USF Libraries
USF Digital Collections

A multilevel modeling analysis of the geographic variability of low birth weight occurrence in Florida

MISSING IMAGE

Material Information

Title:
A multilevel modeling analysis of the geographic variability of low birth weight occurrence in Florida
Physical Description:
Book
Language:
English
Creator:
Green, Joseph William
Publisher:
University of South Florida
Place of Publication:
Tampa, Fla.
Publication Date:

Subjects

Subjects / Keywords:
medical geography
deprivation
multilevel modeling
low birth weight
Dissertations, Academic -- Geography -- Masters -- USF
Genre:
government publication (state, provincial, terriorial, dependent)   ( marcgt )
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )

Notes

Summary:
ABSTRACT: The associations of neighborhood level socioeconomic deprivation and low birth weight were investigated among 1,030,443 singleton live births in the State of Florida between the years 1992 and 1997. Census data for per capita income, unemployment, percent of individuals living below the poverty line, vehicle ownership and educational attainment were used as neighborhood level indicators of socioeconomic status. Additionally, these variables were combined into a deprivation index to measure relative deprivation of neighborhoods across Florida. Birth data were linked to census block groups and tracts, which were used as proxies for low birth weight. Multilevel models were used to model the relationship between the deprivation index and each of the indicators and low birth weight, while adjusting for individual level risk factors.After adjusting for individual level factors no consistent relationship between neighborhood socioeconomic measures and low birth weight could be established. The relationship between neighborhood socioeconomic factors and low birth weight varied across ethnic categories. Among White Non-Hispanics and Hispanics measures of socioeconomic deprivation had a small association with low birth weight. However, for Black Non-Hispanics neighborhood measures had little consistency in predicting the occurrence of low birth weight
Thesis:
Thesis (M.A.)--University of South Florida, 2004.
Bibliography:
Includes bibliographical references.
System Details:
System requirements: World Wide Web browser and PDF reader.
System Details:
Mode of access: World Wide Web.
Statement of Responsibility:
by Joseph William Green Jr.
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains 119 pages.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 001498261
oclc - 57710697
notis - AJU6866
usfldc doi - E14-SFE0000499
usfldc handle - e14.499
System ID:
SFS0025190:00001


This item is only available as the following downloads:


Full Text

PAGE 1

A Multilevel Modeling Analysis of the Geogr aphic Variability of Low Birth Weight Occurrence in Florida by Joseph William Green Jr. A thesis submitted in partial fulfillment of the requirement for the degree of Master of Arts Department of Geography College of Arts and Sciences University of South Florida Major Professor: Steven Reader, Ph.D. Graham A. Tobin, Ph.D. Thomas J. Mason, Ph.D. Date of Approval: October 14, 2004 Keywords: low birth weight, multilevel modeling, deprivation, medical geography Copyright 2004, Joseph William Green Jr.

PAGE 2

Dedication This thesis is dedicated to Joseph Wi lliam Green Sr. and Bonita Marie Green.

PAGE 3

Acknowledgements I would like to express my most sincere gr atitude to the many individuals who have assisted me during my quest to produce this thesis. In particular I would like to thank Dr. Steven Reader for his gui dance, support and encouragem ent. A sincere thank you also goes to my committee memb ers, Dr. Graham Tobin and Dr. Thomas Mason for their contributions and support. A special thank you is given to my family fo r their patience, love and continual support and encouragement. I could not have done it without you; thank you Mom, Dad, and Lara.

PAGE 4

i TABLE OF CONTENTS LIST OF TABLES iv LIST OF FIGURES vii LIST OF MAPS viii ABSTRACT ix CHAPTER ONE: INTRODUCTION 1 Introduction 1 CHAPTER TWO: LITERATURE REVIEW 3 Defining Low Birth Weight 3 Trends 4 Racial Differences in Birth Weight Trends 4 Public Health Implications of Low Birth Weight 5 Risk Factors for Low Birth Weight 6 Theoretical Conceptualization of Levels of Analysis in Health Research 7 Multilevel Analysis 12 Explanation of Multilevel Models 8 Explanation of the Statistical Model Based on A Two-Stage Simple Regression Model 18 Multilevel Analysis for Binary Response 22 Socioeconomic Status and Negative Health Outcomes 23 Mortality Studies 24 Chronic Conditions 24 Socioeconomic Status and Health 25 Index of Deprivation 27 CHAPTER THREE: THEORETICAL FRAMEWORK & METHODOLOGY 34 Theoretical Framework 34 Research Model 35 Research Question 36 Research Hypothesis 36 Description of Data 37 Research Methodology 38 Explanation of Group Level Variables 39 Per Capita Income 39 Availability of an Automobile 40

PAGE 5

ii Linguistic Isolation 40 Poverty Rate 40 Educational Attainment 41 Unemployment Rate 41 Explanation of Indivi dual-Level Variables 44 Ethnicity 45 Age of Mother 46 Smoking 46 Marital Status of the Mother 47 Parity 48 Weight Gain During Pregnancy 49 Gender of Baby 49 Two-Level Random Intercept Logistic Regression Model 50 CHAPTER FOUR: RESULTS 52 Results 52 Individual Level Results 56 Smoking 57 Ethnicity 57 Parity 58 Gain 59 Marital Status 59 Age 60 Gender of Baby 60 CHAPTER FIVE: DISCUSSION 61 Discussion and Interpretation of Results 61 Summary of Findings 61 Major Findings 63 Application of Theory 64 Revision to Theory 65 Hills Criteria for a Causal Relationship Between Neighborhood Deprivation and Low Birth Weight 65 Temporal Relationship 65 Strength 66 Specificity 67 Dose-Response Relationship 67 Coherency 67 Biologic Plausibility 68 Experiment (Experimental Modification) 69 CHAPTER SIX: CONCLUSIONS 71 Consideration of Alte rnative Explanations 71 Consistency with Literature 72 Limitations 73 Strengths 76 Geographic / Public Health Implications 77 REFERENCES 79

PAGE 6

iii APPENDIX A: MULTILEVEL MODEL RESULT TABLES 88 APPENDIX B: MAPS 104

PAGE 7

iv LIST OF TABLES Table 2.1 Low Birth Weight Rate by Race and Hispanic Origin of Mother 5 Table 2.2 Explanation of First Stage Individual-level Equation 20 Table 2.3 Explanation of Sec ond Stage Group-level Equation 21 Table 2.4 Commonly Encountered Grouplevel Indices of Deprivation 32 Table 3.1 Correlations for Data in: 1990 Deprivation Index 43 Table 3.2 Correlations for Data in: 2000 Deprivation Index 44 Table 3.3 Explanation of Variab les used in Construction of Multilevel Models 50 Table A.1 Results of Complete Multi level Model For Year 2000 Census Block Groups Showing Odds Ratios For Individual and Group Level Variables. 88 Table A.2 Results of Complete Multilevel Model For Year 1990 Census Block Groups Showing Odds Ratios For Individual and Group Level Variables. 89 Table A.3 Results of Multilevel Model For White Non-Hispanic Mothers Year 2000 Census Block Groups Showing Odds Ratios For Individual an d Group Level Variables. 90 Table A.4 Results of Multilevel Model For White Non-Hispanic Mothers Year 1990 Census Block Groups Showing Odds Ratios For Individual an d Group Level Variables. 91 Table A.5 Results of Multilevel Model For Black Non-Hispanic Mothers Year 2000 Census Block Groups Showing Odds Ratios For Individual an d Group Level Variables. 92

PAGE 8

v Table A.6 Results of Multilevel M odel For Black Non-Hispanic Mothers Year 1990 Census Block Groups Showing Odds Ratios For Individual an d Group Level Variables. 93 Table A.7 Results of Multilevel M odel For Hispanic Mothers Year 2000 Census Block Groups Showing Odds Ratios For Individual and Group Level Variables. 94 Table A.8 Results of Multilevel M odel For Hispanic Mothers Year 1990 Census Block Groups Showing Odds Ratios For Individual and Group Level Variables. 95 Table A.9 Results of Complete Multilevel Model For Year 2000 Census Tracts Showing Odds Ratios For Individual and Group Level Variables. 96 Table A.10 Results of Complete Multilevel Mothers Year 1990 Census Tracts Showing Odds Ratios For Individual and Group Level Variables. 97 Table A.11 Results of Multilevel Model For White Non-Hispanic Mothers Year 2000 Census Tracts Showing Odds Ratios For Individual and Group Level Variables. 98 Table A.12 Results of Multilevel Model For White Non-Hispanic Mothers Year 1990 Census Tracts Showing Odds Ratios For Individual and Group Level Variables. 99 Table A.13 Results of Multilevel Model For Black Non-Hispanic Mothers Year 2000 Census Tracts Showing Odds Ratios For Individual and Group Level Variables. 100 Table A.14 Results of Multilevel Model For Black Non-Hispanic Mothers Year 1990 Census Tracts Showing Odds Ratios For Individual an d Group Level Variables. 101 Table A.15 Results of Multileve l Model For Hispanic Mothers Year 2000 Census Tracts Showing Odds Ratios For Individual and Group Level Variables. 102 Table A.16 Results of Multilevel M odel For Hispanic Mothers Year 1990 Census Tracts Showing Odds Ratios For Individual and Group Level Variables. 103

PAGE 9

vi LIST OF FIGURES Figure 2.1 Topology of Multilevel Variables 17 Figure 2.2 Hierarchical Data Structure 18 Figure 3.1 Theoretical Framework 35 Figure 3.2 Multilevel Conceptual Model 36

PAGE 10

vii LIST OF MAPS Map B.1 Percent Live Births <250 0g; State of Florida for Years 1992-1997. 104 Map B.2 Miami, Dade County; De privation Index and Low Birth Weight 105 Map B.3 Orange County / Orlando Florida; Deprivation Index and Low Birth Weight. 106 Map B.4 Hillsborough County; Depr ivation Index and Low Birth Weight 107 Map B.5 Jacksonville Florida; De privation Index and Low Birth Weight 108

PAGE 11

viii A Multilevel Modeling Analysis of the Geogr aphic Variability of Low Birth Weight Occurrence in Florida Joseph William Green Jr. ABSTRACT The associations of neighborhood level socio economic deprivation a nd low birth weight were investigated among 1,030,443 singleton live bi rths in the State of Florida between the years 1992 and 1997. Census data for pe r capita income, unemplo yment, percent of individuals living below the pove rty line, vehicle ownership and educational attainment were used as neighborhood level indicators of socioeconomic status. Additionally, these variables were combined into a deprivati on index to measure relative deprivation of neighborhoods across Florida. Birth data were linked to census block groups and tracts, which were used as proxies for low birth wei ght. Multilevel models were used to model the relationship between the deprivation inde x and each of the indicators and low birth weight, while adjusting for individual level ri sk factors. After ad justing for individual level factors no consistent relationship be tween neighborhood socioeconomic measures and low birth weight could be establ ished. The relationship between neighborhood socioeconomic factors and low birth weight varied across ethnic categories. Among White Non-Hispanics and Hispanics measures of socioeconomic deprivation had a small association with low birth weight. However, for Black Non-Hispanics neighborhood measures had little consistency in predic ting the occurrence of low birth weight

