Family Homelessness: An Investigation of Structural Effects

Abstract

Homelessness has often been attributed to personal deficits rather than to socioeconomic (structural) conditions. This study uses a unique dataset or panel containing the number of people in family units in emergency shelters in Missouri by county during five matched summer and winter dates (1993–2001). This panel contains a nearly complete enumeration of emergency shelter populations within central city, suburban, smaller metropolitan, and rural counties in Missouri on these dates. Structural effects on homelessness are assessed using generalized estimating equations. During winter and summer dates, the number of people in families in emergency shelters is positively related to unemployment rates and inversely related to general relief. Economic activity as measured by taxable sales is inversely related to family emergency shelter populations in the summer but not in the winter. Results are consistent with beliefs that family homelessness is a consequence of poor economic conditions and insufficient social welfare support.

KEYWORDS:

• Homelessness
• structural effects
• families
• generalized estimating equations
• Missouri

Introduction

This article looks at the relationship between the number of persons in families ¹ in Missouri emergency homeless shelters and structural conditions during winter and summer months in 1993–2001 (Missouri Association for Social Welfare [MASW], 1993, 1994, 1996, 1998, 2001). The investigation uses county-level data during two sets of point-in-time enumerations, one for the coldest day of a particular year (between January 1 and March 31) and one for a Monday in early summer of the same year. ² The data are a nearly complete enumeration of emergency shelter populations in Missouri, including central city, suburban, smaller metropolitan, and rural counties.

More than 14% of the population of the United States is estimated to have experienced homelessness during their lives, either by way of a stay in a homeless shelter, living on the street, or living in over-crowded and doubled-up conditions (Link et al., 1994). Not only is homelessness a more common experience than many believe but the stresses of homelessness appear to be extraordinary with long-lasting impacts. Hopper (1990, 1995) suggests that “the harrowing simplicity of the term homeless may have outlived its utility” (1991, p. 38), because a continuum of experience embedded in extreme poverty and unstable living arrangements should be the focus of discussion and research. Using New York City data, Culhane (2004) found that the largest age cohort annually experiencing homelessness is children younger than 9 years of age.

Though some research suggests that homelessness can be a consequence of economic conditions, such personal deficits as substance abuse or mental instability are more often identified as causes of homelessness. These personal failings are considered sociopathic or a result of moral turpitude. Homelessness then is a punishment justly administered. Yet, such self-righteousness appears inconsistent with the numbers of children in homeless populations.

Other observers of the homelessness phenomenon prefer to view homelessness as a combination of personal failures, contracting economic opportunity, and structural deficits in the housing supply. Such a view might seem balanced until choices have to be made about where to direct resources and attention to mitigate homelessness. At present, policy appears strongly biased toward a view of homelessness as personal failure, not a result of structural and economic factors. Consequently, policy attention and expenditures ignore relationships among employment creation (jobs and living wage policies), expansion of the low-cost housing supply, rent support, and possible non-market (i.e., public sector) solutions to problems associated with homelessness.

HOMELESSNESS AMONG POPULATIONS: ENUMERATION

Policies directed toward reducing homelessness can be strongly influenced by how the homeless population is counted and defined. The narrower the definition of homelessness and the more limited the time perspective of the count or enumeration, the more likely the homeless population identified will be smaller and will over-represent persons who are chronically homeless, whereas those who are episodically or intermittently without permanent shelter are undercounted. Point-in-time counts almost certainly will over-represent severe psychological and sociopathic conditions within the population and under-represent persons who are episodically homeless owing to economic, insufficient low-income housing, or other structural conditions. An extended discussion involving Dennis Culhane, Anna Kondratas, and others illustrates some of the foundational and methodological issues that arise when enumerating homeless populations.

Though Culhane and Metraux (1991) agreed with Kondratas (1991) that the number of people without homes at any one point in time appeared to be substantially less than advocates for homeless persons had claimed, they noted that single cross-sectional studies, cited by Kondratas (1991), though helpful in avoiding duplicated counts of homelessness, could not capture numbers and characteristics of homeless persons across time. They suggested that longitudinal counts must be done because point-in-time counts will miss many persons who pass through single or intermittent homeless experiences.

Kondratas (1994) responded that homelessness rates could be manipulated by choosing longer and arbitrary time periods for enumeration leading to overestimates of homelessness. Kondratas acknowledged the plausibility of Culhane and Metreaux's structural arguments but found evidence lacking. Kondratas suggested that the proper way to assess homelessness would require a series of point-in-time estimates or counts by consistent methods at regular intervals.

