|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
NEIGHBORHOOD, HOUSING AND WOMEN' S HEALTH DISPARITIES
DINAH PHILLIPS WELCH
A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
Dinah Phillips Welch
First and foremost, I want to acknowledge my family, especially my husband,
David, who has been an unwavering source of support throughout my doctoral education.
In addition to providing emotional support, he has served as both a research assistant and
laboratory consultant. His advice during key time points in my research helped me to
produce quality cortisol results. Dave, I am truly fortunate to have such a supportive and
I am grateful to my children who also supported me through this process with
humor and love that I will always cherish. Their patience, understanding and support kept
me going through the tough times.
Next, I want to thank Dr. Shawn Kneipp for her mentorship and guidance over the
last four years. She has been a source of inspiration since our first meeting. Dr. Kneipp's
high standards are admirable and serve her students well in preparing them to be good
nurse scientists. I thank her for taking the time to be a conscientious and caring mentor.
I gratefully acknowledge my dissertation committee Dr. Nabih Asal, Dr. Kristen
Larsen, Dr. Ichan Huang, and Dr. Sandra Seymour for their expertise and advice
throughout this process.
Finally, I want to thank the National Institutes of Health, National Institute of
Nursing Research and the University of Florida, College of Nursing for partial funding of
this proj ect.
TABLE OF CONTENTS
ACKNOWLEDGMENT S ................. ................. iii.._._. ....
LI ST OF T ABLE S ...._.._ ................ ................. vii...
LI ST OF FIGURE S ........._.._.. .......... ............... viii..
AB SRAC T ................. ................. ix.............
1 INTRODUCTION ................. ...............1.......... ......
Background and Problem Statement .............. ...............1.....
Theoretical Framework............... ...............4
Problem Statement ................. ...............5.................
Purposes of the Study .............. ...............5.....
Limitations ................. ...............6.................
Significance .............. ...............7.....
2 LITERATURE REVIEW ................. ...............9................
Socioeconomic Position and Health ................ ............... .... .. .. ....... ........ 9
Socioeconomic Position and Chronic Disease Health Disparities ......................14
SEP and Women' s Health Disparities ................. ...............15...............
Neighborhood SEP and Health ................. ...............17................
Neighborhoods and Health ................ ...............18........... ....
The Concept of Neighborhood ................. ................ .......... ............1
Neighborhood Disadvantage, Disorder and Health ................. ............. .......20
Neighborhood Social Cohesion and Health ............... .. ...... ..... ..........2
Neighborhood Aspects of Subsidized Housing: Implications for Women' s
Health ................. ...............23.................
Housing and Health ....................... ...............25
Housing and Women's Health............... ...............26.
Housing Policy and Health ................... .. ......... .. ...............27....
What We Know and Gaps in Knowledge and Research ................. .................3 1
SE S and C hroni c Stre ss: The Role of the Hyp othal amic-Pituitary -Adrenal (HPA)
Axi s in Chronic Di sease ................. ..... ........ ......... .... .........3
HPA Axis Physiology and its Role in Chronic Disease Development.......................34
Relationships Among Neighborhood Characteristics, Housing, Chronic Stress,
and H health .............. ...............37....
Summary .............._ ....._. ...............3 8...
3 METHODOLOGY .............. ...............40....
Theoretical Framework............... ...............4
Research Design ............... ...............43....
Population and Sample ..........._.._ ....._. ...............43...
Setting ..........._.... ............._ ...............46....
Human Subj ects Protection ............. .....__ ...._ ............4
Inclusion and Exclusion Criteria .............. ...............46....
Research Variables and Instruments................ ..............4
Maj or Study Variables ........._.___..... .___ ...............48....
Neighborhood Characteristics .............. ...............49...
Neighborhood Economic Disadvantage ....._.__._ ..... ... .__. .. ...._._.........49
Neighborhood Disorder ......__................. .........__ .......... 5
Neighborhood Stress: Crime Exposure .............. ...............50....
Neighborhood Social Cohesion............... ...............51
Housing .........._.... ... .. ....... ... ... ._. ........... .......5
Housing Satisfaction (Perceived Housing Quality) ....._.__._ ........___ ...............52
P erceived Stre s s............... .. ......_ ...............53...
Unfair Treatment and Discrimination .............. ...............53....
Chroni c Stre s s........._... ...... ._ ._ ...............54..
Psychological Distress ........._.._ ..... .___ ...............55.....
Depression ........._... ......___ ...............55....
State-Trait Anxiety .............. ...............55....
General Health ........._.._ ..... .___ ...............56.....
Salivary Cortisol (SC)............... ...............56..
Individual Social Support ................. ...............61................
Study Protocol .............. ...............61....
Statistical Analyses ................. ........... ...............63.......
Statistical Analysis Approach............... ...............63
Specific Aim 1 .................. ...............64..
Issues of Multicollinearity ................. ......... ...............65......
Seemingly Unrelated Regression .............. ...............66....
M ulti-level Analysis .............. ...............67....
Specific Aim 2............... ...............67...
Specify c Aim 3 .............. ...............68....
M missing Data................ ...... ............6
Handling Missing Cortisol Data ................. ...............69................
Handling Missing Survey Data .............. ...............70....
4 RE SULT S .............. ...............73....
Descriptive Results ................. ...............73.......... .....
Description of the Sample .................. .. ......... ...............73. ....
Neighborhood Characteristics of the Sample.................... .. ......... ................74
Stress, Psychological Distress, Health, and Salivary Cortisol Sample
Characteri sti cs ............_. .... ..._ .... .. ..__ ..... .. ...........7
Specific Aim 1: Associations among Neighborhood Characteristics, Stress,
Psychological Distress, Health and Salivary Cortisol............._._.................. .77
Bivariate Analyses of Neighborhood Characteristics, Housing Satisfaction,
Stress, Depression, State Anxiety, Health and Salivary Cortisol ....................78
General Health, Neighborhood Characteristics, Stress, and Psychological
D i stores s................... ........ ._. .............. ........ ..........7
Neighborhood and Individual Level Effects on State Anxiety ...........................80
Depression, Neighborhood Characteristics and Stress ........._..... ..................81
Seemingly Unrelated Regression Analysis of Anxiety and Depression
Regression Equations ............... .. .... .. ....... .... ...... ..... ... ........8
Specific Aim 2: Differences in Neighborhood Characteristics by Housing
Subsidy Type .............. .. .. .. ... ...........................8
Specific Aim 3: Differences in Stress, Psychological Distress, Health and
Salivary Cortisol by Housing Type............... ...............87..
5 DISCUSSION AND RECOMMENDATIONS .............. ...............90....
M aj or Findings. ........... ......_ _. ...............90...
Sample Characteristics ................... ... .. .... .. .. ........9
Specific Aim 1: Relationships between Neighborhood Characteristics Stress,
Psychological Distress, Health and Salivary Cortisol ................. ................. 92
Neighborhood level hypotheses ................. .... ...............92
Discussion Regarding Individual Level Hypotheses .............. ..................94
Discussion Regarding Neighborhood Effects on Psychological Distress,
Health and Salivary Cortisol............... ...............98
Conclusions .............. ... .. ... .... .. .. .. .......9
Specific Aims 2: Differences in Neighborhood Characteristics, Housing
Satisfaction, by Housing Subsidy Type .................. .... .. .......... ..........._.......99
Specific Aim 3: Differences in Stress, Psychological Distress, Health and SC-
AUCg by Housing Type .............. ...............102....
Study Limitations. ................... ... ......... ... ......... .. .. ..........10
Implications for Public Health Nursing Research and Practice. ............... ... ............104
CONSTRUCTS, CONCEPTS,AND OPERATIONAL MECHANISMS ................... ....107
LIST OF REFERENCES ................. ...............110................
BIOGRAPHICAL SKETCH ................. ...............127......... ......
LIST OF TABLES
3-1 Skewness and Kurtosis for Study Variables .............. ...............64....
3-2 Collinearity Diagnostics for Explanatory Variables .............. ....................6
3-3 Example of Missing Cortisol Data for One Participant ................. ............... .....70
4-1 Sample Demographic Profile: (n=67) .............. ...............74....
4-2 Sample Description of Neighborhood Characteristics ................. .....................75
4-3 Stress, Psychological Distress, Health and Salivary Cortisol Scores.......................76
4-4 Salivary Cortisol Scores by Day and Time .............. ...............76....
4-5 Mean Psychological Distress and General Health Scores Compared to National
Norms ................ ...............77.................
4-6 Correlations between Neighborhood Characteristics, Housing Satisfaction,
Psychological Distress, General Health, and Salivary Cortisol .............. ................78
4-7 Bivariate Regression Results for General Health ........................... ...............79
4-8 Bivariate Regression Results for State Anxiety .............. ...............80....
4-10 Neighborhood, Psychosocial, and Individual Effects on Depression (CES-D).......81
4-11 Regression Results for Neighborhood and Psychosocial Measures as Predictors
of Depres si on........._._.._...... .___ ...............83....
4-12 Seemingly Unrelated Regression Analysis of Anxiety and Depression Equations .84
4-13 Simple Regression SC-AUCg .............. ...............85....
4-14 Multiple Regression of Individual Level Characteristics on SC-AUCg .................. 85
4-15 GEE Population Averaged Model of Effects of Neighborhood Characteristics,
Stress and Psychological Distress on Salivary Cortisol .............. .....................8
LIST OF FIGURES
3-1 Socio-biological M odel .............. ...............42....
4-1 Neighborhood Economic Disadvantage (NED) for all Participants ........................75
4-2 Neighborhood Economic Disadvantage (NED) by Housing Subsidy Type ............88
Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy
NEIGHBORHOOD, HOUSING AND WOMEN' S HEALTH DISPARITIES
Dinah Phillips Welch
Chair: Shawn Kneipp
Major Department: Nursing
Humans do not exist in a vacuum; as such, health and illness do not occur entirely
as the result of individual behaviors. People are intrinsically both social and physical
beings and are therefore affected by myriad social factors. Lived experiences vary
tremendously depending on the area one inhabits. The environment constitutes different
contextual aspects that shape one's daily experiences, the social and physical attributes of
neighborhoods and housing are-at the same time-both a product and mediator of larger
social, economic, and political forces. Despite the numerous studies that have established
a clear relationship among neighborhood disadvantage, housing, and health, the
mechanisms by which neighborhoods and housing impact health remain unknown.
Furthermore, a more thorough understanding of how Section 8 and public housing
environments differ is critical, given that the policy intent behind section 8 housing is to
reduce pockets of poverty and its sequelae that have been observed in public housing.
The primary purposes of this study were to examine the relationships among
neighborhood characteristics, perceived stress, psychological distress, and salivary
cortisol secretion among female heads of household with children of low socioeconomic
position (SEP) and to determine the differences in neighborhood characteristics, housing
satisfaction, perceived stress, psychological distress, and salivary cortisol levels in low
SEP female heads of households with children.
Regression analyses indicate that neighborhood characteristics such as disorder,
crime exposure and collective efficacy are associated with increased levels of stress,
psychological distress, and general health. However when individual level factors are
added to models, neighborhood characteristics no longer have an effect on depression,
anxiety, health or salivary cortisol in this group of women. Mann-Whitney U tests
showed a significant difference in neighborhood economic disadvantage by housing type.
Women living in section 8 housing units were located in more economically advantaged
areas (z = -2.552, p<0.05) than women living in public housing. There were no
differences in neighborhood disorder, exposure to crime, nor collective efficacy. Future
studies that replicate this one using a much larger random sample are needed to provide a
better understanding of the impact of neighborhoods and housing on health.
This chapter introduces the background, theoretical framework, and problem
statement. The study purposes and associated hypotheses are stated. Limitations are
acknowledged and the significance of the study is presented.
Background and Problem Statement
Women and children comprise the greatest proportion of people living in poverty in
the United States (U. S.). As of 2004, the official U. S. poverty rate was 12.7%, which is
up from 12.5% in 2003 (U. S. Census Bureau, 2005). In 2004, the poverty rate for
families was 10.2% comprising almost 7.9 million families. Female-headed families
suffer from poverty disproportionately, with 28.4% (nearly 4 million families) living in
poverty compared to 5.5% of married-couple families (3.2 million families) (Institute for
Research on Poverty, 2005).
Socioeconomic position (SEP) is defined as the social and economic factors that
influence what positions) individuals and groups hold within the structure of society"
(Lynch, 2000), and low SEP has repeatedly been associated with poor health outcomes.
For example, the seminal Whitehall study of British civil servants showed that a gradient
in mortality runs across the social hierarchy from the bottom employment grades to the
top (Marmot, 2003; Marmot et al., 1991). Disparities in health for chronic conditions are
more pronounced among women than men, and a steeper gradient in disparate outcomes
exists at the lower end of the economic strata than at the top (Lynch, 2000; Marmot and
Wilkinson, 1999). Studies show, for example, that women's SEP is strongly and
inversely related to cardiovascular disease (CVD) mortality (Wilkinson, 1996). Poor
women are four to five times as likely to rate their health as fair or poor than women with
higher incomes, and middle-aged women in the highest SEP group can expect to live 2.7
to 3.8 years longer than those in the lowest income group (Robert, 1998). Moreover, there
is evidence that the SEP-related disparities of chronic diseases with some of the highest
prevalence, morbidity, and mortality rates among women have actually widened over the
past several years despite efforts to close the gap (Wilkinson, 1997).
Housing costs account for the largest expenditure for most families and serve as an
indicator of one's social and economic standing within society. Disparities in housing
problems are suffered disproportionately in our society and parallel income-related health
disparities. The broad term housing problems can be applied to a wide range of housing
conditions considered to be sub-standard among developed nations. As such, housing
problems include conditions such as high cost burdens relative to income, overcrowding,
poor conditions, and homelessness, among others.
Currently, the U. S. is in the midst of an affordable-housing crisis. Affordable
housing is defined as spending 30% or less of one's income on housing (Green and
Malpezzi, 2003). However, few U. S. citizens are able to pay such a small amount for
housing and many are financially burdened by its high cost. In 2001, one third of the
nation (95 million people) had housing problems. Two-thirds of the people with housing
problems are low-income as defined by federal policy (household income at or less than
80% of the area median). And poignantly, 32% of the low income people with housing
problems were children (Trekson and Pelletiere, 2004).
To put the scope of housing problems on more familiar terrain, the number of
people that experience housing problems exceeds those who lack health insurance
twofold (Joint Center for Housing Studies of Harvard University, 2004). In addition to
unaffordable housing, crowding is on the rise, and nearly 2 million households live in
over-crowded units (Joint Center for Housing Studies of Harvard University, 2004).
Nationally, 47% of renter households lived in unaffordable housing in 2003, which is a
2% increase from 2002.
Studies are beginning to demonstrate that neighborhood and housing characteristics
are independent determinants of health (Kington and Smith, 1997; Steptoe, Lundwall,
and Cropley, 2000). An estimated 72% of all households receiving rental housing
sub sidies--including Section 8 (S 8) and public housing (PH)-are headed by women,
and many are concentrated in lower socioeconomic areas (Adler and Ostrove, 1999)
which have higher rates of chronic stress, anxiety, depression and CVD (Kington and
Smith, 1997; Steptoe and Marmot, 2002). Increasingly, studies have shown that chronic
stress is associated with the development of chronic illnesses such as insulin resistance,
depression, and CVD (Berkman and Kawachi, 2000; Lundberg, 1999). As a result of
these studies, researchers now hypothesize that arousal of the hypothalamic-pituitary-
adrenal (HPA) axis through chronic exposure to stressors in the social and physical
environment (e.g., neighborhood and housing-related stressors) results in "wear and tear"
on physiological systems, contributing to the development of chronic diseases
To further study this phenomenon, epidemiologists have begun employing multi-
level analyses that include both aggregate- and individual-level variables to examine
determinants of SEP-related health disparities. An ecological-to-biological conceptual
framework called for within public health and nursing parallels the conceptual and
methodological approaches proposed in this study. Thus, this study examines whether
neighborhood characteristics and health outcomes differ by housing subsidy type (i.e.,
section 8 and public housing) and whether neighborhood characteristics may contribute
to SEP-related health di spariti es among women through chroni c stre ss-phy si logical
Nancy Krieger's ecosocial theory and Bruce McEwen' s allostatic load model guide
this research. Ecosocial theory seeks to explore the social biological interface through
which environmental factors affect health (Krieger and Davey-Smith, 2004) and calls for
incorporating the use of the concept "embodiment" in order to capture how social
influences (e.g., housing and the built environment) become literally embodied into
physiological characteristics that influence health. Allostatic load is based on the premise
that physical and psychological stressors occur within a social and economic context, and
that there is individual variation in the stress appraisal process as well as behavioral and
emotional coping mechanisms to the perceived stressor (McEwen, 1999).
Combining these two theories into a socio-biological model allows researchers to
simultaneously explore social and biological variables advancing scientific knowledge as
it relates to the understanding of the social-biological interface that may be mediating
relationships among environments (i.e., neighborhoods), chronic stress, and health.
Details on ecosocial theory and allostatic load, and their relevance to this study are
presented in Chapter. 3.