PAGE 12

1 CHAPTER ONE: INTRODUCTION Introduction This study uses a multilevel analysis model to study the simultaneous effects of individual level variables a nd group-level characteristics on low birth weight outcomes. The purpose of this exercise is twofold. First, it is a way to further explore and conceptualize the relationship that exists be tween characteristics of geographic location and low birth weight incidence. Second, it is an investigation of the usefulness of multilevel analysis in predicting and explaining th e variability seen in health outcomes. Additionally, this study brings a geographic pers pective to a public health issue in the hopes of further understanding the relationship between place of re sidence and health outcomes. Low birth weight outcomes were used as the dependent variable in this multilevel analysis. Low birth weight is one of the mo st studied outcomes in public health, in part due to the ready availability of data through vital statistics (Wilcox 2001). It remains a serious public health issue to this day. A lo w weight at birth is associated with an increased risk of mortality during the first ye ar of life and an increased risk of chronic diseases in adulthood, negative developmental and health outcomes (Klein et al. 1989;

PAGE 13

2 Hack 1994; Shino and Berhman 1995; Hard ing 2001; Terry and Susser, 2001). The majority of available studies focus on individu al level variables like maternal nutrition for example. Other studies have attempted to examine ecological fact ors in attempts to determine the contextual influences and spatia l variability of low bi rth weight incidence. (Kieffer et al. 1993; Shiono and Behrman 1995; Cross et al. 1997; Reader 2001). Each approach contributes to the understanding of the variables influencing low birth weight, but limitations do exist with both an individu al and an ecological (or group-level) analysis, which may be overcome through the proper use of a multilevel analysis. Traditional multiple regression models atte mpt to estimate a relationship between an outcome (response or dependent variable) and one or more independent variables (predictor variables). A multiple regressi on model will show an average relationship between the response and predic tor variables assuming that residuals are independent. However, the multilevel structure of some data violates this assumption. Such a violation can commonly be seen in hierarchal or ne sted data (TRAMSS 1999). The social context or geographic place of residence of an expectan t mother is one such example. Therefore, this paper approaches the issue of low birth weight as a multilevel data hierarchy. With mothers and their individual-level risk factor s nested within geogra phic areas of study. The geographic areas of study are census block groups, whic h are used as proxies for neighborhoods.

PAGE 14

3 CHAPTER TWO: LITERATURE REVIEW Defining Low Birth Weight Arvo Ylppo defined the threshold weight for low birth weight (LBW) as a birth weight of <2500 g. This value has continued to be used in the literature to th is day. The purpose of this definition was originally to differentiate between infants carried to full term and those born pre-term. At the time of this de finition, birth weight a nd gestational age were used interchangeably. It was assumed that babies born with a LBW were premature. Because of this assumption the definition of prematurity was considered to be LBW. This definition persisted in the literat ure through the 1960s. (Kiely et al.1994) In the late 1940s, starting with McKewan and Gibson (1947), and continuing through the 1960s, epidemiological evidence began to accu mulate which would clearly define the differences between low birth weight and gesta tional age. It became clear that not all low birth weight babies we re premature. Additionally, some premature babies were not of low birth weight. Researchers began to rec ognize that (LBW) babies could be placed into two groups; babies born pr eterm (earlier than 37 weeks gestation) and those carried full term but which exhibited intrauterine gr owth retardation (IUGR ). In 1961 the World Health Organization (WHO) formally reco mmended against the use of LBW as the

PAGE 15

4 definition for prematurity. Not long after the WHOs recommendation, the use of the term premature was abandoned in favor of the more descriptive term preterm (Kiely et al. 1994) The definition of LBW itself has not change d. However, it now only denotes the weight at birth of an infant. No longer does it carry the assumptions about gestational age. It may be instructive to differentiate between IUGR and preterm birt h when discussing low birth weight when attempting to investig ate or explain potential causes of LBW associated with full term infants of low birth weight. However, for the purposes of this study the outcome of interest is low birth weight, which will incl ude low birth weight babies of normal gestational age and those born preterm. Trends The LBW rate in the United States was 7.7% in 2001. This is an increase from the 2000 rate of 7.6%. The LBW rate for 2001 was th e highest recorded since 1970 (7.9%). The LBW rate decreased in 1985 to 6.75% and has ri sen to current levels. The percent of very low birth weight (VLBW) (birth weight less than 1500 g) births was 1.4%, which is up from 1.27% in 1990 but less than the highest rate of 1.67 in 1981 (NCHS 2001) Racial Differences in Birth Weight Trends The LBW rate among non-Hispanic whites has steadily increased from 5.6% in 1990 to 6.8% in 2001. While the LBW rate among non-Hi spanic blacks decreased from 13.6% in 1990 to 13.1% in 2001. Despite this trend non-Hispanic black mothers remain approximately twice as likely to have a LB W baby as a non-Hispanic white mother. The

PAGE 16

5 LBW rate among Hispanic mothers increase d from 6.1% in 1990 to 6.5% in 2001. The aforementioned numbers include both singlet on and multiple births. Infants born of multiple births are 10 times as likely as a singleton to be of low birth weight. Therefore, the increases seen in non-Hispanic white births are partially due to the increased rate of multiple births (NCHS 2001). Regardless, singleton births among the various racial categories show similar trends (see Table 2.1). Table 2.1 Low Birth Weight Rate by Race and Hispanic Origin of Mother (Singleton Births) (NCHS 2001) 2001 2000 1995 1990 Non-Hispanic White Percent Low Birth Weight 5.77 5.68 5.65 5.29 Non-Hispanic Black Percent Low Birth Weight 13.76 13.9 14.15 14.46 Hispanic Percent Low Birth Weight 6.33 6.3 6.29 6.1 Public Health Implicatio ns of Low Birth Weight While most children born with low birth weight develop no significant health problems, low birth weight babies, as a whole, are more likely to have abnormal growth and development as well as adverse health condi tions (Hack et al. 1994). Numerous studies implicate low birth weight as a predictor fo r cardiovascular diseas e in adulthood (Barker 1992; Kuh and Ben-Shlomo 1997; Terry and Susser 2001; Harding 2001). Low birth weight has been implicated in numerous pr ospective cohort studies as a predictor for other diseases as well (Type II diabetes, breas t and other cancers). Infants born with low birth weight are more likely to have brain injuries, lung and liver disease. Children born

PAGE 17

6 of low birth weight are at an increased like lihood of having learning di sabilities, attention deficit disorders, breathing problems and deve lopmental impairments. (Hack et al. 1994) Reports in developed as well as developing countries support these findings (Terry and Susser 2001). Risk Factors for Low Birth Weight The majority (70%) of low birth weight ba bies are born pre-term (before 37 weeks) (Kieley et al. 1994). There are numerous risk factors for pre-term births, which include; carrying more than once baby, a history of pr e-term births, exposur e to tobacco smoke, environmental stressors, bladder or vaginal infections during pregnancy, or previous abortions. Numerous studies ha ve supported the associations between these risk factors and low birth weight. Kiely et al. (1994) discuss the findings and limitations of many of these studies. Moreover, they have divided the risk factors as follows: Demographic Risk Factors Maternal Age Race and Ethnicity Marital Status Socioeconomic Status Toxic Exposures Risk Factors Cigarette Smoking Alcohol Consumption Illicit Drug Use Ambient Environmental Exposures Pregnancy Risk Factors Maternal Height and Weight Reproductive History Weight Gain During Pregnancy Prenatal Care Parity of 0 or >5

PAGE 18

7 Medical Risks Hypotension Hypertesion/preeclampsia/toxemia Genitourinary abnormalities/surgery Poor obstetric history Maternal genetic factors Infections (including rubella, b acteriuria and cytomegalovirus) When examining the possible in teraction between i ndividual-level characteristics (e.g., maternal health, smoking etc.) and the gr oup-level characteristics (e.g., neighborhood and socio-economic status) it becomes clear that the level of analysis must reflect the interplay between the individual and the group. Multilevel analysis may be a useful tool in examining such interactions. Theoretical Conceptualization of Leve ls of Analysis in Health Research The investigations of health outcomes have traditionally been divided into two distinct levels of analysis; ecologic (a ggregate studies) and individu al-level studies (Greenland 2001). Ecologic studies focus on group-level anal ysis in which the basic unit of analysis is the population. A study that seeks to link average income to mortality rate due to cardiovascular disease would be one example of an ecological study. Ecological studies in this context are generally descriptive a nd hypothesis-generating. Rarely, should such studies be used for the testing of hypotheses (S zklo and Nieto 2000). Such an analysis is generally reserved for the indi vidual-level analysis. Individu al-level studies, like cohort and case control studies, are often used with the underlying idea that disease determinants are best studied at the indi vidual-level (Diez-Roux 2000). I ndividual level analysis is considered the gold-standard in epidemiol ogic studies and as suc h, ecologic studies are

PAGE 19

8 often considered to be a poor surrogate for indi vidual-level analysis. This is due, in part, to the assumption that individual-level outco mes are best explained by individual-level independent variables (DiezRoux 2000; Szklo and Nieto 2000) Neither the ecologic, nor the individual-level st udies take into account th e interaction between the aforementioned levels of analysis simultaneous ly. Rather, they break the analysis down to a common level and ignore the interactions between the two levels. This is largely due to the desire to avoid the ecological bias that can arise from making population based inferences about individual level outcomes, this is know as the ecological fallacy. It is also possible to draw faulty inferences in the opposite direction, when inferences are made about population level outcomes from i ndividual level data; this is known as the atomistic fallacy. Traditional studies of hea lth outcomes generally ignore the individuals interaction with social factors or groups of which they are a part. This is a problem because determinates of diseas e and health alike operate in a larger social context, not just at the individual level (Hox 199 5; Diez-Roux 2000; Greenland 2001). As discussed previously there are two levels of analysis from which an investigator traditionally would work from. There are however several ways that a study can be theorized. Diez-Roux (2000) breaks down four basi c design theories. It is imperative to understand how best to apply one of the follo wing theories before proceeding with an analysis, multilevel or otherw ise. The first way, and most commonly practiced in the epidemiologic literature, woul d be to explain an observed outcome at one level with independent variables at the same level. The second way would be to explain an ecologic-level outcome with individual-level independent variables. However, this could,

PAGE 20

9 if not properly conceptualized, lead to an atomistic fallacy. The third way would be to explain an individual-level outcome with an ecologic-level i ndependent variable, although, this would most likely lead to some form of ecological fallacy. The fourth way a study could be theorized involves the explan ation of an outcome at one level based on variables at various levels as well as interactions between le vels. Multilevel analysis can best be used in the la tter type of analysis. Why have studies of health generally stayed to one level of analysis or another? An historical examination of the causal explanat ions of disease may prove illustrative. As outlined in Courgeau (2003) (see also Ca ssel 1964, Diez-Roux 2003; Pearce 1996; Susser and Susser 1996 for a more complete discussi on of eras and paradigms in epidemiology) epidemiology has seen several distinct paradi gm shifts regarding th e explanation of the causal mechanisms of disease. It is us eful to note the shiftin g paradigms seen in epidemiology closely parallel the theory of paradigm shifts developed by Thomas Kuhn in The Structure of Scientific Revolutions (1962) Kuhn (1962) notes that a hallmark of all mature scientific endea vors is the acceptance of one paradigm, which may last for long periods of time followed by the shifting to a new paradigm, generally caused by a new discovery. Moreover, new paradigms devel op to answer questions that could not be addressed by the previous paradigm. Howeve r, the new paradigms should not be seen as the answer to all questions. This is what the history of epidemiologic inquiry closely resembles when examined historically.

