sign.Controlling for other potential determinants of homelessness, this variable apparently taps

TABLE 3

Regression of Homelessness Rate on Independent Variables

Full Equation Reduced Equation

Variable b B b B

Rent Levela 1.440*** .487 1.395*** .472

Home ownership �.003 �.027

Vacancy rateab .888 .000

Unemployment ratea �.047 �.016

Low-wage jobsa .716 .141 .532 .105

Extreme povertya �.047 �.024

Blacksa �.045 �.071

Hispanicsa .061 .098

Baby boomers .018 .056

1-person households .083*** .272 .084*** .278

PA recipientsa .348 .147 .279* .118

SS receipientsa .036 .013

AFDC paymentb �.907 �.056

MH expendituresa �.039 �.016

Temperature range .002 .021

Precipitation �.005 �.077 �.009** �.149

Mobility rate .008 .075 .012* .114

Transport accessa .111 .081

1/populationc �23.741 �.099 �31.651* �.132

Intercept �11.821 �10.595

Adjusted R2 .342 .354

Note. aExpressed in logarithmic form. bMove decimal point for b three places to left.cMove decimal point for b three places to right.*p< .05. **p< .01. ***p< .001.


the share of the metropolitan population on the brink of privation rather than theadequacy of the local safety net. The last two predictors removed from the stepwiseequation, transportation access (þ) and proportion Hispanic (þ), exhibit signs in linewith our discussion of the demographic composition and transience models respectively,even though they fail to reach conventional significance levels.

Because of the occasional volatility of analyses that rely on macro units like metroareas, we check the robustness of the findings in Table 3 in several ways. First, we searchfor outliers and other types of influential cases that might distort the regression results.Among the better diagnostic statistics for this purpose is Cook’s D, which summarizeshow the estimated regression coefficients (based on all cases) change when a particular caseis deleted (Cook & Weisberg, 1999). Depending on the conservatism of the investigator, Dslarger than .5 or 1.0 are commonly regarded as a cause for concern. Fortunately, no metroarea in our sample generates a D value above .08 for either equation in Table 3.

A second check involves determining if the basic pattern of results holds for classes ofmetro areas. As Table 2 suggests, region may be a worthwhile grouping criterion. Thispossibility receives additional backing from treatments of homelessness that highlight thedistinctiveness of the problem in the Northeast or along the West Coast (e.g., the so-called‘‘Highway 1 homeless’’ in California) or that document regional variation in homelessnessrates and shelter bed availability (Burt, 1992; Lee & Price-Spratlen, 2002; U.S. Depart-ment of Housing and Urban Development, 1984). Accordingly, we have subdivided thesample of metro areas into four regional categories and re-estimated reduced equations foreach (not shown). Rent level (þ) proves to be significant across the board: it is thestrongest predictor in three of the regional equations and appears in all four. Anotherhousing market variable, home ownership (�), shows up in two equations; so does thepercentage of one-person households (þ), in both instances as the second strongestpredictor. Each of these variables takes the hypothesized sign. While some idiosyncraticeffects are sprinkled throughout the equations, the primary structural determinants ofhomelessness in 1990 resemble each other from one region to the next.

A similar conclusion applies when metro areas are grouped by population size, the otherdimension of cross-classification in Table 2. The full distribution of areas has beendichotomized at the 250,000 mark, and separate equations have been run for the twocategories (not shown). A few dynamics underlying homelessness do appear to differbetween the larger and smaller areas. Notably, home ownership (�) reaches statisticalsignificance in the former while one-person households (þ) and precipitation (�) matter inthe latter. What stands out, though, is the persistent positive effect of rent level on thehomelessness rate. It has the dominant impact for smaller metro areas and rivals that ofhome ownership among the larger areas. Moreover, its impact in the size-specific equa-tions, and also in the region-specific ones described above, cannot be traced to a fewinfluential cases based on the low values taken by Cook’s D.

Our final check on the stability of the results in Table 3 is to see if they are sensitive tostrategic modifications of the dependent variable. Using 1990 Census data, we make thenumerator of the homelessness rate more inclusive by adding to the S-Night count: 1) alloccupants of domestic violence shelters, despite the likelihood that many of these personsdepart substantially from the profile of S-Night’s visible homeless, and 2) 5% of the clientsin drug and alcohol abuse group homes, a proportion that matches the share of allalcoholics nationally who are estimated to be homeless (Schutt & Garrett, 1992). Astepwise regression employing this modified rate produces results comparable to thosealready discussed. Rent level (þ) and single-person households (þ) are the best predictorsin the top equation of Table 4, exhibiting effects in the expected direction. Low-wage jobs


(þ) and precipitation (�) once again enter the equation, and the amount of varianceexplained is about the same as in Table 3.

