| Home Page | Resume | Graduation |





Submitted to the Graduate School of Wayne State University, Detroit, Michigan in partial fulfillment of the requirements for the degree of




List of Tables
Evidence for Psychosocial Influences on Health
Health Among the Homeless
Longitudinal Studies of Psychosocial Factors
and Health

Rationale for this Longitudinal Research
Data Analysis
Main Analyses: Baseline Predictors of Health
Main Analyses: Longitudinal Predictors of Health
Main Analyses: Baseline Predictors of Health
Main Analyses: Longitudinal Predictors of Health



Table 1: Background characteristics of homeless adults
Table 2: Final baseline sampling design
Table 3: Intercorrelations among key variables
Table 4: Most frequently endorsed recent symptoms and lifetime health history at baseline
Table 5: T-test assessing change over 12-month follow-up period in illness symptoms
Table 6: Hierarchical multiple regression analysis on baseline recent symptoms: Health behaviors variable set entered last
Table 7: Hierarchical multiple regression analysis on baseline recent symptoms: Psychological variable set entered last
Table 8: Hierarchical multiple regression analysis on baseline recent symptoms: Social variable set entered last
Table 9: Hierarchical multiple regression analysis on recent symptoms at follow-up: Health behaviors variable set entered last
Table 10: Hierarchical multiple regression analysis on recent symptoms at follow-up:
Table 11: Hierarchical multiple regression analysis on recent symptoms at follow-up: Social variable set entered last

To Table of Contents



I would like to dedicate this dissertation to all the homeless people who participated in the research and to the hope that fewer people everywhere will experience episodes of homelessness in the future. I hope this research can contribute to reaching that goal.


I would like to express my thanks to all staff who were instrumental in collecting the data analyzed in this study, such as research assistants and staff at the various facilities where interviews were completed. I am grateful to my mentor and advisor, Paul A. Toro, for all of his help and guidance. I am inspired by his dedication to field research and his interest in issues of social concern. My thanks also go to Mark Lumley, Kathryn Urberg, and Elizabeth Chapleski, whose insightful comments and probing questions challenged me to think more deeply about my findings. Finally, I would like to thank my friends and family for their support, particularly my husband, Stephen Goodfellow, who helps me keep things in perspective and constantly reminds me through his everyday actions that the most important goal is to follow your heart.


It is now well documented that homeless individuals generally suffer from a much greater number of health problems than do members of the housed population (Jahiel, 1992). Such an observation would probably strike most people as intuitively accurate. However, the details and extent of the problem require careful investigation and documentation in order to accurately inform public policy and effectively intervene in the various difficulties faced by the homeless (Kondratas, 1991).

In compiling an overview on the health of the homeless, it is also important to consider more general empirical evidence of the consistent relationship between health and economic privilege (Adler et al., 1994). Within this context, homelessness can be viewed as an extremely low point on the continuum of socioeconomic status (SES). For example, it is well known that homelessness is correlated with marked poverty (Rossi et al., 1986) and that SES is consistently associated with health outcomes.

A recently published study (Sorlie, Backlund, & Keller, 1995) examined data regarding SES from the Census Bureau's interviews with 530,000 adults, who were then followed longitudinally for between four and eleven years. Results revealed that mortality during the follow-up period was significantly predicted by income, education, and occupational status, which are the indicators of SES most frequently used for research purposes. Those with less income, education, and occupational attainment were more likely to die. Also, employment status (e.g. employed, unemployed, houseworker, etc.) was associated with mortality, such that those not in the work force had a greater chance for death.

The above results are consistent with those of many other studies examining the link between SES and health, as reviewed by Adler et al. (1994). As the authors of this review point out, the association with mortality and morbidity is a graded relationship at all levels of SES, not just the highest and lowest extremes. In other words, groups with a small SES discrepancy between them show a parallel differential in their morbidity and/or mortality outcomes. While examining health among the homeless, therefore, it must be remembered that, based on the results of countless previous studies, their health would be expected to be poor, given their low ranking on the SES continuum. The factors responsible for the relationship between SES and health are not well understood, however, since SES is often treated as a control variable in statistical analyses.

Therefore, the following review will examine various areas of research. First, psychosocial factors that may influence health outcomes across the socioeconomic continuum will be discussed. Second, research focusing on the health of homeless adults and comparing it cross-sectionally to the general population and to samples of poverty-stricken housed individuals will be reviewed. Third, longitudinal studies on health among the general population, low SES samples, and homeless individuals will be reviewed. Fourth, research questions on the health of the homeless that require further investigation will be suggested and the rationale for and goals of the present longitudinal study will be offered.

To Table of Contents

Evidence for Psychosocial Influences on Health

Adler et al. (1994) suggest certain factors that could at least partially explain the relationship between health and SES, given that they show linear associations with both SES and important health outcomes. For example, some behaviors have been clearly linked to health and to one or more of the most frequently used indicators of SES (i.e., education, income, and employment status), and therefore, may be important variables to examine in the context of homeless people's health.

One of these behaviors is smoking, which shows an inverse relationship with SES and a direct relationship with morbidity and mortality, most commonly due to cardiovascular disease or cancer (Escobedo et al., 1990; Marmot et al., 1991; Pugh et al., 1991; Remington et al., 1985; Seccareccia, Menotti, & Prati, 1991; Winkleby, Fortmann, & Barrett, 1990).

Another behavior often listed as a health risk is alcohol consumption, although data on its relationship to health outcomes and SES are less consistent than for smoking. Although alcohol may serve as a protective factor against cardiovascular disease when consumed in moderation, alcohol abuse increases risk of liver disease, some cancers, a variety of neurological disorders, and premature death (Adler et al., 1994; Charness, Simon, & Greenberg, 1989; Smith, Cloninger, & Bradford, 1983). Some studies have found correlations reflecting greater alcohol consumption among those of higher socioeconomic standing (Cauley et al., 1991; Marmot et al., 1991; Matthews et al., 1989). Other research provides evidence that 35% of poor people meet criteria for alcohol abuse (Toro, Bellavia, et al., 1995), which is double the rate of 16.4% found in the general population (Regier, et al., 1988).

There has also been extensive research done on psychological factors that may affect health. For example, there is evidence that depression is related to health, most notably cardiovascular disease (Booth-Kewley & Friedman, 1987). Both major depression and general depressive symptoms also show a direct relationship with SES (Kaplan et al., 1987; Murphy et al., 1991).

Another psychological construct that has been investigated in relation to SES and health status is hostility, which is generally thought of as antagonistic behavior towards others based on cynicism and mistrust (Barefoot, Dodge, Peterson, Dahlstrom, & Williams, 1989). Hostility has been found to be inversely related to SES across the continuum of education, occupational status, and income (Barefoot et al., 1991; Scherwitz, Perkins, Chesney, & Hughes, 1991). Significant relationships have also been demonstrated between hostility and hypertension (reviewed in Diamond, 1982), peripheral arterial disease (Joesoef, Wetterhal, DeStafano, Stroup, & Fronek, 1989), and coronary artery disease (reviewed in Diamond, 1982; Barefoot et al., 1983).

Other factors that have been investigated in connection with health include those that reflect the interaction of the individual with influences in the social environment, such as stress (or negative life events) and social support. People experiencing or perceiving stress have been shown to be more susceptible to gastrointestinal disorders (Harris, 1991), cardiovascular disease (Tofler et al., 1990), stroke (Harmesen et al., 1990), and infectious diseases (Cohen, Tyrell & Smith, 1991, 1993; Stone et al., 1992). More stressful life events or greater perception of stress also tend to be reported among lower SES groups than among those with higher SES (Cohen & Williamson, 1988; Dohrenwend & Dohrenwend, 1970; Dohrenwend, 1973; Kessler, 1979; McLeod & Kessler, 1990).

Social support has been shown to buffer the negative effects of stressful life events on health status (Cohen & Wills, 1985; House et al., 1991). Evidence also exists that educational level (the only SES variable that has been analyzed in this context) is inversely related to social support (Ruberman, Weinblatt, Goldberg, and Chaudhary, 1984). In summary, smoking, alcohol abuse, depression, hostility, stress, and social isolation are some factors that may are involved in the relationship between SES and physical health.

To Table of Contents

Health Among the Homeless

In studying the physical health of the homeless, various methods have been employed. Some of the largest samples and most extensive geographical diversity can be found in the data from clinical studies, which gather data on people seeking health care services, namely the Johnson-Pew (Wright & Weber, 1987) and the McKinney (Lewin, 1989) Health Care for the Homeless programs, which offer information on nearly 24,000 and 183,000 homeless adults, respectively. Of the almost 132,500 homeless individuals from the McKinney program whose housing status was precisely recorded, 46.7% were in shelters, 10.7% were in transitional housing, 11.2% were in doubled-up households, and 12.7% were on the street , which reflects a fairly broad representation of homeless people.

Clinical studies' estimates of illness suffer from selection and diagnostic biases, however, since they usually focus on individuals who had some reason to present themselves to receive care. Therefore, clinical samples can be expected to under-represent healthy people, sick people who don't seek care, and individuals who have access to other sources of care. Also, the data that are gathered are not typically standardized, but instead, vary based on the presenting complaint, such that information across individuals is not uniform or comparable (Jahiel, 1992). In addition, a large number of people in clinical samples are seen only once to address a specific, acute concern, thereby leaving chronic conditions undocumented. Therefore, it is only the people with several visit's worth of data who supply more accurate estimates of longer-standing problems (Wright & Weber, 1987).

Another method of collecting data on the health of the homeless has been through population-based studies (i.e., participants not seeking formal services are approached by the researchers). Such studies usually involve smaller samples than do the treatment studies and are restricted to sampling from a few sites in one city, thereby limiting generalizability. They also employ a variety of methods for obtaining self-reported health information (Fischer & Breakey, 1987; Gelberg & Linn, 1988; Ropers & Boyer, 1987). There is evidence that homeless people tend to underestimate the severity of their health problems and that they respond more accurately to questions about specific symptoms versus open-ended questions asking them to describe or evaluate their health (Jahiel, 1992). In addition to self-reported information, some studies include physical examinations and/or laboratory work of varying comprehensiveness (Gelberg & Linn, 1989; Harris, Mowbray, & Solarz, 1994; Jahiel, 1992). According to Jahiel (1992), if physical examinations and laboratory work are done in a thorough manner, relatively minor health problems will be reported at higher rates than they would be among clinical samples. Thorough examinations are also more effective in documenting chronic conditions than are acutely-oriented clinical visits.

Despite differing methodologies, however, all studies on the health of the homeless report rates of both chronic and acute health problems to be much higher, overall, than those recorded for samples from the general population, such as the National Ambulatory Medical Care Survey (Wright & Weber, 1987). For example, some of the most commonly found problems include mild to severe upper respiratory infections, infestations such as lice and scabies, injuries, dental problems such as losing teeth, skin disorders, and peripheral vascular disorders. Given the relative youth of adult homeless samples (median ages are generally in the mid-30's), health problems whose prevalence increases with age, such as hypertension, chronic obstructive pulmonary disease, liver disease, congestive heart failure, and chronic bronchitis, are substantially over represented among the homeless (Gelberg & Linn, 1990, 1992; Jahiel, 1992).

Overall, it appears that increasing age heightens a homeless person's risk of developing the problems listed above. In contrast with the general population, however, the prevalence of seizures, chronic liver disease, hypertension, and diabetes decreases among homeless people who are over 65, suggesting early mortality among homeless people suffering from such conditions (Wright & Weber, 1987). In other words, homeless people seem to develop chronic conditions at earlier ages than do those in the general population and they may tend to die earlier due to these problems. This observation may, at least in part, be due to the fact that racial minorities (particularly African-Americans) are generally over-represented among the homeless (Tessler & Dennis, 1989; Toro, Passero, et al., 1995). Studies on wider cross-sections of the U.S. population have found that blacks tend to be at a general disadvantage with regard to death and disease in comparison with whites until ages beyond 65 years are reached. Among the older age groups of the elderly, however, blacks gain the advantage in health outcomes (Gibson, 1994), just as was seen by Wright & Weber (1987) among the homeless.

Another factor that may be related to health problems among homeless adults is ethnicity. The combined evidence shows that homeless blacks tend to have hypertension more frequently than do homeless whites, which is also true in the general population. The Johnson-Pew data also demonstrated greater risk for seizure disorders and chronic diseases among homeless blacks then homeless whites. In contrast, homeless whites showed higher rates of gastrointestinal ailments, trauma, peripheral vascular disease, and chronic obstructive pulmonary disease than homeless blacks (Wright & Weber, 1987). Other research provides evidence that ethnicity is related to the reasons for becoming homeless, which may affect health outcomes. For example, whites more often become homeless for internal reasons: substance abuse for men and psychiatric illness for women, whereas non-whites lose their housing more frequently due to economic problems (North & Smith, 1994).