Despite the numerous studies that have established a clear relationship among
neighborhood disadvantage, housing, and health, the mechanisms by which
neighborhoods impact health remain unknown. Specific to the development of a research
traj ectory in this area, and adding to the body of knowledge regarding neighborhood
effects on health, determining what neighborhood and housing characteristics may most
affect low SEP women's health risks must precede both targeted neighborhood/housing
aggregate-level interventions and individual-level interventions within the highest risk
neighborhoods. More research is needed that incorporates a socio-biological approach in
order to determine the mechanisms by which neighborhoods "get under the skin" and
contribute to the development of chronic disease.
Purposes of the Study
The specific aims of this study are as follows:
1. To determine the relationships among neighborhood characteristics,
perceived stress, unfair treatment psychological distress, and salivary
cortisol secretion among low SEP female heads of household with
2. To determine the differences in neighborhood characteristics of two
subsidy housing types, specifically section 8 and public housing, in which
low SEP female heads of households with children live; and
3. To examine the differences in housing satisfaction, perceived stress, unfair
treatment psychological distress, and neuroendocrine regulation,
specifically cortisol secretion, in low SEP female heads of households
with children by housing subsidy type (i.e., section 8 and public housing).
The following hypotheses are investigated in this dissertation
1. Significantly higher levels of crime rates, neighborhood disorder,
neighborhood stress, and neighborhood disadvantage will be positively
associated with salivary cortisol.
2. Public housing sites will have significantly more neighborhood disorder,
greater levels of neighborhood disadvantage, higher levels of
neighborhood stress more perceived crime rates, higher obj ective crime
rates, and lower levels of collective efficacy than S8 sites.
3. Women living in PH will experience significantly lower levels of housing
satisfaction, have higher levels of perceived stress, psychological distress,
and greater alterations in salivary cortisol secretion than women living in
This study has several limitations and therefore the Eindings should be interpreted
with caution. First, non-probability sampling limits the generalizability of this study to
other populations. Using a random sample of neighborhoods and a random sample of
participants from each neighborhood would improve the generalizability at the population
level and is vital to conducting epidemiological studies. Second, the small sample size
may account for the lack of significant findings among neighborhood characteristics,
psychological distress, health and salivary cortisol. Third, the research design could be
strengthened by utilizing a longitudinal design that collects physiological measures over
several years as opposed to the cross-sectional repeated measures design used in this
study. Finally, most of the measures of neighborhood characteristics were based on
perceptions of the study participants. Using more obj ective measures of crime rates and
neighborhood disorder would prove useful in future studies. However, perceptions of
one's environment are important factors to consider when investigating behavioral and
physiological responses to stressors. The effect of the social environment results from the
fact that the brain and body are constantly communicating via the autonomic nervous
system and the endocrine and immune systems (McEwen, 2005) Thus, the regulation of
stress-related mediators is dependent upon how a potential stressor is perceived as well as
the individual's capacity to cope with that stressor.
Housing is an important social determinant of health, and housing policy in the
U.S. disproportionately affects women living in poverty. An increased understanding of
relationships among neighborhood, housing, and health has the potential to significantly
improve individual and population health. The data from this study in accordance with
other epidemiological studies in this area indicate that neighborhood disorder and
exposure to crime are important factors to consider regarding women's health.
In line with both the Healthy People 2010 (U. S. Department of Health and Human
Services, 2005) goals and the National Institutes of Nursing Research priorities, nursing
has a commitment to reducing health disparities among disadvantaged groups through its
scientific investigations. This research is consistent with both of these emphases within
the public health arena and nursing. Moreover, there is an emerging interest in the
relation between the built environment (i.e., neighborhoods and housing) and health in
the field of urban planning, Professionals from this field are partnering with public health
practitioners and others to improve neighborhood conditions through multidisciplinary
investigations that aim to improve the publics' health by establishing healthy cities
through more effective public policy (Northridge, Sclar, and Biswas, 2003).
Future research in women's health disparities must include examination of social
and contextual factors that mediate SEP and health in order to develop population-based
interventions (Fleury, Keller, and Murdaugh 2000). Specific to the development of a
research traj ectory in this area, and adding to the body of knowledge regarding
neighborhood effects on health, determining what neighborhood and housing
characteristics may most affect low SEP women's health risks must precede both targeted
neighborhood/housing aggregate-level interventions and individual-level interventions
within the highest risk neighborhoods.
Furthermore, lacking in the literature is knowledge regarding whether the
experiences of women living in S8 housing differ from those in public housing, and how
neighborhood characteristics associated with each of these programs affect health.
Studies that discern whether home environment or neighborhood characteristics of S8
housing differ from public housing in ways relevant to health are needed. A more
thorough understanding of how S8 and public housing environments differ is critical,
given that the policy intent behind S8 housing is to reduce the concentrated pockets of
poverty and its sequelae that have been observed in public housing
Knowledge gained from neighborhood, housing, and health research focusing on
subsidized housing (i.e., public and section 8 housing) policies would provide valuable
data from which to evaluate the impact of housing voucher and mobility programs on
health. In addition, such knowledge can assist public health practitioners to secure
financial resources for improving neighborhood conditions. The inclusion of bio-markers
(such as cortisol, blood pressure, and others) to test specific physiological mechanisms
may provide more in-depth knowledge about physiological pathways that may be
affected by social processes such as housing policies and neighborhood conditions and
how they are embodied into physiological processes and thus produces illness (Acevedo-
Garcia et al., 2004; Krieger and Davey-Smith, 2004).
This chapter presents a literature review that concentrates on four maj or areas
relevant to the study aims. The discussion begins by presenting relevant studies on the
broad topic of the relationships among SEP and chronic disease health disparities,
particularly as they relate to women. Then the discussion will become focused at the
neighborhood level presenting study findings that indicate how neighborhood
characteristics contribute to chronic stress related health disparities. Nested within
neighborhood are a variety of housing subsidy types that may also impact health
outcomes. This study focuses on federally subsidized Section 8 and public housing,
therefore, a brief overview of subsidized housing policy, neighborhood characteristics
associated with subsidized housing, and associations between housing and health are
presented. Finally this review presents hypothesized physiological mechanisms that may
be affected when neighborhoods and housing serve as sources of chronic stress. This part
of the discussion focuses on chronic stress effects on physiology, specifically the hypo-
thalamic-pituitary-adrenal (HPA) axis and the development of chronic disease. Due to the
extensive nature of the literature in the area of socioeconomic position and health, this
section of the literature review will focus on studies conducted in the United States and in
Socioeconomic Position and Health
In the past 20 years, research that focuses on the relationship between
socioeconomic position and health has grown substantially. Some studies compare
morbidity and mortality of different socioeconomic groups within individual countries,
while others contrast health experiences across countries, document the extent of
inequalities, and explore possible explanations of differential health outcomes (Feinstein,
1993). Most studies of SEP and health have focused on individual-level SEP (i.e.,
individual income, education, occupation), and the effects on broad health outcomes such
as morbidity and mortality (Robert, 1998).
In a review of the literature on SEP and health research published from 1970 to
1990, Feinstein reports findings from early seminal work and critiques the methodology
of several studies conducted in the U. S. and England. For example, one study utilized
two data sets the 1960 Matched Records Study and the Chicago Area Study. The Chicago
Area Study collected information on census tracts and will be discussed in the section on
neighborhood SEP and health. The 1960 Matched Records Study linked death certificates
with census information on the educational attainment and household income for 340,000
individuals who died during May-August 1960 in the U. S. These findings show a strong
inverse relationship among whites and non-whites, females and males aged 25-64,
between years of schooling and mortality in 1960. The difference in standardized
mortality rates between the least and best educated subgroups was at least 65% for each
of the four classes (i.e., white/non-white men and white/non-white females) (Feinstein,
1993). Furthermore, this study elucidates the fact that the effects of education and income
are largely independent of one another.
Another important source of evidence supporting the SEP-health relationship
comes from the Black Report which was published in 1980. In 1977 Sir Douglas Black
and other researchers were appointed by the British Government to assess the evidence
on inequalities in health in the United Kingdom (U. K.). The Black Report assesses
inequalities using a classification system of the British population in which the
population is divided into six social classes including professional, intermediate, skilled
nonmanual, skilled manual, partly skilled, and unskilled. Household status is determined
by the occupation of the head of household (Feinstein, 1993). The findings from this
study showed that in 1971, substantial mortality differentials existed in the U. K. and had
in fact widened since 1930. In 1971, the mortality rate among men in the lowest
occupational class (unskilled) was 9.88 per 1,000 as opposed to men in the highest class
which was significantly lower at 3.98 per 1,000. The same trend was found among
women as well. Women in the highest (i.e., professional) occupational class had a
mortality rate of 2. 15 per 1,000 while those in the lowest occupational class had a
mortality rate of 5.31 per 1,000 (Feinstein, 1993).
There are several problematic methodological issues that have been identified in
this area of research. First, the socioeconomic indicators used in these early studies,
particularly income, do not adequately account for the possibility that poor health causes
reduced income rather than low income resulting in poor health. In addition, many early
studies use household income measures which for married households -are generally
the male's income and therefore do not accurately reflect the woman's income or access
to the household income. Thus, the income and health argument holds more validity
when applied to men than when applied to women (Feinstein, 1993). Furthermore, these
studies do not account for the impact of unpaid labor (i.e., household duties in addition to
work) on women's health. Lastly, many researchers in this area believe wealth as
opposed to income to be the superior indicator because the problem of reverse causality is
less likely to affect household wealth, than household income measures, primarily
because household wealth accumulates over time and consequently is less affected by a
single episode of illness (Berkman and Kawachi, 2000; Feinstein, 1993).
After The Black Report was published an explosion of research in health
inequalities followed. Much of the work expanded upon the approach used in The Black
Report to include alternative datasets for more recent years. For example, Whitehead
reviewed evidence from the 1979-83 decennial supplement that showed inequality in
mortality rates across social classes was the same as, or slightly larger than before thus
supporting evidence from the Black Report even after circumventing some of the
methodological weaknesses in previous studies (Feinstein, 1993). In a similar study,
utilizing the same data, the researchers merged the social classes into two different
groups manual and nonmanual--and found similar results that indicate a wide
inequality in heart disease and lung cancer rates between the two groups (Marmot and
The more recent Whitehall I and II Studies of British civil servants conducted by
Michael Marmot and colleagues provide further supporting evidence for the social
inequality and health relationship (Williams and Collins, 1995). The Whitehall I Study
examined mortality rates over 10 years among males aged 20-64. An inverse association
between grade (level) of employment and mortality from coronary heart disease (CHD)
and a range of other causes was observed. Men in the lowest grade (i.e., messengers,
doorkeepers, etc.) had three times the mortality rate than men in the highest grade (i.e.,
administrators) (Marmot, Shipley, and Rose, 1984). After controlling for standard risk
factors such as hypertension, smoking, obesity, and physical inactivity, the lowest grade
worker still had a relative risk of 2.1 for CHD mortality compared to the highest grade
worker (Marmot and Wilkinson, 1999).
The Whitehall II study was designed to assess the effects of the social environment
on health and the causes of social inequalities in health. More specifically, it investigates
the role of stress on health and the extent to which stress might be involved in the social
inequalities in health (Marmot, 2004). The study began in 1985 and included 10,308 male
and female civil servants. Findings were consistent with the first Whitehall study. Clear
employment grade differences in health risk behaviors including smoking, diet, and
exercise, economic and social circumstances, and monotonous work characterized by low
control and low satisfaction were present in both men and women (Marmot et al., 1991).
Furthermore employment grade differences were also associated with CHD (Marmot,
Bosma, Hemingway, Brunner, and Stansfeld, 1997), metabolic syndrome and central
obesity (Brunner, 2000).
In addition to the well known Whitehall Studies described above, numerous
studies, such as the Multiple Risk Factor Intervention Trial, (Wilkinson, 1996)
demonstrate that individual-level SEP disparities exist for many chronic diseases (Fleury,
Keller, and Murdaugh, 2000; Kington and Smith, 1997; Marmot and Wilkinson, 1999).
To date, much of the epidemiological research in the area of individual-level SEP
relationships to chronic disease has focused on CHD (Steptoe and Marmot, 2002).
However, a consistent inverse relationship exists between SEP and multiple health
indicators, such as CVD, diabetes, metabolic syndrome, arthritis, tuberculosis, chronic
respiratory illness, malignant melanoma. (Adler and Ostrove, 1999), and lung and
gastrointestinal cancers (Feinstein, 1993).
Socioeconomic Position and Chronic Disease Health Disparities
Socioeconomic position (SEP) has multiple dimensions that are associated with
health. In virtually every dimension of mental and physical health, people in lower-
socioeconomic groups have poorer health than those in the middle- or upper-income
groups (Dalaker, 2001). Regardless of the SEP indicator used, (such as education,
occupation, housing tenure, income), those who are worse off socioeconomically have
worse health (Marmot and Wilkinson, 1999; Smith, Wentworth, Stamler, and Stamler,
1996). The gradient in morbidity and mortality by SEP for several chronic disease states
has been documented for hundreds of years, observed consistently across studies, within
and across countries and cultures, and persists and is actually increasing today (Lynch,
2000; Marmot and Wilkinson, 1999). For example, in a study that examined the extent of
socioeconomic gradients in all-cause and cardiovascular disease (CVD) mortality among
U. S. men and women aged 25-63 years from 1969 to 1998, the researchers found that
area socioeconomic gradients in all cause and CVD mortality increased significantly over
the last three decades (Singh and Siahpush, 2002). The researchers also found that rates
of all-cause and CVD mortality among men in the lowest area socio-economic group
were 42% and 30% greater in 1969-1970 and 73% and 79% greater in 1997-1998
respectively than those in the highest socioeconomic group. Women in the lowest area
socioeconomic group had rates of all-cause and CVD mortality that were 29% and 49%
greater in 1969-1970 and 53% and 94% greater in 1997-1998 respectively than women in
the highest area socioeconomic group. It is important to note, however, that health
disparities are not observed solely at the extreme ends of the socioeconomic spectrum.
Morbidity and mortality risks increase along each incremental decrease in SEP
The Matched Records Study conducted by Kitagawa & Houser and published in
1975 as previously mentioned shows a persistent inverse relationship between
educational attainment and mortality from heart disease for both men and women and
this relationship is stronger for both sexes aged 25 to 64 (Feinstein, 1993). However, the
relationship between educational attainment and cancers is more complex. Cancers
directly related to smoking such as lung cancer, as well as, stomach, intestinal, and rectal
cancers show a strong inverse relationship to education where other cancers (i.e., prostate
and breast) do not (Feinstein, 1993).
SEP and Women's Health Disparities
Very little research has been conducted that specifically addresses women's SEP
and health. Most research in the area of SEP and health has focused on middle-aged
males or have included both males and females utilizing a cross-sectional descriptive
methodology (Beebee-Dimmer, Lynch, Turrell, Lustgarten, Raghunathan, and Kaplan,
2004). One of the reasons that the relationship between women' s SEP and health has not
received much attention is because of the difficulty in conceptualizing and measuring the
class position of those without direct labor market ties (McDonough, Walters, and
Studies that have focused on women's health in relation to SEP have typically been
limited to the quality of social roles, maj or institutionalized roles, and unequal
distribution of resources while the socioeconomic dimensions of women' s lives remain
relatively unexplored (McDonough et al., 2002). Furthermore, a paucity of SEP and
health related research has been conducted among women residing in the United States
(U. S.). Many of the studies reviewed on individual-SEP and women' s health originated
in other industrialized countries such as Canada (McDonough et al., 2002), Britain
(Arber, 1997; Cooper, 2002; Martikainen, Lahelma, Marmot, Sekine, Nishi, and
Kagamimori, 2004; Stafford, Bartley, Mitchell, and Marmot, 2001) Spain (Artazcoz,
Borrell, Benach, Cortes, and Rohlfs, 2004) and Finland and Japan (Martikainen et al.,
Despite the lack of research that focuses specifically on women' s SEP and health,
studies in the U. S. have consistently shown that women of lower SEP have a higher
prevalence of diabetes (Centers for Disease Control, 2000), report higher levels of social
stress such as recent life events, maj or events, and death events (Turner and Avison,
2003), and have significantly poorer mental health (especially lone mothers with
children) (Macran, Clarke, and Joshi, 1996). In addition, studies that have explored
family demands, employment and health in women have found that among women
workers of low educational level, family demands showed a negative effect on health and
health related behaviors (Artazcoz et al., 2004).
As evidenced by this review, the relationship between SEP and health is well
known. What is missing from the literature is research on the mechanisms by which SEP
affects women's health. While the precise mechanisms that mediate individual-level SEP
and health are unknown, studies indicate it is not only access to material resources that
are important, but that psychosocial factors contribute to health disparities, as well. For
example, there is an interaction among high psychological demand/low control
environments, an increased risk of psychological strain, and physical illness (Berkman
and Kawachi, 2000; Lundberg, 1999; Marco, 2000). Thus, in terms of individual-level
measures of SEP, the pathways through which SEP exerts its influence are not only
through access to material resources (e.g., income or health care services), but may also
operate via psychosocial mechanisms, as more fully described later in the literature
review (see section on Hypothalamic-Pituitary-Adrenal (HPA) axis Physiology and its
Role in Chronic Disease Development). Furthermore, the area in which one lives may
also have a significant impact on health. As will be shown in the following sections, more
research is being conducted that explores the relationship between neighborhoods and
housing and their impact on health, which can account for some of the influence of SEP
Neighborhood SEP and Health
In addition to research on individual-level SEP and health, epidemiologists are now
exploring the SEP-health relationship from an aggregate level. Several studies have found
residents of disadvantaged neighborhoods have worse self-reported health and more
chronic health problems than persons living in higher SEP neighborhoods (Ross and
Mirowsky, 2001). Studies are finding that neighborhood-level SEP indicators have a
significant effect on health independent of individual-level SEP (Bosma, Dike van de
Mheen, Borsboom, and Mackenbach, 2001; Roy, Kerschbaum, and Steptoe, 2001). For
example, in a study of 23 5 residents of 19 lower SEP neighborhoods, researchers found
that neighborhood problems constitute sources of chronic stress that may increase the risk
of poor health (Steptoe and Feldman, 2001).