We also adjust the denominator of the homelessness rate to better capture persons atgreatest risk of becoming homeless. As numerous studies have shown, homeless people aremore likely to be drawn from the ranks of the very poor than from other socioeconomicsegments of the population. In recognition of this selection process, the number of metrodwellers with incomes less than one-half of the official poverty level have been inserted intothe denominator of the rate that serves as the dependent variable in the bottom equation ofTable 4. Changing the rate in such a manner boosts the R2 from 35% (top equation) tonearly 50%. It also reverses the sign of one predictor associated with the safety net model:the availability of public assistance affects the new rate in the expected negative direction,suggesting that welfare programs can keep the poor from slipping into a homeless state.Counterbalancing that ameliorative view of welfare’s role is the positive impact of the size ofAFDC payments on homelessness in 1990, which could be read as support for a ‘‘magnet’’argument (i.e., that higher payments attract disadvantaged families from elsewhere).

At best, however, interpretive ambiguities concerning the safety net model strike us asminor footnotes to the main story told in the bottom equation of Table 4. There, as inprior regression runs, the strongest determinant of the homelessness rate continues to berent level (þ) with one-person households (þ) again ranking second. Independent of otherstructural characteristics, expensive housing markets in which many people live alone arewhat distinguish metro areas with a serious homelessness problem from areas where therate is lower.

TABLE 4

Reduced Equations from Regressions of Modified Homelessness Rates on Independent Variables

Equation/Variable b B

Modified Numerator:

Rent levela .996*** .380

Home ownership �.011 �.096

Low-wage jobsb .607* .135

Hispanicsa .063 .114

1-person households .074*** .275

Precipitation �.007* �.131

1/populationb �23.447* .110

Intercept �6.476

Adjusted R2 .343

Modified Denominator:

Rent levela 1.614*** .468

1-person households .072*** .203

PA recipientsa �.365** �.132

AFDC payment .003* .145

Precipitation �.009** �.121

1/populationb �28.701* �.103

Intercept �11.008

Adjusted R2 .493

Note. Modified numerator includes all persons in domestic violence shelters and 5% of persons in drug/alcohol abusegroup homes as well as those counted during S-Night. Modified denominator is limited to persons with incomes< 50% ofpoverty level.aExpressed in logarithmic form. bMove decimal point for b three places to right.*p< .05. **p< .01. ***p< .001.


CONCLUSION

Our research advances the structural investigation of homelessness in important ways.Despite their perceived shortcomings, the S-Night data that we employed are grounded inobservation, pertain to a more recent year than the homelessness rates from the 1984HUD study, and are available for all metropolitan areas in the US rather than a smallsubset. These advantages enhance the credibility of our efforts to explain intermetropoli-tan variation in homelessness. Admittedly, the picture painted here applies only to 1990.Yet we are confident that most of the effects uncovered would hold up reasonably well atlater points in time.

While conventional wisdom identifies community population size as the primary struc-tural correlate of homelessness, our analysis considers a wider range of antecedents drawnfrom the literature. We find that place-specific characteristics of the housing market arekey. In particular, higher-priced areas (as indexed by the median rent) have higher home-lessness rates irrespective of how those rates are operationalized. But single-person house-holds, who are economically vulnerable and who intensify demand, exert a major influenceon homelessness as well, and precipitation shows up in every equation of Tables 3 and 4.This potent combination of factors is perhaps best exemplified by San Francisco. The pointprevalence of homelessness there can be traced in part to that city’s tight, expensive housingmarket, large live-alone population (nearly 40% of all households), and dry climate.

Of these three characteristics, only the first two seem amenable to policy interventionand only with great difficulty. For example, rents could be made more tolerable throughincentives to stimulate the supply of affordable units and, in the short term, throughsubsidies (e.g., housing vouchers). With regard to single-person households, safety-netprograms that provide emergency support in the case of job loss, illness, or other income-threatening events might diminish vulnerability to homelessness. It is important to recall,however, the extent to which the welfare landscape has changed since 1990. The imple-mentation of the Personal Responsibility and Work Opportunity Act of 1996 greatlyaccelerated the devolution of welfare policies already in progress, increasing differencesacross states. Thus the feasibility of any intervention will depend on how budgetarily andphilosophically acceptable it is within a given jurisdiction. Aspects of the safety net now ineffect in many places (work requirements, time limits on receipt of assistance, etc.) mayactually lessen housing security among certain segments of the poor, becoming a structuralinducement rather than a safeguard against homelessness (Newman & Harkness, 1999).