Gender differences in health status among the adult homeless have generally not been demonstrated in existing studies. Robertson et al. (1985) found that more women than men rated their health as being fair or poor. Other researchers, however, have not found significant gender-related differences in self-reported health problems (Morse, Shields, & Hanneke, 1985) or in results of physical examinations (Harris, Mowbray, & Solarz, 1994; Wright and Weber, 1987). Similarly to distinctive health risks found between races, however, homeless men and women present differing behavior patterns. Women have generally been homeless for less time, spent less time in unsheltered locations, received more income from public assistance, and demonstrated less substance abuse than men. Solitary women, however, are more likely to show alcoholism and schizophrenia and, therefore, may have more risk factors for illness than women who are homeless with their children (North & Smith, 1993).

Causes and patterns of mortality among the homeless have also been investigated by some researchers. In a study on homeless men in Sweden, it was found that death had occurred among the sample at four times the rate of age-adjusted expected mortality in the general population. The largest differential was found for accidents, which occurred at twelve times the rate expected in the general population (Alstrom, Lindelius, & Salum, 1975).

Similar results were found in a study on a sample of 6,308 homeless people who had received mental health or general referral services in Philadelphia over a three year period. Records for the people in the sample who had died were compared with the death records for all other residents during the same period. Results showed a mortality rate among the homeless sample that was 3.5 times that of the general population, with injuries being the most frequent cause of death, followed by heart disease, liver disease, and poisoning. White men and substance abusers showed particular risk for death (Hibbs et al., 1994). Similar evidence of increased rates of deaths related to accidents, violence, and substance abuse (particularly alcohol) among the homeless has been reported in other studies (Hanzlick & Parrish, 1993; Wright & Weber, 1987).

One additional study has addressed the question of whether homeless people's health is significantly different from that of housed poor people. Gelberg, Linn, Usatine, & Smith (1992) compared the physical examination results of a clinical sample of homeless and housed poor people. They found that the homeless individuals were more likely to suffer from dermatological problems, functional limitations, seizures, chronic obstructive pulmonary disease, serious vision problems, foot pain, and grossly decayed teeth. Therefore, there is some evidence that homeless people show poorer health than do poor, housed individuals, thereby adding to the common finding that small decreases in SES correlate with more health problems.

With regard to the possible factors linking SES and health listed above, there is evidence that alcohol abuse is associated with poorer health among homeless adults (Fischer & Breakey, 1987; Wright & Weber, 1987). Smoking (Vredevoe & Brecht, 1992) and lack of social support (Wright & Weber, 1987) have also been shown to correlate with ill health in the homeless, but have not been specifically examined as frequently as has alcohol abuse. Depression has been included as a predictor of health only in combination with other severe mental illnesses, which are related to health among the homeless (Gelberg & Linn, 1988; Wright & Weber, 1987). Therefore, a unique association between depression and health has not been demonstrated in homelessness. Neither has life stress been particularly examined for a relationship with health among the homeless.

In summary, the combined research on the health of homeless adults suggests that age and ethnicity may be significantly related to health. Other psychosocial factors that have been shown to correlate with poor health are alcohol abuse, smoking, and weak social connectedness. Important factors that remain unexplored include depression (separate from other mental illnesses) and life stress.

To Table of Contents

Longitudinal Studies of Psychosocial Factors and Health

In discussing possible inter-relationships between psychosocial variables and health outcomes, the question of causality is of central importance and has implications for optimal research design and methods. Some studies examining psychosocial variables and illness in the general population have employed longitudinal designs, thus providing evidence that certain factors are important predictors of various health outcomes over time. Such longitudinal health research has often focused on a narrowly defined outcome, such as death or the progression of one particular disease process, rather than general health or physical well-being. Since the definition and measurement of general health and well-being is more difficult to establish than that of death or the worsening of a specific disease, researchers have often found it methodologically more tenable to use objective and verifiable dependent variables.

Some longitudinal studies have investigated SES variables' impact on mortality or disease outcome. For example, Marmot et al. (1991) used mortality (predicted by SES) as their dependent variable in a longitudinal design, whereas Pincus and Callahan (1985) measured changes in rheumatoid arthritis symptoms (predicted by educational level). Both of these studies found significant predictive power in their independent variables with regard to their respective measures of health outcome.

Other researchers have examined how psychological phenomena predict death or the worsening of disease. A wide variety of longitudinal studies have found that depression increases the risk of progression of coronary artery disease or death from myocardial infarction (Ahern et al. (1990); Anda et al. (1993); Carney et al. (1988); Frasure-Smith, Lesperance, & Talajic (1993); and Ladwig, Kieser, Konig, Breithardt, & Borggrefe (1991). With regard to hostility, longitudinal research has also tended to focus on heart disease. Three studies interviewed healthy people at baseline and tracked the development of heart disease through their follow-up periods. Results from all three projects provide evidence that higher hostility scores predict increased risk for the development of coronary artery disease (Barefoot, Dahlstrom, & Williams, 1983; Dembroski, MacDougall, Costa, & Gandits, 1989; Shekelle, Gale, Ostfeld, & Paul, 1983).

Longitudinal studies have also focused on the predictive power of social factors with regard to health outcomes. Two studies on heart disease used the social support indices of not living alone and being married, respectively, and demonstrated that these measures significantly predicted lower risk of recurrent cardiac events (in the first case) and death (in the second case) (Case, Moss, & Case, 1992; Williams, Barefoot, Califf, et al., 1992; ). Ruberman et al. (1984) examined the ability of education, stress, and social isolation to predict mortality among survivors of myocardial infarction. All three predictors showed an inverse relationship with mortality over a three-year period. Therefore, these longitudinal research designs have provided evidence that certain psychosocial variables, such as education, SES, depression, hostility, stress, and social support are related to specific health problems or mortality in a prospective manner. Longitudinal prediction of variations in general health in individuals who do not have discreet medical diagnoses, however, has not generally been undertaken.

Research in gerontology has also found some longitudinal evidence for the connection between psychosocial factors and health. Shahtahmasebi, Davies, and Wenger (1992) found that lower SES predicted shorter survival in people 65 years and older. In predicting the development and worsening of disabilities, lower SES predicted greater impairment in people 58 and over, and men and women showed different trajectories of decline (Maddox & Clark, 1992). Rogers, Rogers, and Belanger (1992) found similar results for SES but also showed that higher age, Black ethnicity, and female gender predicted greater disability in people 70 and over. One study on aging focused on a somewhat younger group of men (45 years and over) and found that lower SES, smoking, and excessive alcohol use predicted poorer self-rated health (Hirdes & Forbes, 1993). These studies provide evidence that demographic factors, such as race, age, SES, and gender, as well as health habits, should be examined in longitudinal studies of health.

There has been less longitudinal research on the psychosocial predictors of health outcomes among samples defined by low SES and/or homelessness. One group of researchers focused their prospective research on people in poverty and found that the status of the surrounding area (i.e., a federally designated poverty area) predicted mortality above and beyond the socioeconomic status of the individual (Haan, Kaplan, & Camacho, 1987). No longitudinal studies have been published thus far on the health status of homeless individuals, and, therefore, prospective predictors of health outcomes among the homeless remain unexamined.

To Table of Contents

Rationale for This Longitudinal Research

The literature on the health of homeless adults provides a fairly thorough documentation of the health problems recorded when they seek medical care or when they participate in population-based studies on the topic. The research examining mortality also provides valuable estimates of the shortened life expectancy among the homeless as well as the most commonly documented causes of death. In other words, the research thus far provides cross-sectional snapshots of single points of time in the lives of homeless individuals, be it while they are alive or at the time of their deaths.

Some questions that remain unanswered, therefore, are whether patterns of change can be observed in the health status of homeless adults over periods of time and what factors might influence any such changes. Based on the research results reviewed above, one might expect that beginning at any point in time, homeless individuals would then suffer a steady decline in already poor health status and, ultimately, the shortened life expectancy documented above. There is evidence from one study, however, that self-reported acute health status of homeless adults may improve when measured longitudinally over one-and-a-half years (Toro, Passero, et al., 1995).

One possible explanation for this observed phenomenon is that when people fit the criteria for being homeless and are asked to enter a study, they are at a relatively low point in their lives and have, as it were, nowhere to go but up. Such a floor effect could manifest in the measurement of many psychosocial variables, resulting in a "leveling" phenomenon across participants' scores. As documented by Toro, Passero, et al. (1995), time spent homeless and general level of stress following baseline measurement tend to decline in the following two years, and many other positive outcomes, including acute health symptoms, reflect improvement in the same time period.

Therefore, both homelessness and acute health status may be less static than cross-sectional research would lead one to believe, showing improving and worsening cycles when measured in relatively short time periods but succumbing, ultimately, to the increased risk of premature death documented above. The study that provides evidence for improving health among the homeless, however, did not examine which factors might have influence over such a phenomenon.

The current study examined the magnitude and direction of change in the self-reported health status of homeless adults during a 12-month follow-up period. Further, it investigated which psychosocial factors suggested by prior research showed associations with self-reported health status at the time of baseline measurement and predicted changes in health over the longitudinal follow-up period.

It improved on the methodology of previous research on the health of the homeless in two major ways. First, few studies up to this point have included rigorous methods for obtaining a representative sample of homeless people within a given city. The current study used a probability sample of homeless adults from throughout the Buffalo, NY metropolitan area. Second, few existing studies have used a longitudinal design to follow an identified sample of homeless adults over time. This study employed a longitudinal design, such that changes in acute health status could be examined over the course of a year.

Based on the research reviewed above, several groups of variables were included as factors potentially related to acute health at baseline. The first group of variables were the demographics of age, ethnicity, and gender. Although gender has not consistently been associated with health among the homeless, it was included in this study because of relationships found in other populations and because an association between the two variables would be of importance.

Following demographics, total chronic health problems score was the next variable of interest, given that ongoing illnesses manifest in a variety of symptoms that would most likely account for a portion of the acute health score. It was considered after demographics because particularly age, and possibly gender and ethnicity, could be associated with chronic illnesses.

The third group of variables reflected socioeconomic status, defined both in the traditional sense as education and job income, as well as variables that reflect the socioeconomic well-being of homeless people in particular: length of time homeless, housing transience, and income from public assistance. Given that even small differences along the SES continuum have proven to be important in predicting health in previous studies, the construct was measured in a variety of ways.

The final sets of variables reflected health behaviors (smoking, drug abuse, and alcohol abuse), psychological factors (depression and hostility), and social factors (stressful life events and social support). Each set of these variables was considered in a separate analysis, each controlling for the influence of basic demographics and SES. Past research has examined health behaviors, psychological factors, and social influences on health in isolation, but has not compared the relative associative power of such constructs.

In order to investigate the predictors of change in acute health status at the 12-month follow-up, a similar rationale was employed, with the criterion variable being acute health score at follow-up. The first variable to be considered as a predictor in these analyses, however, was the baseline acute health problems score, such that the correlation between the two scores could be partialled out, effectively leaving the change in symptoms over the 12-month period as the criterion variable for the remaining predictors.

Thereafter, the order of variable consideration was the same as that described for the baseline analyses, comparing the differential strength of health behaviors, psychological factors, and social influences in predicting change in symptoms after having controlled for demographics and SES.

To Table of Contents


The current study is based on the longitudinal follow-up of a baseline probability sample of 420 homeless individuals gathered as part of a research project funded by the National Institute of Mental Health (NIMH) and involved data collected both at baseline and a 12-month follow-up (n=260). The primary purpose of the design was to predict longitudinal self-reported health outcomes using baseline measures of health, demographics, and psychosocial characteristics as predictor variables.

To Table of Contents


The background characteristics of the total sample (see Table 1) are consistent with most recent studies on the homeless (Tessler & Dennis, 1989; Toro, Passero, et al., 1995). The study's sample was relatively young (median age = 31.5 years, SD=8.3) and somewhat more than one quarter white (28%). A majority (74%) was male. While the age distribution of the sample very nearly mirrored that of the general population of the metropolitan area studied (based on 1990 census data, the area population had median age of 32), men and racial minorities were over-represented in the sample (based on 1990 census data, the area was 52% female and 85% white). The educational attainment of the sample was low, with almost half (42%) having failed to complete high school. The average accumulated total time spent homeless since age 16 was 14.5 months.