One of the problems in this area of research is the diversity of indicators used to
measure neighborhood level SEP. Some researchers have aggregated random samples of
individual SEP indicators (i.e., income, education, and occupation to the neighborhood
level) (Bosma et al., 2001). Others have used indicators such as percent unemployed,
percent on public assistance, poverty rates, and percent households headed by females at
the census tract level to determine the neighborhood SEP level (Boardman, Finch,
Ellison, Williams, and Jackson, 2001; Ross and Mirowsky, 2001). Krieger and colleagues
found that among eight studies, four used differently categorized measures of
neighborhood social class composition, education, poverty level, and unemployment rate,
two used measures of average annual family income and two used data on median family
income and educational level (Krieger et al. 2003). Despite the SEP measure used, the
SEP-health disparities relationship persists above and beyond individual socioeconomic
and behavioral factors and therefore warrant further investigation (Steptoe and Feldman,
Neighborhoods and Health
The Concept of Neighborhood
Neighborhood is a concept that has myriad definitions depending on the context in
which it is used. A variety of criteria can be used to define neighborhood, including
historical criteria, geographical criteria, resident's perceptions, and administrative
boundaries (Diez Roux, 2003). In addition, the size and definition of the geographic area
may differ based on the outcome being studied.
Neighborhood has been defined as a place where people can easily walk over and
interact with each other and as a social organization of people who reside within a
geographical boundary (Galster, 2001). Kearns and Parkinson (2001) describe
neighborhood as existing at three different levels. These include the home area (a 5-10
minute walk from one's home), locality, and urban district or region. The predominant
function associated with each level of neighborhood is different. For example, the home
area is where the psychosocial purposes of neighborhood is strongest with the main
functions being, relaxation, connecting with others and fostering attachment and
belonging. Localities or sub-districts (e.g., a public or section 8 housing complex)
function as sites for residential activities and positioning oneself within social networks.
Regional aspects or the larger urban districts of neighborhood (e.g., cities or towns)
provide social and economic opportunities. Importantly, Kearns and Parkinson also note
that at the second level of neighborhood (localities), public or low-income housing can be
subject to social exclusion and discrimination imposed upon them by the larger urban
The attributes comprising neighborhood are dynamic and are the result of past and
current flows of households and resources into and out of a defined geographic space
(Kearns and Parkinson, 2001). For the purpose of this study, neighborhood will be
defined according to Galster's (2001) definition that states, "Neighborhood is the bundle
of spatially based attributes associated with clusters of residences, sometimes in
conjunction with other land uses" (p. 2112). This definition accommodates the structural,
class status, environmental, and social inter-active characteristics of a neighborhood.
Structural aspects include the type, state of repair, density, and landscaping of residential
and non-residential buildings and the presence of sidewalks. Class status characteristics
include income, occupation, and educational composition. The degree of noise pollution,
land pollution, and the amount of litter are included in the environmental characteristics
of neighborhood. Finally, social-interactive characteristics include local and family
networks, degree of inter-household familiarity, type and quality of interpersonal
associations, participation in local organizations, and strength of informal social control
Extant literature in this area of research has generally defined neighborhood using
geographical or administrative boundaries. For example, studies conducted addressing
neighborhood SEP and health outcomes typically identify neighborhood by census tract
area (Boardman et al., 2001; Ross and Mirowsky, 2001) or combine census tracts into
"neighborhood clusters" that are ecologically meaningful and internally homogeneous
(Sampson and Raudenbush, 1997). According to the U. S. Census Bureau, census tracts
are small homogeneous areas in which similar population characteristics, economic status
and living conditions are found (The Public Health Disparities Geocoding Proj ect, 2004).
As evidenced by this review, neighborhood can be defined in a variety of ways.
Therefore, it is important to carefully consider the appropriate spatial scale in regards to
the research questions and variables to be studied (Macintyre, Ellaway, and Cummins,
2002). Neighborhood disadvantage is one example of a variable that is appropriately
measured at the census tract level. On the other hand, social networks may not be
bounded by geographical boundaries and may either be much broader spatially or
Neighborhood Disadvantage, Disorder and Health
Neighborhood disadvantage is a term used to describe socioeconomic position of a
locality. Typically measured at the census tract level, neighborhood disadvantage is most
frequently operationalized by developing an index of various indicators such as family
poverty, male unemployed, educational level, public assistance and female-headed
households. Researchers have used percent poverty level, percent female heads of
households, percent male unemployed, percent on public assistance (Boardman et al.,
2001; Sampson and Raudenbush, 1997), or prevalence of poverty and mother-only
households, college educated residents and homeownership (Ross and Mirowsky, 2001).
Economically disadvantaged neighborhoods are characterized by high poverty
rates. Subsidized and other low-income housing units are frequently concentrated in these
high poverty areas. Neighborhood disadvantage has been positively related to higher
levels of stress, lower social resources, and higher levels of anxiety and depression
(Boardman et al., 2001).
Studies show that neighborhood disadvantage in the U. S. is associated with
increased levels of anxiety and depression among adolescents (Aneshensel and Sucoff,
1996), discrimination in the workplace (Kirschenman and Neckerman, 1991), low
birthweight (Buka, Brennan, Rich-Edwards, and Raudenbush, 2003; Pearl, Braveman and
Abrams, 2001) and heart disease (Diez Roux, Merkin, Arnett, Chambless, Massing,
Nieto, Sorlie, Szklo, Tyroler, and Watson, 2001; Sundquist, Malmstrom, and Johansson,
1999). Furthermore, people living in economically disadvantaged neighborhoods have
reported more frequent experiences of unfair treatment (Schulz, Williams, Israel, Becker,
Parker, Sherman, and Jackson, 2000), higher levels of substance abuse (Boardman et al.,
2001), and higher levels of neighborhood disorder (Ross and Mirowsky, 2001).
Neighborhood disorder is a concept that includes both the physical and social
aspects of a neighborhood. Visible signs of physical disorder include high levels of noise,
dirtiness, abandoned and run-down buildings. Vandalism and graffiti are common in
these areas. Social disorder includes higher crime rates and signs such as fights and
trouble among neighbors, the presence of people hanging out on the streets, drinking and
taking drugs (Ross and Mirowsky, 1999).
Neighborhood disorder can be seen as a chronic stressor among urban communities
that can potentially affect health in a variety of ways. For example, perceptions of higher
neighborhood disorder have been linked to greater depressive symptoms after controlling
for baseline depressive symptoms (Latkin and Curry, 2003). Ross and Mirowsky (2001)
found that people living in disadvantaged neighborhoods with high levels of perceived
disorder have higher levels of fear, report more chronic conditions and worse health and
These studies do not include physiological biomarkers that would illuminate the
mechanisms by which neighborhood stressors impact mental and physical health. Fear
stimulates the release of epinephrine and norepinephrine, followed by the release of
cortisol (Ross and Mirowsky, 2001). Chronic release and exposure to these hormones
have been associated with a variety of illnesses such as hypertension (HTN),
hypercholesterolemia, atherosclerosis, and hyperglycemia (McEwen, 2000). Furthermore,
little is known on how neighborhood social cohesion and social networks may mediate
the impact of neighborhood disadvantage and disorder on health.
Neighborhood Social Cohesion and Health
Social cohesion is a new construct in public health research believed to mediate the
relationship between neighborhood SEP and health. It has been studied predominantly as
an important neighborhood characteristic and has been clearly differentiated from
individual social support in that it represents trust among people with some
geographically defined boundary where one lives. Kawachi and colleagues (1997) found
social cohesion explained 58% of the variance of all-cause age- and SEP-adjusted
mortality, and for 15%-20% of the variance in other CVD mortality. Thus, the role of
neighborhood social cohesion in modulating chronic stress-physiological mechanisms
believed to contribute to health disparities among women deserves further study.
However, the concept of social cohesion is wrought with multiple and confusing
definitions. Authors frequently define social capital and social cohesion in much the same
way (Kawachi, Kennedy, and Prothrow-Stith, 1997). For example, Kawachi has used the
same indicators of social capital to define and measure social cohesion in different
studies (Kawachi, Kennedy, Gupta, and Prothrow-Stith, 1999). Both social cohesion and
social capital have been defined as the level of trust between citizens of a community,
norms of reciprocity and participation in civic organizations that cooperate for mutual
benefit (Kawachi, Kennedy, and Glass, 1999). Forrest and Kelly (2001) describe social
cohesion as being "about getting by and getting on at the more mundane level of
everyday life" (p. 2127).
While social phenomenon within neighborhoods such as those represented by
social cohesion are potentially relevant areas for intervention to improve health, nurse
scholars (Drevdahl, Kneipp, Canales, and Dorcy, 2001) and others (Buka et al., 2003;
O'Campo, 2003) have urged caution in assuming increasing social cohesion is a
simplistic or complete remedy for reducing the health disparities observed to occur
among neighborhoods with 'more' or 'less' cohesion. Nonetheless, when considering one
(of several) contributing factors relevant for examining mechanisms underlying health
disparities, the consistency and magnitude of associations observed to date deem it a
worthwhile concept to investigate. For the purposes of this study a socially cohesive
neighborhood is defined as the extent of trust and social interaction within the
neighborhood (Beauvais and Jenson, 2002; Buka et al., 2003; Sampson and Raudenbush,
Neighborhood Aspects of Subsidized Housing: Implications for Women's Health
Nested within neighborhoods are a variety of housing types such as single-family
and multi-family units, owned versus rental housing as well as subsidized housing. High
poverty neighborhoods tend to have more low-income subsidized rental housing units
such as public and section 8 housing (Pendall, 2000). Public housing (PH) consists of
housing units owned and operated by the government and typically houses very low-
income families. PH residents pay rent based on income with a minimum of $25.00 to
$50.00 up to 30% of their monthly income (less deductions allowed by regulations) (U.
S. Department of Housing and Urban Development, 2003b). Section 8 housing is a
voucher program in which recipients look for rental housing in the private market.
Recipients of S8 contribute approximately 30% of their monthly income toward housing
costs with S8 programs paying the remainder of a defined payment standard (U. S.
Department of Housing and Urban Development, 2003a). Nationally, only 14.8% of S8
voucher recipients live in high poverty neighborhoods (Turner, Popkin, and Cunningham,
1999), however, while 53.6% of PH residents live in high poverty neighborhoods.
In the United States, more than 5 million families are living in substandard housing
(Bashir, 2002). Despite often-deplorable conditions, housing is often the highest
expenditure for poor families. Increases in fair market rent (gross rent estimates for a
specified area that include shelter cost and utilities except telephones) are far exceeding
increases in income, particularly among low-income families. When families are forced
to spend most of their income on housing, other important needs such as food, clothing,
health care, and emotional stability suffer (Bashir, 2002).
Approximately 4.8 million subsidized housing units are available in the U.S., with
2.4 million households participating in the Section 8 program, 1.3 million households
living in public housing units, and the remaining 1.1 million units comprised of other
types of housing assistance (U. S. Department of Housing and Urban Development,
2002). Of all available subsidized units in the U.S., 96% are occupied. The occupied units
are comprised of 58% minority households and 70% female-headed households such as
single-mother households and elderly women living alone (U. S. Department of Housing
and Urban Development, 1998).
Housing and Health
Decent, affordable housing not only impacts individual and family health; it is also
the building block of healthy neighborhoods, shaping the quality of life in communities.
Improved housing can lead to better outcomes for individuals and society at large. The
relationship between housing and health has been a long-standing issue in the field of
public health. As early as 1872 in a series of essays entitled "The Housing Question",
Freidrich Engels discussed the relationship between housing conditions and poor health.
He argued that the conditions of the poor, working class areas in cities viewed as
breeding grounds for epidemics could not be ignored without impunity. While the
conditions creating the kind of infectious epidemics Engels addressed have been brought
under control in today's industrialized countries and cities, the spatial concentration of
socioeconomic groups is still observable (Dunn, 2000).
Extant research on housing and health has been mainly concentrated in four areas:
1) the disadvantage of individuals who are already in poor health in the housing market
and their self-selection into substandard housing conditions, which may in turn account
for any observed association between poor housing and poor health; 2) health status and
access to health care for homeless persons; 3) pathological aspects of dwellings as the
presumed cause of both physical and mental health outcomes, and 4) studies that
specifically examine the stresses associated with unaffordable and/or inadequate housing
Regarding physical health, the literature provides evidence on associations among
overcrowding, dampness and mold, indoor pollutants, infestations, and inadequate
heating and infectious respiratory diseases, asthma, rhinitis, eczema, and heart disease
(Marsh, Gordon, Heslop, and Pantazis, 2000). Other studies have shown relationships
between stress, mental health, and housing. For example, in a study on housing stressors,
social supports and psychological distress, researchers revealed that housing stressors are
significantly associated with psychological distress and that living in substandard housing
is an independent and added source of stress to the lives of people with lower incomes
(Smith, Smith, Kearns, and Abbot, 1993). Missing from the literature is research on the
mechanisms by which housing affects physical and mental health.
Housing and Women's Health
Inequalities in women's health that parallel income inequalities are related to
housing conditions in which women live. The negative effects of poverty or near-poverty
on health are often mediated or reinforced by substandard housing. In one case study, a
single mother living in public housing described the physical manifestations and social
consequences of substandard housing that she believed contributed to poor physical and
mental health in women (Welch, 1997) Although there are inherent limitations to single-
case studies, other studies with samples of over 300 women living in public housing
substantiate this finding (Edin and Lein, 1997; Rollins, Saris, and Johnston-Robledo,
2001; Wasylishyn and Johnson, 1998). Interviews of women living in public housing
conducted by Rollins and colleagues (2001) highlighted problems such as structural
damage and safety issues. Nicolas and JeanBaptiste (2001) used focus group sessions to
learn about the experiences and perceptions of women who receive public assistance
(including public housing subsidies). Some of the maj or themes that emerged included
feelings of shame and disrespect, an insecure future, and a sadness regarding life's
outcomes (Nicolas and JeanBaptiste, 2001). Similar themes were identified through
interviews with 13 pregnant women living in PH by McAllister and Boyle (1998),
including discontent, struggling to make ends meet and loneliness. These three core
themes depict the consequences of poverty and living in low-income housing. The
women in this study also viewed housing assistance as degrading and stigmatizing, and
notably, were much more concerned about the violence in their neighborhood than they
were about their current pregnancy (McAllister and Boyle, 1998).
These four studies provided excellent descriptions of the experiences and
perceptions of living in public housing. However, these studies do not address whether
the experiences of women living in S8 housing might differ from those in PH and how
neighborhood characteristics affect health. Whether home environment or neighborhood
characteristics of S8 housing differ from PH in ways relevant to health remain unknown.
A more thorough understanding of how S8 and PH environments differ is critical, given
that the policy intent behind S8 housing is to reduce the concentrated pockets of poverty
and its consequences that have been observed in PH. In addition, few studies have
incorporated an approach that examines the physiological mechanisms by which
neighborhoods and housing may impact health and either exacerbate or attenuate SEP-
related health disparities among women. Nurse researchers, however, are essentially
silent in this domain, even though public health nurses have an extensive history of
addressing neighborhood-related concerns in relation to health (Lundy and Janes, 2001).
Housing Policy and Health
Home ownership is the American dream. Viewed as part of the transition to
adulthood, owning one's home is a common goal for many young men and women. Thus,
the maj ority of housing assistance policies and programs have focused on
homeownership. Unfortunately, due to a combination of factors such as the lack of
affordable housing and an inadequate living wage, many Americans are unable to
become homeowners. Many low-income workers, particularly single-women with
children, are unable to own their own homes and must not only rent, but also rent
utilizing federal housing subsidies. Moreover, minorities and poor women who must
depend on housing subsidies to maintain shelter are often stereotyped as lazy, ignorant,
unkempt or destructive, and thus marginalized, discriminated against and located apart
from the main stream community and its higher level resources (Hays, 1995).
Since the emergence of federal housing policies in the 1930s and the passage of the
Wagner-Steagall Housing Act in 1937, low-income housing has been characterized by
many factors leading to stigmatization and marginalization. By 1942, the United States
Housing Authority (USHA) had built over 100,000 units in over 140 cities (Von
Hoffman, 1996). The construction of public housing utilizing minimal design elements
reflective of the moderns style contributed to the distinctive yet negative image that came
to be associated with public housing. Functional yet austere looking designs and the
placement of high density multi-family housing complexes in super-blocks also
contributed to the distinctive image of public housing Therefore, a sharp contrast to the
types of residences detached single-family homes that most Americans occupied
emerged that in time would stigmatize public housing (Von Hoffman, 1996).
Further investigations of past housing policies reveal blatant discriminatory
language and practice. The 1934 Housing Act, focused on homeownership, guaranteeing
loans for mortgages through government appropriations. However the funds for these
loans were restricted to the production of new and existing homes for a single owner.