Most proposals to ‘‘do something’’ about the homelessness problem strike us as overlynarrow in scope. Indeed, the more general message conveyed by our results is that nosingle structural variable or set of variables adequately accounts for homelessness on itsown. Tables 3 and 4 attest to the complex mix of forces operative at the metropolitanlevel, drawing predictors from each of the six explanations and documenting a majority ofeffects in the hypothesized direction. Even the inverse of population size, though not oftheoretical interest, takes the expected sign: larger communities have higher homelessnessrates than smaller ones. We suspect that size has no direct causal connection to home-lessness but rather is associated with other characteristics omitted from the analysis.Consider, for the sake of illustration, Jencks’s (1994) argument that the emerging crackepidemic was responsible for the rise of the new homeless during the 1980s. If crack and itsconsequences are more widespread in larger metro areas, then size may simply be asurrogate, confirming that our models are not fully specified. Put another way, there aresurely additional independent variables that deserve attention in future work and thatcould boost the moderate R2 values reported here.


Among such variables are programmatic steps being taken at the local level. Oursafety net measures are admittedly crude, targeting poverty and mental health issuesinstead of homelessness per se. This inability to control for the influence of homelessassistance programs prevents us from ruling out an alternative interpretation: that ourdependent variable, insofar as its numerator is dominated by people in shelters, can beconstrued as an indication of a metropolitan area’s response to homelessness ratherthan as a gauge of problem magnitude. According to Culhane et al. (1996), themeaning of enumeration results depends on the elasticity of the shelter bed supply.Enumerations better reflect the true rate of homelessness in communities whereincreases in shelter demand can be met and where shelter policies are liberal interms of admission criteria and the length of stay permitted. Information on thesedetails is not available for the 335 metropolitan areas in our sample at the time of theS-Night count. Nevertheless, we are reassured by the significant expansion of sheltercapacity nationally prior to 1990 (Burt, 1992; Jencks, 1994) and by the range ofhomeless housing programs revealed in a subsequent survey of municipal governments(Berman & West, 1997).

Ultimately, of course, neither a structural approach like ours nor its individual-levelcounterpart does full justice to explaining homelessness. The former disregards theimmediate circumstances responsible for a particular homeless episode; the lattercannot explain differences in prevalence across metro areas or other settings. Asmost scholars now recognize, it is the intersection of structural and individual influ-ences that matters. Thus, the types of personal deficits commonly cited as proximalcauses (e.g., mental illness, weak educational background, domestic violence, socialisolation) can be expected to shape the likelihood of becoming homeless contingent onthe community context in which people live. In similar fashion, contextual conditionssuch as rent level are likely to register different effects when someone has accumulatedmore rather than fewer deficits.

From a methodological standpoint, what is required for satisfactory synthesis is ananalysis in which the contributions of both structural and individual factors areestimated. Thus far, no data set has been up to the task, and for good reason: thesample would have to include pools of vulnerable people in multiple locations forwhom homeless or non-homeless outcomes are recorded after contextual and indivi-dual characteristics have been measured. If indicators of severity (number of spells,duration, etc.) were to replace the homeless/non-homeless dichotomy as the focus, atrue cross-level treatment might be possible with the National Survey of HomelessAssistance Providers and Clients (Burt et al., 2001). Similarly, should service providersin enough jurisdictions adopt standardized automated record systems (Metraux et al.,2001), multi-level influences on the sequence of entries into and exits from a homelessstate could be estimated. Unfortunately, in both instances shifts in the dependentvariable would also alter the nature of the research problem. Barring unforeseendevelopments, the integration of structural and individual perspectives on the causesof homelessness seems likely to remain more of a theoretical than an empiricalendeavor.

ACKNOWLEDGEMENT: We thank Douglas Anderton, Dennis Culhane, Lauren Krivo, and the

anonymous referees for their helpful comments on previous drafts. Support for this project has been

provided by the Population Research Institute of The Pennsylvania State University, which receives core

funding from the National Institute of Child Health and Human Development (Grant 1-R24-HD41025).


REFERENCES

Appelbaum, R. P. (1990). Counting the homeless. In J. A. Momeni (Ed.), Homelessness in the United

States, Vol. II: Data and issues (pp. 1–16). New York: Greenwood.