PAGE 21

10 The miasmatic theory of disease causa tion, formalized by Lancisi (1717), was the prevalent paradigm from antiquity through the 19th century (Courgeau 2003). In this paradigm of disease causation, unseen environmenta l factors were the culp rits of disease. More specifically, diseases were caused when the soil, air or water was bad due to the decay of organic matter. This explanation of disease causation focused on the aggregate level of analysis. Sanitary conditions of popul ations were studied and related to disease outcomes in hopes of preventing the spread of disease. The miasmatic theory of disease causation was eventually replaced by the germ theory of disease causation. The impetus for the para digm change was spurred by advances in microbiology, especially Pasteurs discoveri es. Epidemiologists focused more on discovering disease causing agents and less on the environment and aggregate-level studies. This paradigm pe rsisted until the middle of th e twentieth century (Courgeau 2003; Diez-Roux 2003). Although the germ theory of disease was inst rumental in eradicat ing several diseases through the development of vaccines, chronic disease began to occupy the focus of scientists. Thus, began the chronic disease or risk factor para digm, which focused on individual level variables and study designs. The majority of studies conducted in this manner focus on biomedical and behavioral factors and their interactions (all individuallevel). This has often been referred to as the web of causation or multi-causal model of epidemiology (Diez Roux 2003).

PAGE 22

11 The chronic disease model of causation is be ing reconceptualized by a growing number of epidemiologic researcher s (Diez-Roux 2003; 2001; 2000). Th is is not because of its lack of efficacy; it is seen as an incomp lete explanation of the causes of disease, especially chronic disease. The new paradi gm should assimilate all of the previous paradigms along with a new approach, one that also takes the social context of disease into account. The current discussion and formulation of this new paradigm focuses on how best to conceptualize the social c ontext of disease (WHO 1998; Diez-Roux 2000; Diez-Roux 2001; Diez-Roux 2003) Or, to put it another way, how can factors like social context be included in the anal ysis of disease causation. Another reason for the focus on one level of an alysis at the expense of others is the largely positivistic nature of epidemiologi c research and to some extent medical geography. This is due in large part to the st udy of the diseases themselves. As illustrated previously the diseases and the humans with the disease of study we re treated to some extent as automata with a definite cause a nd effect relationship. This is not entirely incorrect, however it must be realized that humans are social beings within a larger context. By removing the focus from the so lely empirical nature of past research a broader understanding of health outcome s and their context may be gained. The social sciences have long recognized th e interaction between individuals and their surroundings. It is a central component to ma ny disciplines within th e social sciences. Gliddens structuration theory and the critical realist id eology are two examples of theoretical frameworks from the social sc iences which place an individual within a

PAGE 23

12 broader context and account for the multilevel nature of simultaneous individual and group level interaction. (Duncan et al. 1996) Gliddens structuration theory highlights the interplay between individuals and social structure, which will eventually produce certain socio-cultural structures and contexts as well as manife stations of social behavior. (Giddens 1984) Critical real ism rejects the notion that explanations of phenomena are transhistorical and transcultural. Rather, they are place and time specific and as such are contextually influenced. (Duncan et al 1996; Bhaskar 1975) A robust theoretical framework combining Giddens structuration theory and a critique from the realist perspective would further aid in the clarification and con ceptualization of contextual health effects. Once the context is fully and accurately theorized for then the quantitative analysis of the relationship between the in dividual and context can be undertaken. While qualitative theories as to the nature of individual-context intera ction are part of the solution to including context in public health research a, qu antitative approach is still necessary. This led to the development of multilevel analysis (Blalock HM 1984; DiezRoux 2000; Hox 2002.) Multilevel Analysis Multilevel analysis is a statistical methodol ogy that is commonly us ed on data with a hierarchical structure (Hox 1995; Sullivan et al. 1999; Diez-Roux 2000). A hierarchical data structure contains a sequence of vari ables that contain or are contained by one another. For example, a simple two level hierarchical data structure would contain individual level variables nested within a gr oup level variable. Indi vidual level variables

PAGE 24

13 are characteristics of an indi vidual such as age, sex or race. Group level variables are characteristics of a particular group that individual may belong to and may influence individual level outcomes for those indivi duals. The most common example given in multilevel literature is academic performan ce of children in different classrooms (Hox 1995). In this example, the characteristics of the individual child ren are the individual level variables. Those children are contai ned in classrooms the groups, to which the children are members. The purpose of conducting a multilevel analysis is to account for individualand group-level e ffects on an individual-lev el outcome simultaneously. Multilevel analysis attempts to show how th e context and contextual variability of an individual will affect the out come under study (Diez-Roux 2000). This type of analysis can be easily applied to a geographic analys is, where individuals are nested within a particular geo-political unit, or even health research where in dividuals are nested within a larger socio-demographic unit (Hox 1995; Duncan et al. 1996; Diez-Roux 2000; Greenland 2000). As stated previously multilevel analysis wa s originally developed as a way to model student performance in school and classroom settings (Hox 1995), it has since been used to address a number of hea lth related outcomes in both publ ic health and epidemiology, with admittedly mixed results. The first use of multilevel analysis in a public health context was by Wong and Mason (19 85) and was further refined in Entwisle et al. (1986). These studies investigated how country a nd individual level variables affect an individuals fertility and contraception use. Logistic multilevel regression models were used to model World Fertility Survey data. These studies found that micro-level and, to a

PAGE 25

14 much lesser degree, macro-level socioeconomic factors effected contraception use. The subsequent usage of multilevel analysis in public health and epid emiology can generally be divided into several categories; health services research, alcohol and drug abuse research and geographic and soci al determinants of health. Health services research is concerned with the availability and utilization of health services as well as the influence hospital and health care provider characteristics have on health outcomes and patient satisfaction. Mu ltilevel analysis has been used extensively in this area of research to gauge the perf ormance of various types of health services (Duncan et al. 1997; Entwisle et al. 1997; Duncan et al. 1999; Plote and Tager 2002; Merlo et al. 2001; Merlo et al. 2003). Analysis of health services appears to be a logical area for the application of a multilevel analysis. By its very nature health care service data are hierarchical. Indi vidual patients are nested within hospitals or care providers. Moreover, individual level fact ors may interact with group level factors to influence utilization of health care se rvices. In an example of one such study, Entwisle et al. (1997) used a multilevel analysis to address how accessibility to family planning services affects contraceptive choice in Nang Rong, Th ailand. They found that travel time, road composition, relevance of alternative choices and local history of services were all influences on contraceptive choice. Merlo et al. 2001 utilized a logistic multile vel regression to analyze the interaction between individual level and institutional leve l effect on heart failure outcomes in 90 hospitals in Sweden. The study focused on th e variation in short-term prognosis (30 day

PAGE 26

15 mortality) following a hospital stay for hear t failure that appeared to exist between different hospitals in Sweden. The results of their multilevel analysis showed that individual factors played a much great er role than contextual effects. Merlo et al. 2003 used multilevel analysis to address potential cont extual effect of neighborhood of residence on an individuals use of hormone replacement therapy (HRT) and anti hypertensive medication (AHM). This study is an example of the current trend in multilevel analysis, in which the effects of small geographic areas on individual level outcomes are examined. The findings of this study were mixed with regard to the effect neighborhood has on use of the therapies. No neighborhood effects were found for the use of AHM while women living in neighborhoo ds with low social participation were less likely to utilize HRT. Studies of drug and alcohol abuse also use multilevel analysis to examine the characteristics of various contexts such as fam ily or peer groups and their interaction with individual level factors (Re ijneveld 1998; Wang et al. 1998 ; Rountree and Clayton 1999). For example, Wang et al. (1998) used a multilevel analysis to examine the influence of social context on the sharing of needles among intravenous drug users (IDU). They found that risk behaviors such as shari ng needles are associated with individual characteristics as well as the social context IDUs are nested within. The bulk of recent studies utilizing a multilevel analysis address the social determinants of health (Duncan et al. 1996; Sundquist et al. 1999; Diez-Roux 1999; Diez-Roux et al.

PAGE 27

16 2001; Pearl et al. 2001; Rauh et al. 2001; Ahern 2002) Such studies focus on the interaction of physical, social and psychological environments and individual level risk factors in hopes of better understand ing chronic disease outcomes. Studies of social determinants of diseas e generally focus on some form of spatial inequity, usually socioeconomic status. In the majority of such studies the neighborhood of residence or some larger geographic area is used as the context-level variable. Neighborhoods vary in their so cioeconomic environment. Much research is currently being focused on the role this may have in influencing an individuals health. For example, Raugh et al. (2001) utilized a logistic multilevel regression analysis to determine the effects maternal age, race a nd poverty had on low birth weight outcomes. They found that community poverty had a si gnificant effect on low birth weight outcomes in New York City. Diez-Roux et al. (2001) examined the infl uence of neighborhood of residence on the incidence of coronary heart disease when c ontrolling for individual le vel factors. This study compared several neighborhoods in the US using data from the Atherosclerosis Risk in Communities Study. They used census block groups as a proxy for neighborhoods. Their findings showed th at living in a disa dvantaged neighborhood (independent of individual risk factors) is associated with and increased incidence of coronary heart disease.

PAGE 28

17 An explanation of the variables of interest for a multilevel analysis can be found in Larzfeld and Menzel (1961), Swanborn (1981) and Hox (1995). Variables are conceptualized in the following way by Hox (1995) (figure 2.1): Figure 2.1 Topology of Multi level Variables (Hox 1995) Note: Aggregations of data at a higher level are denoted with a while disaggregations are denoted with a symbol. 1st Level 2nd Level 3rd Level Nth Level absolute analytical relational structural contextual global analytical relational structural contextual global relational contextual Hox (1995) explains the variables as follows: First-level variables are nested inside second-level and first and second-level are nest ed inside third level and so on. Within each level there are several types of variab les. Absolute and global variables are variables that refer to the part icular level of definition. Ab solute variables are variables that are only unique to the indi vidual. Relational variables describe the relationship of the units in one level to one a nother. Analytical variables refer to th e distribution of an absolute or global variable at a lower leve l. Structural variables account for the

PAGE 29

18 distribution of relational variable s at a lower level. Contex tual variables account for the higher-level context within which, the lower level variables are located. Explanation of Multilevel Model Before summarizing the multilevel model it is im portant to be able to conceptualize the hierarchical data structure of the variable s contained within the model. The notation developed by Bryk and Raudenbush (1992) for tw o and three level hier archical data has been used in several other publications e xplaining multilevel analysis (Gatsonis et al. 1995 Sullivan et al. 1999; Diez-Roux 2000). Th e notation contained in both papers is both straight forward and easily carried over in a multilevel model. Therefore, to adhere to convention and utility, it is the same notati on used here. For a two level hierarchical data structure, individuals comprise the first level (Level 1). Individuals are part of, or nested in groups or contexts (Level 2). At level 2 or the gro up level there can be J number of units or groups. Within level two there can be nj individuals in each of the level two groups. This relationship is graphically represented in figure 2.2. Figure 2.2 (From Sullivan et al. 1999) Hierarchical Data Structure Level 2 (Group or Context) Level 1 (Individual) Explanation of the Statistical Model Ba sed on a Two-Stage Simple Regression Model. In a two-level hierarchical structure, such as the one illustrated in figure 2.2 with a continuous dependent variable, a two stage mode l is constructed with an individual-level J 1 2 nj 2 1 2 N2 1 1 2 n1