To Table of Contents

Table 1
Background characteristics of homeless adults (N=420)

n %

18-29 161 38
30-39 168 44
40-62 91 18
Male 310 74
Female 110 26
White 116 28
African-American 286 68
Other 18 4
Less than high school 175 42
High school graduate 140 33
Training beyond high school 105 25
Total months homeless (as adults)
1-3 months 217 51
4-12 months 102 24
13 months-21 year 101 25
Housing Moves
1-6 217 51
6-12 168 40
12-35 168 40
Mental Health and Substance Use
Depression 85 20
Alcohol Abuse 235 56
Drug Abuse 206 56
Smoking 349 83

To Table of Contents


The sampling method of the baseline NIMH-funded study was adapted from that of Farr et al. (1986), with the final design being determined by two preliminary steps, a key informant survey and a series of sampling surveys, both having taken place in Erie County, N.Y. (the current U.S. Census definition of metropolitan Buffalo, which has a total population of 1.0 million). Key informants were defined as the people most knowledgeable about the county's homeless population and were recruited from four types of agencies: police departments (n=21), mental health and substance abuse treatment facilities (n=44), neighborhood-based community organizations (n=51), and agencies explicitly serving the homeless, such as emergency shelters and soup kitchens (n=44). A structured 15-30 minute phone interview was completed with each informant, during which he/she was asked to indicate the geographical sector of the county with which he/she was most familiar (from a total of 13 sectors). While focusing on the chosen sector, each informant was asked to estimate the incidence of homelessness there, as well as the exact whereabouts and characteristics of that sector's homeless population. The estimates showed that only 14% of the homeless people in the county within the past year were located in nonurban sectors. Based on these low numbers, numerous subsequent attempts by the research staff to locate homeless individuals in nonurban areas, and the practical problems in sampling homeless people in these areas, it was decided that recruitment for the study would be done only in the urban and semi-urban areas shown to have substantial homeless populations.

Based on the data obtained from key informants, nonoverlapping proportions were estimated of homeless people over the course of a year who used shelters (32%), went to food programs but no shelters anytime in the past year (38%), were admitted to psychiatric hospitals or other inpatient settings but not the above sites (6%), accessed out-patient mental health or substance abuse treatment facilities, drop-in centers, or referral agencies but not the above sites (14%), and survived on the street, taking advantage of none of the above services (10%). These results, showing that the majority of homeless people used either shelters or food programs, are consistent with those obtained by Farr et al. (1986). The hierarchy of these proportions served as the preliminary estimates for the sampling factors that determined the types of sites at which to recruit a representative sample of the area's homeless population.

The appropriate sampling factors were finalized by conducting sampling surveys (n=597) on 91 different occasions at 51 different sites from all of the six organizational categories listed above. The sampling surveys were completed by approaching individuals at a particular service site and asking if they would be willing to complete a brief survey concerning their housing situation, thereby determining the proportion of homeless people encountered at each site. An individual was considered homeless if he/she was staying or had stayed at a homeless shelter in the past month, had no housing, or was staying with friends or relatives and could not provide rent.

If a person met the criteria for homelessness, he/she was asked further questions concerning the locations and frequencies of his/her service use over the past year. The data on those found to be homeless were used to determine the proportion of participants to be recruited at each type of site and which specific sites should be included in the recruitment plan for achieving a representative sample. According to the data on service use gathered, a median use frequency value for each site was also computed. In order to avoid the increased likelihood that high-frequency service users would be approached later for recruitment, a common problem in research that recruits participants from service settings, the median frequency values were used during the subject recruitment period to enroll low-frequency service users as 50% of the full sample of participants (see Toro, 1993, for further details on the sampling design).

The process of participant recruitment was arranged in a non-overlapping hierarchy based on the results of the key informant survey and the sampling surveys (see Table 2), which indicated participants should be recruited based on the following sampling factors: 73% (N=306) from shelters, 20% (N=85) from soup kitchens, 4% (N=15) from in-patient psychiatric and substance abuse treatment facilities, and 3% (N=14) from out-patient service sites. According to the sampling survey data, only 0.5% of the homeless people questioned indicated that they were living on the street and had used none of the services higher in the list during the prior year. This was a much smaller proportion than that suggested by the data from the key informants (10%). Since sampling survey data indicated that subjects who met this criteria made up less than 1% of the total homeless population, it was decided that such people would not be sought for interviews without much risk to the representativeness of the sample.

Recruitment of participants (n=420) involved asking homeless people who fit the frequency criterion if they would be interested in completing a 3-4 hour interview in return for a $20 reimbursement. Each member of the sampling team had an up-dated list of participants, so that replications of interviews could be avoided.

Interviews were completed in previously arranged rooms on-site, with the requirement that each individual's privacy and confidentiality be maintained. The informed consent process indicated to participants that answering each question was a voluntary process, that they had the right to withdraw at any time, and that their identity would not be exposed to anyone who was not a part of the research team. Each person was informed that follow-up interviews would take place every three months for twelve months after the baseline interview was completed, with brief (10-20 minute) interviews being completed at 3 and 9 months and full-length interviews being completed at 6 and 12 months. An additional 20 participants were interviewed and then re-interviewed one week later, in order to assess the test-retest reliability of various measures.

Table 2

To Table of Contents

Final Baseline Sampling Design

Step 1

Step 2

Step 3
Preliminary a Sampling Surveysb Final Finalc
Factor Factor Number
Done/Asked Homeless
Type of Site
Shelters .32 150/187 150 (25%) .730 306
Food Progs. .38 1,165/1,645 240 (40%) .202 85
In-Patient .06 79/83 51 ( 9%) .032 15
Out-Patient .14 454/589 92 (15%) .031 14
Streets .10 112/154 64 (11%) .005 0
Total 1.00 1,960/2,658 597 (100%) 1.00 420

aBased on 160 Key Informant Surveys.

bSampling Survey data were collected at 51 different sites.

cThe 420 interviews were done over 16 months at 23 different sites (4/92-7/93)


In addition to the information on how the homeless were distributed across types of sites (e.g., shelters, food programs, etc.), the sampling surveys also provided useful data regarding the overlap of service use within sampling sectors. Based on the sampling survey data, a number of shelters, food programs, and out-patient referral agencies were excluded as recruitment sites because the number of nonoverlapping homeless individuals (i.e., people who had not used any other sites within a given sampling sector) found at those sites was very small. Any site was generally excluded if it yielded no more than 0.5% unique contribution to the total homeless population. The number of recruitment sites chosen in each sector, along with the specific services offered by each, are described below.

Shelters. A total of seven shelters were identified as providing sites for the recruitment of sizable numbers of homeless people. In addition to a large, all-male shelter, the sites include two shelters designed for females and families, one small shelter which housed males and females, a domestic violence shelter, and a shelter which housed young men under the age of 21. The baseline interviews completed at shelters numbered 306, and the refusal rate was 3% (N=8).

Food Programs. Nine food programs were identified as unique sites for recruitment of sizable numbers of homeless people. Aside from meeting the criteria for homelessness, participants were recruited only if they had not used a shelter in the past twelve months. Baseline interviews done at food programs numbered 85. Among those who met the inclusion criteria, 6% (N=5) refused to participate. Five others (6%) refused before eligibility could be determined.

In-Patient Psychiatric and Substance Abuse Facilities. In addition to obtaining participants from the area's state psychiatric facility, recruitment was also done at a large substance abuse inpatient facility, an inpatient unit operated by the Veterans Administration, and a unit operated by a local supervised housing agency which cares for the homeless mentally ill on an emergency basis. Individuals were asked to participate if they were homeless at admission or at any time in the month prior to admission, were deemed by the staff to be both willing and able to complete the interview, were not considered dangerous or incapable of giving informed consent, and had not used a shelter or food program in the previous year. Baseline interviews (n=15) were conducted in the hospital prior to discharge. Among those who met inclusion criteria, one patient (7%) refused to participate.

Outpatient Programs. The sites included were the Erie County Department of Social Services, the Housing Assistance Center, and a case management program for placing the homeless in housing. Following the method of including only non-overlapping service users, participants were recruited only if they had not used services in categories higher in the hierarchy over the previous year (n=14). One person (7%) refused participation after eligibility had been determined.

This sampling design offered substantial improvements over past research on homelessness. First, the key informant surveys provided a systematic basis for the process of gathering sampling surveys, which in turn resulted in carefully refined sampling factors that were applicable to recruitment at each site. Second, a broad geographical area was covered with multiple sites throughout the whole metropolitan area being sampled (as opposed to sites only in the skid row area, for example). Third, the sampling plan included a broad range of homeless individuals, such as those using inpatient and outpatient treatment facilities (in addition to users of shelters, for example). Fourth, as described above, 50% of the interviewed subjects were low-frequency services users, thereby preventing the common research problem of high-frequency service users being over-represented. Fifth, to avoid seasonal differences, the brief surveys and the final sample of 420 baseline interviews were collected throughout the four seasons.

In order to achieve as high a follow-up rate for ensuing interviews as possible, research staff employed several tracking techniques. First and foremost, interviewers strove to develop personal rapport with each participant while maintaining a professional relationship. With this in mind, interviewers made contact with participants about once a week in the first month following baseline and then at least once a month thereafter through the remainder of the follow-up period. As interviewers became familiar with the subjects they had recruited, they could record the typical habits of each person and any movements around the city he/she might make. Second, each participant was asked at baseline to provide the names of some collateral contacts or significant others who would be fairly up-to-date on his or her whereabouts at any given time. If subjects themselves could not be located for follow-up meetings, collateral contacts could then be called on the telephone to inquire about them. Third, even if participants missed interviews their interviewers continued to try to locate them, since gaps in follow-up data were considered better than no data at all. Interviewers often found that participants were motivated to stay in contact based on the payments they received for each interview and the friendly, respectful human contact the researchers provided.

Follow-up interviews with participants for whom tracking had been successful were conducted at any appropriate sites that were mutually convenient for interviewers and subjects. For subjects who were still homeless, the same sites in which the baseline interviews had been conducted could be used again. People who had acquired housing could meet with interviewers in public places, such as restaurants, or in the same sites where they had been interviewed previously. Brief (10-20 minute) follow-up interviews, for which subjects were paid $10, were completed at 3 and 9 months. Full-length follow-up interviews, for which subjects were paid $20, were completed at 6 months (n=301, follow-up rate=%72) and 12 months (n=260, follow-up rate=%62).

To Table of Contents


Demographics and Background

A survey of demographic characteristics such as gender, age, ethnicity, and educational achievement was administered at baseline only (see Appendix __).

Homelessness and Employment History

Housing, Income, and Services Timeline (HIST). The HIST was originally developed in two previous studies on homelessness (Toro, Bellavia, et al., 1995; Toro, Passero, et al., 1995) and was further developed for this research to provide a reliable method for assessing a person's life history (since the age of 16 years) in five major domains: housing, homelessness, employment, income, and utilization of social services. A number of strategies were used to improve recall of life events, including the use of data elicited in other life domains to provide anchors for estimating when the events in other domains occurred. The brief interviews at 3 and 9 months following baseline were intended to improve recall of HIST information for longitudinal measurement, as well as to facilitate interviewer-participant contact and thereby reduce attrition. The HIST is based on the Life Event Calendar technique, which has demonstrated accurate participant recall of events previously in a five-year longitudinal study (Freedman et al., 1988).

In the current analyses, the baseline calculations of total time homeless (across all episodes), total number of housing moves (i.e., housing transience), total income from legal employment, and total income from all forms of public assistance, all since age 16 years, were used as predictor variables. Test-retest correlations for these HIST variables based on the pilot reliability sample (n=20) were as follows: housing transience (0.98), total time homeless (0.73), income from wages (0.94), and income from public assistance (0.81).

Depression and Substance Abuse Diagnoses

Diagnostic Interview Schedule (DIS). The DIS was used to assess mental health status. Assessing mental illness among the homeless has become controversial (Snow et al., 1986; Susser, Conover, & Struening, 1989; Toro & Wall, 1991; Wright, 1988). Commonly used indicators of mental illness in existing studies have included: (a) a diagnosis generated by a standardized clinical interview such as the DIS, (b) a score on a symptom checklist, (c) past history of psychiatric hospitalization, and (d) global observer assessments. Applied to the homeless, each of these methods has its merits in terms of reliability, validity, and/or practicality. The standardized clinical interview method tends to yield the most conservative estimates (as low as 10-15% for the severe psychiatric disorders of schizophrenia and major affective disorders), while the less rigorous methods yield higher estimates (as high as 78%).