Culhane and Lee (1997) later presented a longitudinal intra-city study of both New York and Philadelphia correlating shelter admissions with neighborhood demographic characteristics. They offered their study as a contribution to a structural model of homelessness. Lee, Price-Spratlen and Kanan (2003) later used the 1990 U.S. Census S-Night enumeration employing a structural model that acknowledged individual-level proximal determinants of homelessness within an interurban context. Both studies demonstrated structural connections to homelessness using different methodologies.

The Missouri homeless shelter data, described later, offer an opportunity to use a consistent statewide series of point-in-time counts, as suggested by Kondratas, to assess whether structural conditions have profound effects on homeless populations. This assessment is done using a combined cross-sectional–time series panel analysis of families in emergency homeless shelters by county, consistent with the methodological insights of both Kondratas and Culhane. This county-level perspective incorporates data from intra-city, interurban, suburban, small metropolitan, and rural areas.

We have focused on homelessness in families in shelters, believing that this fast growing and large (in terms of lifetime incidence) segment of the homeless population is most sensitive to structural conditions (Burt & Aron, 2000; Culhane, 2004; Institute for Children and Poverty & Homes for the Homeless, 2005). Additionally, homeless families are of special interest in that family instability can have profound and long-lasting effects on persons and are a matter of grave concern to students of human development. Evidence indicates that high levels of poverty among youth in the United States tend to be maintained into later life (Rainwater & Smeeding, 2003).

HOMELESSNESS AMONG POPULATIONS: DEFINITIONS

Definitions of who should be counted as homeless among populations have evolved over time (Barak, 1991; Bassuk, Browne, & Buckner, 1996; Blau, 1992; Culhane, 1997; Hewitt, 1996; Hopper, 1991; Jencks, 1994; Kozol, 1988). The problems involved in counting homeless persons, as previously noted, are considerable; deciding who is to be considered “homeless” is a formidable but essential task. The current meaning of homelessness is constructed around the concept of a person without shelter. People literally without shelter should be counted as “homeless,” and this would include persons living on the street, in abandoned buildings, in cars, and other places not meant for human habitation.

Extensions of the meaning of “homelessness” have been more debatable. Most would consider persons living in homeless shelters to be homeless. Additionally, most would define homeless persons to include persons who would otherwise have no place to sleep and are temporarily housed in a motel or hotel by a social service provider. Somewhat more controversial inclusions in the definition of “homeless persons” are people who are temporarily doubled-up with family or friends because they have nowhere else to stay. Such persons are often counted as homeless by advocates. Counts of persons in doubled-up living arrangements were conducted in the 1990s by random telephone surveys, but there has been little research into characteristics of this population (Burt & Cohen, 1989a; Cordray & Pion, 1991; Link et al., 1995; Link et al., 1994; Rossi, 1989; Wright, 1989; Wright & Devine, 1995).

More controversial still is inclusion of persons living in inadequate, unsafe and/or unsanitary housing as “homeless.” Over-crowded/doubled-up and/or unsafe/unsanitary housing conditions are most commonly endured by people in rural areas as homeless shelters in towns and cities relieve some of these conditions (Hill, 2001; Vissing, 1996). Ellickson (1990), among others, suggests that homeless shelters may “cause” observed homelessness because these institutions entice people from marginal living arrangements that they might otherwise tolerate. Hopper's (1991) concern, however, over the “harrowing simplicity” of the term homeless distracting our attention from the housing conditions of extreme poverty seems especially well-placed in this instance and also points to the dilemma of special funding for homelessness resulting in the dumping of clients from main stream systems and alternative living arrangements (Kondratas, 1994). Indeed, some advocates have argued for shelters to be considered housing (Nunez, 2004b).

Household members who willingly leave their homes to escape domestic violence are often viewed as homeless, and the U.S. Department of Housing and Urban Development (HUD) allows homelessness funds allocated under the McKinney/Vento Act to be spent to shelter survivors of domestic violence. HUD also allows McKinney/Vento funds to be used for short-term rent or utility assistance to avoid utility shutoff or eviction. In addition, HUD's McKinney/Vento funds provide permanent housing for homeless people with disabilities and for persons who are chronically homeless, defined to be single persons with disabilities who have experienced long-term homelessness or multiple episodes of homelessness (Applebaum, 1987; Baum & Burnes, 1993; Cordray & Pion, 1991; Cousineau & Ward, 1992; Edin, 1992; Hopper, 1991, 1995; Horowitz, 1989; Marcuse, 1990; Rossi, 1987, 1989; Rossi, Wright, Fisher, & Willis, 1987; Straw, 1995; White, 1992).