PAGE 30

19 (level 1) regression model constructed for each group (contained in level 2) and a group level (level 2) model for each of the appr opriate group-level covariates (Diez-Roux 2000 and Sullivan et al. 1999). The first stage of a multilevel analysis is: Yij b0 j b1 jIij ij In the above equation Yij is the outcome variable or dependent variable for the i th individual unit nested within the j th group (in level 2). The next term, b0j, is the intercept for the j th unit in the group-level (level 2). b1 j is the regression coefficient associated with the individual level variable Iij, which is the level 1 covariate of the i th individual in the j th group. The symbol ij is the random error for the individual-level (level 1) associated with the i th individual-level (level 1) unit nested within the j th group (level 2). Individual level errors are a ssumed to be normally distributed with a mean of 0 and a variance of 2. The second stage equations of a multilevel anal ysis focus on the groups (level 2) as the unit of analysis. In this stage each of th e groupor contextspecific regression coefficients ( b0j and b1 j) are considered to be dependent variables and are modeled as a function of group level va riables. Further explanation of va riables can be seen in table 2.4 and 2.5. b0j = 00 + 01 Cj + Uoj Uoj ~ N ( 0, 00 ) b1j = 10 + 11Cj + U1j U1j ~ N( 0, 11)

PAGE 31

20 Table 2.2 Explanation of First stage indivi dual-level equation (from Diez-Roux 2001) Equation Term Explanation Assumptions Yij Outcome variable for i individual nested within j group unit Two level hierarchical data structure. Continuous dependent variable with an approximately normal distribution. b0j Intercept for the j group-level unit b1 j Regression coefficient associated with the individual level predictor Iij for the j group-level unit Iij Individual-level covariate of the i individual in the j group ij Individual-level error coefficient for the i individual in the j group Errors within each group are assumed to be independent and normally distributed with a mean of 0 and a variance of 2 Where b0j and b1j are the context specific regression coefficients carried over from the first equation. In this stage they are modeled as group-level variables. Cj is the grouplevel or contextual covariate Uoj and U1j are errors in the gr oup level equations (also know as macro errors) and are assumed to be normally dist ributed with a mean 0 and variances of 00 and 11 The variable Uoj measures the unique deviation of each group from the overall intercept 00 after accounting for the effect of C j. The variables 00 and 11 are variances of the gr oup intercepts and group sl opes after accounting for the group level variable C j, 01 represents the covariance between intercepts and slopes. 01 is positive as the intercept increases and the slope increases.

PAGE 32

21 Table 2.3 Explanation of second stage gr oup-level equation (from Diez-Roux 2001) Equation term Explanation Assumptions b0j Intercept for the j group unit 00 overall mean intercept adjusted for Cj 01 overall slope adjusted for Cj Cj Group-level covariate or predictor Uoj Random effects of the j group-level unit on the intercept adjusted for C assumed to be normally distributed with a mean 0 and variances of 00 b1j Slope for the j group unit 10 regression coefficients associated with the group level predictor C relative to the group level unit on the intercept 11 regression coefficients associated with the group level predictor C relative to the group level unit on the slope U1j Random effects of the j group-level unit on the slope adjusted for C assumed to be normally distributed with a mean 0 and variances of 11 Multilevel analysis summarizes the distributi on of the group-specific coefficients in terms of two parts. One is fixed and unchanging and the other is a random pa rt that varies from group to group. Group level errors are assume d to be independent across contexts and independent of individual-level errors.

PAGE 33

22 The above models can be combined into a final random-effects model that will include: Fixed Effects of: Random Effects of: Group level variables 01 Random intercept Uoj Individual level variables 10 Random Slope U1jIij And their interaction on 11 Individual level errors ij The individual-level outcome Yij The combined equation: Yij = 00 + 01 Cj + 10 jIij + 11CjIij + Uoj +U1jIij + ij The combined equation uses c ovariates from both stages ( C and I ) along with a term ( 11CjIij) that is considered to be cross-level and a complex error term (Uoj +U1jIij + ij) to model the interaction between levels. The errors in the combined model show a complex interaction in which individual-le vel errors are depende nt upon the group in which they are nested. Thus, the assumption of independent normally distributed errors in standard regression models is violated and sp ecial techniques must be used to estimate parameters. Multilevel Analysis for Binary Response Health data often measure incidence or outco me and as such often is a qualitative or discrete measurement. The multilevel model will differ slightly for a discrete dependent variable. For example, in this study low birth weight will be a bina ry variable where the outcome is =1 if low birth weight and =0 fo r a normal or high birth weight. In such an instance a linear regression model cannot be used because the error terms will be

PAGE 34

23 heteroskedastic, not normally distributed (due to there being only two values) and use of a linear regression could potentially lead to probabilities greater than 1 and less than 0. Therefore, it is necessary to use a logistic multilevel regression analysis. A logistic regression analysis is a non-lin ear transformation of the basi c linear model. This will constrain the estimated probabilities to fall between 0 and 1. The transformation of the linear multilevel analysis to a non-linear lo gistic multilevel analysis for a dichotomous dependent variable and continuous predictors will appear as follows (Goldstein and Rasbash 1996; Barbosa and Goldstein 2000; Rice 2001): For a binary outcome where: Yij= 0 1 0 Yij if otherwise And the probability of observing a Yij= 1 is: P(Yij = 1 x1ij) = P(Yij > 0 | x1ij) = P( ij > b0j b1 jIij + ij)= F(b0j+ b1 jIij + ij) = i j The first step is to define the logit link function at either the i ndividual or the contextual level. The logit model for an individual level equation: i j =f (b1 jIij + ij) = {1+ exp(-[ b1 jIij + ij])}-1 Where i j is the expected value of the response variable and f is a non-linear function of the linear predictor b1 jIij. For the Combined equation of group and individu al-level variables the above equation is placed in a multilevel framework (Goldstein 1996): Yij = i + ijzij Where zij = ij ij ijn / ) 1 (

PAGE 35

24 Socioeconomic Status and Ne gative Health Outcomes Risk factors for low birth we ight include individual-level medical factors as well as individual and group level demographic risk factors. The demographic risk factors include low socioeconomic status of the moth er, low educational attainment, age, race and marital status (Kiely et al. 1993). While educ ational level, age, ra ce and marital status can all be obtained from vital records the socioeconomic status of the mother is not readily available and must be obtained via al ternative methods. Yet, many researchers feel that socioeconomic status of the mo ther one of the most important factors influencing low birth weight as well as many other health outcomes (Pickett and Pearl 2001). Studies have shown that socioeconomic status influences health outcomes even amongst those with a high socioeconomic st atus (Macintyre 1994) Macintyre (1994) found less advantaged individuals had poorer health outcomes that did the more advantaged, even when the population of study wa s of individuals with a relatively high socioeconomic status. Including socioeconomic-status in the study of health outcom es serves a dual purpose. Primarily, it is a way to account for the influe nce of the structural location of groups and individuals within a population. Additiona lly, socioeconomic status accounts for the context of exposures that may be protectiv e or detrimental to a group or individual throughout the life course (Brown et al. 2004). Due to the increas ed interest in contextual influences on health outcomes, several studies in the past decade have illustrated that social and economic depriva tion are direct influences on negative health outcomes.

PAGE 36

25 (Brown et al. 2004; Marmot 2002; Lynch et al 1997; Krieger et al. 1997; Kaplan et al. 1996; Haan et al. 1987; Backlund et al. 1996; Haan et al. 1996). Mortality Studies Several studies have demonstrated a statistic al relationship between all cause mortality and those living in areas of lo w socioeconomic status. (Anders on et al. 1997; Smith et al. 1998; Haan et al.. 1987; Kaplan et al. 1997; LeClere et al. 19 97). This relationship has been explored with various single and com posite indices in both the UK and US. Smith et al. (1998) report a relative risk of 1.34 in men for all cau se mortality in the areas of highest socioeconomic depriva tion, as calculated by the Cars tairs index of deprivation and a relative risk of 1.26 for women when compared to those liv ing in non deprived areas. Additionally, Anderson et al. (1997) report a relative risk of low versus highincome men equal to 1.26 for white men and 1.49 for black men in their study of black and white adults from the US Nati onal Longitudinal Mortality Study. Chronic Conditions The outcomes of the chronic condition studies were similar to the mortality studies. Several different indi ces were used, as were differe nt methodologies and different populations. Diez-Roux (1997) repo rted significant influence of socioeconomic status on high blood pressure readings in the US. Smith et al. (1998) reporte d similar findings in their study in the UK Pickett and Pear l (2001) conducted a liter ature review of all publications prior to 1998 referencing neighborhood socioe conomic status in developed countries. Twenty-five studies were identified that met their criteria. The criteria were

PAGE 37

26 that the study must be publis hed in English, for devel oped countries, adjusted for individual income level and found through keyw ord search on Index Medicus. Of those 25, 23 were found to have statistically si gnificant relationships between neighborhood socioeconomic deprivation and negative health outcomes. Additionally, Subramanian and Kawachi (2004) conducted a literature review of multilevel studies of income inequality and health. While not a complete measure for soci oeconomic status of an area or an individual, it is in f act, generally agreed upon in th e literature that socioeconomic deprivation is composed of both social and ma terial deprivation, income inequality is a good measure of economic deprivation and as su ch may shed some light on the influence of economic deprivation as compared to soci al deprivation. Subramanian and Kawachi (2004) identified 21 studies in their revi ew. Of the 21 identified studies, 10 found significant relationships betw een income deprivation and negative health outcomes Socioeconomic Status and Health There exists a paucity of explanations as to the combination of individual and ecologic factors responsible for the effect of socio economic status on health. However, the literature suggests that at least some of this is due to ecological influences, more specifically the neighborhood of residence of an individual The neighborhood in which a person lives may influence health outcomes in a number of ways. The availability of healthcare services, lack of infrastructure, stress due to crime and poverty, absence of places to exercise safely, prevalent attitude s regarding health and healthy lifestyles as well as availability of healt hy foods all are neighborhood vari ables that may influence an individuals health (Pickett and Pearl 2001).

PAGE 38

27 An alternative explanation may be that pe rsons of a more econo mically and socially deprived area may perceive their place in th e social environment to be lower. Thus, bringing into play not only physical and economic effects on individuals but psychosocial as well (Hawe 1998; Lynch et al.. 1998; Lync h and Kaplan 1997). Th e implications of this are that the neighborhood effects may exhibit effects in broad contexts over and above the most commonly studied variables. The idea of broad contextual effects can be found prominently in community psychology research and in architecture and design theo ry. The most commonly applied terminology to this phenomenon is the activ ity setting or participatory place-making. This is the idea that a place can have different meaning to different people based due to the multiuse nature of an area. Hawe (1998) uses school s as an example. A classroom in a school is used for children during the day and for co mmunity meetings at night. Thus, different individuals may have different perceptions of the same place and its influence on them. This may hold true for and individuals nei ghborhood context as well in that individuals relate and respond to their en vironment based on the distri bution of certain components, like wealth, physical resources, time spent in a particular location, the people in that location, symbols and roles in dividuals relate to or part icipate in (Haw e 1998; ODonnell et al. 1993). Geography also uses similar ideas within the place inte gration theoretical framework (Dovey, 1985).