Tragically, during this time period, discrimination was prominent and many lenders
practiced red lining, refusing to loan in certain areas of town based on race and ethnicity.
Redlining was supported by the Federal Housing Administration (FHA). The FHA
manuals used by lenders instructed them to avoid areas where discordant racial groups
resided. In addition, the FHA encouraged developers to establish deed restrictions
prohibiting black owners and residents. Discrimination in the housing market became
prevalent. Consequently, property values of minority neighborhoods plummeted, and
neighborhood segregation by race and income was perpetuated (Orlebeke, 2000).
Further alienation and marginalization of low-income and minority persons was
inadvertently propelled by the design of new public housing units. Unfortunately, high-
rise buildings turned into PH disasters due to lack of funding for building design, basic
amenities and maintenance, isolation and alienation from surrounding neighborhoods and
the lack of public space (e.g., parks) (Orlebeke, 2000). Moreover, during the 1950s and
1960s inner city neighborhoods, now termed ghettos, continued to carry a negative
stigma. Burdened with cycles of poverty, lack of formal education, lack of economic
stability, inadequate housing and a maj or reduction in federal funding, life in these areas
become unsafe and unhealthy (Von Hoffman, 1996).
Current trends in housing policy paint a bleak picture for housing subsidy
programs. In 1971 members of Congress argued that high cost, shoddy construction, poor
administration, failure to help low-income families and lack of planning on a
metropolitan scale were only a few reasons for serious restructuring and reformation of
housing subsidy policies and programs. In 1973, President Nixon called for a moratorium
on subsidized housing production. Since then the development of three different program
types vouchers, block grants and tax credits--have become the primary means of
providing support for rental housing.
In the early seventies, the Nixon administration introduced the Section 8 voucher
program which gave the recipient the option of choosing a unit costing more than the
FMR and paying the difference out of his/her own pocket (Hays, 1995). Voucher
programs result in increased mobility of recipients to better neighborhoods that are less
socially and economically distressed with greater employment opportunities. One study
showed that only 14.8 % of certificate and voucher recipients live in high-poverty
neighborhoods (more than 30% poor), compared with 53.6 % of public housing residents
(Newman and Schnare, 1997). Furthermore, they have lower rent burdens enabling them
to use more of their income on food, clothing and health care needs (Newman and
Debates regarding the best use of scarce federal housing dollars often focus on
arguments between housing production and rental assistance through voucher type
programs. The original purpose of public housing programs was to provide housing for
poor working families in urban inner city areas as a means of improving slums. However,
several issues associated with public housing (such as, housing design, poor maintenance,
residential segregation, and placement of low-income housing in economically distressed
areas) have contributed to social inequalities. The economic and racial segregation of
poor families to the poorer less desirable areas of cities associated with federal housing
policies and programs beginning in the 1930s unfortunately persist to this day. The social
inequalities associated with housing subsidy have significant implications for the health
of women and the neighborhoods where they live. Marginalization, discrimination,
substandard housing, and housing located in disadvantaged neighborhoods may serve as
chronic stressors that catalyze a cascade of events that in time may lead to poor physical
and mental health.
What We Know and Gaps in Knowledge and Research
While research in this area has found consistent associations between housing,
poverty, and health, the pathways and mechanisms by which the social aspects of these
phenomena produce physiological alterations are not well known (Dunn, 2000). Few
studies have incorporated an approach that examines the physiological mechanisms by
which neighborhoods and housing may impact health and either exacerbate or attenuate
SES-related health disparities among women. Lacking in the literature is knowledge
regarding whether home environment or neighborhood characteristics of various
subsidized housing types differ and whether the experiences of women living in various
subsidized housing types differ in ways relevant to health. A more thorough
understanding of how subsidized housing (S8 and PH) environments differ is critical,
given that the policy intent behind S8 housing is to reduce the concentrated pockets of
poverty and its sequelae that have been observed in PH. Furthermore, very little research
explores housing as a factor in the social production of health inequalities. Population
health studies are needed that explore the relationships among housing, social capital,
social cohesion, income/wealth inequalities and women's health from a life course
perspective (Berkman and Kawachi, 2000; Dunn, 2000). Studies should address the
social-biological interface, thus sorting out the mechanisms by which the social aspects
of housing are embodied into physiological characteristics that impact health (Acevedo-
Garcia et al., 2004).
SES and Chronic Stress: The Role of the Hypothalamic-Pituitary-Adrenal (HPA)
Axis in Chronic Disease
This section of the literature review highlights two areas: the first provides a brief
overvi ew of normal hyp othal ami c-pituitary -adrenal (HPA) axi s functi on and its
hypothesized role in chronic disease development and the second reviews literature
specific to changes in normal HPA axis function by SEP and chronic stress exposure.
Human studies have established relationships between psychosocial stressors and
physiologic stress involving the HPA axis (Linden, Rutledge, and Con, 1998)
Furthermore, there is increasing evidence that characteristics of lower socioeconomic
environments are associated with excessive HPA activation (Seeman and McEwen, 1996)
that may lead to the development of chronic conditions that have high morbidity and
mortality rates (Rosmond and Bj orntorp, 2000). Responses of the HPA to stress allow
organisms to achieve allostasis, the ability to obtain stability through change, which is
required for survival (McEwen, 1998). McEwen and others (Seeman, Singer, Rowe,
Horwitz, and McEwen, 1997) have proposed the cumulative effects of adapting to
stressors (predominantly through pronounced HPA activation) may be quantifiable using
the concept of allostatic load as an index of wear and tear on the body over time.
Accumulation of allostatic load is hypothesized to play a role in the pathogenesis of
chronic diseases and is a useful concept for considering the relationships among
socioeconomic status, the psychosocial stressors of single-mothers, physiologic stress
arousal patterns, and related disparities in health (McEwen, 1998). Animal studies have
been particularly useful in determining neuroendocrine pathways of the chronic stress
and health relationship due to the ability to eliminate selection bias (Kneipp and
Drevdahl, 2003). Nonhuman primate studies indicate that dominant social status in a
stable environment is associated with less HPA activation (Sapolsky and Mott, 1987),
higher HDL levels (Sapolsky, 1989), and less coronary atherosclerosis in both males and
females (Kaplan, Manuck, Clarkson, Lusso, and Taub, 1982). Reciprocally, animals that
are socially subordinate, socially isolated, or in other socially-stressful situations
consistently demonstrate greater HPA activity (Kalin and Cames, 1984). In the
immediate postpartum period, for example, Bahr, and colleagues (Bahr, Pryce, Dobeli,
and Martin, 1998) found female gorillas living under more stressful environments in
captivity (e.g., being harassed by other adult and juvenile gorillas) had higher urine
cortisol levels and less physical contact with their infants, suggesting the social
environment affects parenting behavior and infant bonding via stress-related mechanisms.
Several rat studies (Gelsema, Schoemaker, Ruzicka, and Copeland, 1994; Roy et al.,
2001) support a relationship between chronically stressful environments, psychological
distress, and CVD.
The SEP of women' s lives in relation to chronic stressors and disease outcomes has
received little attention, even though there is increasing evidence from human studies that
chronic stress and HPA axis alteration independent of behavioral or lifestyle factors
exists (Julius and Nesbitt, 1996). Other studies show that characteristics of lower SEP
environments are associated with altered HPA activity believed to be involved in the
development of chronic conditions that have, at the ecological level, been associated with
lower SEP (e.g., CVD, DM, and asthma) (Bjorntorp, Holm, and Rosmond, 1999;
Wamala, Lynch, and Kaplan, 2001). In a study of diumal cortisol patterns in healthy
mothers of toddlers, investigators found that individual differences in cortisol secretion
patterns could be predicted from medical, demographic, contextual (home and work
demands), and psychological variables (Adam and Gunnar, 2001).
While many studies have consistently found chronic stress effects on cortisol
patterns, others have not found a relationship (Smyth, Margit, Ockenfels, Gorin, Porter,
Kerschbaum, Hellhammer, and Stone, 1997). One reason for this may be that research
done in this area has focused on two types of cortisol secretion in response to stressors:
(1) those that are short-lived and occur immediately in response to acute, laboratory
stressors and (2) those that reflect changes in diurnal secretion patterns in response to
chronic, or ongoing, stress exposures. Since the specific aims of this research are to
examine chronic stressors in relation to neighborhood context and health, the focus of this
discussion will be on alterations in the HPA axis from chronic stress.
HPA Axis Physiology and its Role in Chronic Disease Development
A complex system, the HPA axis regulates the release of many different hormones.
These hormones have either a stimulatory or inhibitory effect on many body functions.
The following discussion focuses specifically on the release of the glucocorticoid,
cortisol in response to stimulation of the HPA axis. When stimulated, the parvocellular
neurons within the paraventricular nuclei of the hypothalamus release corticotropin-
releasing hormone (CRH), AVP, and other factors. The portal system transports these
factors to the anterior pituitary, activates corticotrophs and stimulates the secretion of
adrenocorticotropin hormone (ACTH). The systemic blood system transports ACTH to
the adrenal glands. The adrenal cortex then synthesizes and secretes glucocorticoids
(Campeau, Day, Helmreich, Kollack-Walker, and Watson, 1998). Most importantly
researchers hypothesize that exposures to stressors initiate this cascade of events and is
one of the hypothesized mechanisms involved in mediating the SEP-health relationship.
Glucocorticoids are known to have metabolic, immunologic, anti-inflammatory,
and growth inhibitory effects on the body. They also influence levels of awareness and
sleep patterns (McCance and Huether, 1998). However, the main function of
glucocorticoids is to promote conditions that assist the body to adapt to adverse
situations. Therefore, glucocorticoid receptors are widely dispersed. The most potent
glucocorticoid is cortisol. Cortisol supports increased energy requirements during periods
of stress by facilitating the mobilization of free fatty acids (FFA) found in adipose tissue
in the form of triglycerides. The increase in FFA inhibits utilization of glucose in the
peripheral tissues. Cortisol stimulates the release of gluconeogenic enzymes, specifically
phosphoenolpyruvate carboxykinase, which regulates the rate of gluconeogenesis. In
addition, cortisol also functions to mobilize amino acids from proteins in skeletal muscle
The main function of glucocorticoids is to promote conditions that assist the body
systems to adapt to adverse situations. There is evidence to indicate changes in cortisol
play a pivotal role in the development of diabetes and CVD. One function of cortisol is to
support increased energy requirements during periods of stress by facilitating the
mobilization of FFA found in adipose tissue, which may contribute to the development of
insulin resistance and, ultimately, Type 2 Diabetes Mellitus (Bjorntorp et al., 1999).
Cortisol is perhaps most widely known for its immunosuppressive effects, and evidence
now suggests inflammatory processes modulated by cortisol output may play a role in the
development of atherosclerosis and CVD (Yudkin, Kumari, Humphries, and Mohamed-
Normal cortisol has a diumal rhythm with a peak occurring in the early morning
and a nadir in the early evening. Under periods of stress, cortisol is released acutely to
assist the body to adapt to its external and internal demands. Exposure to chronic
stressors, however, results in alterations in cortisol secretion that persist over time. It is
this change in pattern of cortisol secretion that has been most associated with SEP,
chronic psychosocial distress, and the eventual development of select chronic diseases
(Lovallo and Thomas, 2000; Raber, 1998). Changes in HPA response and, specifically,
the normal diurnal pattern of cortisol secretion to stress, may result in pathological
changes that lead to the development of select chronic diseases (McEwen, 1998; Raber,
1998). For example, Plat and coworkers describe how prolonged hypothalamic
stimulation from a stressor might result in abnormally high levels of cortisol secretion in
the early evening. In addition they found that evening elevations in cortisol were
associated with delayed hyperglycemic effects, stimulation of lipolysis and increased
concentrations of free fatty acids that have been associated with CVD (Plat, Leproult,
L'Hermite-Baleriaux, Fery, Mockel, Polonsky, and Van Cauter, 1999). In another study,
researchers found that a stress-related cortisol secretion pattern with a flattened curve--
depicting a loss of adaptability to stimuli--was strongly correlated with elevated body
mass index, waist-hip ratios, blood pressure, heart rate, triglycerides, total and low-
density lipoproteins, insulin, glucose, and visceral fat mass (Bjorntorp et al., 1999).
Therefore, they postulate that stress-related cortisol secretion along with an impaired
regulation of the HPA axis, are connected to physiologic alterations associated with
chronic disease development.
Relationships Among Neighborhood Characteristics, Housing, Chronic Stress, and
Studies have demonstrated that neighborhood characteristics play a significant role
in determining the type and intensity of daily stress experienced and therefore are
important social determinants of health (Boardman et al., 2001; Wasylishyn and Johnson,
1998). Adjusted for individual-level SEP, living in high poverty neighborhoods has been
associated with increased daily stressors such as increased exposure to drugs (Boardman
et al., 2001) and violent crime (Sampson and Raudenbush, 1997). Studies indicate lower-
income women view the stress in their lives as maj or determinants of not only overall
health status but also of other chronic diseases. For example, focus groups conducted by
researchers with low-income African-American women to examine their awareness of
and concern for CVD found they considered CVD to be associated with stress and low
SEP (Behera, Winkleby, and Collins, 2001). Similarly another study found that low-
income women with mental health problems were most interested in stress management
strategies indicating that they view stress as an important aspect of psychological health
(Alvidrez and Azocar, 1999).
Additional qualitative studies with women living in low SES neighborhoods in
Detroit highlighted that women linked stressors directly related to neighborhood
characteristics (Schulz, Parker, Israel, and Fisher, 2001). Furthermore, the cumulative
effect of chronic stressors such as safety issues and unfair treatment was strongly
associated with symptoms of depression, while financial and family stress showed the
strongest relationships with poorer self-reported health status (Schulz et al., 2001). A
recent study conducted by Buka et. al. (2003) examined neighborhood economic
disadvantage, neighborhood support, and infant birth weight in 343 neighborhoods. They
found neighborhood economic disadvantage alone accounted for 80.8% of the between
neighborhood variance infant birth weight for African-American mothers and 76.3% for
White mothers while controlling for individual risk factors (including maternal age,
education, smoking during pregnancy, and receipt of prenatal care). When neighborhood
social support was added to the model, the addition of this explanatory variable to
economic disadvantage accounted for 90.9% of the between neighborhood variance in
infant birth weight for Whites.
These results indicate that stressors produced as a consequence of living in
economically disadvantaged neighborhoods have significant implications for health,
regardless of individual behavioral factors. What remains, unknown, however, is whether
and how chronically stressful environments of low-income housing have physiological
consequences that contribute to chronic disease development, and whether the
environments of PH and S8 differ in ways that are relevant for health (Buka et al., 2003).
As described in this literature review, neighborhood-level characteristics have a
significant effect on health above and beyond individual-level factors. Studies have
shown that neighborhood-level factors can produce environments that promote chronic
stress and poor health. However, we do not have a clear understanding of the social-
biological interface that can provide evidence of the physiological mechanisms by which
environments (i. e., neighborhoods) contribute to poor health and chronic illness. The
study of social and biological variables at the same time is intrinsically valuable because
we are as humans, both social and biological (Brunner, 2000). More research is needed in
order to understand what constitutes an unhealthy environment and how it "gets under the
skin" to produce illness (Taylor, Repetti, and Seeman, 1999). This study aims to
contribute to this body of research by investigating social and biological variables
simultaneously, by beginning to explain the social-biological processes by which
neighborhoods and housing impact women's health disparities.
As stated in Chapter 1, this study is guided by a combination of an ecosocial
paradigm and the allostatic load model. Together, they allow for exploration of the social
- biological interface through which environmental factors affect health. Krieger's
ecosocial theory and McEwen' s allostatic load model guide the development of the
conceptual framework used in this study (Krieger and Davey-Smith, 2004). Krieger and
Davey-Smith (2004) call for incorporating the concept "embodiment" in order to capture
how social influences (i.e., housing and the built environment) become literally embodied
into physiological characteristics that influence health. The concept of "embodiment"
simultaneously embraces biologic and social processes while avoiding the trap of
equating "biologic' with "innate" and without assuming the soma is governed exclusively
by the psyche. In addition, as Krieger and Davey-Smith state, "this new scholarship
emphasizes how actualization and suppression of people' s agency, that is, their ability to
act within their bodies, intimately depends on socially structured opportunities for, and
threats to, their well-being" (pg. 95). Thus keeping the concept of embodiment in mind,
the conceptual framework developed for this study draws from multiple disciplines such
as public health, sociology, and medicine.
The second theoretical model used to derive the socio-biological conceptual
framework is McEwen's Allostatic Load model (McEwen, 1998). This model is based on
the premise that physical and psychological stressors occur within a social and economic
context and that there is individual variation in the stress appraisal process as well as
behavioral and emotional coping mechanisms to the perceived stressor (McEwen, 1999).
McEwen describes four key propositions of his allostatic load model. First, the brain is
the integrative center for coordinating the behavioral and neuroendocrine responses
(hormonal, autonomic) to challenges. Second, there are considerable differences in
coping with challenges based on interacting genetic, developmental, and experiential
factors that predisposes persons to react differently physiologically and behaviorally to
events throughout life. Third, inherent within the neuroendocrine and behavioral
responses to challenge is the capacity to adapt (allostasis). However, while these
physiological processes are protective in the short term, inefficiency or alterations in the
ability of the neuroendocrine system to turn on and off responses leads to cumulative
negative effects over time. Fourth, allostasis has a price defined as allostatic load -that
reflects the cumulative negative effects or the wear and tear on bodily systems from being
forced to constantly adapt to various psychosocial challenges and adverse environments
(i.e., disadvantaged neighborhoods).