Appelbaum, R. P., Dolny, M., Dreier, P., & Gilderbloom, J. I. (1991). Scapegoating rent control: Masking

the causes of homelessness. Journal of the American Planning Association, 57, 153–164.

Baker, S. G. (1996). Homelessness and the Latino paradox. In J. Baumohl (Ed.), Homelessness in America

(pp. 132–140). Phoenix: Oryx.

Berman, E. M., & West, J. P. (1997). Municipal responses to homelessness: A national survey of

preparedness. Journal of Urban Affairs, 19, 303–318.

Blau, J. (1992). The visible poor: Homelessness in the United States. New York: Oxford University Press.

Bohanon, C. (1991). The economic correlates of homelessness in sixty cities. Social Science Quarterly, 72,

817–825.

Burt, M. R. (1992). Over the edge: The growth of homelessness in the 1980s. New York: Russell Sage.

Burt, M. R., Aron, L. Y., & Lee, E. (with Valente, J.). (2001). Helping America’s homeless: Emergency

shelter or affordable housing? Washington, DC: Urban Institute Press.

Cook, R.D., &Weisberg, S. (1999).Applied regression including computing and graphics. NewYork: JohnWiley.

Cousineau, M. R., & Ward, T. W. (1992). An evaluation of the S-Night street enumeration of the

homeless in Los Angeles. Evaluation Review, 16, 389–399.

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., & Kuhn, R. (1998). Patterns and determinants of public shelter utilization among

homeless adults in New York City and Philadelphia. Journal of Policy Analysis and Management,

17, 23–43.

Culhane, D. P., Lee, C., & Wachter, S. M. (1996). Where the homeless come from: A study of the prior

address distribution of families admitted to public shelters in New York City and Phildelphia.

Housing Policy Debate, 7, 327–365.

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. Population Research

and Policy Review, 18, 219–236.

Devine, J. A., & Wright, J. D. (1992). Counting the homeless: S-Night in New Orleans. Evaluation Review,

16, 409–417.

Dworsky, A. L., & Piliavin, I. (2000). Homeless spell exits and returns: Substantive and methodological

elaborations on recent studies. Social Service Review, 74, 193–213.

Edin, K. (1992). Counting Chicago’s homeless: An assessment of the Census Bureau’s ‘street and shelter

night.’ Evaluation Review, 16, 365–375.

Elliott, M., & Krivo, L. J. (1991). Structural determinants of homelessness in the United States. Social

Problems, 38, 113–131.

Federal Task Force on Homelessness and Severe Mental Illness. (1992). Outcasts on Main Street.

Washington, DC: Interagency Council on the Homeless, U.S. Department of Health and Human

Services.

Firebaugh, G., & Gibbs, J. P. (1985). User’s guide to ratio variables. American Sociological Review, 50,

713–722.

First, R. J., Rife, J. C., & Toomey, B. G. (1994). Homelessness in rural areas: Causes, patterns, and trends.

Social Work, 39, 97–108.

Fitchen, J. M. (1992). On the edge of homelessness: Rural poverty and housing insecurity. Rural

Sociology, 57, 173–193.

Garfinkel, I., & Piliavin, I. (1994). Trends in the size of the nation’s homeless population during the 1980s: A

surprising result. Institute for Research on Poverty Discussion Paper, No. 1034–94. Madison:

University of Wisconsin.

Greenberg, M. H., & Baumohl, J. (1996). Income maintenance: Little help now, less on the way. In

J. Baumohl (Ed.), Homelessness in America (pp. 63–77). Phoenix: Oryx.


Hardin, B. (1996). Why the road off the street is not paved with jobs. In J. Baumohl (Ed.), Homelessness in

America (pp. 46–62). Phoenix: Oryx.

Hernandez, R. (1994, December 4). In their backyard: Homeless shelter for men in rural area provokes

outrage. New York Times, pp. 45, 55.

Hoch, C., & Slayton, R. (1989). New homeless and old: Community and the skid row hotel. Philadelphia:

Temple University Press.

Honig, M., & Filer, R. K. (1993). Causes of intercity variation in homelessness. American Economic

Review, 83, 248–55.

Hopper, K. (1992a). Counting the New York homeless: An ethnographic perspective. In P. O’Malley

(Ed.), Homelessness: New England and beyond (pp. 771–791). Amherst, MA: McCormack Institute

of Public Affairs, University of Massachusetts.