PAGE 39

28Index of Deprivation The renewed interest in context and its e ffect on health places specific importance on geographic variation of soci al and economic deprivation. Approaching neighborhood effects on health outcomes from a geographic pe rspective requires an area level analysis. There are several benefits to this approach, primarily, the availability of census data. Census data allow for relatively quick and co st effective analysis of geographic areas. Additionally, they allow for the accurate linkage of data to existing political boundaries. Area level analysis also allows the research er to quickly visualize spatial patterns of socioeconomic status and compare that with sp atial patterns of dis ease through the use of GIS mapping technology. This may provide the opportunity to dete rmine the necessary delivery of health care and identify areas of high risk for adverse health outcomes. Kreiger et al.. (2003) have found that census based measur ements of socioeconomic deprivation are useful, when li nked to individual level record s, (geocoded) at predicting adverse health outcomes. Numerous other st udies have supported the utility of an areabased measurement of deprivation (Carstai rs 2000; Reienveld et al. 2000; Townsend 1987; Jarman 1984). There are, however, drawbacks to the area level analysis of so cioeconomic status on health outcomes. The main criticism of area level analysis is they still do not properly deal with the ecological fallacy Put another way, many studies do not accurately address whether the effect seen is a compositional one or a contextual one (Subramanian and Kawachi 2004). As previously discussed the ec ologic fallacy is basi cally the incorrect assumption that all individuals living in a gi ven area share identical characteristics with

PAGE 40

29 that area. The ecological fallacy must be taken into consideration but area level analysis must not be completely ruled out. Taking individual level vari ables and aggregating them may create the opposite problem, commonl y referred to as the atomistic fallacy. However, it is the contention of this res earch that multilevel analysis is a methodology particularly suited to overcome both the eco logical and atomistic fallacy when used properly (see introduction). Additionally many studies do not properly examine the methodological issue of compositional vers us contextual effects (Diez-Roux 2004; Subramanian and Kawachi 2004). Thus it is necessary for a researcher to develop a way to account for neighborhood effects. One such solution is to devel op an index ranking small neighborhoods or area levels of deprivation. Data are readily available for pre-defined, political boundaries. However, data are almost non-existent for more difficult to define areas. Additionally, some areas have moving or ephemeral bounda ries (e.g.,; social groups). Thus, the question becomes; where does one obtain the data for such groups and once obtained how those data can be linked to the individual? (Diez-Roux 2004) Thus, it should be noted that for a more accurate investigation of the distribution of causal factors, and to avoid potential misspecification, the groups shoul d be more rigorously defined (Diez-Roux 2004; Pickett and Pearl 2001). This is the most difficult aspect of multilevel analysis. Raudenbush and Sampson (1999) have begun to a ddress this question with a statistical methodology called Ecometrics, which promises to assess the validity of ecological contexts, specifically neighbor hood settings. Through the us e of interviews, direct observation of multiple observers and video an alysis of neighborhoods, sources of error

PAGE 41

30 in multilevel analysis are highlighted and c onsidered (Raudenbush and Sampson 1999). This approach, which is indeed promising, is time consuming and not well suited for an investigation such as this one. A composite measurement of deprivation is necessary to assess a geographic areas socioeconomic status. Such a measure should co mbine data from a number of variables in a way that places a particular area along an axis of depriv ation ranging from the most deprived (poverty) to the least deprived (affl uence). What this implies is that particular values for the variables making up a given index are more desirable than others. That is to say that it is more desirable for an indi vidual to be employed and to have a car, for example (Carstairs 2000). Additionally, Kriege r et al. (2003) define three criteria that should guide the development of a deprivation index. The researcher should have an existing definition and conceptual framework of socioeconomic position and social class from which to work. Additionally, literatu re supported evidence for the detrimental health effects of material de privation is necessary for the meaningful application of a deprivation index. Finally a de privation index should consis t of a measure or measures that can be compared over time and space. Socioeconomic indices of depr ivation are less common in the literature in the United States than in the UK. This is however, be ginning to change with the advent of new multilevel methodologies. Through a literature search, 28 common area based socioeconomic measures were identified (Tab le 2.6). The findings of this search are consistent with Krieger et al (2003) with a few minor additi ons. What is of particular

PAGE 42

31 interest is the lack of measures of social deprivation and the focus on material influences of deprivation. This is especially common in the US literature. While the argument can be made from a Marxist framework, that the material influences the social, the explanation is, however, most likely that the US decennial census is biased toward material measurement. That is most of th e variables contained within the U.S. Census measure the material. This may be affi rmation of the Marxist viewpoint or the reinforcement of the focus on material attainment of US society as a whole. Or conversely it may only be that theses values are most easily and reli ably measured. Data from the decennial U.S. census does not lend itself to the creation of a socioeconomic deprivation index. At best a composite index of material deprivati on can be derived from several variables found in the Summary Tape F ile 3. As such, numerous studies have attempted to create socioeconomic indices through varying methodologies By far, the Townsend Deprivation Index (Townsend 1988) a nd the Carstairs Index (Carstairs 2000) are both indices commonly used in the UK to measure relative material deprivation (Carstairs 2000) are the most commonly encountered compos ite deprivation indices in the literature. Numerous indi ces have been created that ar e highly correlate d with both indices (Krieger et al. 2003). However, due to the differences between the US and UK census there is some difficulty in directly applying an index created in and specifically for the UK in the US. Additionally, indices created for European populations may not be applicable to US populations due to the hom ogeneity of the socioeconomic status of some European countries. Therefore caution mu st be exercised befo re applying an index to a population for which it was not designe d (Pearl and Pickett 2001; Reijnveld 1998).

PAGE 43

32Table 2.4 Commonly Encountered Group-level Indices of Deprivation (from Pickett and Pearl 2001) Index Type of Measure Variables Working Class Krieger et al. 1997 Material Deprivation % of persons employ ed in non-supervisory roles as a percent of persons employed in one of the US Census occupational groups Unemployment Krieger et al. 2003 Material Deprivation % of individual s >16yrs in the labor force who are unemployed Median Household Income Krieger et al. 2003 Material Deprivation Median household income (1989) $30056 Low Income Material Deprivation % of households with income < 50% of the US median household income High Income Krieger et al. 2003 Material Deprivation % of households with income >400% of the US median household income Below Poverty US Census Bureau 1997 Material Deprivation % of persons be low the federally defined poverty line. Average equaled $12647 for a family of 4 in 1989 Expensive Homes Krieger et al. 2003 Material Deprivation % of owner occupied homes >400% of the US median value of owned homes Educational Attainment Krieger et al. 2003 Social Status/Material Potential % of individuals >25 years old with less than a 12 grade education (low). Conversely % of individuals > 25 years old with at least 4 years of College (high) Crowding Krieger et al. 2003 Social Environment/Material Deprivation Percentage of households with >1 person per room Socio-economic Position 1 Krieger et al. 2003 Material Deprivation % of individuals below poverty level, working class, and expensive homes Scio-economic Position 2 Krieger et al. 2003 Material Deprivation % of individuals below poverty level, working class, and high income Socio-economic Economic Position Index Krieger et al. 2003 Material Deprivation % working class, % unemployed, % below poverty level, % individuals with low educa tional attainment, expensive homes, and median household income Carstairs (UK) Carstairs and Morris 1991 Material Deprivation Male unemployment, automobile ownership, social class, crowding Variables are not weighted Jarman (UK) Jarman 1983 Needs For Primary Care Services Unemployment, low social class, unskilled labor, overcrowding, single parent household, # children under 5yrs., pensioner living alone, moved in past year, ethnic minority. Variables are weighted. Townsend (UK) Townsend 1987 Material Deprivation Unemployment, lo w social class, not owner occupied, lacking amenities DoE (UK) DoE 1995 Needs for local authority services Unemployment, overcrowding, lacks amenities, children in unsuitable accommodations, children in low earner households, not in educational system, low income support recipients, low educational attainment, derelict land. Deprivation Index (US) Andrulis et al.. 2001 Material Deprivation Poverty rate, vi olent crime rate, unemployment rate, educational attainment, per capita income, ability to speak English. Variables are not weighted. Care Need Index (Sweden) Sundquist et al. 2003 Material Deprivation Elderly living alone, foreign-born people, unemployed people, single parents, reside nts who have moved, people with low economical, status, children under age 5. Variables are weighted Mayer-Jencks Material Hardship Measure Mayer & Jencks1989 Material Deprivation Calculates a fam ilys income to needs ratio, including; healthcare and food affordability. Gini Coefficient Gini 1912 Material Deprivation Income inequality (half of the arithme tic average of the absolute differences between all pairs of incomes in a population normalized on mean income) Robin Hood Index Material Deprivation/Income/Inequality The proportion of money that must be transferred from the rich to the poor to achieve equality.

PAGE 44

33Table 2.4 (Continued) Commonly Encountered Group-level Indices of Deprivation (from Pickett and Pearl 2001) Index Type of Measure Variables Thiel Entropy Thiel 1967 Material Deprivation/Income Inequality Measure of income inequality derived from information entropy theory Socioeconomic Deprivation index (US) Sing et al. Material Deprivation Principal component an alysis selected variables: Educational attainment, occupational stat us, Median family income, income disparity, median home value, median gross rent, unemployment rate, occupied houses with telephone, occupied houses w/o complete plumbing Atkinson Atkinson 1970 Material Deprivation/Income Inequality Calculates equity density av erage income, which is the measurement of per capita in come which if enjoyed by everybody would make total welfare exactly equal to the total welfare generated by the actual income distribution Cogdon Index Social Deprivation Mobility of I ndividuals, number of single person households for persons <65, and private renting Index of Multiple Deprivation (IMD) Jordan et al. 2000 Material and health Deprivation/Access to services 32 Variables measuring income, employment, health deprivation, disability, educa tion, skills, training, housing and geographical access to services. Diez-Roux et al.. Diez-Roux et al. 2001 Material Deprivation Variables selected through factor analysis. Log of median household income, log of median value of housing units, % of households receiving inte rest, dividend or net rental income, % of adults (>25yrs) who completed high school, % of adults who completed co llege, occupational status. Variables are not weighted. US CDC Index of Local Economic Resources Casper et al. 1999 Access to material resources White collar employment, unemployment, and family income.

PAGE 45

34 CHAPTER THREE: THEORETICAL FRAMEWORK & METHODOLOGY Theoretical Framework The multilevel model used in this research assumes a theoretical framework as seen in figure 3.1. This framework is modified fr om the conceptual model developed by Duncan et al. (1996) figure 3.2. In th e theoretical model, an indivi dual level respon se (low birth weight in this case) is directly influen ce by individual level factors (race, smoking and parity, for example). Individual level factors are influenced by contextual level variables (in this case neighborhood socioeconomic status ). Additionally, this study proposes that the individual level outcome may not only be influenced by individual level factors but by contextual level variables as well. What this suggests is that a persons neighborhood of residence will influence the development of certain risk factors for low birth weight as well as directly influence low birth weight outcomes.

PAGE 46

35 Figure 3.1 Theoretical Framework (Modified from Duncan et al. 1996). Lines of influence are in red. Contextual levels are in blue shapes. The qualitative equivalent in geography can be seen in the critical re alist explanation of spatial variability. The likelihood of contex tual variation requir es methodology, which accounts for the variability across time and space. That is to say that, human beings will behave differently at different times and under different circumstances (Peet 1998). The geography of a particular area is as important as the individu als that comprise it. The spatial landscape has the ability to in fluence individual level outcomes. Research Model The research model for this study follows the proposed multilevel structure conceptual model found in Duncan et al. (1998) figure 3.2. In this conceptual model, individual level responses are nested with in individual that are nested within groups. This model could be extended to include groups nested within regions. Additionally groups may be nested within different times as in a rep eated measures or longitudinal study. SocioEconomic Individual Risk Factors Level3 Context Level2 Individual Level1 Birth wei g ht

PAGE 47

36 Figure 3.2 Multilevel Conceptual Model (Duncan et al.. 1996) Level 3 Place Level 2 Person Level 1 Response Research Question What are the relative roles of individual-level, and ecologic, risk factors in explaining the geographic variability in the occurrence of low birth weight outco mes in the State of Florida? Research Hypothesis It is hypothesized that, after ad justing for individual level va riables, the odds of having a low birth weight child are higher for a mother living in a neighborhood with a high deprivation index score than a mother living in a neighbor hood with a low deprivation index score. Additionally, a statistically si gnificant portion of the spatial variation of low birth weight outcomes in the State of Florida is due to neighborhood effects.