Accumulation of allostatic load is hypothesized to play a role in the pathogenesis of
select chronic diseases, such as insulin resistance, atherosclerosis, increased susceptibility
to infections and memory loss (Bjorntorp, Holm, and Rosmond, 1999; McEwen, 1998;
McEwen, 2000). Since this model addresses the fact that daily stressors occur within a
social and economic context, it is a useful framework for considering the relationships
among SEP, the psychosocial stressors of single mothers, physiologic stress arousal
patterns, and their noted disparities in health (McEwen, 1999). As such, the impetus in
this model is to move from the individual back to populations and consider the average
properties of groups of individuals classified according to measures of SEP attending to
not only the social and cultural factors that influence health, but the potential
physiological mediators found in these relationships. However, McEwen's model lacks
detail regarding the socio-economic aspects of neighborhood and health.
Combining Krieger' s ecosocial theory and McEwen' s allostatic load model enables
the researcher to simultaneously explore social and biological variables. This allows for
advancing scientific knowledge as it relates to the understanding of the social-biological
interface that may be mediating relationships among environments (i.e., neighborhoods),
chronic stress, and health. Relationships among environment, social, psychological, and
physiologic factors relevant to this study in relation to the Allostatic Load Model are
illustrated in Figure 3-1. Key constructs and concepts and methods of operationalization
are explained in the section on maj or study variables below. The final outcome (CVD) is
in grey because it was not explored in the present study. It is only an example of one
possible outcome that could be explored using this framework.
Figure 3-1: Socio-biological Model
Hypothesized Mechanisms Contributing to Disease Development
This study utilized a cross-sectional design in which physiological measures were
obtained six times a day for 2 days in a sample of 84 women. Of the 84 who participated,
complete data were available for only 67 participants 23 from PH and 43 from S8
housing. The relationships among housing type, neighborhood characteristics, stress,
psychological distress and salivary cortisol are examined by specific aim 1. To meet
specific aims 2 and 3, differences in neighborhood characteristics, stress, psychological
distress and salivary cortisol between women living in S8 and PH are explored.
Population and Sample
The population investigated in this study includes 18 to 45 year old women who are
heads of households and have at least one child 18 years-old or less living with them.
This age group was selected because it is representative of the target population to be
studied. Based on data from the department of Housing and Urban Development, almost
two-thirds of those living in subsidized housing are between the ages of 18 and 62 (U. S.
Department of Housing and Urban Development, 1998). A subcategory of female
children (as defined by NIH) those who arel8 to 21 years of age and are mothers --, are
included in this study. While considered by many as children, young (18- to 21- years-
old) mothers often live in disadvantaged neighborhoods and poor housing environments.
Having the same adult responsibilities as any parent, they experience many daily
stressors, associated with adult responsibilities in maintaining family safety and stability.
The specific aims of this study are to examine the potential of chronic disease
development as a result of cumulative stress associated with adult family responsibilities
and being female heads of households within a neighborhood context.
The sample was selected based on their exposure to either S8 or PH environments.
Research participants were recruited from the area in cooperation with the Gainesville
Housing Authority and S8 housing managers. In Gainesville, Florida and within a 10
mile perimeter of the city limits, there are 1,062 S8 housing "slots". Approximately 400
people per year attend S8 housing orientation provided by the Gainesville Housing
Authority. Ten to fifteen percent of people who move into S8 housing move from PH
units (Dolder, C. personal communication, June 26, 2002). During a typical application
process, of the 400 interested people, 150 applications are processed and reviewed;
approximately 5% to 10% are able to move into S8 housing.
The sample was recruited by posting flyers at the local housing agencies, rental
units participating in housing subsidy programs, the university, health science center,
local hospitals and primary care offices, social service agencies, and churches. In
addition, the principal investigator (PI) attended community meetings such as the Black
on Black Crime Task Force, and tenant/neighborhood associations in neighborhoods
where the sample was located and informed them of the research and recruited interested
persons. Addresses were obtained from the Gainesville and Alachua County Housing
Authority and letters were disseminated tol500 section 8 and public housing addresses.
Most of the public housing participants were recruited by going door to door.
Sample size determination was based on a power analysis to ensure a power of 0.80
is achieved. A sample size of 49 subj ects per housing type was needed to detect a
difference of .40 in the outcome measures (i.e., salivary cortisol, chronic stress, and
psychological distress). For multiple regression analyses a total of 107 women were
needed. Power for this study was not achieved due to an inadequate sample size.
A total of 84 women participated in this study. The original sample size of 107
women was not attained due to several issues. Despite compensation with a $30.00 gift
certificate to Wal-mart, recruitment of women living in public housing was difficult.
Though data was not collected as to why women chose not to participate, based on their
comments, fear of getting in trouble with the local housing authority is one possible
reason many women did not participate in the study. Also, many of the women refused to
collect saliva. Some were repulsed by the idea of collecting saliva, while others voiced
concern about what would happen to it after the study was over. Fear and mistrust despite
efforts to assure women that their privacy and confidentiality would be maintained was
believed to be a maj or factor in not achieving the desired sample size. Also the study was
limited to participants who lived within a 10-mile radius of Gainesville, thus limiting
Of the 84 women who participated, 14 did not have usable cortisol data (defined as
missing more that 2 time points in one day). These cases were deleted. Survey and
cortisol data were imputed for the 70 participants remaining. Three participants failed to
answer over 50% of the questions from one measure leaving a Einal sample size of 67
women. It was decided a priori that if a participant failed to answer more than 30% of the
items in a measure, she would be excluded from data analysis. Up to 10% missing data
on a measure may be considered small, while 40% missing data is considered to be high
(Musil, Warner, Yobas, and Jones, 2002). Recommendations for handling missing data in
nursing research are limited. Decisions regarding the appropriate methods to deal with
missing data are based on the pattern, level (subject or item) and amount of data that are
missing (Kneipp and McIntosh, 2001; Patrician, 2002). The robustness of certain
imputation techniques is often dependent on extent and amount of missing data.
Therefore, these factors should be considered in order to minimize estimation error and
response bias (Fox-Wasylyshyn and El-Masri, 2005). Imputation methods used in this
study are described in detail later in this chapter.
The study was conducted in a naturalistic setting (i.e., community of residence).
The PI or her research assistant met with participants in their homes or other settings as
preferred by the participants.
Human Subjects Protection
Approval for this study was obtained from the University of Florida Health Science
Center Institutional Review Board prior to any subj ect recruitment or data collection. All
subj ects signed an informed consent form and were given a copy prior to enrollment in
the study. Data collection took place in the participant's residential neighborhood.
Confidentiality was maintained by use of a code for each subject. All files were kept in a
locked file cabinet in the researcher' s office. Saliva samples were also coded and stored
in a freezer in the college of nursing wet lab, which is locked at all times and has limited
access by select faculty, staff, and research assistants.
Inclusion and Exclusion Criteria
Inclusion criteria include:
a) Living in public or section 8 housing for at least 1 month,
b) Able to speak and read English,
c) Between the ages of 18 and 45 years old,
d) A mother who is head of the household with a child living in the same
household who is 18 years old or less.
Exclusion criteria include:
a) Age greater than 45 years old,
b) Diagnosis of an autoimmune disorder,
c) Pregnant or breastfeeding,
d) Taking antidepressant, anxiolytic, or steroid-based medications.
e) Working the night shift (from the hours of 11:00 pm to 7:00 am).
The inclusion criteria were selected because this study focuses on women living in
section 8 and public housing units and how these areas may serve as stressor for women
responsible for maintaining family safety and stability. Women living in an area for at
least one month have had time to assess their neighborhood regarding crime, disorder and
other characteristics. Women 18 years-old up to 45 years-old are representative of most
of those who live in subsidized housing as previously described. The study was limited to
participants who could read and speak English due to financial constraints related to
hiring a translator.
These exclusion criteria were selected because they are known to alter cortisol
levels and may alter responses to stress, depression and anxiety measures. Studies have
produced controversial results regarding differences in salivary cortisol based on age
group. Studies have demonstrated systematic differences are present in early morning
salivary cortisol in which decreasing cortisol concentrations are positively correlated with
age (Kirschbaum and Hellhammer, 1992). The lowest mean value of 11.6 nmol/1 was
found in the age group between 59 and 64 years. The effect of pregnancy on salivary
cortisol is controversial. Some studies have shown increases in salivary cortisol secretion
while others have not (Kirschbaum and Hellhammer, 1992). In addition, the sample
upper age limit of 45 years old will reduce confounding effects of menopause and chronic
disease development on physiological measures.
As previously described in the literature review, cortisol has a diurnal pattern with
a peak 45 minutes to one hour after awakening and a nadir just before bedtime.
Alterations in sleep quality and quanity and working the night shift have been shown to
affect the HPA axis and therefore alter cortisol secretion patterns (Leproult, Copinschi,
Burton, and Van Cauter, 1997; Spiegel, Leproult, and Van Cauter, 1999).
In addition to altering salivary cortisol levels, antidepressants and anti-anxiety
agents may influence how a participant responds to questions regarding stress, depression
and anxiety leading to underestimation and response bias. Therefore persons taking anti-
depressant or anti-anxiety agents were excluded from the study.
Research Variables and Instruments
A demographic data sheet was used to collect information such as age, marital
status, race, household type, number and ages of children, individual income, education,
and occupation, chronic diseases diagnoses, medication use, height, and weight. In
addition information was obtained on current address, living situation (i.e., living with
others or others living with them), housing type, length of time in current dwelling, rent
assistance per month, public assistance, receipt of food stamps and other sources of
income via public assistance resources, such as child care and transportation. Additional
data regarding smoking history, alcohol intake, and menstrual cycle phase and regularity
were obtained, as well.
Major Study Variables
This section provides detailed information on the measures used in this study. A
quick overview of each of the dependent and explanatory variables is provided in the
table in the Appendix.
Neighborhood is defined according to Galster' s (2001) definition that states,
"Neighborhood is the bundle of spatially based attributes associated with clusters of
residences, sometimes in conjunction with other land uses" (p. 2112). This definition is
broad and quite abstract. It includes several aspects of neighborhood such as the
structural, class status, environmental, and social inter-active characteristics of
For the purposes of this study, the term neighborhood characteristics include
information on neighborhood economic disadvantage (measured at the census tract level),
perceived neighborhood disorder, exposure to crime, and neighborhood cohesion. Each
of these measures is described more fully below.
Neighborhood Economic Disadvantage
Neighborhood economic disadvantage was obtained using census tract level data
from the 2000 census. Census tracts are designed to be demographically homogenous
with stable boundaries over time and generally contain between 3000 and 8000 resident
(Boardman, et. al., 2001). Extensive research by Krieger and colleagues has shown that
socioeconomic data obtained at the census tract level performs better at detecting
economic gradients expected than measures at the county or state level (Krieger et al.,
For the purposes of this study, neighborhood economic disadvantage is an index
measure of percent family poverty, percent of female headed households, male
unemployment rate, and percent of families receiving public assistance. The four values
were summed to create the neighborhood disadvantage measure in which higher numbers
indicate greater disadvantage, with scores ranging from 0 to 12. This measure has been
used in studies that investigated the relationship between neighborhood disadvantage and
health in adult samples much like the sample in this study (Boardman et al., 2001;
Sampson and Raudenbush, 1997). Prior research by Sampson & Raudenbush (1997)
demonstrated that these characteristics are highly interrelated and load on one single
factor that can be described as neighborhood disadvantage (a= 0.97).
The concept of perceived neighborhood disorder includes both social and physical
signs indicating a lack of order in the neighborhood. Areas with high levels of disorder
are characterized by deviance, noise, vandalism, drug use, crime, trouble with neighbors
and other incivilities (Ross and Mirowsky, 1999). This study measured perceived
neighborhood disorder using an index that measures physical signs of disorder such as
graffiti, vandalism, noise, and abandoned buildings, and social signs such as crime,
people hanging out on the street, and people drinking or using drugs. It also includes
reverse-coded signs of neighborhood order such as safety, people taking care of their
houses and apartments or watching out for each other. The perceived neighborhood
disorder scale consists of 15 items on a four point Likert scale that ranges from order on
the low end (15) to disorder on the high end (60) of the continuum. This scale has an
alpha reliability of .915 (Ross and Mirowsky, 2001).
Neighborhood Stress: Crime Exposure
Neighborhood Stress is defined as exposure to a range of events and conditions in
one's proximal environment that are capable of eliciting stressful emotions (e.g., fear,
anger, depression) and that may exacerbate disease processes or undermine health
(Ewart, 2002). The City Stress Index (CSI) was developed by Ewart (2002) and is used as
a self-report measure to assess perceived neighborhood disorder and exposure to crime.
The CSI is an 18-item measure with scores ranging from 18 to 72. Low scores indicate
lower levels of neighborhood stressors. This measure can be completed by persons with
an eighth grade reading-level. It has good validity and reliability with the neighborhood
disorder and exposure to violence portions of the scale having a Chronbach' s alpha of .88
and .85 respectively (Ewart, 2002). The recent development of the tool limits the data on
use in other populations such as adults, but the reading level and use in urban dwelling
adolescents make it a useful measure for this study. Permission to use this measure was
obtained from Craig Ewart, a professor in the Center for Health and Behavior at Syracuse
University (C. Ewart, personal communication, December 7, 2004).
Neighborhood Social Cohesion
Social cohesion refers to the level of trust, extent of connectedness and solidarity
among groups in society (Kawachi and Berkman, 2000; Sampson and Raudenbush,
1997). For some, the neighborhood may become an extension of home for social
purposes and becomes important in identity terms possibly leading to a high degree of
interaction among community members (Forrest and Kearns, 2001). Neighborhood social
cohesion was measured using a using 5 conceptually related items that ask participants
whether or not people in the neighborhood willing to help others, get along with each
other, share the same values, can be trusted, and whether or not they agreed they lived in
a close-knit neighborhood (Sampson and Raudenbush, 1997). The items were scored on a
5-point Likert scale. Scores may range from 0 to 25 with higher scores indicating greater
levels of cohesion. The reliability with which neighborhoods can be distinguished on
neighborhood social cohesion ranges between 0.80 to 0.91 (Sampson and Raudenbush,
Nested within neighborhoods is the construct of housing. Without further
definition, housing can refer to several different types of housing such as high or low-
income housing, public or section 8 (government subsidized housing), or rental versus
owned housing. For use in this study, housing will be further defined as subsidized rental
housing (public and section 8 housing).
Housing Satisfaction (Perceived Housing Quality)
Satisfaction with one's housing was measured using one item from HIUD's
Customer Service and Satisfaction Survey (U. S. Department of Housing and Urban
Development, 2003c). The Customer Service and Satisfaction Survey was developed in
consultation with housing industry representatives and public housing resident leadership
groups. This survey consists of 20 items in addition to six optional demographic
questions that were not used in this study. This survey is designed to be both an
assessment of current resident opinions regarding their housing quality and a
management tool to identify areas of concern (U. S. Department of Housing and Urban
Development, 2004). Housing satisfaction was determined by asking participants, "How
satisfied are you with your unit/home?" Responses to these questions are based on a 5-
point Likert Scale and ranged from, "Does not apply, very dissatisfied, dissatisfied,
satisfied to very satisfied." Higher scores are indicative of more satisfaction. This
measure has not been used in research, thus, validity and reliability has not been reported.
The term stress has many definitions depending on the context in which it is used.
Hans Selye, a pioneer in the development of stress theory, developed the concept of stress
using a response-based orientation (Lyon, 2000). For some stress is good in that it
produces excitement, anticipation, and challenge; for others, the same stressor is bad,
producing an undesirable state characterized by worry, frustration, chronic fatigue, and
inability to cope (McEwen, 2005). Stress is defined in this study as an undesirable state
based on one's perceptions of situations and events such as neighborhood crime and
disorder, or unfair treatment which evokes certain emotional, behavioral, and nonspecific
Perceived Stress was measured using Cohen's (1983) Perceived Stress Scale. This
scale measures the degree to which situations in one's life are appraised as stressful. The
Perceived Stress Scale is a widely used and accepted measure with good validity and
reliability. Responses to these questions are based on a 5 point Likert scale asking
participants to respond to their feelings and thoughts over the last month. It has good
internal consistency with a Chronbach's alpha of .84 .86. Scores range from 0 to 56
with lower scores indicating less stress (Cohen, Kamarck, and Mermelstein, 1983).
Unfair Treatment and Discrimination
Unfair treatment was assessed using The Interpersonal Mistreatment Scale
developed by Williams and colleagues (1997). These items were developed to assess how
often in their day-to- day lives persons experience a variety of forms of interpersonal
mistreatment. The framework consisted of poor interpersonal treatment and made no
reference to race, prejudice, or discrimination (Guyll, Matthews, and Bromberger, 2001;
Williams, Yu, Jackson, and Anderson, 1997). The Interpersonal Mistreatment Scale
consists of 10 items on a 4-point Likert scale. Scores range from 10 to 40 with higher
scores corresponding to more frequent experiences of mistreatment. This measure has
demonstrated good internal consistency with a Cronbach's alpha of .76 to .86 (Guyll et
al., 2001; Williams et al., 1997).