Broadly speaking, the category of “homelessness” has been used to encompass three populations: people who are (1) sheltered by an agency either with provision of an actual place of habitation or financial support, (2) people who are living on the streets and in the rough, and (3) people doubled up with relatives or acquaintances owing to lack of regular shelter. These populations can overlap; persons sleeping in the rough one night may be in a shelter the next. People without regular and adequate shelter, however, are likely to remain in only one of the foregoing categories in a single evening.

People tend to experience homelessness in three ways. First, people who are chronically homeless spend a substantial portion of their lives (even until death) in unstable shelter arrangements. They tend to be single and male. At any one time, the chronically homeless population may be 20% of the sheltered population and much of the unsheltered homeless population. Second, there are people who will experience homelessness episodically and may in time either find safe harbor or join the ranks of the chronically homeless population. These people may be 35% of the sheltered population at any one time. Third, the rest of the people who experience homelessness do so briefly prior to transitioning to more stable living arrangements. Transitioning may take some people a short time, and others much longer. The actual percentages of people in these categories will vary by methodology and working definitions applied to research (Burt et al., 1999; Culhane & Kuhn, 1997).

The relative proportion of these groups in the homeless population also appears to change periodically, although the absolute number of homeless persons has grown over time. Though the chronically homeless population tends to be primarily single males, the episodically homeless population appears to have varying proportions of people who are single or in familial groups. The transitory homeless population tends to be people in families and single persons who come out of doubled-up living situations (Burt, 2001; Burt & Cohen, 1989a, b; Link et al., 1994; Link et al., 1995; Phelan & Link, 1999; Rossi, 1989; Wright, 1989).

Many ethnographic studies document the debilitating process of the loss of ability of persons to negotiate their environments which often precedes homelessness (Connolly, 2000; Friedman, 2000; Hill, 2001; Kozol, 1988; Newman, 1999; Nunez, 2004a; Vissing, 1996). Homelessness itself is seriously traumatizing (Jahiel, 1992a). Effects are manifested in mental and physical illnesses, addictions, and in other dysfunctions (Blau, 1992; Downer, 2001; Van Ry, 1993; Whitman, Accardo, & Sprankle, 1992). People living on the economic margin who already have a serious mental or physical (often undiagnosed) illness are at-risk for chronic and episodic homelessness (Grigsby, Baumann, Gregorich, & Roberts-Gray, 1990; Koegel, Burnham, & Farr, 1988; Kuhn & Culhane, 1998).

Given existing economic and social structures, many now believe that 15% to 20% of the U.S. population is vulnerable to possible homelessness. Any disadvantage in the competition for a shrinking and inadequate supply of low-cost housing can force persons or families into homelessness. (Burt, 1991, 1992a, b; Culhane & Lee, 1997; Culhane, Lee, & Wachter, 1996; Hamburg & Hopper, 1992; Hopper, 1991; Hopper, Susser, & Conover, 1985; Jahiel, 1992a, b, c; Koegel, 1992; McChesney, 1990). Indeed, experiences of homelessness may precede and contribute to pathologies often considered to be causes of homelessness. As noted, the dataset available for this study permits the effects of structural and economic conditions on homelessness among families in emergency shelter to be directly assessed.

OBSERVATIONS ON DEBATES ABOUT CAUSES OF HOMELESSNESS

Despite increasing insights into structural problems and interrelationships among employment, housing supply, and housing costs that have arisen during debates about counts and definitions of homelessness, the prevailing perspective toward homelessness remains focused on point-in-time counts with a heavy emphasis on individual pathologies rather than social and economic conditions (Bane & Ellwood, 1986; Gladwell, 2006; Shinn, 1992; Sosin, Piliavin, & Westerfelt, 1990). Many homeless people living on the streets are conspicuous, troubled, and more often seen by the public than those in shelters or substandard housing (Snow, Andersen, & Koegel, 1994). Studies of homeless people have chronicled pathology and disaffiliation (Arce, Anthony, Vergare, & Shapiro, 1983; Argeriou & McCarty, 1990; Koegel et al., 1988; Scanlon, 1989; Shinn, Knickman, & Weitzman, 1991), whereas Hollywood has plotted comedies and tragedies around pathology and dysfunction. Indeed, one suspects that pathology, dysfunction, and eccentricity capture public attention more than complex associations among structural barriers to decent housing and improved living conditions.