Hopper, K. (1992b). Counting the homeless: S-Night in New York. Evaluation Review, 16, 376–388.

Hopper, K., & Milburn, N. G. (1996). Homelessness among African Americans: A historical and

contemporary perspective. In J. Baumohl (Ed.), Homelessness in America (pp. 123–131). Phoenix:

Oryx.

Jencks, C. (1994). The homeless. Cambridge: Harvard University Press.

Kasinitz, P. (1984). Gentrification and homelessness: The single room occupant and the inner city revival.

Urban and Social Change Review, 17, 9–14.

Koegel, P. M., Burnam, A., & Baumohl, J. (1996). The causes of homelessness. In J. Baumohl (Ed.),

Homelessness in America (pp. 24–33). Phoenix: Oryx.

Koegel, P. M., Burnam, A., & Morton, S. C. (1996). Enumerating homeless people: Alternative strategies

and their consequences. Evaluation Review, 20, 378–403.

Law, R., & Wolch, J. R. (1991). Homelessness and economic restructuring. Urban Geography, 12, 105–136.

Lee, B. A. (1991). The Nashville method. In C. M. Taeuber (Ed.), Enumerating homeless persons: Methods

and data needs (pp. 139–146). Washington, DC: U.S. Bureau of the Census.

Lee, B. A., & Price-Spratlen, T. (2002). The geography of homelessness in American communities: Con-

centration or dispersion? Unpublished Paper. University Park, PA: Department of Sociology,

Pennsylvania State University.

Lichter, D. T., & Crowley, M. L. (2002). Poverty in America: Beyond welfare reform. Population Bulletin,

57(2), 1–36.

Link, B., Phelan, J., Bresnahan, M., Stueve, A., Moore, R., & Susser, E. (1995). Lifetime and five-year

prevalence of homelessness in the United States: New evidence on an old debate. American Journal

of Orthopsychiatry, 65, 347–354.

Main, T. J. (1996). Analyzing evidence for the structural theory of homelessness. Journal of Urban Affairs,

18, 449–457.

Martin, E. (1992). Assessment of S-Night street enumeration in the 1990 Census. Evaluation Review, 16,

418–438.

Metraux, S., Culhane, D., Raphael, S., White, M., Pearson, C., Hirsch, E., Ferrell, P., Rice, S., Ritter, B.,

& Cleghorn, J. S. (2001). Assessing homeless population size through the use of emergency and

transitional shelter services in 1998: Results from the analysis of administrative data from nine U.S.

jurisdictions. Public Health Reports, 116, 344–352.

Morrow-Jones, H. A., & van Vliet, W. (1989). Homelessness in Colorado. In J. A. Momeni (Ed.),

Homelessness in the United States, Vol. I: State surveys (pp. 21–37). New York: Praeger.

National Institute of Mental Health. (1989). Deinstitutionalization policy and homelessness: A report to the

U.S. Congress. Washington, DC: U.S. Department of Health and Human Services.

Newman, S. J., & Harkness, J. (1999). The effects of welfare reform on housing. In S. J. Newman (Ed.),

The home front: Implications of welfare reform for housing policy (pp. 29–80). Washington, DC:

Urban Institute Press.

O’Flaherty, B. (1996). Making room: The economics of homelessness. Cambridge, MA: Harvard University

Press.

Rand McNally. (1990a). 1990 commercial atlas and marketing guide (121st ed.). Chicago: Rand McNally.

Rand McNally. (1990b). Road Atlas: United States, Canada, Mexico, 1990 (67th ed.). Boston: Rand

McNally.


Redburn, F. S., & Buss, T. F. (1986). Responding to America’s homeless: Public policy alternatives. New

York: Praeger.

Ringheim, K. (1990). At risk of homelessness: The roles of income and rent. New York: Praeger.

Robinson, F. G. (1985). Homeless people in the nation’s capital. Unpublished Paper. Washington, DC:

Center for Applied Research and Policy, University of the District of Columbia.

Rosenheck, R., Bassuk, E., & Salomon, A. (1999). Special populations of homeless Americans. In

L. B. Fosburg & D. L. Dennis (Eds.), Practical lessons: The 1998 National Symposium on Home-

lessness Research (pp. 2.1–2.31). Washington, DC: U.S. Department of Housing and Urban

Development, U.S. Department of Health and Human Services.

Rossi, P. H. (1989). Down and out in America: The origins of homelessness. Chicago: University of Chicago

Press.