Chronic stress, the cumulative load of minor, everyday stressors, can have long-
term consequences (McEwen, 1998). The effects of chronic stress may be exacerbated by
unhealthy behaviors such as lack of physical activity, high calorie, high fat diets, smoking
and alcohol use. Chronic Stress was measured in this study using the Trier Inventory for
the Assessment of Chronic Stress (TIC-S) (Schlotz and Schulz, 2004). This measure is a
comprehensive measure of chronic stress that comprises nine dimensions including work
overload, social overload, overextended at work, lack of social recognition, work
discontent, social tension, performance pressure at work, performance pressure in social
interactions, social isolation, and worry propensity. This measure is included in this study
in addition to the perceived stress scale because in addition to being more comprehensive,
it asks people to answer questions based on their experiences for the last 3 months.
Stressors experienced for this amount of time have more of a chronic component than
stress experienced for only one month. Furthermore, the TIC-S includes specific
dimensions which allow researchers to pinpoint specific areas (work or social life) that
may be considered stressful, unlike other measure used in this study. Responses are based
on a 5 point Likert scale. The TIC-S has demonstrated good internal consistency with a
Cronbach' s alpha of .76 to .91 and a split-half reliability of .79 to .89. Permission was
obtained from William Schlotz a professor at the University of Trier, Department of
Psychobiology in Johanniterufer Germany to use the short version of the Trier Inventory
for the Assessment of Chronic Stress (TICS-S) (Schlotz and Schulz, 2004).
Psychological Distress is defined as a discomforting emotional state experienced by
an individual in response to one or more stressors or demands that is manifest by a
change in baseline stable emotional state to one of anxiety, depression, demotivation,
irritability, aggressiveness, or self-depreciation (Ridner, 2004). In this study,
psychological distress was measured by using scales that measure depressive
symptomology and state anxiety. In addition to serving as independent variables that
influence the outcome variable health and salivary cortisol, these variables will also be
dependent variables when addressing the impact of neighborhood stressors and mental
The Center for Epidemiological Studies of Depression Scale (CES-D) is a 20-item,
self-report scale that measures depressive symptoms in the general population
(Weissman, Sholomskas, Pottenger, Prussoff, and Locke, 1977). It includes six major
symptom areas: (1) depressed mood; (2) guilt-worthlessness; (3)
helplessness/hopelessness; (4) psychomotor retardation; (5) loss of appetite; (6) sleep
disturbance. Responses are based on a 4-point Likert scale. Validity and reliability of this
scale has been supported in previous studies Internal consistency and reliability using
Cronbach's alpha has ranged from 0.85 to 0.91 (McDowell & Newell, 1996).
The Speilberger State-Trait Anxiety Inventory for Adults Form Y (STAI Form Y-
1) was used to measure anxiety. The STAI has been used extensively in research and
clinical practice. It comprises separate self-report scales for measuring state and trait
anxiety. The state portion of the scales consists of 20 statements that evaluate how
respondents feel at the moment they are completing the survey. The trait portion of the
scale consists of 20 statements that assess how people generally feel (Spielberger,
Gorsuch, Lushene, Vagg, and Jacobs, 1983). Responses are based on a 4-point Likert
scale. This measure had demonstrated good internal consistency with a Cronbach's alpha
of .86 to .95.
General Health was measured using one item from the SF-12v2TM SUTVey form
(Ware, Kosinski, and Keller, 1996). Participants were asked, "In general would you say
your health is poor, fair, good, very good, or excellent?" Answers were based on a four-
week recall. Scores ranged from 1 to 5 respectively. These scores were transformed to a 0
to 100 scale and compared with national norms for women of the same age group (Ware,
Kosinski, Turner-Bowker, and Gandek, 2002). Cronbach's alpha for the SF-12v2TM
survey ranges from 0.73 0.77 in the general population of women ages 18 to 44 years
old (Ware et al., 1996).
Salivary Cortisol (SC)
SC is a widely accepted method for measuring physiological responses to acute
laboratory induced stress and perceived chronic stress. It highly correlates with serum
(blood) and urine cortisol levels and offers stress-free, non-invasive sampling, easy
collection and storage (Kirschbaum and Hellhammer, 1994). However, cortisol levels are
affected by a variety of factors such as an acute stressor, smoking, drugs (such as steroid-
based medications, contraceptives, anti-depressants and anxiolytics), a high protein meal,
lack of sleep and the luteal phase of the menstrual cycle (Kirschbaum and Hellhammer,
1992). These factors were controlled for in the exclusion criteria and saliva collection
protocol, or by incorporating them as covariates in statistical models. Samples were
analyzed using the HS-Cortisol High Sensitivity Salivary Cortisol Enzyme Immunoassay
Kit. This kit requires minimal saliva volume (25 CIl), detects < 0.012 to 3.0 Cll of cortisol,
has a serum-saliva correlation of r = .94, p <.0001 (Salimetrics, 2005). It was designed as
a superior alternative to resolve problems associated with serum-based
radioimmunoassay and other salivary immunoassays (Schwartz, Granger, Susman,
Gunnar, and Laird, 1998).
Participants provided 12 cortisol samples consisting of 6 samples per day for 2
days. Specimens were collected over 2 days based on expert recommendations from the
John D. and Catherine T. MacArthur Research Network on Socioeconomic Status and
Health. Though somewhat controversial, the more measurements in a day for a greater
number of days (at least 3 to 4) allows for a more reliable measurement of "trait" daily
concentration of cortisol (AUC). The advantage of using multiple days is that it helps to
control the unreliability of one day's data, which can underestimate the cortisol
relationship to outcomes (Stewart and Seeman, 1999).
A period of 2 days was decided upon for several reasons. First, prior research
experience with a similar population suggested that data collection for a period of time
longer than two days would be unrealistic. The day to day turmoil experienced by many
in this population precludes prolonged daily data collection. Furthermore, daily data
collection for 4 to 6 days places a significant burden on the participants in addition to
their daily routines and responsibilities. Finally, biological specimen collection and
analysis is costly. Materials and supplies for collection and analyses of biological
specimens can be quite expensive, and participants should be compensated appropriately
for the time commitment and burden placed upon them as study participants. Therefore,
financial constraints also prevented more frequent or prolonged data collection.
Specimen collection was timed based on each participant' s time of awakening with
the first sample to be collected upon awakening (T1). The remaining 5 samples were
collected at 30 minutes I hour, 4, 9, and 11 hours after waking (T2-T6). This method of
salivary cortisol collection is preferred since the time of cortisol peak is not dependent
upon the absolute time nor is it influenced by daylight; it is dependent on wake-up timing
of each individual (Immuno-Biological Laboratories, 2004; Stewart and Seeman, 2000).
The total area under the curve (AUCg) with respect to ground as described by
Pressner and colleagues (2003) was examined in terms of its relationship to the
independent variables in this study. The formula used for the AUCg is derived from the
trapezoid formula (Pressner, Kirschbaum, Meinlschmid, and Hellhammer, 2003). The
formula used to calculate AUCg is presented below in Equation 3-1.
This AUCg calculation takes into account change over time of each measurement and the
distance of the measures from zero (the level at which the changes over time occur and
results in a measure that is more related to total hormonal output (Pressner et al., 2003).
Researchers at the MacArthur Research Network on Socioeconomic Status and Health
agree that AUC is the most widely accepted measure whereas diurnal rhythm, or diurnal
'pattern' analysis is more controversial (Stewart and Seeman, 2000).
However, AUCg is not without limitations. Though AUCg is a summarized index
for repeated measures over time, it is not sensitive to fluctuations of repeated measures.
For example, if two persons have completely different patterns of cortisol levels relative
to time, they may get the same AUCg. In addition, the AUCg approach does not take into
account the correlations among repeated outcome measures within a specific person.
Given these limitations, generalized estimating equations (GEEs) will additionally be
used to examine the relationships among the independent variables and salivary cortisol.
GEEs provide a general framework for the analyses of continuous, ordinal,
polychotomous, dichotomous, and count-dependent data, and relax several assumptions
of traditional regression models. GEEs represent an extension of the generalized linear
model (GLMs) to accommodate correlated data. GLMs assume that the dependent
variable can be expressed as a linear function of the independent variables. It also
assumes that the variance of the dependent variables is a known function of its
expectation (thus allowing relaxation of the homoscedasticity assumption). Other
assumptions of the GEE method include: (1) the number of clusters be relatively high (a
rule of thumb is no fewer than 10, possibly more than 30, and (2) the observations in
different clusters be independent, although within-cluster observations may correlate.
Hence, GLMs do not require the specification of the form of the distribution, but only the
relationship between the outcome mean and the explanatory variables and between the
mean and the variance (Ghisletta and Spini, 2004).
GEE is a marginal (or population averaged) as opposed to a cluster-specific (or
subj ect-specifie, conditional) method. Population average models model the average
response over the subpopulation that shares a common value of the predictors as a
function of such predictors. Population average parameters represent the averaged effect
of a unit change in the predictors for the whole population. The GEE approach specifies a
working correlation matrix for the vector of repeated measures from each participant to
account for the dependency among the repeated measures. The working correlation can
be assumed to be the same for all participants, reflecting average dependence among the
repeated measures over participants. Several working correlation structures can also be
specified, including independent, exchangeable, autoregressive, and unstructured
correlation. The standard errors are derived from what is called the sandwich estimator of
the covariance matrix of the regression coefficients. The main advantage of GEEs is that
the calculation of the standard errors for the regression coefficients is robust even if the
specifications of the correlation structure is incorrect or if the strength of the correlation
between repeated outcomes varies somewhat from person to person. Although the use of
robust standard errors ensures that regression inferences are consistent regardless which
correlation structure is chosen, however, there is no straightforward way in GEE models
to truly determine the best correlation structure to use (Ghisletta and Spini, 2004).
GEE is not without limitations. First the technique is asymptotic, hence requiring
large total sample sizes for unbiased and consistent estimation. Second, in applications to
empirical data, sensitivity analyses of different specifications of the intracluster
correlation matrix are advised. Finally, GEE methodology assumes missing completely at
random data, because GEEs do not specify the full conditional likelihood. However,
GEEs do no yield a great deal of bias with missing at random data (Ghisletta and Spini,
Given the limitations of AUCg and the advantages of GEE methods, GEE will
also be used to examine the relationships among neighborhood characteristics,
psychological distress, stress, and salivary cortisol.
Individual Social Support
Individual social support is considered a covariate in this study. The amount and
type of individual social support one has may possibly offset the lack of support that may
exist in the area in which one lives. Therefore, it is important to control for the amount of
individual social support when examining the effects of neighborhood social cohesion on
health. Individual-level social support was measured in this study using the International
Support Evaluation List General Population Form (ISEL-GP) (Cohen, Mermelstein,
Kamarck, and Hoberman, 1985). ISEL-GP consists of 40 items designed to assess the
perceived availability of four separate functions of individual social support (tangible,
appraisal, self-esteem, and belonging). Responses are measured on a 4-point Likert scale.
In other studies Cronbach's alpha has been reported as 0.88 and 0.90 and test-retest
reliability coefficients = 0.87 (Cohen et al., 1985).
Participants were screened for inclusion in the study. For those who met inclusion
and exclusion criteria and agreed to participate, a time and place was agreed upon for the
participant to meet with the PI. At this initial meeting, the consent to participate in
research was reviewed with the participant, who then signed the informed consent form
approved by the Institutional Review Board Human Subj ects Committee at the University
of Florida. At this point, the PI or research assistant covered the requirements of the study
in detail. The PI or research assistant left the Salivettes for saliva collection and the
questionnaire and scheduled a time to pick up the saliva and completed questionnaire.
To provide a sample, participants were given 12 tubes called Salivettes@ (The
Sarstedt Group, 2003). Six tubes for each day of collection were provided in separate
plastic bags. Saliva collection times were based on the time of awakening and were
collected 30 minutes, 1, 4, 9, and 11 hours after waking. Participants were instructed to
collect the first saliva specimen (upon awakening) before rising and getting out of bed.
Upright positions may significantly increase salivary cortisol concentrations (Hennig et
al., 2000). They were also instructed to place a cotton roll in their mouths, chew on it
until it became saturated, and place it in the salivette. Participants were instructed not to
brush their teeth, smoke, eat, or drink anything at least two hours prior to collection
because the factors have been shown to alter salivary cortisol concentrations
(Kirschbaum, Read, and Hellhammer, 1992). They were also instructed to place the
salivette tubes in the freezer at the end of each day. After being collected by the
researcher or research assistant, samples were centrifuged and stored frozen (-200 C).
Before analysis samples were thawed and mucins were precipitated from the specimens
at 3000 rpm for 15 minutes. Cortisol was measured by using Expanded Range High
Sensitivity Salivary Cortisol Enzyme Immunoassay Kit (Salimetrics, 2005). All analyses
were conducted according to the manufacturer' s directions. Samples with greater than
30% coefficient of variation (CV) were rerun. Inter- and intra-assay CV% was less than
15%. Inter-assay CV is based on the high and low controls of 28 plates. Intra-assay CV
was based on eight high and eight low control duplicate samples.
Participation in this study required a commitment of completing a survey that took
approximately 1 to 1.5 hours to complete and completing saliva collection for 2 days.
Each saliva collection was estimated to take a maximum of 5 minutes which would entail
an additional 1 hour of the participant' s time. Given the time burden placed on the
participants, a $30.00 gift certificate to Wal-mart was given to each participant that
completed data collection. A $30.00 gift certificate was used as opposed to cash so it
would not be counted as income, placing the participants at risk for loosing food stamp
supplements, housing subsidy or other Einancial assistance through the Temporary
Assistance for Needy Families program, if they were receiving these supports.
Data were analyzed using Stata 9.0 statistical software. Two-tailed tests were used
in all cases and an alpha level of .05 was selected a priori to determine significance.
Descriptive statistics were tabulated for all variables.
Statistical Analysis Approach
Before any analyses were conducted, the primary outcome variables (general
health, state anxiety, depression and SC-AUCg) were examined for normality. Skewness
and kurtosis tests for normality were used to examine the distribution of all study
variables (see table 3-1 below). General health was transformed to a 0 to 100 scale.
General health and SC-AUCg were both positively skewed; therefore, log
transformations were conducted based on the ladder of powers (Hamilton, 2006).
Depression was also positively skewed, but was transformed using square root
transformation. Choosing a transformation method for each outcome variable was based
on analyses using ladder of powers. This test combines the ladder of powers with tests of
normality specificallyy the skewness/kurtosis test in Stata) and reports whether the result
is significantly non-normal (Hamilton, 2006). The transformation with the lowest Chi
square and a normal distribution was chosen because most statistical procedures work
best when applied to variables that follow a normal distribution.
Table 3-1 Skewness and Kurtosis for Study Variables
Neighborhood Characteristics Skewness Kurtosis
Neighborhood Economic Disadvantage 0.000 0.089
Neighborhood Disorder 0.305 0.456
Neighborhood Stress 0.039 0.515
Neighborhood Social Cohesion 0.474 .669
Housing Satisfaction 0.001 0.184
Unfair Treatment 0.212 0.210
Perceived Stress 0.780 0.527
Chronic Stress 0.732 0.629
ISEL social support 0.501 .556
Anxiety 0.130 0.510
Depression 0.645 0.007
General Health 0.001 0.357
Salivary Cortisol (SC) 0.000 0.001
Specific Aim 1
The first aim of this study was to determine the relationships among neighborhood
characteristics, perceived stress, psychological distress, and salivary cortisol levels
among low SEP female heads of households with children living in either section 8 or
public housing. More specifically, this research sought to examine whether neighborhood
characteristics had an independent effect on the outcome variables
Bivariate correlation and multiple regression analyses were used to determine
significant relationships among the study variables. First simple regression analyses were
conducted and nonsignificant explanatory variables were not added to subsequent
models. Standard multiple regression analyses were utilized to determine associations
between neighborhood characteristics, depression, anxiety, health, and SC-AUCg above
and beyond individual level predictors. For all multiple regressions, assumptions were
tested by examining normal probability plots of residuals and scatter diagrams of
residuals versus predicted residuals. No violations of normality, linearity, or
homoscedasticity were detected. There was no evidence of influential outliers based on
stem and leaf plots and studentized residuals. In addition, perceived stress, chronic stress,
anxiety and depression were examined for multicollinearity. Finally seemingly unrelated
regression is used to compensate for cross-equation error correlation between the anxiety
and depression equations (Chen, Ender, Mitchell, and Wells, 2006).
Issues of Multicollinearity
Multicollinearity can occur in multiple regression analysis when independent
variables are too highly intercorrelated (Polit, 1996) and is associated with unstable
estimated regression coefficients (Chatterj ee, Hadi, and Price, 2000). A thorough
investigation of multicollinearity will involve examining the value of R2 that results from
regressing each of the predictor variables against all the others. Table 3-2 shows
collinearity diagnostics for all possible explanatory variables. The relationship between
explanlatory variables, R would be closeC to 1, andC LI theIIC~ vaiance LLV inflation fato (VIF)
would be large. Values of VIF greater than 10 is indicative of collinearity problems
(Chatterj ee et al., 2000). Tolerance defined as 1/VIF is used also used by many
researchers to check on the degree of collinearity. A tolerance value lower than 0. 1 means
that the variable considered is a linear combination of other independent variables (Chen
et al., 2006). In addition, a condition number a commonly used index of global
instability greater than or equal to 10 is an indication of global instability. The
condition index number for the variables noted in table 3.2 is 4.66. No problems with
collinearity were identified.