PAGE 48

37Description of Data For areas with greater than 40 live births the 2000 US Census block group, and tract were used as proxies for neighborhood of residence. Block groups with le ss than 40 live births were aggregated to the Census 2000 Tract le vel. The 2000 census was used because for some births only a 2000 block group or Track were identifie d. This is due to the changing boundaries between the 1990 and 2000 cen sus. Census variables were obtained from the U.S. Census web site as dbf files. The census data were used to represent the group level variables. The variables obtained were per capita income, number of unemployed, linguistic isolation, number of individuals living be low the federal poverty level, number of individuals with an automob ile, and level of education. In addition to the Census data, Vital Statistics birth data for 1992-1997 from birth cer tificates were used as measures of individual level variables. Th e Vital Statistics data contains gestation parity, gain during pregnancy, as well as age, smoking status, education, race, and marital status of the mother. Methodology Birth record data were obtained for all singleton births in the state of Florida for the years 1992-1997 (n =1,030,443). For identification pur poses, each individual record was given a birth identification number. The individua l records were linked to the census block group of residence (average of 1000 persons) of the mother. Census block groups were used as proxies for neighborhoods. This me thodology is similar to that found in DiezRoux et al. (2001) in their study of ne ighborhood influence on the incidence of cardiovascular disease. This methodology wa s chosen for two reasons. First, the

PAGE 49

38 predefined boundaries of the U.S. Census allow for easy data linkage. Without this linkage the task of assigning census level da ta to areas with alternatively constructed boundaries would be extremely difficult and time consuming. Pickett and Pearl (2001) noted that this was a commonly accepted me thodology. Moreover, of the studies they examined in their critical review, all but one use data linked to census boundaries. However, for a more accurate investigation of the distribution of causal factors, the neighborhood of residence should be more carefully designed and conceptualized (Pickett and Pearl 2001). In addition to using Census Block Groups, i ndividual birth records were linked to the 1990 and 2000 U.S. Census Tract boundaries. The purpose of this was to compare the results obtained from the block group-level model and to possibly determine which level (Block Group or Census Tract ) the deprivation index explai ned the variance seen in low birth outcomes across groups. A material deprivation index for all block groups in the state of Florida for 1990 and 2000 was created. This study was only concerne d with material deprivation due to the nature of the US Census data. Census data are demographic and mate rial in nature, there are no direct measures of social capital or so cial environment. Da ta from the Summary Tape File 3A were used to construct an inde x. Similar to Andrulis et al. (2004), poverty status, educational attainment linguistic isolation, per cap ita income, and unemployment rate were included in the cons tructed index. However, A ndrulis et al. (2004) included crime rates in their index. There are two r easons this is not included in the constructed

PAGE 50

39 index. Most importantly is the geographic sc ale of the data. Crime rates are generally only available for MSAs. (Andrulis et al. 2004) Calculated thei r index for MSAs as proxies for cities. This research is more concerned with neighborhood contextual effects. Therefore, data that only exist at la rger aggregate areas is of little use to this study. In addition to the index variables defi ned by Andrulis et al. (2004) this study has included two additional variab les: vacancy rates and auto mobile ownership. The explanation of variable choice is expl ained in the following paragraphs. Explanation of Group Level Variables Per Capita Income In their literature review, Subramanian and Kawachi (2004) found a potential relationship between income distribution and health out comes. They hypothesize that although some studies show strong statistical relationships between low income and negative health outcomes, the failures of these to adequately explain the causal mechanism of incomes influence on health and the failure of others to find such a relationship is due to the fact that income is only one dime nsion of deprivation and as such other factors should be considered. Therefore, the annual per cap ita income of each block group was obtained from the US Census STF3 and incl uded in the model of deprivation Availability of an Automobile Several studies have shown that lack of available transpor tation plays a strong role in influencing health outcomes (Rittner and Kirk 1995; Melnikow et al.. 1997; Williamson and Fast 1998; Takano and Nakamura 2001). The lack of adequate transportation may

PAGE 51

40 act as a barrier to the receipt of health serv ices as well as influencing access to food and social networks (Bostock 2001; Brown et al. 2004 ). As Florida has little in the way of effective public transportation I am includi ng the availability of automobiles in the deprivation index. The data regarding access to one or more vehicles was obtained from the U.S. Census STF3. The percent of i ndividuals with access to no automobile was calculated from the Census data. This was then added to the de privation index. Linguistic Isolation Linguistic isolation has been shown to influence self-care behaviors and health literacy. Additionally patients with a la nguage barrier, specifically Sp anish-speaking Latinos, are less likely to have a regular source of health care and are less likely to report satisfaction with their health care (Brown 2004; Fiscella et al. 2002; Sc hur ad Albers 1996; Hu and Covell 1986). Thus, linguistic isolation may be an influence on pre-natal care and understanding of healthy practices during pregna ncy. It is estimated that only 40% of Latinas utilized prenatal car e in the District of Columb ia (Kaiser Family Foundation). Poverty Rate Krieger et al. (2003) conducted an analysis of single and composite measures of socioeconomic deprivation on childhood lead poisoning and low bi rth weight. For the outcome of low birth weight their study they report an odds ratio of 2.08 (1.98 to 2.19, 95% Confidence Interval (CI)) for mothers living in poverty in the state of Massachusetts as defined by the US Census Bureau. Additi onally, an odds ratio of 1.97 (1.65 to 2.13, 95% CI) was reported for mothers living in poverty in Rhode Island (Krieger 2003).

PAGE 52

41 Educational Attainment The level of educational attain ment has been used as a predictor for low birth weight as part of a composite index or independently in several studies (Andrulis 2004; Krieger et al.. 2003; Pearl et al. 2001; Pick ett and Pearl et al. 2001). Krie ger et al. (2003) reports a low birth weight odds ratio of 1.97 (1.86 to 2.08 95% CI) for singleton births in Massachusetts to mothers of low educational a ttainment and an odds ratio of 1.91 (1.65 to 2.22 95% CI) for singleton bi rths in Rhode Island. Lower educational attainment has been observe d to negatively influence health literacy (Gazmararian et al. 1999; Baker et al. 1998). Hea lth literacy is linked to health status. Patients with lower educational attainment are more likely to be admitted to the hospital than their more educated counterparts and less likely to be able to recognize signs and symptoms before a serious problem develops (Gazamararian et al. 1999; Baker et el 1998; Williams et al. 1998; Baker et al. 1997). Unemployment Rate Unemployment rate is one of the most co mmonly included metrics in studies of area based socioeconomic influence on negative hea lth outcomes (Andrulis 2004; Brown et al. 2004; Krieger et al. 2003; Pi ckett and Pearl 2001; Pearl 2001). For mothers living in areas of high unemployment the odds ratio for low birth weight is 1.72 (1.61 to 1.84 95% CI) (Massachusetts) a nd 1.51(1.37 to 1.67 95% CI) (Rhode Island) (Kriger et al. 2003). Additionally, Epstein et al (1985) report that less effective patient-provider

PAGE 53

42 communication is observed when the patient is of a lower occupational status. This may, in turn, influence antenatal behaviors and care practices. Once all variables were obtained and mani pulated a Z-score was calculated for each variable. Where the Z score x / where x is the block groups individual value of a variable, is the aver age for the block groups, and is the standard deviation of the variable for the city. The sum of a block gr oups Z-scores is a bl ock groups index of deprivation. The larger the score the more de prived an area is assumed to be (Andrulis 2004). Once the z-scores were calculated, a correla tion matrix was created for the 1990 variables and 2000 variables using S-Plus to determin e if certain variables were measuring a variable more than once or if there was correlation between variables. A table of Pearson Product-Moment Correlation Coefficients ( r ) was created to exam ine the degree of the linear relationship between all possible combinat ions of the coefficients that comprise the deprivation index. The cal culated correlation coefficients may be equal to any number between .00 and 1.00. A score of .00 re presents a perfect negative relationship between two variables while a score of 1.00 represents a perfect positive relationship. The results can be seen in tables 3.1 and 3.2. When examining the variables in the 1990 depr ivation index, the st rongest relationships can be seen occurring between Vehicle Own ership (Z-VehOwn) and Poverty, with a coefficient of .774. Poverty also showed a strong correlation with educational attainment

PAGE 54

43 as well as income and unemployment. E ducation attainment also showed a strong correlation with a number of vari ables, particularly income. Educational attainment and per capita income show a correlation coeffi cient of -.647. Additionally, education and vehicle ownership also showed a strong correlation with a score of .633. The remainder of the variables showed a moderate to small degree of correlation. The variables in the 2000 de privation index showed some correlations. Poverty and vehicle ownership showed the strongest correl ation (r=.623). Educa tional attainment and unemployment had a correlation coefficient of .574. Other variable combinations showed a moderate to sma ll degree of correlation. As evident in table 3.1 the 1990 index shows a good deal of correlati on between variables while the 2000 index does not. Therefore, the 1990 index will only be used as a reference to compare with the 2000 index. This is to avoid unduly weighing certain areas over others. The use of an index with a hi gh degree of correlation amongst its composite variables would artificially inflate the depriva tion z-score. The indivi dual variables that make up each index will be also used as the gr oup level variable in the multilevel analysis to determine the effect each has on low birth weight. Table 3.1 Correlations for data in: 1990 Deprivation Index Z-LingIso Z-EduAttainZ-Veh OwnZ-PovertyZ-Percap Z-Unemp Z-LingIso 1.00 .522 .306 .268 -.260 .260 Z-EduAttain .522 1.00 .633 .720 -.647 .559 Z-VehOwn .306 .633 1.00 .774 -.421 .574 Z-Poverty .268 .720 .774 1.00 -.568 .639 Z-Percap -.260 -.647 -.421 -.568 1.00 -.436 Z-Unemp .260 .559 .574 .630 -.436 1.00

PAGE 55

44 Table 3.2 Correlations for data in: 2000 Deprivation Index Z-LingIso Z-EduAttainZ-Veh OwnZ-PovertyZ-Percap Z-Unemp Z-LingIso 1.00 .424 .076 .109 -0.075 .216 Z-EduAttain .424 1.00 .368 .475 -.310 .547 Z-VehOwn .076 .368 1.00 .623 -.273 .287 Z-Poverty .109 .475 .623 1.00 -.390 .347 Z-Percap -0.075 -.310 -.273 -.390 1.00 -.197 Z-Unemp .216 .547 .287 .347 -.197 1.00 In addition to group-level variables individua l level variables were obtained from vital records data. Dummy variables were cons tructed for each of the individual level variables as the data were obtained in categor ical format. Once the data were re-coded and converted from database IV format to tab delimited format it was imported into MLwiN for multilevel analysis and modeling. The conversion was necessary to allow for the proper hierarchical data structure to be set-up. Explanation of Individual Level Variables The individual level variables included in the multilevel models are: o Ethnicity of the mother o Age of the mother o Smoking, o Marital status, o Parity o Weight gain during pregnancy o Gender of Baby

PAGE 56

45Ethnicity Ethnicity of the mother was also included as an individual-level variab le due to the racial and ethnic disparity in low bi rth weight outcomes. Black Non-Hispanic mothers give birth to 13 times more low birth weight in fants than their white counterparts. Among Hispanics low birth outcomes ar e 6.5% while white outcomes are 1.1%. Therefore, it is important to adjust for ethnicity when s eeking out the causes of low birth outcomes (Kiely et al. 1993). Odds ratios were calculated by taking the e xponents of the coeffici ents of the ethnic category variables from a single-level mode l. The odds of a Black Non-Hispanic mother having a low birth weight baby are 2.27 times that of a White Non-Hispanic mother. While the odds of a Hispanic mother was not much higher that that of a White NonHispanic mother (1.06 vs. 1.00). The predicated probability of a Black Non-Hispanic mother having a low birth weight outcome is 10.7 % compared to 5.3% in Hispanics (which is comparable to White Non-Hispanics a 5%). The data clearly show the difference in lo w birth weight outcomes amongst the three ethnic categories. This is fu lly supported by numerous studi es in the low birth weight literature (Kiely et al. 1993). It is worth noting that the difference in Hispanic and White Non-Hispanic low birth outcomes is smaller that the nationally reported difference.