The numbers of people who are “homeless” has been hotly contested, with advocates seemingly exaggerating limited data and others assuming that true homelessness was rare. Point-in-time counts of the number of homeless persons reported have ranged from 200,000 to more than 3,000,000 persons (Burt & Cohen, 1989a; U.S. Department of Housing and Urban Development, 1984; Hombs & Snyder, 1982; Kondratas, 1991). Link et al. (1994) in a well-done study found a surprisingly high lifetime and 5-year prevalence of homelessness. Their study was based on a telephone survey of 1,507 households and, despite potential undercounting using this methodology, 14.0% of the total sample (by projection, 26,000,000 people) had been homeless at some point in their lives, and 4.6% of the sample (by projection, 8,500,000 people) had been homeless during the 5 years prior to the interview. When the investigators used only people in shelters or on the streets as reference points, the lifetime incidence of homelessness was estimated to be 7.4% (by projection, 13,500,000 people) with a 5-year incidence of 3.1% (by projection, 5,700,000).

At nearly the same time, Culhane & Lee (1997) also released a period prevalence study of turnover rates for shelters in New York City (1988–1992) and Philadelphia (1990–1992). They found that during 1992, approximately 1% of each city's population had used its shelters. During the 3 years of 1990 to 1992, not quite 3% of Philadelphia's population requested shelter whereas more than 2% of New York's population used shelters. Of New York's population, 3.27% passed through their public shelter system during the 5 years of the study. Burt (1994) found these measures of annual incidence and prevalence to be consistent with data from other parts of the country. Burt (2000) estimated a point-in-time range of 446,000 to 840,000 homeless persons, noting that structural problems were clearly indicated by such high incidence.

The fact that such high estimates of prevalence and incidence of homelessness, however defined, have been found in multiple studies raises grave doubts about attributing homelessness solely or, perhaps, even primarily to personal failures, pathology, or sheer madness. Structural effects and public and private policies that sustain what appears to be growing numbers of persons, including families and children, without regular shelter require intensive study. The Missouri data allow us to examine some structural effects in a new and more comprehensive fashion.

The General Research Question

The overriding or general research question addressed in this study is whether the number of people in familial groups in emergency shelters in Missouri counties correlates with structural or economic factors, including changes in public income maintenance support, during the five winter and the five summer dates of the MASW Censuses of Shelter Providers for Homeless People in Missouri (1993, 1994, 1996, 1998, 2001). Recognizing that answering such a general question is difficult, seven specific study hypotheses are addressed.

Specific Study Research Hypotheses

The specific research hypotheses examined for winter and summer sets of data are whether:

1. The number of homeless people in familial groups in emergency shelters in Missouri counties is positively associated with the unemployment rate.

2. The number of homeless people in familial groups in emergency shelters in Missouri counties is negatively associated with the taxable sales revenue per population.

3. The number of homeless people in familial groups in emergency shelters in Missouri counties is positively associated with fair market rents. ³

4. The number of homeless people in familial groups in emergency shelters in Missouri counties is positively associated with the number of beds in shelters per county population. ⁴

5. The number of homeless people in familial groups in emergency shelters in Missouri counties is positively associated with: (a) the size of population in the county and (b) population growth.

6. The number of homeless people in familial groups in emergency shelters in Missouri counties is inversely associated with General Relief, a proxy measure for income maintenance benefits. ⁵

7. The number of homeless people in familial groups in emergency shelters in Missouri counties is lower in the earlier 1990s than in later slower periods of economic growth such as 2001.

Methods

Data Used in this Study: The Missouri Censuses of Homeless Shelters

As a result of earlier work (Low Income Housing Task Force, 1987; Task Force on the Homeless Mentally Ill, 1988; Task Force on Low Income Housing, 1989; Task Force on Survival, 1985), the Missouri Housing Development Commission contracted with the MASW to provide the State of Missouri with data for their homelessness reports to HUD. The MASW (1993, 1994, 1996, 1998, and 2001) conducted five censuses engaging the same principle investigator for all five. These censuses of shelters were conducted for summer and winter dates.

Each time an inventory of all shelters in the state was created, systematic queries were done of the number of people sheltered and their familial status. Also collected was information as to the relative percentages of subpopulations served and open-ended comments to questions. Data reported concerning numbers and familial status were obtained from shelter records.