Roth, D., Toomey, B. G., & First, R. J. (1992). Gender, racial, and age variations among homeless

persons. In M. J. Robertson & M. Greenblatt (Eds.), Homelessness: A national perspective (pp. 199–

211). New York: Plenum.

Schutt, R. K., & Garrett, G. R. (1992). The homeless alcoholic: Past and present. In M. J. Robertson &

M. Greenblatt (Eds.), Homelessness: A national perspective (pp. 177–186). New York: Plenum.

Shinn, M., Weitzman, B. C., Stojanovic, D., Knickman, J. R., Jiminez, L., Duchon, L., James, S., &

Krantz, D. H. (1998). Predictors of homelessness among families in New York City: From shelter

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

Slater, C. M., & Hall, G. E. (Eds.). 1992. 1992 county and city extra: Annual metro, city and county data

book. Lanham, MD: Bernan.

Smith, A. C., & Smith, D. I. (2001). Emergency and transitional shelter population: 2000. U.S. Census

Bureau, Census 2000 Special Reports, Series CENSR/01–2. Washington, DC: Government Printing

Office.

Snow, D. A., & Anderson, L. (1993). Down on their luck: A study of homeless street people. Berkeley:

University of California Press.

Stark, L. R. (1992). Counting the homeless: An assessment of S-Night in Phoenix. Evaluation Review, 16,

400–408.

Stefl, M. E. (1987). The new homeless: A national perspective. In R. D. Bingham, R. E. Green, &

S. B. White (Eds.), The homeless in contemporary society (pp. 46–63). Newbury Park, CA: Sage.

Taeuber, C. M., & Siegel, P. M. (1991). Counting the nation’s homeless population in the 1990 Census. In

C. M. Taeuber (Ed.), Enumerating homeless people: Methods and data needs (pp. 92–122). Washing-

ton, DC: U.S. Bureau of the Census.

Timmer, D. A., Eitzen, D. S., & Talley, K. D. (1994). Paths to homelessness: Extreme poverty and the urban

housing crisis. Boulder, CO: Westview.

Timmer, D. A., & Knottnerus, J. D. (1989). Homelessness in Florida. In J. A. Momeni (Ed.), Home-

lessness in the United States, Vol. I: State surveys (pp. 39–56). New York: Praeger.

Tucker, W. (1990). The excluded Americans: Homelessness and housing policies. Washington, DC: Regnery

Gateway.

U.S. Department of Health and Human Services. (1990). Expenditures and sources of funds for mental

health organizations: United States and each state, 1986. Mental Health Statistical Note No. 193.

Rockville, MD: National Institute of Mental Health.

U.S. Department of Health and Human Services. (1992). Annual Statistical Supplement, 1992, to the Social

Security Bulletin. Washington, DC: Social Security Administration.

U.S. Department of Housing and Urban Development. (1984). A Report to the Secretary on the Homeless

and Emergency Shelters. Washington, DC: Office of Policy Development and Research.

U.S. General Accounting Office. (1991). 1990 Census: Limitations in methods and procedures to include the

homeless. Washington, DC: General Government Division.

Vissing, Y. M. (1996). Out of sight, out of mind: Homeless children and families in small-town America.

Lexington, KY: University of Kentucky Press.

Weaver, R. K. (2000). Ending welfare as we know it. Washington, DC: Brookings Institution Press.

Wolch, J. R., & Dear, M. J. (1993). Malign neglect: Homelessness in an American city. San Francisco:

Jossey-Bass.


Wolch, J. R., Dear, M., & Akita, A. (1988). Explaining homelessness. Journal of the American Planning

Association, 54, 443–452.

Wong, Y. I. (1997). Patterns of homelessness: A review of longitudinal studies. In D. P. Culhane &

S. P. Hornburg (Eds.),Understanding homelessness: New policy and research perspectives (pp. 135–164).

Washington, DC: Fannie Mae Foundation.

Wright, J. D., & Devine, J. A. (1995). Housing dynamics of the homeless: Implications for a count.

American Journal of Orthopsychiatry, 65, 320–329.

Wright, J. D., & Lam, J. (1987). Homelessness and the low income housing supply. Social Policy,

17(Spring), 48–53.

Wright, J. D., Rubin, B. A., & Devine, J. A. (1998). Beside the golden door: Policy, politics, and the

homeless. New York: Aldine de Gruyter.


Determinants of Homelessness in Metropolitan Areas

Documents

Transcript of Determinants of Homelessness in Metropolitan Areas