Table 3-2 Collinearity Diagnostics for Explanatory Variables
Variable VIF Tolerance R2 COndition
Neighborhood Economic Disadvantage 1.13 0.885 0.11 1.0
Neighborhood Disorder 1.45 0.692 0.31 1.75
Neighborhood Stress 1.95 0.514 0.49 2.05
Neighborhood Social Cohesion 1.67 0.599 0.40 2.41
Unfair Treatment 1.51 0.663 0.34 2.65
Perceived Stress 1.67 .0598 0.40 2.78
Chronic Stress 2.27 0.440 0.56 3.22
Social Support 1.99 0.504 0.50 3.41
Depression 2.87 0.350 0.65 3.50
Anxiety 3.16 0.317 0.68 4.66
Seemingly Unrelated Regression
Pairwise correlation of anxiety and depression revealed that these two measures
had a strong correlation (r, 0.74; p-value < 0.001). Though problems with collinearity
when these variables were used as explanatory variables were not revealed as mentioned
above, it was suspected that when anxiety and depression were used as dependent
variables in separate equations, the regression errors may be correlated. Correlation of
errors in regression models may lead to underestimation of the regression coefficients
(Chen, Ender, Mitchell, and Wells, 2005). Seemingly unrelated regression allows
researchers to estimate both models simultaneously while accounting for the correlated
errors at the same time, leading to more appropriate standard errors (Chen et al., 2005).
Unlike traditional multivariate regression, seemingly unrelated regression allows one to
estimate equations that do not have the same set of predictors, allowing more flexibility
in model estimation approaches. The estimates provided for the individual equations are
the same as the ordinary least squares estimates. A Chi- Square test is used to determine
the overall fit of the model (Chen et al., 2005). Seemingly unrelated regression was used
instead of multivariate regression because the explanatory variables differed for the two
Typically, when data are nested as in this study persons nested within
neighborhoods and the study is examining the contextual effects of neighborhoods on
individual health, multi-level analyses are warranted (Diez-Roux, 2000). Multi-level
analysis allows researchers to examine neighborhood-level variation in health among
populations (Merlo, Chaix, Yang, Lynch, and Rastam, 2005) and to test hypotheses about
how variables measured at one level (neighborhoods) affect relations occurring at another
(individual) level (Raudenbush and Bryk, 2002). It is intuitive that people living in the
same neighborhood or in neighborhoods with similar characteristics will have
comparable health characteristics. Therefore, when examining neighborhood contextual
effects on individual health, variation in neighborhood characteristics is essential. Given
the lack of variation in neighborhood economic disadvantage in this study (as determined
in Aim 2), multi-level statistical analyses could not be conducted.
Specific Aim 2
The second aim of the study was to determine the differences in neighborhood
characteristics of two subsidized housing types, specifically section 8 and public housing,
in which low SEP female heads of households with children live.
Assumptions for using t-tests include random sampling, a normal distribution, and
homogeneity of variance. Skewness and kurtosis tests for normality as shown in Table
3.1 were used to examine the distribution of all outcome variables. Variance comparison
tests for each of the neighborhood variables by housing subsidy type showed that the
homogeneity of variance assumption was not violated. Neighborhood economic
disadvantage was negatively skewed. Group comparison t-tests were used to determine
the differences in neighborhood disorder, neighborhood stress, and neighborhood social
cohesion by housing type. The Mann-Whitney U-test, the non-parametric analogue of the
t-test, (Polit, 1996) was used to examine differences neighborhood economic
disadvantage by housing subsidy type.
Specific Aim 3
The final aim of the study was to examine the differences in housing satisfaction,
perceived stress, psychological distress, and salivary cortisol levels, in low SEP female
heads of households with children by housing subsidy type (section 8 and public
Again, assumptions as noted in the previous section were analyzed for violation.
General health, housing satisfaction, depression, and SC-AUCg were significantly
skewed as shown in Table 3.1. SC-AUCg, general health and depression were
transformed as previously described. Group comparison t-tests were performed to detect
differences in SC-AUCG, depression, perceived stress, chronic stress, anxiety, social
support, and health by housing subsidy type. Variance comparison tests showed no
violations in homogeneity of variance by housing subsidy type. Mann-Whitney Utests
were used to examine the differences in housing satisfaction by housing subsidy type
because housing satisfaction was a one-item question measured on an ordinal scale and
was not normally distributed.
Missing data were present in several study variables including salivary cortisol.
Item non-response occurs when a participant does not respond to a question or questions
on a survey, which is the case for the missing data in this study. Several methods to deal
with missing data are available to researchers depending on the pattern of missing data
including case mean substitution, sample and group mean substitution, hot-deck
imputation, regression and multiple imputation (Fox-Wasylyshyn and El-Masri, 2005;
Patrician, 2002). Case-wise single item imputation using multinomial logistic regression
analysis was used to impute data in this study. This method was chosen because it uses a
respondent' s scores on non-missing values within a scale or subscale to predict missing
values. This approach takes into account that missing values may differ based on
differences in individual characteristics. The outcome variables (item codes) were
categorical with more than two categories; polytomous or multinomial logistic regression
was preformed to predict the missing value in a subscale. Regression imputation uses
knowledge of the available data to predict values of missing data. The underlying
principle is that missing data items can be predicted by other items in the measure or
subscale, the resulting regression equation can be used to predict missing values
(Patrician, 2002). More specific information is provided regarding missing data patterns
in the section Handling Missing Survey Data.
Handling Missing Cortisol Data
Of the 84 participants, 14 did not have usable cortisol data (defined as greater than
2 time points missing in one day). These cases were deleted and not used in data
analyses. Of the 70 participants remaining, 49 had complete cortisol data on Day 1 and
46 on Day 2. For statistical analyses missing data at Days 1 and 2 T2 were replaced by
the average values from the preceding and following samples. For example, from table 3-
3 below, one participant was missing cortisol data on Day 1 T2 and TS, and on Day 2 T3.
The average of T1 and T3 on Day 1 was used to replace the missing data point.(0.219 +
0.44)/2 = 0.66/2 = 0.33. Therefore, 0.33 was the value used to replace the missing data
point for the Day1 at T2. Equation 3-2 provides a formula for calculating the average for
a T2 data point.
Table 3-3: Example of Missing Cortisol Data for One Participant
Time of Day Day 1 Day 2 Replacement Value
T1 Awakening 0.219 0.094 --
T2 30 min after waking 0.225 0.33
T3 60 min after waking 0.44 0.44
T4 4 hr after waking .552 0.272 --
T5 9 hr after waking 0.166 0.166
T6 11 hr after waking 0.199 0.122 --
Equation 3.2: Formula for Calculated T2 SC for Days 1 and 2
T2 = T1 + T3/2
When data were missing at any time other than T2, the value from the same time point on
the preceding or following day was taken. For example, from table 3-3 the participant
was missing cortisol data on Day 2, T3 this missing data point was replaced with the
value at the same time from the preceding day (0.44).These techniques have been utilized
in other studies (Odber, Cawood, and Bancroft, 1998). After imputing SC data, a total of
70 participants were retained for data analysis.
Handling Missing Survey Data
Once cases were deleted due to missing cortisol data, missing survey data was
imputed. No more than 20% of the data were missing from any study measure. Before
imputing data, missing value patterns were determined by dummy coding missing data
for each participant with 0 = no missing data and 1 = at least one missing data point.
Study participants were grouped on whether or not missing data was present and two-
sample t-tests were performed on each study variable. Creating a missing data dummy
code and computing t-test comparisons between respondents and non-respondents is
often used to determine if non-responders differ on any of the items in the data set
(Wasylyshyn and El-Masri, 2005). A significant difference between respondents and non-
respondents indicates an association, and rules out the possibility the data are missing
completely at random (MCAR) (Wasylyshyn and El-Masri, 2005). Missing data points
are said to be MCAR if the probability of missing data on one variable is not related to
the value of that variable or is not related to other variables in the data set (Patrician,
2002). Because the state anxiety was statistically significantly different (t= -1.99, df 70,
p=0.05), and the individual social support scale approached significance, (t=-1.88, df, 61.
p=0.06), data from this study were determined to be missing at random (MAR). MAR
occurs when the probability of a missing data point in one variable is not related to the
value of that variable (Patrician, 2002). Each measure used in this study was divided into
its appropriate subscales, if present, and multinomial regression analyses were performed
for each item missing within a measure based on items present within the subscale.
Case-wise multinomial regression imputation was used to predict missing values.
This method ascribes the respondent' s predicted score based upon the items that are
present within in the missing score subscale for that respondent. The primary advantage
of this technique is that it acknowledges differences across cases (respondents) and
maximizes any one respondents own data from items in a given subscale. Also, imputing
item-level missing data retains the inter-subj ect variability across summed scores because
the maj ority of information from each participant is retained with measurements and their
sub scales. Using single value regression to replace missing values is most useful when
data are 10% 40% incomplete (Wasylyshyn and El-Masri, 2005).
After data imputation for each study variable was complete, three additional
participants had to be withdrawn from the study due to excessive missing data, leaving a
Einal sample size of 67.
In summary, after imputing missing data, 67 women were included in data analysis.
Specific aim 1 was addressed by bivariate analysis, standard multiple and multivariate
regressions, and GEE. Specific Aims 2 and 3 are examined by using t-test and the Mann-
Whitney Utest depending on the type of variable under study and whether the normality
assumption was met.
The first aim of this study was to examine the relationships among neighborhood
characteristics, perceived stress, psychological distress, and salivary cortisol levels among
low SEP female heads of households with children. The second aim was to examine the
differences in neighborhood characteristics by housing subsidy type (i.e., public and
section 8 housing). Finally, this study sought to determine if participants who lived in
public versus section 8 housing differed in terms of stress, psychological distress, general
health and salivary cortisol levels.
This chapter first presents descriptive results, including means, standard deviations,
and frequency data for each variable. The hypotheses posed in Chapter 1 are addressed
using parametric and nonparametric tests.
Description of the Sample
As described in Chapter 3, data analysis included a final sample size of 67 women.
Most of the participants in this study were black, single, had a high school education or less
and one to two children. The mean age was 30 years old. Over half of the participants
reported their main daily activity as either looking for work or keeping house and raising
children. Mean gross income was $486.50/month. Sixty-four percent of the participants
lived in section 8 housing and less than one-third were receiving direct financial assistance
through the Temporary Assistance for Needy Families Program (TANF). Table 4-1 below
provides a detailed description of the sample.
Table 4-1: Sample Demographic Profile:
Less than 9th grade 4
Less than 12th grade 21
High School Diploma 9
General Education Diploma 10
Some College/Training 16
Associates Degree 5
Number of Children
Work Full-time 9
Work Part-time 10
School Full-time 4
School Part-time 4
Work and School Part time 6
Keep House/Raise Children 30
Public Hosing 24
Section 8 Housing 43
% Mean (SD)
-- 30.33 (8.31)
Neighborhood Characteristics of the Sample
Two-thirds of the participants lived in neighborhoods with the greatest amount of
economic disadvantage that were characterized by high rates of disorder, and exposure to
crime. (See figure 4-1).
1.493 1.493 1.493
0 3 6 9 12
Figure 4-1: Neighborhood Economic Disadvantage (NED) for all Participants
Over 50% of the participants scored above the mean on neighborhood disorder while 25%
scored will over the mean of 37. 15 on neighborhood stress indicating that they perceived
their neighborhoods as areas with high rates of and crime. In addition, these women also
reported higher rates of social cohesion which is not surprising given that studies have
shown that neighborhood social cohesion may buffer the effects of neighborhood disorder
and stress (Ross and Jang, 2000). See Table 4.2 on the following page.
Table 4-2: Sample Description of Neighborhood Characteristics.
Variables Mean (SD) Range
Neighborhood Economic Disadvantage 10.16 (12.66) 1-12
Neighborhood Disorder 36.52 (17.98) 16-55
Neighborhood Stress (Crime Exposure) 37.15 (11.89) 18-72
Neighborhood Social Cohesion 13.31 (13.86) 5-25
Stress, Psychological Distress, Health, and Salivary Cortisol Sample Characteristics
Table 4-3 provides mean scores and ranges for all individual level psychosocial and
Table 4-3: Stress, Psychological Distress, Health and Salivary Cortisol Scores
Salivary Cortisol (AUC) ug/dl
Salivary cortisol levels vary based on the time at which the sample is taken. The
ranges of salivary cortisol in this sample of women are within the ranges for healthy
women of the same age group reported by other investigators (Kirschbaum, Read, and
Hellhammer, 1992). Table 4-4 provides mean scores and ranges for salivary cortisol
measures by day and time.
Table 4-4 Salivary Cortisol Scores by Day and Time
Salivary Cortisol (ug/dl) Mean (SD)
30 minutes after waking
60 minutes after waking
4 hours after waking
9 hours after waking
11 hours after waking
30 minutes after waking
60 minutes after waking
4 hours after waking
9 hours after waking
11 hours after waking
Overall, this sample of women reported higher levels of anxiety and scored lower on
general health compared to national norms (Spielberger, Gorsuch, Lushene, Vagg, and
Jacobs, 1983; Ware, Kosinski, Turner-Bowker, and Gandek, 2002; Weissman, Sholomskas,
Pottenger, Prussoff, and Locke, 1977) indicating poorer health. (See Table 4-5). They
scored well above the cut off of 16 on the CES-D, which indicates depressive symptoms
are high enough to suggest clinical depression with a mean of 24.73.
Table 4-5: Mean Psychological Distress and General Health Scores Compared to National
Sample Mean Norm for females of same age
State Anxiety 44.67 (a 11.95) 35.20 (a 10.61)
Depression 24.73 (a 11.77) >16 Suggests clinical depression
General Health 37.21 (a 27.69 49.84* 52.11** 51.01***
(a 10.62) (A 9.86) (A 8.70)
* National norms for women 18-24 years old
** National norms for women 25-34 years old
*** National norms for women 35-44 years old
Specific Aim 1: Associations among Neighborhood Characteristics, Stress,
Psychological Distress, Health and Salivary Cortisol
The first aim of this study was to determine the relationships among neighborhood
characteristics, perceived stress, psychological distress, and salivary cortisol levels among
low SEP female heads of household with children 18 years old or less. It was hypothesized
that higher rates of neighborhood disorder, exposure to crime, and neighborhood economic
disadvantage, and elevated levels of stress would be positively associated with depression,
state anxiety, and salivary cortisol and negatively associated with general health. More
specifically, this research investigated whether neighborhood characteristics had an effect
on any of the outcome when individual level factors (perceived stress, unfair treatment,
chronic stress, and social support) were added to the model.
Bivariate Analyses of Neighborhood Characteristics, Housing Satisfaction, Stress,
Depression, State Anxiety, Health and Salivary Cortisol
Based on bivariate correlations the hypotheses for specific aim 1 are partially
supported. Neighborhood disorder (ND), neighborhood stress (NS), and neighborhood
social cohesion (NSC) have significant weak to moderate positive associations with
depression, chronic stress, and unfair treatment. Only ND and NS were positively
associated with perceived stress. NSC had a positive, but weak association with housing
satisfaction and a weak negative correlation with chronic stress. Housing satisfaction also
had a weak negative association with unfair treatment. (See Table 4-6). Neighborhood
economic disadvantage (NED) was not associated with any of the outcome variables in this
study. The remainder of this section is ordered based on the outcome variable under study.
First, predictors of general health are presented, followed by state anxiety, depression and
Table 4-6: Correlations between Neighborhood Characteristics, Housing Satisfaction,
Psychological Distress, General Health, and Salivary Cortisol
NEDa ND NS NSC Housing a
Housing Satisfaction a -0.033 0.23 (0.06) -0.20 0.30*
State Anxiety -0.12 0.22 (0.06) 0.37** -0.14 -0.11
Depression (sqrt) -0.17 0.29* 0.39** -0.24* -0.06
General Health (log) -0.07 -0.09 -0.04 .019 0.057
Unfair Treatment -0.19 0.41*** 0.35** -0.29* -0.25*
Perceived Stress -0. 14 0.26* 0.39** -0. 19 -0.02
Chronic Stress -0.15 0.48*** 0.52*** -0.25* -0.04
Salivary Cortisol (SC)
AUCg ug/dl (log ) 0.07 -0.09 -0.11 0.13 -0.08
Note: *p< 0.05 **p <0.01 ***p<0.001; n =67
-0.40** 0.14 0.11
8.85 -0.68 -0.13
Note: *p < 0.05 **p < 0.01 ***p<0.001; df (1, 65); n = 67
General Health, Neighborhood Characteristics, Stress, and Psychological Distress
As previously stated, general health was log transformed to obtain a normal
distribution. Bivariate regression analysis revealed that unfair treatment and smoking
significantly impacted health in this sample of women. However, the magnitude of the
effect of unfair treatment is quite small (adj. R2 = 0.05, F (1,65 = 4.37. p-value <0.05)
accounting for only five percent of the variability in general health. Smoking (adj. R2
0. 11, F (1, 65 = 8.85. p-value <0.01) accounted for 1 1% of the variation in health. As
shown in table 4-7 below, none of the other variables in this study had a significant effect
on general health.