The survey method used two point-in-time counts. Data were collected regarding numbers of individuals and families served during the coldest day in Missouri between January 1 and March 31 and for a Monday in early summer. This study reports on only the numbers of people in families in emergency shelters in Missouri, by county, during the winter dates and the summer dates of the years mentioned. Data were obtained by county for some structural factors or variables and income maintenance benefits during the same time periods as the surveys. The units of analysis used in this article are Missouri counties that had emergency homeless shelters during any of the five censuses of homeless shelters.

Census return rates ranged between 93.8% and 100% (mean, 98%) of the emergency homelessness shelters in the State of Missouri. There are 114 counties in Missouri plus the City of St. Louis. Of these 115 entities, 52 had at least one emergency shelter during one of the five censuses. In total, there were 198 different shelters operating at some time during the five censuses; the number of shelters during any one census period ranged from 90 to 123.

This study uses census information about the number of persons in family groups in emergency shelters during these censuses. The data available for analysis, therefore, consist of a panel of shelters for the 5 years mentioned organized by county. It is a censored data set, meaning that some shelters did not function at all time periods during which the census was taken.

Statistical Procedures

Statistical procedures and the software used permitted corrections to be made in standard errors for clustering by county and for censoring in the data set (StataCorp, 2006). Two panel analyses, one for winter and one for summer, were done using generalized estimating equations (GEE) ⁶ and semi-robust standard errors.

GEE are often used with panel data to correct for correlated errors over time and clustering. Corrections using GEE allow parameters to be estimated for independent variables controlling for longitudinal effects and clusters, in this case clustering by county (Hardin & Hilbe, 2003). GEE are considered the longitudinal procedure of choice in panel analysis, except where systematic relationships with time are anticipated (Duncan, Duncan, & Strycker, 1999). The data set for this study has been examined, and there is no evidence of systematic relationships among the dependent variable, the independent variables, and time.

GEE analysis also conveniently allows standard errors of coefficients to be corrected for correlated effects. This produces what are called “semi-robust” estimates of standard errors. Without such corrections, estimates of standard errors are biased, and tests of statistical significance are unreliable.

For the purposes of this study, the probability at which null hypotheses (H0) are to be rejected is p ≤ .05. Variables that had a one-tailed level of significance of p ≤ .05 are noted because, despite convention, most of the hypotheses examined in this study are one-tailed in that a specific directional relationship was anticipated. Tables herein report only z statistics and p-values; more detailed information can be obtained from the authors.

Study Variables Defined for the GEE

The dependent and independent variables used in this study are exhibited and defined below. Unfortunately, good structural variable data are often limited, and this is the case in this study. A number of the independent variables also might be considered “covariates.” For example, County Emergency Shelter Beds per Mean Population serves as a covariate allowing the other GEE parameters to be estimated while statistically controlling for bed availability.

Dependent variable

Individuals in Families in Shelters per Mean County Population is the number of individuals in family units in emergency shelters (MASW, 1993, 1994, 1996, 1998, 2001) divided by the mean county population in 1990 and 2000 (Missouri Department of Economic Development, 2003a). A family is defined in the MASW censuses as having at least one parent or guardian and one child younger than 18, or a homeless pregnant woman, or a homeless person in the process of securing legal custody of a person younger than 18, or children younger than 18 who have children.

Independent variables or covariates

Economic factors

County Unemployment Rate is the rate of unemployment for that county and year (Missouri Department of Economic Development, 2003b).

County Taxable Sales per Mean Population is the value of taxable sales for the fiscal quarter of a particular census date for that county (Missouri Department of Revenue, 2003) divided by the county population, which is the mean of the 1990 and 2000 federal censuses.

County Fair Market Rent is the Fair Market Rent (U.S. Department of Housing and Urban Development [HUD], 2003) as set by HUD for a two-bedroom apartment for a particular county and time period. ³

Population factors

County Emergency Shelter Beds per Mean Population is the number of emergency shelter beds of a county (MASW, 1993, 1994, 1996, 1998, 2001) divided by the average population of that county between the 1990 and 2000 federal censuses (Missouri Department of Economic Development, 2003a). ⁴

County Population Average is the mean of the county population in the 1990 and 2000 U.S. censuses (Missouri Department of Economic Development, 2003a).

County Population Growth Rate is the rate of growth for a county between 1990 and 2000 federal censuses (Missouri Department of Economic Development, 2003a).

Income maintenance factor

County General Relief is the dollar amounts for that county and year and “consists largely of general assistance, refugee assistance, foster home care and adoption assistance, earned income tax credits, and energy assistance.” (U.S. Department of Commerce, 2003, footnote 5 for table CA35 Local Area Personal Income)

Year effects indexed to 2001

Y93-Y94 are dichotomous variables (0, 1) that measures year effects on the dependent variable by comparing 1993, 1994, 1996 and 1998 to 2001, thereby statistically “controlling” for year effects on all variables in the GEE.