PAGE 57

46 Age of Mother The age of the mother during pregnancy has been shown to influence the likelihood of a low birth weight outcome. Wo men who have a first time pregnancy under the age of 17 and over the age of 35 are more likely to have a low birth weight outcome when compared to women between the ages of 18 and 34. The age of the mother in years at time of birth was divided into three categories. The standard Census categories of 12-14, 15-18, 19-24, 25-29, 30-34, 35-39, 40-44 and 45+ were collapsed to create the three categories us ed in this study. The reference category used was 18 to 34 years of age. Ages 12-17 was the next category and ages 35-54 was the final category. The results of the tabulation and modeling in the MLwiN software package showed 14.5% of all low birth weight babies were born to mothers un der the age of 17 years old. Additionally, 11.3% were born to mothers over the age of 35. The odds ratio for a low birth weight occurrence for a mother under th e age of 17 years was 1.77 while the odds ratio for a mother over the age of 35 was 1.18 compared to a mother between the age of 18 and 34 years old. Smoking A strong association between smoking and an increase of low birth weight outcomes has been reported in the literat ure and in larger studies a dose response curve has been

PAGE 58

47 established (Keffier et al 1993). For example the Cent ers for Disease Control and Prevention conducted a study of risk factor s during pregnancy on low-income mothers for the years 1978-1998. This study showed that amongst women who had low birth weight babies (6.9%), there was a high pr evalence of smoking during pregnancy (29.7%) (Fichtner et al. 1990). First, the relationship betw een low birth weight (binary outcome) and smoking habit of the mother (also a binary variable) was examined. Smoking status was categorized as either non-smoking or smoking. The non-smoking category was used as the reference category. A tabulation of per cent low birth weight births by smoking status of the mother was generated in MLwiN. This tabulation showed that amongst women who had a low birth weight baby, the preval ence of smoking was 10.7%. This also showed that women who smoke are almost twice as likely to have a low birth we ight baby that a nonsmoking mother. Odds ratios were calculated by taking the exponents of the coefficients of the smoking category variables The odds of a mother who smokes having a low birth weight baby is 1.93 times that of a non-smoking mother. From these results it is clear that the percentage of low birth weight infant s born to mothers who smoked in higher than those born to non-smoking mothers although no t as high as Fitchner et al. (1990). Marital Status of the Mother The marital status of the moth er during the pregnancy is a cat egorical variable with the responses being either married or not married. Generally, unwed mothers have a slight increase in the likelihood of having a low birth weight baby (Kiely 1993). The

PAGE 59

48 tabulation of marital status of the mother shows a slight increase of the result of the single-level logistic regressi on. Among low birth weight outcomes 9.2% were to unwed mothers compared to 4.9% among married mothers. Parity Parity, defined in this study, as the number of previous live births, has been shown to have an effect on low birth weight outcomes Women with low parity (one previous birth) are at a decreased risk of having a low birth weight outcome when compared to women who are primiparious. However, the CDC reports that primiparous women have a 23% greater risk of a low birth weight out come when compared to multiparous women (Kiely et al. 1993). The relationship between low birth weight outcomes and parity was examined through the tabulation of parity categories and the r unning of a single level model in MLwiN. The results from MLwiN show that a woman with one previous live birth has a decreased risk of having a low birth weight infant as does a mo ther with two to four previous live births when compared to a woman who is primipar ous. The odds ratio for a low birth weight outcome amongst primiparous women was 1.43 in this study. On the opposite end of the parityspectrum, women with a parity of gr eater than 5 had an odds ratio of 1.46 when compared to women with one to four previous births.

PAGE 60

49 Weight Gain During Pregnancy According to Kiely et al. (1993) severa l studies have found significant correlation between very low weight gain and low birth weight outcomes. For the purposes of this study weight gain during pregnancy is a cate gorical variable divide d into 3 categories, low, good and excessive. Weight gain was measured in pounds gained during pregnancy. To examine the effects of various levels of gain on a level 1 birth outcomes the weight gain was divided into categories as seen in table 3.3. Women with low weight gain comprised 15.7% of the low birth weight outcomes. The odds ratio for women of low we ight gain was 3.85. Excessi ve weight gain, by contrast, seemed to have a protective effect. The odds ratio for women with excessive weight gain during pregnancy was 0.54. Gender of Baby Of all low birth weight outcomes in the study sample 7% were female babies and 5% were male. The odds ratio for female babies be of low birth wei ght was 1.19. Thus female babies are at a sligh tly increased likelihood of be ing of a low birth weight. In addition to the individual level variable s the group level depriv ation index variable was included. Alternative models were cons tructed using the indivi dual level variables and one of the variables comprising the deprivat ion index. The definition of the variables used variables used the building of the multilevel model are explained in table 3.3.

PAGE 61

50Table 3.3 Explanation of Variables used in the construction of multilevel models. Variable Name Definition Birth Identifying code fo r each birth (level 1) Block_grp Identifying code for each block group (level 2) Gender Gender of baby. 1=Male; 2=Female LBW Low Birth Weight (Outcome) 0=birth weight >2500gm; 1=<2500gm Smoking Smoking status of the mother. 0=nonsmoker; 1=smoker Depind Deprivation Index calculated for each block group (addition of z-scores) Marital Marital Status of the mother 0=not married; 1=married AgeCat Categorical variable for age 1=12-17; 0=18-34; 2=35+ GainCat Categorical variab le for gain. 1=low birth (0-15 lbs gain); 0=(15-30 lbs gain) 2= excessive (30+ lbs) ParCat Categorical vari able for parity. 1= Primiparous; 0= 1 or more previous births Cons Constant vector Denom Denominator vector Pov Z-Score for percent individual living below the poverty line (level 2) Edu Z-Score for percent individuals with less than a high school education Veh Z-Score for percent of individuals without an automobile (level 2) Inc Z-Score for per captia income (level 2) Emp Z-Score for percen t individuals in the work-force that are unemployed Two-Level Random Intercept Logistic Regression Model The above variables were combined along with the deprivation index to fit a multi-level logistic regression model. The purpose of this was to allow for group level (block group

PAGE 62

51 or census tract) effects on the probability of a low birth weight outcome. The purpose of creating a two level model is to allow the effect of th e group level variables, the deprivation index or variables comprising the index, to vary across groups. The groups in this study are either census block groups or census tracts. The tables 4-17-4.72 outline the results from running the two level random coefficient model. A total of 55 models were run. The models run were for all r ecords using the 1990 depr ivation index, the 2000 deprivation index, the binary deprivation index (based on the 2000 deprivation index) as well as one model for each of the variables that went into making the deprivation index. After the full models were then run one for each of the three ethnic categories was run. The results of the model are reported as th e odds ratios for both the individual level variables and the group level va riables. The odds ratio is th e odds an exposed individual develops the outcome divided by the odds an unexposed individual develops the outcome. Thus any value greater than 1 sugge sts that an exposed i ndividual has greater odds of developing the outcome than an unexposed.

PAGE 63

52 CHAPTER FOUR RESULTS Results The population of study was all singleton live-bir ths in the state of Florida for the years 1992-1997. As a result the total number of records was equal to 1,030,443. Of these births 51.2% (527,916) were male babies and 48.8 % (502,527) were female. The number of infants born to unwed mothers was 663,865 while the number born to married mothers was 964,014. The average age of th e mother giving birth during the time period 1992-1997 was 26.131 years with the youngest moth er being 12 years old and the oldest being 54 years old. The average parity was 1.02 previous births. The lowest parity was zero previous births and th e highest was 22. Mothers on average gained 22.28 pounds during pregnancy the lowest gain was zero pounds and the highest gain was 98 pounds. The study population was divided into three ethnic categories, White Non-Hispanic, Black Non-Hispanic and Hispanic. White N on-Hispanic births comprised 58.5% of the study sample (603171 births) compared to 55.2% statewide for 1990-2000 and 64% nationwide. Black Non-Hispanics comprise d 23.8% of the sample population (244924

PAGE 64

53 births) compared to 23.2% statewide a nd 12% nationwide for the years 1990-2000. While Hispanic births were 17.7% of the sample (182348 births) compared to 22 % of births statewide and 17% nationwide for the years 1990-2000 (Grigg et al. 2000). Births to mothers who smoke were 13.7% of the sample population (141,769 births) compared to 21.9% statewide and 22.5% na tionwide for the years 1990-2000. It should be noted however, that 607 bi rths had no record of smoking history at all. Low birth weight births were 6.4% of the sample population (66,429 births) compared to 8% statewide and 7.6% nationa lly (Grigg et al..2000). A summary of the results of the multilevel models for the block group level models can be seen in Appendix A. Table A.1 shows an odds ratio of 1.08 for the 2000 deprivation index when adjusting for individual-level factors. The 2000 poverty z-score and the 2000 per capita income z-score s howed larger odds ratios. The 2000 poverty z-score had an odds ratio of 1.27 while the 2000 per capita income z-score had an odds ratio of 1.20. It should be noted that the results for models run for the 2000 variable s were significant at the =.01 level (p =.0132). Additionally the residual variance, the variance that can naturally be expected by movi ng from one block group to the next, was found to be less than 2%. Table A.2 additionally shows results for the 1990 models. While the results for these models were statistically significant a =.01. The p value was equal to 0.0185. The

PAGE 65

54 odds ratios were closer to 1 than the 2000 models. For comparison the 1990 deprivation index odds ratio was 1.08 while the 1990 poverty z-score was 1.02 and the 1990 per capita income z-score was 1.12. The results of table A.1 show that a mother living in an ar ea of high poverty, as defined by the 2000 census, has a 1.27 times increase d odds of having a low birth weight outcome. While a person living in an area with low per capita income z-scores has a 1.2 times increased odds of having a low birth weight outcome. Tables A.3-A.8 also shows the results of th e models stratified by ethnic category. For White Non-Hispanic mothers, there was a 1. 16 times increased odds of having a low birth weight baby living in a neighborhood with a high score on the 2000 deprivation index (odds ratio=1.16). Additionally, a 1.13 times increased odds of havi ng a low birth weight baby was found for White Non-Hispanic mo thers living in neighborhoods having a high value on the 2000 educational atta inment z-score. The residual variance for these models was less than 2% and the p value was less than .01. For Hispanic mothers the highest increase odds of having a low birth weight baby were found amongst those living in a neighborhood with a high 2000 deprivation index (odds ratio 1.12) and areas with high values for the 2000 no vehicle z-so re (odds ratio 1.12). Additionally mothers living in areas with hi gh values for unemployment had a 1.28 times increased odds of having a low birth weight baby.