The rigorous DIS was chosen for this study to operationalize severe mental illness and substance abuse and dependence. It was originally developed for the National Institute of Mental Health's Epidemiologic Catchment Area (ECA) program and allows lay interviewers to obtain self-reported data which can be scored by computer to yield lifetime and current psychiatric disorders based on DSM-III criteria (Eaton & Kessler, 1985; Regier et al., 1988).

The DIS was chosen for the current research because of (a) its prior use in several studies on the homeless (Farr et al., 1986; Toro & Wall, 1991; Toro, Bellavia, et al., 1995); (b) its ability, unlike other assessment methods, to distinguish severe mental illness from substance abuse; (c) extensive reliability and validity data, including comparability with expert diagnoses (Robins et al., 1981); and (d) the availability of normative comparison data.

Because the complete DIS would take over an hour for many participants, several diagnostic categories were eliminated for this study: various anxiety-related disorders, sexual dysfunctions, anorexia nervosa, pathological gambling, tobacco use, organic brain syndrome, and anti-social personality disorder. The DIS sections retained were those dealing with: major affective disorders (depression and mania), alcohol abuse and dependence, and drug abuse and dependence. The DIS was administered at baseline and again at the twelve-month follow-up. The baseline lifetime diagnoses of major depression, alcohol abuse, and drug abuse were utilized in the current analyses.


SCL-90-R symptom checklist. Many have argued that symptom checklists measure current "psychological distress" or "demoralization" among the homeless, unlike structured clinical interviews such as the DIS, which provide reliable diagnostic information (Lovell, in press; Robertson, 1992; Toro & Wall, 1991). The SCL-90-R is such a symptom checklist and yields nine subscales, one of which is Hostility.

Acceptable internal consistency and test-retest reliability have been demonstrated for the SCL-90-R and all of its subscales, including Hostility, and extensive evidence on concurrent and discriminative validity exists (Derogatis, 1977). The one-week test-retest reliability coefficient for the current study was found to be .78 for the Hostility subscale. Full length and brief (10- to 53-item) forms of the SCL-90-R used in several recent studies have found the homeless to have significantly higher scores than normative samples (Morse & Calsyn, 1986; Mowbray, Solarz, Johnson, Phillips-Smith, & Combs, 1986; Solarz, 1986; Sosin, Colson, & Grossman, 1988; Toro & Wall, 1991).

Life Stress

Modified Life Events Interview (MLEI). The MLEI is an 85-item checklist that measures events that would be considered to be stressful in a number of life domains. These include social relationships, housing situations, employment, education/job training, and mental and physical health. It was developed by Lovell (1984) specifically for use with homeless populations. Calculations from the reliability sample indicated that the instrument has good test-retest reliability (r=0.84).

For the purpose of the current study each participant's score on the MLEI was calculated excluding the five items that deal with physical health, thereby eliminating content overlap with the Physical Health Symptoms Checklist described below. Therefore, the maximum range for scores used in the current analyses was 0 - 80.

Perceived Social Support

Interpersonal Support Evaluation List (ISEL). The 40-item ISEL (Cohen, Mermelstein, Kamarck, & Hoberman, 1985) has four sub-scales, each intended to measure the availability of a different type of social support: tangible, concerning the provision of material aid; appraisal, the belief that one has people to turn to for advice on one's problems; self-esteem, the belief that one's status is equal to that of friends; and belonging, concerning access to people with whom one can engage in activities. Across several studies, alpha coefficients for the four subscales have ranged from .62 (self-esteem) to .82 (appraisal), and two-day test-retest reliability coefficients have ranged from .67 (belonging) to .84 (appraisal).

The ISEL has shown consistently moderate correlations with other measures of social support and with various outcome measures, including self-esteem, health behavior, and psychological symptoms (Cohen et al., 1985). To improve variability, a four-point scale (1=completely false, 2=somewhat false, 3=somewhat true, 4=completely true) was used on ISEL items, rather than the usual dichotomous scale. Using this scaling in a one to two week test-retest reliability assessment (n=31), Toro, Bellavia, et al., (1995) obtained the following coefficients: .85 for tangible, .77 for appraisal, .66 for self-esteem, .62 for belonging. The total score was used for these analyses.

Health Status

Physical Health Symptoms Checklist (PHSC). The PHSC is a checklist made up of 59 items dealing with acute symptoms, 9 items inquiring about chronic conditions, and 10 items regarding access to and use of health care services. All items are answered true or false. The acute and chronic sections were adapted from the Health Problems Checklist (Schinka, 1989). The access and utilization items were adapted from a variety of other health checklists. Data from the reliability sample revealed that the total score on the acute symptoms section showed a test-retest reliability of .85

The acute symptoms section of the PHSC contains four items to be administered only to females, since they cover menstruation problems, menopause, and abnormalities of the breasts. These items were eliminated from the scoring for the purposes of the current analyses, such that a spurious association between gender and number of acute symptoms would be avoided. Therefore, the total number of acute symptoms for each participant was based on responses to the 55 items administered to both sexes. Scores on the section covering access to and use of health care services were not examined in the current study. Current cigarette smoking was assessed by a yes-or-no question in conjunction with the PHSC.

To Table of Contents

Data Analysis

A total of 18 variables from the baseline data and one from the 12-month follow-up interview (the health symptom total) were used for the present analyses. Demographic variables used were gender (0=male, 1=female), age, and ethnicity (1=white, 2=non-white). Variables reflecting socio-economic status were education (0=less than high school, 1=high school diploma, 2=training beyond high school), lifetime income from work, lifetime income from public assistance, lifetime total homelessness (across all episodes), and housing transience (moves since age 16). For the multivariate analyses described below, education was dummy coded in the same manner as it would be for an ANOVA. All SES variables except educational level were obtained from the HIST.

Health behavior variables were smoking (0=no, 1=yes; taken from the PHSC), DIS lifetime diagnosis of drug abuse (0=no, 1=yes), and DIS lifetime diagnosis of alcohol abuse (0=no, 1=yes). Psychological factors were measured by DIS lifetime diagnosis of major affective disorder (0=no, 1=yes) and hostility from the SCL90R (continuous scale). Social factors were represented by MLEI stressful events (continuous scale) and total perceived social support from the ISEL (continuous scale). Zero-order correlations were calculated between all variables.

Only participants who had provided data for every measure were included in the analyses. Therefore, the baseline sample was 413, since six people had missing data on at least one measure and one outlier was excluded based on an extreme score on the PHSC. At follow-up, the sample was 257, since the outlier described remained excluded and two people had missing data.

Attrition analyses were done in order to determine if the subjects who could not be located for 12-month follow-up interviews (n=156) demonstrated any significantly different characteristics at baseline from those who remained in the study at 12-month follow-up (n=257). An analysis of variance (ANOVA) was performed for each of the 10 continuous baseline variables, and a Chi-Square was performed for each of the seven categorical baseline variables to compare the two groups and to determine if any significant baseline differences would be indicated.

Next, a t-test was run to examine the pattern of change in acute health scores over the 12-month follow-up period. In this analysis, the difference between baseline and follow-up symptom totals was tested to determine if the mean change was significantly different from zero and, if so, the direction that it indicated. This analysis was designed to assess whether homeless people tend to show improvement on this measure of health symptoms (as found by Toro, Passero, et al., 1995).

To Table of Contents

Main Analyses: Baseline Predictors of Health

For the first set of the study's main analyses, three hierarchical multiple regression equations were evaluated to determine baseline relationships between acute health symptoms and gender, ethnicity, and age in the first step, followed by chronic health problems in the second step. The SES variables of education, job income, public assistance income, housing transience, and time homeless were entered in the third step. Thereafter, the sets of health behaviors (smoking, alcohol abuse, and drug abuse), psychological characteristics (depression and hostility), and social factors (stressful events and social support) were entered in each equation, respectively, in order to compare their association with health, above and beyond demographics and SES. Within each hierarchically ordered set of variables, individual variables were entered step-wise into the equations.

To Table of Contents

Main Analyses: Longitudinal Predictors of Health

For the second set of main analyses, longitudinal data were considered. The dependent variable in these analyses was the total number of acute symptoms at follow-up. In the first step in these longitudinal multiple regression analyses, the baseline acute health score was entered in order for regressed change in health symptoms to be the dependent variable for all the predictors to follow (Cohen & Cohen, 1975). This approach was taken instead of using a simple change score as the dependent variable, since such a score is correlated with the prescore and may produce spurious and misleading results. In contrast, when the postscore is regression-adjusted for the prescore, the result is a residualized change score that is statistically independent of the prescore (Cohen & Cohen, 1975).

Three longitudinal multiple regression equations were created in the same order as the baseline equations, with demographics entered next after the prescore, followed by chronic health problems, then SES variables. The sets of health behavior variables, psychological characteristics, and social factors were entered respectively in the final steps of each equation in order to compare their relative predictive power.

To Table of Contents


Of the 10 ANOVAs and 7 Chi-Squares performed to investigate potential differences between the baseline sample and the 257 participants who completed 12-month interviews (i.e., attrition effects), the results for age and life stress were statistically significant. These results indicate that the younger a subject was (F(1,411) = 5.15, p<.05) and the more stressful events he/she had experienced (F(1,411) = 3.94, p<.05), the less likely he/she was to complete a 12-month interview . Zero-order correlations between all variables are presented in Table 3.

The health symptom endorsements (see Table 4) at baseline showed that the most frequent affirmative answers indicated general health problems, such as trouble sleeping (59%), getting tired easily (40%), back pain (40%), and poor health (28%). Other frequently endorsed problems were dental symptoms, such as toothache (30%) and gums bleeding after brushing (27%), and colds or upper respiratory symptoms, such as runny or stuffed up nose (40%), painful breathing (18%), wheezing (18%), frequent cough (20%), and coughing up blood or mucus (9%).

With regard to chronic health problems, endorsements were as follows: serious head injury (33%), hypertension (25%), seizure disorder (7%), heart attack (4%), HIV positive (3%), diabetes (2%), cancer (2%), and stroke (1%).

To Table of Contents

Table 4

Baseline endorsements of acute health and chronic health

problems (N=413)

Acute Health Problems

n %

General Health

Trouble Sleeping 244 59

Get Tired 166 40

Back Pain 163 40

Poor Health 117 28

Dental Problems

Toothache 122 30

Gums Bleed after Brushing 110 27

Colds and Upper Respiratory Infections

Runny or Stuffed-up Nose 166 40

Frequent Cough 84 20

Painful Breathing 76 18

Wheezy Breathing 74 18

Coughing up Blood or Mucus 37 9

Chronic Health Problems

Serious Head Injury 136 33

Hypertension 102 25

Seizure Disorder 27 7

Heart Attack 16 4

Tested HIV Positive 12 3

Cancer 9 2

Diabetes 7 2

Stroke 5 1

The t-test done to examine the change in symptom score over the 12-month follow-up period (see Table 5) revealed that the mean number of symptoms endorsed at follow-up was about four less than the average endorsement at baseline, which led to a statistically significant result (t = -7.07, p<.01).

To Table of Contents


Table 5

T-test assessing change over 12-month follow-up period

illness symptoms (N=257)

Baseline Follow-up Mean Change Std Dev T

9.05 5.42 -3.63 8.24 -7.07**



Main Analyses: Baseline Predictors of Health

The three multiple regression equations using the baseline acute health symptoms score as the dependent variable allowed a comparison between its associations with health behaviors, psychological factors, and social variables, above and beyond demographic characteristics, chronic health diagnoses, and socio-economic situation (see Tables 6, 7, and 8 for the stepwise entry order of individual variables).

Results indicated that no demographic characteristics were significantly associated with baseline acute health status. Chronic health problems did show a significant relationship with acute problems (F(1,408) = 117.25, p<.01), such that those with more chronic health problems also endorsed more acute symptoms. Among socio-economic indicators, only housing transience showed a significant relationship with acute health problems (F(1,407) = 9.04, p<.01), indicating that people with adult histories of less stability in housing reported more illness symptoms.

Comparison of the associative strength of the three sets of variables entered following demographic, chronic illness, and SES variables indicated that psychological and social factors added 11.8% (F(2,400) = 37.37, p<.01) and 11.3% (F(2,400) = 35.72), p<.01), respectively, to the variance accounted for in the dependent variable (see Tables 6 and 7). Within the set of psychological factors, both hostility (F(1,401) = 56.08, p<.01) depression (F(1,400) = 16.49, p<.01) showed significant individual associations. Within the set of social factors, both stressful events (F(1,401) = 51.01, p<.01) and low social support (F(1,400) = 18.28, p<.01) contributed significantly. The health behavior measures accounted for only 1.02% of the variance in acute health problems (see Table 8), which did not reach statistical significance.