STUDY RESULTS

The study results are reported further by winter and summer season. As noted, two GEE were estimated, one for each season. (Study hypotheses are stated on page 177).

Winter Results

Four hundred and one observations were available across 48 counties; some shelters existed in summer only and some in winter only. ⁷ A minimum of 1 observation or shelter per county and a maximum of 106 observations per county, with an average of 8.4 observations per county, were available for this panel analysis. The Wald χ² was 124.06, with a p ≤ .0001 that the independent variables are unrelated to the dependent variable in the GEE. Table 1 summarizes winter results.

Economic Factors

1. The unemployment rate is positively associated with the number of people in families in emergency homeless shelters per county, adjusted for county population in Missouri. This variable has a two-tailed probability (p-value) less than .001 and a z score of 4.14. The hypothesis is supported. Unemployment rates appear to be strongly associated with the number of people in families in shelters.

   TABLE 1 GEE Results for the Winter (1993–2001; n = 401)

   Independent variables	z	p
   Economic factors
   County Unemployment Rate	4.14	<.001
   County Taxable Sales/Mean Population	.66	.509
   County Fair Market Rent	.60	.548
   Population factors
   County Emergency Shelter Beds/Mean Population	8.13	<.001
   County Population Average	1.09	.275
   County Population Growth Rate	−1.14	.255
   Income Maintenance factor
   County General Relief	−2.67	.008
   Year Effects Indexed to 2001
   1993	−3.08	<.002
   1994	−1.91	.028*
   1996	−1.55	.120
   1998	−0.91	.365
   Wald χ^2 = 124.06, p < .0001Dependent Variable: Individuals in Families in Shelter per Mean County Population

2. The taxable sales per population are not associated with the number of people in families in emergency homeless shelters, adjusted for covariates. This hypothesis is not supported.

3. The fair market rents are not associated with the number of people in families in emergency homeless shelters, adjusted for covariates. This hypothesis is not supported.

Population Factors

1. The number of emergency shelter beds per mean population of a county is positively and significantly associated with the number of families in emergency homeless shelters. The z score is 8.13 with a p-value less than .001. The association between bed availability and family use of shelters is very strong. This hypothesis is supported.

2. Neither size of population in the county nor population growth is associated with the number of families in shelters during the winter months. Neither (a) nor (b) of this hypothesis is supported.

Income Maintenance

1. General relief is inversely and significantly associated with the number of families in emergency homeless shelters. The z score is −2.67, and the p-value is .008. General relief is strongly associated with a reduction of the number of families in shelters during the winter. This hypothesis is supported.

Year Effects

1. Using 2001 as the index year, the numbers of families in shelters was inversely and significantly lower in 1993 and 1994 with negative but non-significant coefficients in 1996 and 1998. Z scores and p-values ranged from a high of −3.08 to a low of −0.91, with p-values ranging from 0.002 to 0.365.

Summer Results

Four hundred and one observations were available across 49 counties; some shelters existed in summer only and some in winter only. ⁷ A minimum of 1 observation per county and a maximum of 102 observations per county, with an average of 8.2 observations per county, were available. The Wald χ² was 588.69, with a probability (p-value) less than .001 that the independent variables are not associated with the dependent variable. Table 2 summarizes summer results.

Economic Factors

1. The unemployment rate was positively associated with the number of people in families in emergency homeless shelters per county, adjusted for county population in Missouri and other covariates. This variable had a two-tailed p-value of .020 and a z score of 2.33. The hypothesis is supported. Unemployment rates were associated with the number of people in families in shelters.

   TABLE 2 GEE Results for the Summer (1993–2001; n = 401)

   Independent variables	z	p
   Economic factors
   County Unemployment Rate	2.35	.020
   County Taxable Sales/Mean Population	−1.81	.035*
   County Fair Market Rent	−.49	.627
   Population factors
   County Emergency Shelter Beds/Mean Population	22.19	<.001
   County Population Average	3.06	.002
   County Population Growth Rate	−.16	.874
   Income Maintenance factor
   County General Relief	−2.80	.005
   Year Effects Indexed to 2001
   1993	−3.79	<.001
   1994	−3.12	.002
   1996	−2.72	.007
   1998	−1.21	.226
   Wald χ^2 = 588.69, p < .001Dependent Variable: Individuals in Families in Shelter per Mean County Population

2. The taxable sales per population were negatively and significantly associated with the number of people in families in emergency homeless shelters. This variable had a one-tailed p-value ≤.0355 and a z score of −1.81. The hypothesis is supported but mildly, controlling for the other independent variables. This proxy measure of the level of retail economic activity showed a modest association with a decrease in the number of people in families in shelters during the summer.