Results for General Health
B SE Adj R2 F
-0.006 0.021 0.001 0.08
Table 4-7: Bivariate Regression
-0.43 0. 17
Perceived Stress (PSS)
Chronic Stress (TCSI)
Individual Social Support
Number of children
0.008 0.03 2.90
0.003 -0.01 0.45
0.008 0.05 4.37
0.004 0.01 1.89
0.005 0.02 2.18
0.003 0.004 1.23
0.006 0.007 1.49
0.15 0.01 0.73
0.05 -0.01 0.06
0.04 -0.01 0.08
0.00 -0.01 0.17
Multiple regression analysis of the effects of smoking and unfair treatment on health
showed that both variables are significant predictors of health (adj. R2 = 0. 11, F (2, 64 =
6.76. p-value <0.01) (table not shown). Potential confounding variables considered were
age, race, marital status, number of children living in the household, income, and smoking.
None of these variables (other than smoking) were significantly associated with general
health in bivariate regression analyses.
Neighborhood and Individual Level Effects on State Anxiety
Bivariate analyses (see Table 4-8) show that neighborhood stress (disorder plus exposure to
crime) significantly affects state anxiety accounting for 13% of the variation (Adj. R2 0. 13,
F (1, 65) = 10.56, p-value 0.002). No other neighborhood characteristics had an impact on
Table: 4-8: Bivariate Regression Results for State Anxiety
F 95% CI
-0.67 0.55 0.023
0.33 0.18 0.036
0.37** 0.11 0.13
-0.26 0.23 0.004
0.38 0.23 0.03
1.01*** 0.18 0 .33
0.30*** 0.06 0.28
-0.35*** 0.06 0.37
-0.04 0.18 -0.01
3.27 2.09 0.02
-1.3 0.10 0.01
-0.005 0.003 0.01
4.6 4.02 0.005
-1.7 1.25 0.01
** p<0.001: df 1, 65 n = 67
Neighborhood Social Cohesion
Perceived Stress (PSS)
Chronic Stress (TCSI)
Individual Social Support
Number of children
* p <0.05, ** p <0.01, *"
The Einal model for state anxiety included neighborhood stress, perceived stress, chronic
stress and individual social support. Covariates considered were, age, marital status, race,
B SE Adj. R2
education, income, and number of children living in the household. None of these suspect
covariates had a significant impact on state anxiety and were not included in the final
model. Neighborhood stress, perceived stress, chronic stress and social support were added
to the final model. Once individual level characteristics were added to the model,
neighborhood stress no longer had an effect on state anxiety. Together perceived stress and
social support accounted for almost 50% of the variation in state anxiety (Adj. R2 = 0.49, F
(4, 62) = 17.12, p < 0.001). See Table 4-9 below
Table 4-9 Effects of Neighborhood and Individual Level Characteristics on State Anxiety
Variable B SE 95% CI
Neighborhood Stress -0.13 0.11 -0.22 0.20
Individual Level Factors
Perceived Stress (PSS) 0.60 ** 0.18 0.22 0.93
Chronic Stress (TCSI) 0.11 0.64 -0.01 0.24
Individual Social Support (ISEL) -0.21*** 0.63 -0.33 --0.08
Note: n = 67 F(4, 62) = 17.12 Adj. R2 = 0.49***
* p <0.05, ** p <0.01, *** p<0.001:
Depression, Neighborhood Characteristics and Stress
Bivariate analyses show that neighborhood disorder (Adj R2 0.07, F (1, 65) = 6.16, p-
value < 0.05), neighborhood stress (Adj. R2 0. 14, F (1, 65) = 11.42, p-value < 0.01) and
neighborhood social cohesion (Adj. R2 0.04, F (1, 65) = 4.04, p-value < 0.05) have mild
effects on depression scores in this group of women. See table 4-10.
Individual factors such as perceived stress (Adj. R2 0.26, F (1,65) = 24.42, p-value <
0.001), chronic stress (Adj. R2 0.39, F (1,65) = 43.32, p-value < 0.001), unfair treatment
(Adj. R2 0. 14, F (1,65) = 11.92, p-value < 0.01), and individual social support (Adj. R2
0.26, F (1,65) = 24.27, p-value < 0.001) also significantly impact depression.
Due to the small sample size and the large number of variables, significant neighborhood
and individual level variables in Table 4-11 were put into separate multiple regression
models. Variables that continued to have a statistically significant effect on depression
were put in the final model. Table 4-11 illustrates the first model and shows which
neighborhood variables remain significant predictors of depression. When all neighborhood
level variables were added to the model, only neighborhood stress remained significant
(Adj. R2 0. 13. F (3. 63) = 4.3 8, p -value < 0.01). Therefore, neighborhood stress was placed
in the final model.
Table 4-10: Neighborhood, Psychosocial, and Individual Effects on Depression (CES-D)
Variable B SE Adj. R2 F 95% CI
Neighborhood Economic -0.08
Neighborhood Disorder 0.05*
Neighborhood Stress 0.04**
Neighborhood Social -0.05*
Perceived Stress (PSS) 0.1***
Chronic Stress (TCSI) 0.04***
Unfair Treatment 0.08**
Individual Social Support -0.03***
Marital Status 0.43
Number of children -0.20
Note n = 67 df (1, 65)
* p <0.05, ** p <0.01, *** p<0.001:
0.06 0. 14
0.03 0. 13
The second model consists of significant individual level factors from Table 4-10
above. As shown in Table 4-11, only perceived and chronic stress (Adj. R2 0.47. F (4. 62) =
15.85, p -value < 0.01) continue to be significant predictors accounting for almost 50% of
the variation in depression. In the final model neighborhood stress no longer has a
significant effect on depression. For this sample of women, perceived and chronic stressors
(Adj. R2 0.45, F (3. 63) = 18.85, p -value < 0.001) are more important predictors of
depression than neighborhood disorder and exposure to crime.
Table 4-11: Regression Results for Neighborhood and Psychosocial Measures as Predictors
Variable B SE 95% CI
Model 1 Neighborhood Level Factors
n = 67, F (3, 63) = 4.3 8; Adj. R2 0. 13** '
Neighborhood Disorder 0.006 0.25 [-0.45 0.06]
Neighborhood Stress 0.45* 0.19 [ 0.07 0.83]
Neighborhood Social Cohesion -0.03 0.03 [-0.09 0.02]
Model 2 -Individual Level Factors
n = 67, F (4, 62) = 15.85; Adj. R2 0.47**
Perceived Stress (PSS) 0.05* 0.02 [0.01 0.09]
Chronic Stress (TCSI) 0.02** 0.007 [0.008 0.04]
Unfair Treatment 0.25 0.02 [-0.02 0.07]
Individual Social Support (ISEL) -0.01 0.007 [-0.02 0.003]
Final Model Combined
n = 67, F (3, 63) = 18.85; Adj. R2 0.45**
Neighborhood Stress 0.01 0.15 [-0.29 0.31]
Perceived Stress (PSS) 0.06** 0.02 [0.02, 0.09]
Chronic Stress (TCSI) 0.03*** 0.007 [0.02, 0.04]
* p <0.05, ** p <0.01, *** p<0.001:
Seemingly Unrelated Regression Analysis of Anxiety and Depression Regression
The correlation matrix of residuals for anxiety and depression was 0.52. Breusch-
Pagan test of independence revealed that the residuals from the two equations above are not
independent (Chi Square 18.24, p-value < 0.001). Table 4-12 shows seemingly unrelated
regression results for anxiety and depression regression models. Again, with individual
level factors added to the model, neighborhood stress is no longer a predictor of anxiety or
depression. Both perceived stress and chronic stress remain significant predictors of
depression and anxiety. Social support is also a significant predictor of anxiety with lower
levels of social support associated with higher levels of anxiety.
Possible confounding variables included weight, smoking, hours of sleep, and presence of
an acute stressor, and menstrual cycle phase. None of these factors were significantly
associated with SC-AUCg. The only individual level factors associated with SC-AUCg
were unfair treatment (Adj. R2 0. 13, F (1. 65) = 11.26, p value < 0.01) and weight (Adj.
R2 0.04, F (1. 65) = 4.05, p -value < 0.05).
Table 4-12 Seemingly Unrelated Regression Analysis of Anxiety and Depression
Equation Obs. Arms. RMSE R2 Chi2
CESD 67 3 0.95 0.47 60.12 0.000
State Anxiety 67 4 8.22 0.52 71.32 0.000
B SE Z p-value 95% CI
Neighborhood Stress 0.002 0.14 0.08 0.94 [-0.27 0.30]
Perceived Stress 0.06 0.02 2.98 0.003 [0.02, 0.09]
Chronic Stress 0.03 0.007 4.49 0.000 [0.17 ,0.044]
Neighborhood Stress 0.0005 1.25 -0.10 0.92 [-2.59 ,2.34]
Perceived Stress 0.62 0.17 3.64 0.000 [0.28 ,0.95]
Chronic Stress 0.13 0.06 2.21 0.03 [-0.26 ,-0.06]
Social Support -0.16 0.05 -3.08 0.002 [-0.26 -0.06]
When these variables were added to a multiple regression model both remained
significant predictors of salivary cortisol (Adj. R2 0.21, F (2. 62) = 9.32, p -value < 0.001)
and accounted for 20% of the variability in mean salivary cortisol over the day. See table 4-
14. Contrary to the hypothesis that unfair treatment (stress) would be positively associated
with SC-AUCg, for each 0.04 point increase in unfair treatment SC-AUCg decreased by
one unit (ug/dl).
Given the limitations regarding the lack of sensitivity of SC-AUCg to differences in
individual cortisol levels over time and the correlation of repeated salivary cortisol
measures within each person as discussed in chapter 3, general estimating equations (GEE)
were also used to examine the relationships among neighborhood characteristics, stress,
psychological distress and salivary cortisol.
Table 4-13: Simple Regression SC-AUCg
Variable B SE Adj. R2 F 95% CI
0.02 0.03 -0.01
Individual Social Support
Menstrual Cycle Phase
0.29 -0.05 0.08
-0.17 0.04 -0.01 0.19 -0.09 0.06
0.04 0.05 -0.003 0.76 -0.5 0.13
* p <0.05,
*** p<0.001; df 1, 65
** p <0.01,
Table 4-14: Multiple Regression of Individual Level Characteristics on SC-AUCg
Variable B SE 95% CI
Unfair Treatment -0.045*** 0.01 -0.07 -0.02
Weight -0.004* 0.002 -0.007 -0.0008
Note: n =65 F(2, 62) =9.32 Adj. R2 0.21***
* p <0.05, ** p <0.01, *** p<0.001:
As stated in chapter 3, in addition to using standard regression, GEE was used to
examine the relationship between neighborhood characteristics, stress, and psychological
distress and repeated measures of salivary cortisol. Using GEE methodology, neighborhood
characteristics do not have an effect on salivary cortisol. In this sample of women,
perceived stress and unfair treatment are negatively associated with salivary cortisol. As
seen in the final model, the GEE approach yields more conservative results compared to
standard multiple regression using AUCg. For each unit change in unfair treatment and
perceived stress salivary cortisol decreases by 0.02 and 0.03 units respectively, after
controlling for other potentially confounding psychosocial and physiological stressors (p
<0.05). See table 4-15.
Table 4-15: GEE Population Averaged Model of Effects of Neighborhood Characteristics,
Stress and Psychological Distress on Salivary Cortisol
Variable B SE p 95% CI
Neighborhood Characteristics Multiple Regression
Number of observations 804; number of groups 67; Wald Chi2 7. 02; p = 0. 22
Neighborhood Economic Disadvantage -0.007 0.03 0.80 -0.066 0.50
Neighborhood Disorder 0.09 0.01 0.47 -0.016 0.03
Neighborhood Stress -0.01 0.008 0.18 -0.025 0.005
Neighborhood Social Cohesion 0.007 0.02 0.77 -0.040 0.053
Individual Level Characteristics Multiple Regression
Number of observations = 804; number of groups = 67; Wald Chi2 1 6.42; p = 0. 02
Unfair Treatment -0.031 0.013 0.02 -0.06 0.005
Perceived Stress -0.29 0.013 0.04 -0.05 0.002
Chronic Stress 0.005 0.005 0.27 -0.004 0.014
Individual Social Support 0.0001 0.005 0.98 -0.009 0.009
Depression 0.009 0.010 0.41 -0.012 -0.009
Anxiety -0.007 0.011 0.53 -0.027 0.014
Final Model Controlling for Individual SES, physiological factors and health behaviors
Number of observations 768; number of groups = 64; Wald Chi2 1 9. 73; p 0. 01
Unfair Treatment -0.02 0.012 0.03 -0.05 -0.002
Perceived Stress -0.025 0.011 0.02 -0.05 -0.003
Monthly Income -0.0001 0.0002 0.64 -0.0004 0.0003
Weight -0.003 0.002 0.07 -0.006 0.0003
Menstrual Cycle Phase 0.06 0.11 0.60 -0.16 0.27
Smoking Packs per day -0.14 0.19 0.47 -0.52 0.24
Number of Children in household -0.029 0.063 0.65 -0.15 0.09
Specific Aim 2: Differences in Neighborhood Characteristics by Housing Subsidy
The second aim of this study was to determine the differences in neighborhood
characteristics of two subsidized housing types, specifically section 8 and public housing,
in which low SEP female heads of households with children live. It was proposed that
public housing sites would have significantly more neighborhood disorder, greater levels of
neighborhood disadvantage, higher levels of neighborhood stress, higher reports of crime
exposure, and lower levels of neighborhood social cohesion than section 8 housing sites.
Group comparison T-test and Mann-Whitney U-test were used to test whether
neighborhoods differed by housing subsidy type.
Eighty percent of the women living in public housing lived in the most economically
disadvantaged neighborhoods, while a little over one half of those living in section 8
housing lived in the poorest areas. Figure 4.2 illustrates the differences in neighborhood
economic disadvantage by housing subsidy type.
As shown in table 3-1, skewness and kurtosis tests for normality showed that NED is
significantly skewed to the left. Therefore, the Mann-Whitney Utest was used to test
whether there were differences in NED by housing subsidy type. The hypotheses were
partially supported. Women living in section 8 housing units were located in more
economically advantaged areas (z = -2.552, p<0.05) (table not shown). No differences in
neighborhood disorder, exposure to crime, or collective efficacy by housing type were
found in this sample of women.
Specific Aim 3: Differences in Stress, Psychological Distress, Health and Salivary
Cortisol by Housing Type
The final aim of this study was to examine the differences in housing satisfaction,
perceived stress, psychological distress, and salivary cortisol levels, in low SEP female
10 15 0
I II I
heads of households with children by housing type. It was purported that women living in
public housing would experience significantly lower levels of housing satisfaction, have
higher levels of perceived stress, psychological distress, and greater alterations in salivary
cortisol secretion than women living in section 8 housing.
2 3262 326
Graphs by SQ_S8PH
Figure 4.2: Neighborhood Economic Disadvantage (NED) by Housing Subsidy Type
The outcome variables were housing satisfaction, perceived stress, chronic stress,
state anxiety, depression and SC-AUCg. Housing satisfaction is an ordinal variable;
therefore the Mann-Whitney test was used to examine differences in housing satisfaction
by housing type. T-tests were used for all other variables. There were no differences in any
of the outcome variables by housing subsidy type. The hypotheses for this specific aim
were not supported.
The hypotheses for specific aims one and two were partially supported. Study
results did not support specific aim three. The women in this study have higher rates of
state anxiety and depression, and lower levels of general health compared to national norms
for the same age group. Neighborhood disorder and crime exposure were mildly to
moderately associated with increased levels of perceived stress, unfair treatment, chronic
stress, depression and anxiety. However, the neighborhood effects on depression and
anxiety became statistically insignificant when perceived stress, unfair treatment, chronic
stress and other individual level covariates were added to the model. The following chapter
provides a detailed discussion on the study results, discusses the limitations of the study
and implications for public health nursing research and practice.
DISCUSSION AND RECOMMENDATIONS
This chapter presents maj or study findings, addresses study limitations and
discusses implications for public health nursing research and practice. First, findings
regarding sample characteristics are discussed. Then maj or Eindings for each specific aim
and associated hypotheses are presented. Next study limitations are acknowledged.
Finally, implications for public health nursing research and practice are discussed.
This study is unique in its design and attempts to examine the associations among
housing type, neighborhood characteristics, stress, psychological distress, health, and the
hypothalamic-pituitary-adrenal axis (HPA axis), specifically salivary cortisol. Salivary
cortisol samples were collected for two days in women living in section 8 or public
housing while in their natural setting going about their daily routine. To date only one
study has examined neighborhood characteristics (neighborhood socioeconomic status) in
relation to the HPA axis, specifically cortisol levels; however, that study examined
cortisol as a response to an acute stressor, as opposed to basal levels in relation to chronic
stress exposures. Kapuku, Trieber and Davis (2002) colleagues examined the association
between neighborhood socioeconomic status (SES), cardiovascular function, and plasma
cortisol in response to laboratory-induced stress in a sample of 24 black males 16 to 25
years old. They found that family SES was related to baseline serum cortisol level (partial
r = .46, p<.05), but the correlation between neighborhood SES was not statistically
significant (Kapuku, Treiber, and Davis, 2002). Other studies have examined individual-