To Table of Contents

Table 6

Hierarchical Multiple Regression Analysis on Baseline Acute Symptoms: Health Behaviors Variable Set Entered Last

Demographics r sr2 F df

Gender .05 .05 .003 1.11 (1,411)

Age .05 .05 .003 1.14 (1,410)

Race -.01 -.01 .000 0.05 (1,409)

Chronic Illnesses

Chronic Problems .46 .50 .222 117.25** (1,408)

Socioeconomic Indicators

Moves .21 .15 .017 9.04** (1,407)

Education --- --- .004 1.84 (2,405)

Income -.03 -.04 .001 .37 (1,404)

Time Homeless .08 .02 .000 .15 (1,403)

Public Assistance .10 .02 .000 .15 (1,402)

Health Behaviors

Alcohol Abuse .14 .09 .006 3.45 (1,401)

Smoking .06 .05 .003 1.37 (1,400)

Drug Abuse .09 .04 .001 .71 (1,399)

Overall Health

Behaviors Set --- --- .010 1.84 (3,399)

Note. The sr2s are squared semi-partial correlation coefficients indicating the percent of unique criterion variance accounted for by each predictor variable. The s are the associated standarized regression coefficients and indicate the direction of the relationship between symptoms and each predictor variable. The Fs test whether the s (as well as the sr2s) differ significantly from zero.

*p<.05 **p<.01

To Table of Contents

Table 7

Hierarchical Multiple Regression Analysis on Baseline Acute Symptoms: Psychological Variable Set Entered Last

Demographics r sr2 F df

Gender .05 .05 .003 1.11 (1,411)

Age .05 .05 .003 1.14 (1,410)

Race -.01 -.01 .000 0.05 (1,409)

Chronic Illnesses

Chronic Problems .46 .50 .222 117.25** (1,408)

Socioeconomic Indicators

Moves .21 .15 .017 9.04** (1,407)

Education --- --- .004 1.84 (2,405)

Income -.03 -.04 .001 .37 (1,404)

Time Homeless .08 .02 .000 .15 (1,403)

Public Assistance .10 .02 .000 .15 (1,402)

Psychological Factors

Hostility .42 .32 .092 56.08** (1,401)

Depression .32 .18 .026 16.49** (1,400)

Overall Psycho-

logical Set --- --- .118 37.37** (2,400)

Note. The sr2s are squared semi-partial correlation coefficients indicating the percent of unique criterion variance accounted for by each predictor variable. The s are the associated standarized regression coefficients and indicate the direction of the relationship between symptoms and each predictor variable. The Fs test whether the s (as well as the sr2s) differ significantly from zero.

*p<.05 **p<.01

To Table of Contents

Table 8

Hierarchical Multiple Regression Analysis on Baseline Acute Symptoms: Social Variable Set Entered Last

Demographichs r sr2 F df

Gender .05 .05 .003 1.11 (1,411)

Age .05 .05 .003 1.14 (1,410)

Race -.01 -.01 .000 0.05 (1,409)

Chronic Illnesses

Chronic Problems .46 .50 .222 117.25** (1,408)

Socioeconomic Indicators

Moves .21 .15 .017 9.04** (1,407)

Education --- --- .004 1.84 (2,405)

Income -.03 -.04 .001 .37 (1,404)

Time Homeless .08 .02 .000 .15 (1,403)

Public Assistance .10 .02 .000 .15 (1,402)

Social Factors

Stress .42 .32 .084 51.01** (1,401)

Social Support -.28 -.18 .029 18.28** (1,400)


Social Set --- --- .113 35.72** (2,400)

Note. The sr2s are squared semi-partial correlation coefficients indicating the percent of unique criterion variance accounted for by each predictor variable. The s are the associated standarized regression coefficients and indicate the direction of the relationship between symptoms and each predictor variable. The Fs test whether the s (as well as the sr2s) differ significantly from zero.

*p<.05 **p<.01

Main Analyses: Longitudinal Predictors of Health

As described above, the dependent variable in the multiple regression equations dealing with the longitudinal data was acute health symptoms score at 12-month follow-up (see Tables 9, 10, and 11). The baseline score on the same measure was entered on the first step, thereby establishing residualized change in score as the criterion for the remaining independent variables. This first step did indicate a significant relationship between the baseline and the 12-month acute health symptoms scores (F(1,255) = 60.90, p<.01).

Upon entry of demographic variables, no significant associations were found. Number of chronic conditions endorsed at baseline also showed no significant predictive power regarding change in acute symptom total. Within the set of socio-economic variables, once again, housing transience proved a significant predictor of worsening health (F(1,250) = 5.03, p<.05). No other socio-economic variables proved to be significant predictors of change in symptoms.

In comparing the added contributions of the health behaviors, psychological factors, and social interaction, it was demonstrated that the psychological characteristics (see Table 9) accounted for the greatest amount of variance in change in acute health status (2.48%, F(2,243) = 4.19, p<.05), with depression, but not hostility, reaching significance as a predictor of worsening health (F(1,244) = 5.90, p<.05). Social factors added only 0.66% to the variance accounted for in the dependent variable (see Table 10), followed by the set of health behavior variables which accounted for only 0.24% of the variance in symptom change (see Table 11). These last two sets did not achieve statistical significance.

To Table of Contents

Table 9

Hierarchical Multiple Regression Analysis on Follow-up Acute Symtpoms: Health Behaviors Set Entered Last

Demographics r sr2 F df

Baseline Symptoms .44 .45 .193 60.90** (1,255)

Race -.10 -.10 .011 3.33 (1,254)

Age .13 .09 .007 2.29 (1,253)

Gender .06 .06 .003 .97 (1,252)

Chronic Illnesses

Chronic Problems .31 .12 .009 3.05 (1,251)

Socioeconomic Indicators

Moves .26 .15 .015 5.03* (1,250)

Income -.05 -.11 .008 2.61 (1,249)

Time Homeless -.04 -.11 .009 2.89 (1,248)

Education --- --- .001 .11 (2,246)

Public Assistance .08 -.03 .001 .24 (1,245)

Health Behaviors

Drug Abuse .08 .05 .002 .69 (1,244)

Smoking .02 .01 .000 .06 (1,243)

Alcohol Abuse .07 .01 .000 .03 (1,242)

Overall Health

Behaviors Set --- --- .002 .26 (3,242)

Note. The sr2s are squared semi-partial correlation coefficients indicating the percent of unique criterion variance accounted for by each predictor variable. The s are the associated standarized regression coefficients and indicate the direction of the relationship between symptoms and each predictor variable. The Fs test whether the s (as well as the sr2s) differ significantly from zero.

*p<.05 **p<.01

To Table of Contents

Table 10

Hierarchical Multiple Regression Analysis on Follow-up Acute Symptoms: Psychological Variable Set Entered Last

Demographics r sr2 F df

Baseline Symptoms .44 .45 .193 60.90** (1,255)

Race -.10 -.10 .011 3.33 (1,254)

Age .13 .09 .007 2.29 (1,253)

Gender .06 .06 .003 .97 (1,252)

Chronic Illnesses

Chronic Problems .31 .12 .009 3.05 (1,251)

Socioeconomic Indicators

Moves .26 .15 .015 5.03* (1,250)

Income -.05 -.11 .008 2.61 (1,249)

Time Homeless -.04 -.11 .009 2.89 (1,248)

Education --- --- .001 .11 (2,246)

Public Assistance .08 -.03 .001 .24 (1,245)

Psychological Factors

Depression .26 .15 .018 5.90* (1,244)

Hostility .24 .10 .007 2.45 (1,243)

Overall Psycho-

logical Set --- --- .024 4.19 (2,243)

Note. The sr2s are squared semi-partial correlation coefficients indicating the percent of unique criterion variance accounted for by each predictor variable. The s are the associated standarized regression coefficients and indicate the direction of the relationship between symptoms and each predictor variable. The Fs test whether the s (as well as the sr2s) differ signifcantly from zero.

*p<.05 **p<.01

To Table of Contents

Table 11

Hierarchical Multiple Regression Analysis on Follow-up Acute Symptoms: Social Variable Set Entered Last

Demographics r sr2 F df

Baseline Symptoms .44 .45 .193 60.90** (1,255)

Race -.10 -.10 .011 3.33 (1,254)

Age .13 .09 .007 2.29 (1,253)

Gender .06 .06 .003 .97 (1,252)

Chronic Illnesses

Chronic Problems .31 .12 .009 3.05 (1,251)

Socioeconomic Status

Moves .26 .15 .015 5.03* (1,250)

Income -.05 -.11 .008 2.61 (1,249)

Time Homeless -.04 -.11 .009 2.89 (1,248)

Education --- --- .001 .11 (2,246)

Public Assistance .08 -.03 .001 .24 (1,245)

Social Factors

Stress .23 .12 .007 2.18 (1,244)

Social Support -.13 -.00 .000 .00 (1,243)


Social Set --- --- .001 1.09 (2,243)

Note. The sr2s are squared semi-partial correlation coefficients indicating the percent of unique criterion variance accounted for by each predictor variable. The s are the associated standarized regression coefficients and indicate the direction of the relationship between symptoms and each predictor variable. The Fs test whether the s (as well as the sr2s) differ signifcantly from zero.

*p<.05 **p<.01

To Table of Contents


The overall health symptoms and chronic illness profile endorsed by the current sample of homeless adults corroborate the findings of previous studies, which have recorded upper respiratory infections, dental problems, hypertension, and trauma as some of the most frequently encountered health problems among the homeless (Gelberg & Linn, 1990, 1992; Jahiel, 1992; Wright & Weber, 1987). This provides evidence of convergent validity for the PHSC, indicating that its most highly endorsed health problems correspond to the overall categories of illness diagnosed most frequently by other researchers.

To Table of Contents

Main Analyses: Baseline Predictors of Health

The cross-sectional baseline analysis in this study revealed that basic demographics were not associated with acute health problems among the homeless. This corroborates previous research findings that homeless men and women show roughly equal rates of illness. The lack of association between ethnicity and health symptoms may be due to the phenomenon evidenced in previous research that different races among the homeless have different patterns of illness vulnerability. These differing patterns may result in roughly comparable total numbers of symptoms, however. In contrast to previous research, age was not significantly associated with more illness symptoms in this representative sample. This finding could be due to the fairly young age of today's homeless (mean = 33 years in this study), which may not include a broad enough range to detect the known increases in health problems that come with advancing age in the general population.

The more chronic illnesses a subject endorsed, however, the higher number of acute symptoms he/she tended to report (after the influence of demographics, including age, had been accounted for). This finding makes intuitive sense and most likely reflects the symptomatic manifestations of the various chronic health problems endorsed.

Traditional indicators of socio-economic status did not relate significantly to health symptoms above and beyond demographics and chronic health problems in this sample of homeless adults, although housing transience did. The lack of findings for traditional SES variables may reflect the restricted range within this sample, although educational attainment varied from less-than-high school to beyond high school and lifetime work income varied from negligible amounts up to totals reflecting substantial employment histories. Housing transience, a measure more directly applicable to the precariously housed and homeless, was significantly associated with illness symptoms. This may be due to the fact that the variable reflects general instability, recurrent uprooting, and the stress involved in such frequent moves that could take a toll on health.

In comparing the additional contributions of health behaviors, psychological characteristics, and social factors (after having controlled for demographics, chronic health problems, and SES variables), it was found that social and psychological factors accounted for significant amounts of variance in health symptoms, while health behaviors did not. Both individual variables within the psychological set, hostility and depression, showed significant relationships with greater health symptomatology. Similarly, both variables in the social factors set, stressful events and social support, contributed significantly to the variance explained in health symptoms.

The abuse of substances, a well-known illness promoter, made up the block of variables accounting for the least variance in baseline health symptoms in comparison to psychological and social factors. With regard to the insignificant relationship of smoking, alcohol abuse, and drug abuse with symptoms, it is possible that the health manifestations of these behaviors have not yet developed by this time in the participants' lives, given the fairly young average age of this representative sample of homeless adults (mean = 33). Another possible reason for this nonsignificant result is the relatively high base rates of these poor health habits among the homeless. In the current sample, 56% and 49% have satisfied diagnostic criteria for alcohol and drug abuse, respectively, during at least one period in their lives, while 83% currently smoke tobacco. Therefore, it may be that there is not enough variance within these measures of health habits for detection of relationships with illness symptoms in this segment of the population. When wider cross-sections of society have been studied, relationships have often been found between over-indulgence in substances and illness (Adler et al., 1994; Charness, Simon, & Greenberg, 1989; Smith, Cloninger, & Bradford, 1983).