3. Fair market rents were not associated with the number of people in families in emergency homeless shelters, adjusted for covariates. This hypothesis is not supported.

Population Factors

1. The number of emergency shelter beds per mean population of a county was significantly associated with the number of families in emergency homeless shelters. The z score was 22.19, with a p-value less than .001. The association between bed availability and family use of shelters in the summer was very strong. This hypothesis is supported.

2. The larger the population of the county, the larger the number of people in families in emergency homeless shelters, adjusted for covariates. The z score was 3.06, and the p-value was .002. The number of families in shelters was associated with county population size. Population growth, however, had no significant association with the number of families in shelters. In this study, part a of the hypothesis is supported; part b is not.

Income Maintenance

1. General relief was inversely and significantly associated with the number of families in emergency homeless shelters. The z score was −2.80, with a p-value of .005. General relief was strongly associated with a reduction of the number of families in shelters during the summer. This hypothesis is supported.

Year Effects

1. Using 2001 as the index year, the numbers of families in shelters was inversely and significantly lower in 1993, 1994, and 1996, with a negative but non-significant coefficient in 1998. Z scores and p-values ranged from a high of −3.79 to a low of −1.21, with p-values of <.001 to .226.

SUMMARY AND DISCUSSION

Economic Factors

The Unemployment Rate is related to the number of homeless people in family groups in Missouri emergency shelters by county for both winter and summer dates, controlling for the other variables in the study. This relationship is positive in that as unemployment increases, so do the number of people in families in emergency shelters.

Fair Market Rents are not significantly related to the number of homeless people in family groups in Missouri emergency shelters by county for either the winter or summer dates, controlling for the other variables in the study. The market fluctuations of rental costs at the fortieth and fiftieth percentiles may not appropriately describe housing markets faced by homeless families. Housing rents also are likely to be a poor proxy measure of housing available to potentially homeless people. Indeed, there may not be an effective market for housing accessible to low-income and unemployed persons in Missouri, or elsewhere for that matter. Therefore, housing markets may be segmented with Fair Market Rents unrelated to demand or need of low-income persons.

Taxable sales are not related to the number of homeless people in family groups in Missouri emergency shelters by county during the winter dates but are related to the number of sheltered people in families in the summer, controlling for the other variables in the study. This finding is consistent with data that suggest many families who are taken in by family and friends during the cold season will leave in search of work in warmer weather. If the economy is more robust, these families will have a better chance of finding employment that will support them in housing when they leave their winter “refuge.” Additionally, employment opportunities among poorly sheltered persons may not be correlated with winter taxable sales if these job opportunities arise in informal and irregular employment sectors that are seasonally related to better weather in summer months. A further possibility is that there is likely to be more competition between the single street population and families for shelter space during the winter dates regardless of the robustness of the economy. This could attenuate correlations between winter months, taxable sales, and the number of persons in families in shelters.

Population Factors

The population growth rate of counties is not related to the number of people in families in emergency shelters by county during winter or summer dates, controlling for the other variables in the study. This finding may arise because faster-growing counties do not have more shelter space than slower-growing counties of the same size. Overall shelter availability may not respond quickly to population growth. Shelter availability was found to be one of the strongest variables associated with the number of families in shelters.

The number of people in families in emergency shelters by county is related to the size of a county's population during summer months in Missouri controlling for the other variables in the study but not for winter months. This finding is somewhat contrary to popular perceptions that are often shaped by the number of homeless people who are visibly discernable in the larger cities and towns (i.e., on the streets). The homeless people who are visible in public locations are probably not in family groups, and many of these people may not even use emergency shelters.

The number of bed spaces per county population is directly related to the number of people in families in emergency shelters by county during both winter and summer months in Missouri, controlling for the other variables in the study. This finding suggests that as more beds are available, more people in families appear to fill them. This could be indicative of the large size of an inadequately housed population who are living in overcrowded and doubled-up conditions. Even emergency shelters may be preferable to present living conditions for many families when shelters and shelter space are available.

Income Maintenance Benefits

County General Relief was inversely related to the number of people in families in Missouri emergency shelter on both the winter and summer dates. For both winter and summer, as the amount of county payments for general relief decreased, families in emergency shelters increased. It appears that relief for which County General Relief is a proxy measure in our study tends to reduce homelessness by families in shelters. Further research is needed on the cost-benefits of public assistance, adequate housing, and long- and short-term effects of homelessness on families.

Year Effects

The number of people in families in Missouri counties in emergency shelters was significantly lower in most periods prior to 2001, controlling for the other independent variables. Indeed, the number of people in families in Missouri counties in emergency shelters was significantly lower in 1993, 1994, and 1996 relative to 2001 for the summer dates, controlling for the other independent variables. The data also are strongly consistent with an economic explanation for homelessness among families, as in each of the years from 1993 through 2001, the pattern and size of GEE coefficients suggested smaller sheltered populations of families when rates of economic growth were rising in the 1990s (Stiglitz, 2003).

LIMITATIONS OF THE STUDY

Even in this unique data set, the data are insufficient to examine more than a few effects of structural conditions on homelessness. Moreover, the data are only from Missouri, one state; however, it should be noted that Missouri is diverse in terms of its demographics, and it is often identified as a bellwether state for that reason. Data from individual occupants of shelters are not available; therefore, the analysis relies heavily on shelter records aggregated at a county level and county socioeconomic and population data. The analysis, therefore, may be subject to ecological biases, but this is difficult to determine owing to generally inadequate longitudinal data from individual shelter occupants. The analysis examines only emergency shelter use by families as defined in the MASW censuses (1993, 1994, 1996, 1998, and 2001); other data might have permitted examination of populations in other destabilized living arrangements. One consequence of this limitation is that the relative weights or contributions of structural versus individual or person-specific factors contributing to homelessness cannot be directly assessed. Thus, the results reported in this study illustrate the absolute—not relative—importance of structural variables on homelessness in families.

CONCLUSION

The theory that family homelessness is related to economic conditions is strongly supported by this study. More attention should be given to how existing economic and other public policies can result in increases in family homelessness, contributing to the further growth of chronic homelessness, and how better policies and a more equitable economy may alleviate homelessness and consequent social ills.

Notes

Note. *One-tailed test; otherwise all tests are reported as two-tailed.

Note. *One-tailed test; otherwise all tests are reported as two-tailed.

1. A family is defined in the MASW censuses as having at least one parent or guardian and one child younger than 18, or a homeless pregnant woman, or a homeless person in the process of securing legal custody of a person. The number of individuals in family units came from the MASW censuses and is recorded by shelter by county. The number of people and their familial status were among those questions answered from shelter records. The county population used in this study is the mean of the 1990 and 2000 U. S. censuses.

2. The coldest day, January 1 through March 31, statewide, was ascertained ex post from National Weather Center records. Shelter directors indicated that occupancies during the summer were higher on Mondays. Four of the census Mondays were the last in June; the first Monday in July was used in 1993.

3. FMRs are generally set one time per year though there were two years in this study in which rents were set twice. FMRs are based on the preceding U. S. decennial census, adjusted with information from the American Housing Surveys (AHS) and random digit dialing (RDD) telephone surveys. FMRs are generally set at the fortieth percentile rent drawn from the distribution of rents of all recent movers. Newly built housing (less than 2 years old), public housing and substandard housing are excluded. FMRs include all utilities except telephone. Starting in January 2, 2001, 39 areas had their FMRs increased to the fiftieth percentile to “promote residential choice, help families move closer to areas of job growth, and deconcentrate poverty” (U. S. Department of Housing and Urban Development, 2003b, p. 60084). The St. Louis and Kansas City metropolitan areas were included among the 39 areas whose FMRs were increased to the fiftieth percentile.

4. The number of shelter beds in a county, as with the number of people in families in a shelter, is taken from shelter records for the MASW censuses.

5. Martha Burt, principal research associate in the Urban Institute's Center on Labor, Human Services and Population, reviewed our initial test results that showed no relationship between the dependent variable and the income maintenance independent variables (personal communication, November 12, 2004). She suggested that Food Stamps, Family Assistance, and SSI would be invariant at the county level and might mask possible relationships with General Relief. Further tests revealed a high degree of correlation among the income maintenance variables. We therefore included only General Relief in this analysis.

6. GEE was described by Liang and Zeger (1986) for use with correlated cross-sectional longitudinal data sets.

7. Owing to the availability of socioeconomic and demographic data, the county is the unit of analysis. Observations of shelters are aggregated by county.