Given the cross-sectional nature of these baseline analyses, a causal interpretation of the findings is not warranted, since there is no temporal distance between most of the independent variables and the criterion and none of the variables could be under experimental control or manipulation. Hostility, stress, lack of social support, and depression could all just as well be developments catalyzed by poor health status. In other words, symptoms of illness could engender hostility, encourage more stressful experiences, and cause social contacts to withdraw due to the excessive burdensomeness of a participant's health problems. Also, it is equally possible that some third, unmeasured variable caused the poor health as well as the hostility, stress, social isolation, or depression to develop.

Another caveat cautioning against over-interpretation of the baseline findings involving hostility, stressful events, and lack of social support is that the method of assessing each construct was similar in nature to the method used to inquire about health symptomatology. In each case, the measure used was a questionnaire that posed a specific list of events, feelings, or symptoms to be answered by the participant in a forced-choice or Likert-scale format. Therefore, some of the association seen between the predictors and the criterion could be a reflection of shared method variance between the questionnaires.

Depression holds up better against the problem of shared method variance, since each participant provided information during the structured interview process of the DIS that classified lifetime diagnostic status with regard to this disorder. Although it is reliant on self-reported symptoms, the DIS is in a very different format than the checklists comprising the other measures and, therefore, probably shares less method variance with them.

Controversy also exists surrounding the methodological value and accuracy of self-reported health. In the one study that has addressed the issue as it applies to the homeless, researchers found that self-reported ratings of symptoms, chronic health problems, and functional health status did accurately identify the homeless people in the poorest health, when confirmed by physical examinations (Gelberg & Linn, 1989). Of concern, however, was the fact that homeless people were unreliable in the assessment of certain aspects of health, such as vision, and were frequently unaware of subjectively unnoticeable conditions, such as hypertension (Gelberg & Linn, 1989). As mentioned above, the participants' endorsements on the PHSC in the current study provide evidence of the measure's content validity, since the most frequently indicated illness categories correspond with diagnoses made in previous studies (Gelberg & Linn, 1990, 1992; Jahiel, 1992; Wright & Weber, 1987).

Other researchers, however, have found evidence that self-reports of overall health in the general population are associated with psychological well-being, such that both factors may distort one another (Hooker & Siegler, 1992; Johnson, Stallones, Garrity, & Marx, 1991). Evidence that psychological characteristics may lead to biases on self-reported health measures has been documented by some research groups. For example, a higher degree of neuroticism has been shown to partially account for higher levels of self-reported somatic distress, while documented, verifiable symptoms of illness also account for some of the somatic distress reported (Costa & McRae, 1985,1987; Watson & Pennebaker, 1989). It has been suggested that more neurotic individuals tend to subjectively experience more distress and, therefore, report more problems on self report inventories, in general. With regard to self-report health measures, specifically, more objective signs reflecting organically based phenomena (e.g., fever, swollen glands, cold sores) may be less influenced by a neurotic style than vaguer, more subjective symptoms, such as fatigue or loss of appetite (Tomakowsky, 1994; Watson & Pennebaker, 1989).

To Table of Contents

Main Analyses: Longitudinal Predictors of Health

The longitudinal follow-up analyses identifying predictors of change in health symptoms over 12 months provide results that can be interpreted with more confidence, given that they address many of the methodological concerns cited above. The current results provide more clarity regarding the direction of the effects, since the predictors were measured at baseline, whereas the criterion variable was the change that had occurred in symptom endorsements in the twelve months that had passed since baseline. Therefore, the criterion could not have conceivably influenced the predictors. It is still possible, however, that the predictors as employed in this study are not, in fact, the actual causal agents, but that some unmeasured variable is determining the observed associations between predictors and criterion.

The possible influences of moderating or mediating variables, response biases, and personality styles have been reduced in the longitudinal analyses, however, since the baseline health symptoms score was entered first into the regression equations at follow-up, thereby leaving only residualized change in health symptoms as the criterion for the remaining predictors. The effects on health symptoms scores attributable to measurement method, response biases, and personality characteristics would be expected to remain fairly constant over twelve months, yielding the same influences on both the baseline and follow-up PHSC scores. When the variance of the baseline score is removed from that of the follow-up score, the effects of all of those potential nuisance variables that could have influenced the self-reports at both time points are statistically controlled. In fact, the baseline health symptoms accounted for a generous 19.28% of the variance in the follow-up score.

As in the results at baseline, demographic variables did not significantly predict change in health symptoms. While chronic health problems were significantly related to acute problems at baseline (following the entry of demographics into the equation), they did not account for a significant amount of variance in the change in acute problems during the follow-up period.

After controlling for demographics and chronic health problems, the residualized change in health scores was significantly predicted by housing transience (as in the cross-sectional analyses). No other SES variables proved to be significant predictors of change in symptoms. In contrast to the more widely used indicators of SES (i.e., income and educational achievement) that are applicable to all levels of privilege, housing transience can be viewed as an SES variable that captures more accurately the quality of life of homeless and poverty-stricken people, specifically. In addition, results of this study indicate that instability of living conditions for homeless people also increases their vulnerability to illness. These results are particularly striking in that they achieve statistical significance even within a sample of people who have experienced at least one episode of homelessness and who have generally had a number of housing moves in their adult lives (mean = 12 moves). This adds credence to the assertion by previous researchers that even small differences in economic privilege correspond to differences in future mortality or illness.

The comparison between the added contribution of psychological characteristics, social factors, and health behaviors showed that they accounted for only 2.48%, 0.66%, and 0.24% of the variance in change in health, respectively. The psychological set was the only one that maintained statistical significance at baseline and follow-up. Individuals who fit diagnostic criteria for depression at any time prior to baseline showed worsening health symptomatology in comparison to the non-depressed subjects. As stated above, this finding emerges even after having controlled for background characteristics and response biases that might cause depressed people to reflect more negatively on their health status than do the non-depressed. Hostility, however, did not maintain the significant association as a longitudinal predictor of symptom changes that it had demonstrated with baseline health problems.

The negative effect that depression may have on health status could be the result of one or more processes (Carney, Freedland, Rich, & Jaffe, 1995). First, it may be that depression causes the immune system to weaken and, therefore, fight off illness less effectively. Second, depression may cause one to lose the initiative to care for one's health. For example, depression could lead to attempts to self-medicate through substance abuse, to adopt an unhealthy diet, to lessen the amount of exercise one gets, or to stop following one's medical care routine for an illness. Third, a person could have a physiological predisposition to developing depression and physical health problems, with depression simply tending to appear earlier in life.

Given the fairly large compilation of existing studies on the psychosocial correlates of health in the general population, it is somewhat surprising that so few significant results were obtained in the current study, particularly longitudinally. The sample sizes of 413 at baseline and 257 at follow-up are sufficiently large (and larger than most studies on homelessness to date) to provide statistical power for the analyses. In fact, some of the significant findings involved very small effects. For example, housing transience accounted for 1.7% of the variance in baseline health symptoms and 1.5% of the variance in symptom changes when entered in the multiple regression equations after demographics and SES variables had been controlled for, yet these relationships achieved statistical significance. Therefore, if the sample size had been larger, even smaller effects might have reached significant statistical levels. The practical implications of such small effects, however, would be questionable.

One possible explanation for the relative lack of longitudinal results is that the time frame of twelve months follow-up is simply too short a period to document predictors of change in health status. Although increases or decreases in number of symptoms experienced could conceivably occur quickly, perhaps marked change sufficient for more statistically significant results may require longer follow-up.

Another possible explanation for non-findings in the current study is that some of the baseline psychosocial variables used to predict change in health over twelve months may not capture the most relevant time frame. More specifically, the following variables are calculated based on cumulative data since age 16 for each subject: housing transience, employment income, time homeless, public assistance income, drug abuse, alcohol abuse, and depression. It is possible that these variables would predict health symptoms and symptom changes more powerfully if they were calculated so as to reflect more recent life circumstances, such as what had gone on within just the 12 months prior to baseline. For example, maybe an impact of alcohol abuse on health would emerge if alcoholism occurring only within the past year were considered, as opposed to the presence of alcoholism at any time since adolescence.

The current design also does not allow for the examination of factors occurring since baseline that may have stronger influence on acute, and therefore quite variable, health. For example, it is possible that the amount of time spent homeless since baseline measurement, rather than preceding baseline, is a stronger predictor of acute health changes. Such an approach, however, would require a correlational analysis, thereby making it more difficult to tease apart causal influences.

Another challenge to the study of the process of change in the social sciences is the unreliability inherent in such measurement. In the physical sciences, the quantification of change is generally quite reliable, unlike in the social sciences, since changes in measurements such as temperature, weight, and volume can be accurately expressed in simple difference scores. The factors that complicate the measurement of change in the social sciences are individual differences in change between subjects and measurement error (Cohen & Cohen, 1975). Although the two-week test-retest reliability of the Physical Health Symptoms Checklist was shown to be adequate at .85 in a separate reliability sample, perhaps the measurement error involved was sufficient to prevent the emergence of results. It is also possible that individual differences in change in this sample formed an obstacle against an accurate measurement of true change.

The conceptual clarity of the construct of health is also a difficult empirical question. As was stated earlier, existing studies have mainly relied on very observable and verifiable health-related events, such as death, myocardial infarctions, or cancer reoccurrences. Some studies in gerontology have used measures that quantify progression of overall disability, which possibly involves slightly more subjective judgment. In the current study, however, the concept of health employed as the dependent variable is that of self-reported, discreet symptoms. Since this is not a commonly used measure of health in research, the amount of change to be expected in such a symptom profile is not well understood. In addition, in this study there was no independent confirmation available of the self-reported symptoms, which calls into question the accuracy of the participants' perceptions of their health.

Mentioned above is also the suppression of statistical relationships that could have been due to restriction of range. Truncated range among any sample of homeless people could be due to two factors. First, homeless people are generally among those with the fewest resources in this society. Therefore, predictors that might be associated with change in health symptoms in a wider sample of the population, such as traditional indicators of socio-economic status, simply may not contain enough variance in this sample of homeless people to lead to very many statistically detectable relationships. In other words, the effect of homelessness may be that it acts as a "leveler" on many variables, in that they become difficult to use statistically due to the extremeness of the group.

Also, when people are recruited for a study while currently homeless, it is often a low point in their lives which may manifest as a floor effect in the measurement of many psychosocial variables. Since many people experience cyclical homelessness, they experience fluctuating conditions. If the baseline of a study begins during a period of current homelessness, the subjects have, in a manner of speaking, nowhere to go but up, at least over the short term. This is supported by the finding that the number of health symptoms decreased in a statistically significant way (an average of nearly four symptoms) between baseline and follow-up measurement, which contradicts the overall worsening that one might intuitively expect. Such a generally improving trend in this sample may have "washed out" potentially significant results.

Attrition effects could have had some impact on the study's results as well. Although two significant attrition effects out of the 17 tests done do not exceed chance expectations at an alpha level of .05 (Sakoda, Cohen, & Beall, 1954), those participants who were successfully followed over 12 months (n=257) were significantly younger (mean = 34 years) and affected by more stressul events (mean = 16 events) than were those who were lost to attrition (mean age = 32 years; mean stress = 18 events). Although both of these attrition differences are fairly small, it could be that the loss of younger, more stressed participants contributed to a decreased range in the follow-up analyses, thereby masking potential findings. It is difficult to speculate on the degree and direction of such a loss as it applies to change in illness symptoms, however, given that the pattern of younger, more stressed individuals does not lead to an intuitive or theoretical prediction regarding the health changes they might experience. It should be noted, however, that the results of the current study may not be generalizable to such young, highly stressed homeless people who are difficult to track longitudinally.

Despite its limitations, the results of the current study corroborate the findings of previous studies that homeless people report a wide range of health problems. They further indicate that the homeless people who have experienced more instability in housing or have suffered from episodes of major depression show worsening self-reported health when measured over a one year follow-up. The worsening predicted by these variables was in contrast to the average improvement in health symptoms documented for the sample as a whole during the follow-up period.

There are several implications for public policy and intervention that can be drawn from these results. The associations of housing transience and depression with health symptoms cross-sectionally and these variables' prediction of worsening health over time provide a consistent pattern of findings. Many people assume that when one refers to mental illness among the homeless that schizophrenia, not mood disorder, is of primary concern. Typically, however, the rate of phychosis in homeless samples (3%-5%) is much lower than that of depression (20%-30%) (Snow et al., 1986; Susser, Conover, & Struening, 1989; Toro & Wall, 1991; Wright, 1988). Given that depression is a much more easily treatable disorder than is thought disorder, it would make sense to put much more emphasis on diagnosis and alleviation of depression in the homeless population than is currently done. Such efforts would largely need to focus on improved access to mental health care for homeless people, which is currently often unattainable (Rosenblatt, 1992). Treatment programs might also include strong components of outreach work, since depressed individuals might not be expected to have the motivation or energy to seek treatment on their own, especially if they are overwhelmed by the circumstances of being homeless or have no health insurance. Based on the results of this study, efforts toward diagnosis and treatment of depression could also lead to improved physical health in homeless people.

Mental health treatment would be difficult to carry out in the milieu of many homeless people's lives, however, partly because they tend to move around so often. In this study, housing transience itself was a risk factor for poor and worsening health, with the current sample reporting and average of 12 moves and as many as 35 moves in their adult lives. There are a number of reasons why homeless people move often. First, since homeless people are generally quite poor, many have been evicted because of inability to pay rent at one time or another. Second, if they are relying on the charity of friends who provide them with shelter, many people find that generosity wears thin over time or choose not to over-stay their welcome. Third, many shelters have limits to the length of time that people can stay at them, which keeps people traveling between various shelters if they are homeless for any substantial period of time. Based on the results of this study, it can be concluded that the mere sheltering of homeless people on a night-by-night basis may not sufficiently safeguard their health. If they are in circumstances that force them to be moving often, they may be at greater risk for developing illness. Therefore, the goal of facilitating stable shelter and/or housing could be expected to better protect physical health among the homeless than do many aspects of the current circuit of emergency shelters.

Future studies may provide further illumination on the question of the psychosocial predictors of physical health among the homeless if comparison groups are employed. For example, if participants representing a wider range of socio-economic status were included, perhaps further evidence of SES's impact on health would be revealed. Also, the factors that predict worsening or improving health may be different for different groups based on their level of economic and social privilege. Knowledge of such differences could influence public policy decisions regarding homelessness specifically and the broader public health of the country in general.

To Table of Contents


Adler, N. E., Boyce, T., Chesney, M. A., Cohen, S.,

Folkman, S., Kahn, R. L., & Syme, S. L. (1994).

Socioeconomic status and health: The challenge of the

gradient. American Psychologist, 49, 15-24.

Ahern, D. K., Gorkin, L., Anderson, J. L., et al. (1990).

Biobehavioral variables and mortality or cardiac arrest

in the Cardiac Arrhythmia Pilot Study (CAPS). American

Journal of Cardiology, 66, 59-62.

Alstrom, C. H., Lindelius, R., & Salum, I. (1975).

Mortality among homeless men. British Journal of

Addictions, 70, 245-252.

Anda, R., Williamson, D., Jones, D., et al. (1993).

Depressed affect, hopelessness, and the risk of ischemic

heart disease in a cohort of U.S. adults. Epidemiology,

4, 285-294.

Barefoot, J. C., Dahlstrom, W. G., & Williams, R. B., Jr.

(1983). Hostility, CHD incidence, and total morality: A

25-year follow-up study of 255 physicians. Psychosomatic

Medicine, 45, 59-63.

Barefoot, J. C., Dodge, K. A., Peterson, B. L., Dahlstrom,

W. G., & Williams, R. B., Jr. (1989). The Cook-Medley

Hostility Scale: Item content and ability to predict

survival. Psychosomatic Medicine, 51, 46-57.

Barefoot, J. C., Peterson, B. L., Dahlstrom, W. G.,

Siegler, I. C., Anderson, N. B., & Williams, R. B.

(1991). Hostility patterns and health implications:

Correlates of Cook-Medley Hostility Scale scores in a

national survey. Health Psychology, 10, 18-24.

Booth-Kewley, S., & Friedman, H. S. (1987). Psychological

predictors of heart disease: A quantitative review.

Psychological Bulletin, 101, 343-362.

Carney, R. M., Freedland, K. E., Rich, M. W., & Jaffe, A.

S. (1995). Depression as a risk factor for cardiac events

in established coronary heart disease: A review of

possible mechanisms. Annals of Behavioral Medicine, 17,


Carney, R. M., Rich, M. W., Freedlan, K. E., Sarni, J.,

TeVelde, A., Sineone, C., & Clark, K. (1988). Major

depressive disorder predicts cardiac events in patients

with coronary artery disease. Psychosomatic Medicine, 50,


Case, R. B., Moss, A. J., Case, N., et al. (1992). Living

alone after myocardial infarction: Impact on prognosis.

Journal of the American Medical Association, 267, 515-


Cauley, J. A., Donfield, S. M., LaPorte, R. E., &

Warhaftig, N.E. (1991). Physical activity by SES in two

population- based cohorts. Medicine and Science in Sports

and Exercise, 23, 343-352.

Charness, M. E., Simon, R. P., & Greenberg, D. A. (1989).

Ethanol and the nervous system. New England Journal of

Medicine, 321, 442-451.

Cohen, J. & Cohen, P. (1975). Applied multiple regression/

correlation analysis for the behavioral sciences.

Hillsdale, NJ: Lawrence Erlbaum Associates.

Cohen, S., Mermelstein, R., Kamarck, T., & Hoberman, H.

(1985). Measuring the functional componenets of social

support. In I. Sarason & B. Sarason (Eds.), Social

Support: Theory, research, and applications. Dordrecht,

The Netherlands: Martinus Nijhoff.

Cohen, S., Tyrell, D. A. J., & Smith, A. P. (1991).

Psychological stress in humans and susceptibility to the

common cold. New England Journal of Medicine, 325, 606-


Cohen, S., Tyrell, D. A. J., & Smith, A. P. (1993).

Negative life events, perceived stress, negative affect,

and susceptibility to the common cold, Journal of

Personality and Social Psychology, 64, 131-140.

Cohen, S., & Williamson, G. M. (1988). Stress and

infectious disease in humans. Psychological Bulletin,

109, 5-24.

Cohen, S., & Wills, T. A. (1985). Stress, social support

and the buffering hypothesis. Psychological Bulletin, 98,

310- 357.

Costa, P. T., & McCrae, R. R. (1985). The NEO Personality

Inventory Manual. Odessa, FL: Psychological Assessment

Resources, Inc.

Costa, P. T., & McCrae, R. R. (1987). Neuroticism, somatic

complaints, and disease: Is the bark worse than the bite?

Journal of Personality, 55, 299-316.

Dembroski, T. M., MacDougall, J. M., Costa, P. T., Jr., &

Grandits, G. A. (1989). Components of hostility as

predictors of sudden death and myocardial infarction in

the Multiple Risk Factor Intervention Trial.

Psychosomatic Medicine, 51, 514-522.

Derogatis, L. R. (1977). SCL-90-R (revised version) Manual

I. Baltimore, MD: John Hopkins University School of


Diamond, E. L. (1982). The role of anger and hostility in

essential hypertension and coronary heart disease.

Psychological Bulletin, 92, 410-433.

Dohrenwend, B. P. (1973). Social status and stressful

events. Journal of Personality and Social Psychology, 28,


Dohrenwend, B. S., & Dohrenwend, B. P. (1970). Class and

race as status-related sources of stress. In S. Levine &

Scotch (Eds.), Social Stress (pp. 111-140). Chicago:


Eaton, W. W., & Kessler, L. G. (1985). Epidemiologic field

methods in psychiatry: The NIMH Epidemiologic Catchment

Area Program. San Diego, CA: Academic Press.

Escobedo, L. G., Anda, R. F., Smith, P. F., Remington, P.

L., & Mast, E. E. (1990). Sociodemographic

characteristics of cigarette smoking initiation in the

United States. Journal of the American Medical

Association, 264, 1550-1555.

Farr, R. K., Koegel, P., & Burnam, A. (1986). A study of

homelessness and mental illness in the skid row area of

Los Angeles. Los Angeles: Los Angeles County Department

of Mental Health.

Fischer, P. J., & Breakey, W. R. (1987). Profile of the

Baltimore homeless with alcohol problems. Alcohol Health

& Research World, 11, 36-61.

Frasure-Smith, N., Lesperance, F., & Talajic, M. (1993).

Depression following myocardial infarction: Impact on

6-month survival. Journal of the American Medical

Association, 270, 1819-1825.

Freedman, D., Thorton, A., Camburn, D., Alwin, D., & Young-

DeMarco, L. (1988). The life history calendar: A

technique for collecting retrospective data. In Clifford

C. Cloggs (Ed)., Sociological Methodology. American

Sociological Association.

Gelberg, L., & Linn, L. S. (1988). Social and physical

health of homeless adults previously treated for mental

health problems. Hospital and Community Psychiatry, 39,


Gelberg, L., & Linn, L. S. (1989). Assessing the physical

health of homeless adults. Journal of the American

Medical Association, 262, 1973-1979.

Gelberg, L., & Linn, L. S. (1990). Differences in health

status between older and younger homeless adults. Journal

of the American Geriatric Society, 38, 1220-1229.

Gelberg, L., & Linn, L. S. (1992). Demographic differences

in health status of homeless adults. Journal of General

Internal Medicine, 7, 601-608.

Gelberg, L., Linn, L. S., Usatine, R. P., & Smith, M. H.

(1990). Health, homelessness, and poverty: A study of

clinic users. Archives of Internal Medicine, 150, 2325-


Gibson, R. C. (1994). The age-by-race gap in health and

mortality in the older population: A social science

research agenda. Gerontologist, 34, 454-462.

Greenblatt, M. (1992). Deinstitutionalization and

reinstitutionalization of the mentally ill. In M. J.

Robertson, & M. Greenblatt (Eds.), Homelessness: A

national perspective (pp.47-56). New York, NY: Plenum


Haan, M. N., Kaplan, G. A., & Camacho, T. (1987). Poverty

and health: Prospective evidence from the Alameda County

study. American Journal of Epidemiology, 125, 989-998.

Hanzlick, R., & Parrish, R. G. (1993). Deaths among the

homeless in Fulton County, GA. Public Health Reports -

Hyattsville, 108, 488-491.

Hibbs, J. R., Benner, L., Klugman, L., Spencer, R.,

Macchia, I., Mellinger, A. K., & Fife, D. (1994).

Mortality in a cohort of homeless adults in Philadelphia.

New England Journal of Medicine, 331, 304-309.

Harmesen, P., Rosengren, A., Tsipogianni, A., & Wilhelmsen,

L., (1990). Risk factors for stroke in middle-aged men in

Goteborg, Sweden. Stroke, 21, 23-29.

Harris, S. N., Mowbray, C. T., & Solarz, A. (1994).

Physical health, mental health, and substance abuse

problems of shelter users. Health & Social Work, 19,


Harris, T. O. (1991). Life stress and illness: The question

of specificity. Annals of Behavioral Medicine, 13, 211


Hirdes, J. P., & Forbes, W. F. (1993). Factors associated

with the maintenance of good self-rated health. Journal

of Aging and Health, 5, 101-122.

Hooker, K., & Siegler, I. C. (1992). Separating apples from

oranges in health ratings: Perceived health includes

psychological well-being. Behavior, Health, and Aging, 2,


House, J. S., Kessler, R., Herzog, A. A., Mero, R.,

Kinney, A., & Breslow, M. (1991). Social stratification,

age, and health. In K. W. Scheie, D. Blazer, & J. S.

House (Eds.), Aging, health behaviors, and health

outcomes (pp. 1-32). Hillsdale, NJ: Erlbaum.

Jahiel, R. I. (1992). Health and health care of homeless

people. In M. J. Robertson, & M. Greenblatt (Eds.),

Homelessness: A national perspective (pp.133-163). New

York, NY: Plenum Press.

Joesoef, M. R., Wetterhal S. F., DeStafano, F., Stroup, N.

E., & Fronek, A. (1989). The association of peripheral

arterial disease with hostility in a young, healthy

veteran population. Psychosomatic Medicine, 51, 285-289.

Johnson, T. P., Stallones, L., Garrity, T. F., & Marx,

M. B. (1991). Components of self-rated health among

adults: Analysis of multiple data sources. International

Quarterly of Community Health Education, 11, 29-41.

Kaplan, G. A., Roberts, R. E., Camacho, T. C., & Coyne, J.

C. (1987). Psychosocial predictors of depression:

Prospective evidence from the Human Population Laboratory

studies. American Journal of Epidemiology, 125, 206-220.

Kessler, R. C. (1979). Social status and psychological

distress. Journal of Health and Social Behavior, 20, 259-


Kondratas, A. (1991). Ending homelessness: Policy changes.

American Psychologist, 46, 1226-1231.

Ladwig, K. H., Kieser, M., Konig, J., Breithardt, G., &

Borggrefe, M. (1991). Affective disorders and survival

after acute myocardial infarction. European Heart

Journal, 12, 959-964.

Lewin, (1989). The health needs of the homeless: A report

on persons served by the McKinney Act's Health Care for

the Homeless Program. Washington, DC: National

Association of Community Health Centers.

Lovell, A. M. (1984). Modified Life Events Interview.

Unpublished Manuscript, Epidemiology of Mental Disorders

Research Department, New York State Psychiatric

Institute, New York, NY.

Lovell, A. M. (in press). Classification and its risks: An

analysis of psychiatric categorization in homelessness

research. In A. Rosenblatt (Ed.), Homelessness (Vol. III:

A critique of American health and welfare policy: Making

sure the solution isn't the problem. Albany, NY: The

Nelson A. Rockefeller Institute of Government.

Maddox, G. L., & Clark, D. O. (1992). Trajectories of

functional impairment in later life. Journal of Health

and Social Behavior, 33, 114-125.

Marmot, M. G., Smith, G. D., Stansfeld, S., Patel, C.,

North, F., Head, J., White, I., Brunner, E., & Feeney, A.

(1991). Health inequities among British civil servants:

The Whitehall II study. Lancet, 337, 1387-1393.

Matthews, K., Kelsey, S., Meilhan, E., Kuller, L., & Wing,

R. (1989). Educational attainment and behavioral and

biologic risk factors for coronary heart disease in

middle-aged women. American Journal of Epidemiology, 129,

1132- 1144.

McLeod, J. D., & Kessler, R. C. (1990). Socioeconomic

status differences in vulnerability to undesirable life

events. Journal of Health and Social Behavior, 31,


Morse, G., Shield, N. M., & Hanneke, C. R. (1985). Homeless

people in St. Louis: A mental health program evaluation,

field study, and follow-up investigation. Jefferson City,

MO: Missouri Department of Mental Health.

Morse, G., & Calsyn, R. J. (1986). Mentally ill homeless

people in St. Louis: Needy, willing but underserved.

International Journal of Mental Health, 14, 74-94.

Mowbray, C., Solarz, A., Johnson, B. S., Phillips-Smith,

E., & Combs, C. J. (1986). Mental health and homelessness

in Detroit: A research study. Lansing, MI: Michigan

Department of Mental Health

Murphy, J. M., Olivier, D. C., Monson, R. R., Sobol, A. M.,

Federman, E. B., & Leighton, A. H. (1991). Depression and

anxiety in relation to social status. Archives of General

Psychiatry, 48, 223-229.

North, C. S., & Smith, E. M. (1993). A comparison of

homeless men and women: Different populations, different

needs. Community Mental Health Journal, 29, 423-431.

North, C. S., & Smith, E. M. (1994). Comparison of White

and non-White homeless men and women. Social Work, 39,


Pincus, T., & Callahan, L. F. (1985). Formal education as a

marker for increased mortality and morbidity in

rheumatoid arthritis. Journal of Chronic Diseases, 38,


Pugh, H., Power, C., Goldblatt, P., & Arber, S. (1991).

Women's lung cancer mortality, socio-economic status and

changing smoking patterns. Social Science and Medicine,

32, 1105-1110.

Rappaport, J., Seidman, E., Toro, P. A., McFadden, L. S.,

Reischel, T. M., Roberts, L. J., Salem, D., A., Stein, C.

H., & Zimmerman, M. A. (1985). Finishing the unfinished

business: Collaborative research with a mutual help

organization. Social Policy, 15, 12-24.

Regier, D. A., Boyd, J. J., Burke, J. D., Jr., Rae, D. S.,

Myers, J. K., Kramer, M., Robins, L. N., George, L. K.,

Karno, M., & Locke, B. Z. (1988). One-month prevalence of

mental disorders in the United States: Based on five

epidemiologic catchment area sites. Archives of General

Psychiatry, 45, 977-986.

Remington, P. L., Forman, M. R., Gentry, E. M., Marks, J.

S., Hogelin, G. C., & Trowbridge, F. L. (1985). Current

smoking trends in the United States: The 1981-1983

behavioral risk factor surveys. Journal of the American

Medical Association, 253, 2975-2978.

Robertson, M. J. (1992). The prevalence of mental disorder

among the homeless: A review of the empirical literature.

In R. Jahiel (Ed.), Homelessness: A prevention-oriented

approach. Baltimore, MD: Johns Hopkins Press.

Robertson, M. J., Ropers, R. H., Boyer, M. A. (1985). The

homeless of Los Angeles County: An empirical evaluation.

Los Angeles, CA: UCLA School of Public Health.

Robins, L. N., Helzer, J. E., Croughan, J., & Ratcliff, K.

S. (1981). National Institute of Mental Health Diagnostic

Interview Schedule: Its history, characteristics, and

validity. Archives of General Psychiatry, 38, 381-389.

Rogers, R. G., Rogers, A., & Belanger, A. (1992).

Disability-free life among the elderly in the United

States: Sociodemographic correlates of functional health.

Journal of Aging and Health, 4, 19-42.

Ropers, R. H., & Boyer, R. (1987). Perceived health status

among the new urban homeless. Social Science and

Medicine, 24, 669-678.

Rossi, P. H., Fisher, G. A., Willis, G. (1986). The

condition of the homeless in Chicago. Amherst, MA:

University of Massachusetts.

Ruberman, W., Weinblatt, E., Goldberg, J., & Chaudhray, B.

(1984). Psychosocial influences on mortality after

myocardial infarction. The New England Journal of

Medicine, 311, 552-559.

Sakoda, J. M., Cohen, B. H., & Beall, G. (1954). Test of

significance for a series of statistical tests.

Psychological Bulletin, 51, 172-175.

Scherwitz, L., Perkins, L., Chesney, M., & Hughes, G.

(1991). Cook-Medley Hostility Scale and subsets:

Relationship to demographic and psychosocial

characteristics in young adults in the CARDIA study.

Psychosomatic Medicine, 53, 36-49.

Schinka, J. A. (1989). Health Problems Checklist.

Odessa, FL: Psychological Assessment Resources, Inc.

Seccareccia, F., Menotti, A., & Prati, P. L. (1991).

Coronary heart disease prevention: Relationship between

socio-economic status and knowledge, motivation and

behavior in a free-living male, adult population.

European Journal of Epidemiology, 7, 166-170.

Shahtahmasebi, S., Davies, R., Wenger, G. C. (1992). A

longitudinal analysis of factors related to survival in

old age. Gerontologist, 32, 404-413.

Shekelle, R. B., Gale, M., Ostfeld, A., & Paul, O. (1983).

Hostility, risk of coronary heart disease, and mortality.

Psychosomatic Medicine, 45, 109-114.

Smith, E. M., Cloninger, C. R., & Bradford, S. (1983).

Predictors of mortality in alcoholic women: A prospective

follow-up study. Alcoholism: Clinical and Experimental

Research, 7, 237-243.

Snow, D. A., Baker, S. G., Anderson, L., & Martin, M.

(1986). The myth of pervasive mental illness among the

homeless. Social Problems, 33, 407-423.

Solarz, A. L. (1986). Follow-up study in a temporary

shelter for the homeless: A look at quality of life and

social supports. Unpublished doctoral dissertation,

Michigan State University.

Sorlie, P. D., Backlund, E., & Keller, J. B. (1995). US

mortality by economic, demographic, and social

characteristics: The National Longitudinal Mortality

Study. American Journal of Public Health, 85, 949-956.

Sosin, M. R., Colson, P., & Grossman, S. (1988).

Homelessness in Chicago: Poverty and pathology, social

institutions and social change. Chicago: School of

Social Service Administration, The University of Chicago.

Stone, A. A. L., Bovbjerg, D. H., Neale, J. M., Napoli, A.,

Valdimarsdottir, H., Cox, D., Hayden, F. G., & Gwaltney,

J. M., Jr. (1992). Development of the common cold

symptoms following experimental rhinovirus infection is

related to prior stressful life events. Behavioral

Medicine, 13, 70-74.

Susser, E., Conover, S., & Struening, E. L. (1989).

Problems of epidemiologic method in assessing the type

and extent of mental illness among homeless adults.

Hospital and Community Psychiatry, 40, 261-265.

Tessler, R. C., & Dennis, D. L. (1989). A synthesis of

NIMH-funded research concerning persons who are homeless

and mentally ill. Unpublished manuscript, Social and

Demographic Research Institute, University of

Massachusetts, Amherst, MA.

Tofler, G. H., Stone, P. H., Maclure, M., Edelman, E.,

Davis, V. G., Robertson, T., Antman, E. M., Muller, J.

E., & The MILIS Study Group. (1990). Analysis of possible

triggers of acute myocardial infarction (The MILIS

Study). The American Journal of Cardiology, 66, 22-27.

Tomakowsky, J. (1994). Optimism and coping in relation to

health status in individuals with human immunodeficiency

virus. Unpublished doctoral dissertation, Wayne State


Toro, P. A. (August, 1993). A systematic, three-step

methodology for representative sampling of homeless

persons. Presentation at the 101st Annual Convention,

American Psychological Association, Toronto, Canada.

Toro, P. A., Bellavia, C. W., Daeschler, C. V., Owens, B.

J., Wall, D. D., Passero, J. M., & Thomas, D. M. (1995).

Distinguishing homelessness from poverty: A comparative

study. Journal of Consulting and Clinical Psychology, 63,


Toro, P. A., Passero, J. M., Bellavia, C. W., Daeschler,

C. V., Wall, D. D., & Thomas, D. M. (August, 1995).

Evaluating an innovative program for homeless adults:

Results of a field experiment. Presentation at the 103rd

Annual Convention, American Psychological Association,

New York, NY.

Toro, P. A., & Wall, D. D. (1991). Research on homeless

persons: Diagnostic comparisons and practice

implications. Professional Psychology: Research and

Practice, 22, 479-488.

Vredevoe, D. L., Brecht, M. L., Shuler, P., & Woo, M.

(1992). Risk factors for disease in a homeless

population. Public Health Nursing, 9, 263-269.

Watson, D., & Pennebaker, J. W. (1989). Health complaints,

stress, and distress: Exploring the central role of

negative affectivity. Psychological Review, 96, 234-254.

Williams, R. B., Barefoot, J. C., Califf, R. M. et al.

(1992). Prognostic Importance of social and economic

resources among medically treated patients with

angiographically documented coronary artery disease.

Journal of the American Medical Association, 267, 520-


Winkleby, M., Fortmann, S., & Barrett, D. (1990). Social

class disparities in risk factors for disease: Eight-year

prevalence patterns by level of education. Preventive

Medicine, 19, 1-12.

Wright J. D. (1988). The mentally ill homeless: What is

myth and what is fact? Social Problems, 35, 182-191.

Wright, J. D., & Weber, E. (1987). Homelessness and health.

New York, NY: McGraw-Hill.

To Table of Contents





December, 1995

Adviser: Paul A. Toro, Ph.D.

Major: Psychology (Clinical)

Degree: Doctor of Philosophy

This study identifies baseline variables that predict change in self-reported physical health status among a sample of 420 homeless individuals over 12 months of longitudinal measurement. The subject pool consists of a representative sample of homeless men and women in Buffalo, New York gathered with methodological care across diverse sites. The representative sample and the prospective design with a low rate of attrition for a homeless sample (38% over 12 months) offer methodological improvements over most past research on health among the homeless. The comprehensive health checklist used has demonstrated reliability (test-retest r=.85) and is based on a measure used in a number of prior studies on he homeless. Hierarchical multiple regression analyses were done on total symptoms at baseline and 12-month follow-up using a variety of demographic, social, and psychological variables as predictors. Baseline results suggested that statistically significant (p<.05) correlates of greater symptoms were poor health history, high housing transience, depression, hostility, high stress, and low social support. Results at 12-month follow-up indicated that after controlling for baseline symptomatology, statistically significant (p<.05) predictors of poorer health outcomes were high housing transience and depression. These results have implications for identifying the homeless people most at risk for poorer health over time, who, therefore, have the greatest need for policy initiatives and clinical interventions involving housing and physical and mental health care.

| Home Page | E-mail: lrowland@wwnet.com | Resume | Graduation |