PAGE 66

55 Black Non-Hispanic mothers only showed a 1.04 times increase in odds of having a low birth weight baby in areas with high 2000 deprivation inde x scores. The neighborhood level variable which showed the most influe nce on the odds of having a low birth weight outcome amongst Black Non-Hispanic mothers was linguistic isolati on. Mothers living in a neighborhood with high values for the 2000 linguistic isolati on z-score had a 1.12 times increase in odds of having a low birth weight baby. The results of the multilevel models for the tract level models can be seen in tables A.9 through A.16 found in appendix A. The re sults show the odds ratios for both the individual level variables and the group level variables. The results are similar to the results seen in the block group level results, however, the odds ratios for the tract level variables are closer to 1. Table A.9 shows an odds ratio of 1.18 for a mo ther living in an area with a high value for the 2000 poverty z-score. While a mother liv ing in an area with a high 2000 deprivation index value had an odds ratio of 1.07 after ad justing for individual le vel variables. The residual variance for the census tract level model was 3% and the results were significant at the =.01 level. The results for the ethnically stratified m odels, seen in tables A.11 though A.16 in Appendix A, for the census tract level, were also similar to the block group results with all odds ratios being closer to 1 as well. Wh ite non-Hispanic mothers living in areas with

PAGE 67

56 high scores on the 2000 deprivati on index had an odds ratio of 1.14. It is worth noting however, that mothers living in areas with low educational attainment had an odds ratio of 1.12. Black Non-Hispanic mothers had a 1.07 increased odds of having a low birth weight baby in areas with high scores on th e 2000 deprivation inde x. Moreover, Black Non-Hispanic mothers had a 1.12 increased odds of having a low birth weight baby in areas with high values for the 2000 linguistic isol ation z-score. Hispan ic mothers were at a 1.11 times increased odds of having a low bi rth weight baby in ar eas with high scores on the 2000 deprivation index. In addition, Hi spanic mothers living in areas with high values on the 2000 no vehicle z-score had a 1.11 times increased odds of having a low birth weight baby. Individual Level Results The multilevel model constructed for this research included individual level variables. The purpose for including indivi dual level variables was to adjust for the effects these variables have on the outcome and to examine the effect the explanatory variables had on the overall model. The inclusi on of individual level variable s is an important aspect of multilevel modeling. Individual level contro lling factors prevent ecological bias from occurring in the model. Additi onally they aid in the examina tion of the group level data. If group level variables were to have s how strong influence on the outcome, the individual level variables for that particular model would ha ve shown a large decrease in their influence as seen by a drop in the repor ted odds ratio. This did not occur in this research. In fact, the individual level variables remained relatively stable across all of the

PAGE 68

57 models, indicating a small effect of the group le vel variables. The i ndividual level results show that the variables that primarily infl uence on low birth weight outcomes area at the individual level and additiona lly, group level variables do not play a strong role in influencing low birth outcomes. There were a few minor exceptions, which are discussed in the following paragraphs. Smoking Smoking had the largest positive effect on the probability of a mother having a low birth weight baby. Odds ratios for the effect of smoking ranged from 2.25 in the White NonHispanic block group model to 2.08 in the Bl ack Non-Hispanic model (see appendix A). This is wholly in-line with other findings in the literature, see Ki ely et al. (1994) for a complete discussion. Generally, the influe nce of smoking on the outcome was modified slightly by the inclusion to group-level variab les, as is to be expected in a multilevel logistic model. However, the effect varied very little with the in clusion of each of the group-level variables. It is im portant to note that the strengt h of the effect of smoking in the models in this research suggest that it is the primary risk factor for low birth weight outcomes. Ethnicity For the non-ethnically stratified models a Bl ack Non-Hispanic ethn icity category had the second largest positive effect on the probability of a mother having a low birth weight baby. Odds ratios for Black Non-Hispanics ranged from 2.0 to 1.87 (appendix A) in the 2000 and 1990 block group full models. Kiel y et al (1994) repor t that Black Non-

PAGE 69

58 Hispanics have the highest rates of low birt h weight babies when compared to other ethnic categories. It is interesting to note that the strength of the influence of ethnicity varied little across the various group level variable mode ls with the exception of 1990 and 2000 poverty. The inclusion of this group level variable caused the largest decrease in the effect of ethnicity on the outcome. This suggests a diff erential effect of the group level variable for poverty across ethnic categor y. Consequently, this highlights the need to run ethnically stratified models to examin e the variation of indi vidual and group-level predictors across all ethnic categories. Parity After stratifying by ethnicity, parity category of the moth er was the individual-level variable with the third larges positive effect on the probability of a mother having a low birth weight baby, behi nd weight gain during pregnancy. This research divided parity categories into primiparous or multiparous. Primiparity was defined as never having previously giving birth. Multiparous mothers were defined as mothers having previously given birth to one or more babies. Parity ra nged from no previous bi rths to twenty-two. Like smoking, the influence of parity varied little for each of the group-level variables included in the models. The odds ratio for parity ranged from 1.77 amongst White NonHispanics to 1.41 amongst Bl ack Non-Hispanics in the 2000 census block group models (appendix A). It is important to note that parity did not vary with the use of different group-level variables. It di d however vary across ethnic categories with the smallest effect seen in Black Non-Hispanics.

PAGE 70

59 Gain Weight gain during pregnancy was previously determined to be an important influence on low birth weight outcomes (see li terature review for complete discussion). This research showed that low weight gain was highly in fluential on the outcome. Low weight gain had an odds ratio ranging from 1.93 in th e 2000 block group level models run for Black Non-Hispanic mothers. The group in which low gain had the smallest influence was Hispanic mothers. Amongst Hispanic mother s the contribution of low weight gain was an increased odds of 1.31 of having a low bi rth weight baby. Ga in was only slightly affected by the use of per capita income as a group level variable, slightly decreasing it influence. This suggests the potential for va riation across different income groups, which were not included in this research. Therefor e, it should be noted that the inclusion of individual-level income vari ables would further affect the influence of gain on the outcome. Excess gain status among all models contributed a protec tive influence to the overall likelihood of having a low birth weight baby. The odds ratios for excess gain ranged from 0.43 to 0.36. Excess gain was less protective among Black Non-Hispanic mothers. Marital status The marital status of the mother was a sm aller overall influen ce on the model. Among White Non-Hispanic mothers the influence of being in a not marred category was the strongest. The odds ratio was 1.38 for White Non-Hispanic mothers with little variation

PAGE 71

60 for each of the group level predictors. Am ong Black-Non Hispanic mothers the effect was less pronounced with the highest odds ratio being 1.22. Age Age of the mother showed a somewhat moderate influence on the overall model. A mother in the age category of 35+ had a m oderate influence on the model (odds ratio of 1.40-1.35. This influence was strongest am ongst Hispanic mothers and weakest amongst Black Non-Hispanic mothers. A mother younger than 18 years had a comparable influence on the overall model with odds ratios ranging from 1.4 to 1.38. The strongest influence was seen amongst Hispanic mothers and the weakest amongst Black NonHispanic mothers. Gender of Baby The gender of the baby had the least influence on the models. The odds ratio ranged from 1.18-1.11. The odds ratio of 1.18 was seen in the complete 2000 model. The smallest influence was seen in Hispanic mo thers (odds ratio =1.11). There was little variation in the influence of gender when compared across models run for each of the group level variables.

PAGE 72

61 CHAPTER FIVE DISCUSSION Discussion and Interpretation of Results: Summary of Findings A total of 66,429 low birth weight, singleton babies were born duri ng the study period of 1992-1998. Thus, singleton low birth weight births represent 6% of all singleton births in the State of Florida during this period. This is close to the nationally cited figure for all races of 6.9% of live births. For white nonHispanics the percent of low birth weight babies was 5% for the study period. Black Non-Hispanics low bi rth weight outcomes were 10.7% of live births. Additionally, among Hispanics in the sample 5.3% of live singleton births were low birth weight. The results of the multilevel model for depriv ed versus non-deprived block groups, after adjusting for individual-level factors, showed a small a ssociation betwee n living in a deprived neighborhood and low bi rth weight outcomes. What the models do not agree on is the measure of deprivation. Some mode ls showed stronger associations between low birth weight and depriv ation when deprivation was measured as a single variable, rather than the constructed deprivation index.

PAGE 73

62 Women living in a deprived neighborhood show ed an odds ratio 1.08. This corresponds to a 1.08 times increase of having a low birth weight outcome in a deprived neighborhood. However, four of the six vari ables that comprise the deprivation index showed a larger increased odds of a low bi rth weight outcome than the constructed deprivation index. The z-score measures of poverty, low educational attainment, unemployment and low per cap ita income all had odds ratios larger than the 2000 deprivation index (see Appendix A, Table A.1 for values). Among White Non-Hispanics the odds ratio increased to a 1.16 times increase. Among Black Non-Hispanics the odds increase 1.04 times and for Hispanic women the deprivation index had a 1.12 times increase in the odds of a low birth outcome. The differences seen across ethnic groups may be due to some variable not measured in this study. The individual variables that make up the de privation index were used as grouplevel fixed coefficients to compare the efficacy of using a pre-defined measure of deprivation versus a constructed measure. Of these va riables the z-score for per capita income from the 2000 census showed the stronge st association with low birt h weight outcomes in the overall model. This is consistent with the findings of the literature (Pickett and Pearl 2001; Subramanian and Kawachi 2004). The fact that measures of economic deprivation show the strongest relationship to low birth weight births is not surprising particularly because Subramanian and Kawachi (2004) report that income is a strong determinant of health at both the individual level and the aggregate level.

PAGE 74

63 The association between group variables wa s much weaker among Black non-Hispanics than among White Non-Hispanics and Hispanics. The odds ratio for the z-score of per capita income was 1.07. This may hint at ot her factors at play in the determination of health outcomes. Residential segregation, ac cess to health services stress at both the individual level and neighborhood and perceived socioeconomic status may also play an important role. This may be reflected in th e decreased efficacy of any of the group level variables in predicting low birt h outcomes. It has been wi dely reported that Black NonHispanics report lower perceived socioec onomic status and e xperience residential segregation at higher rates th an White Non-Hispanics and Hi spanics (Brown et al. 2004; Krieger and Smith 2004). Major Findings The strength and nature of the relationship between materi al deprivation and low birth weight outcomes in the State of Florida varies among different ethnic groups and with the use of different indicators. The strongest associations are found when per capita income or percent living below poverty z -scores are used as group-level variables. However, these associations are still very small. All group levelvariables showed a stronger relationship to low birth we ight outcomes among White N on-Hispanic residents of Florida. Conversely, the same variables show ed almost no relationship to the outcome in Black Non-Hispanic residents. The cons tructed deprivation i ndex showed a small association with low birth we ight outcomes among all Flor ida residents (odds = 1.08) with per capita income showi ng the strongest association.

PAGE 75

64 The findings do not show the same degree of association between deprivation and low birth weight as do the published results in the current literature. In California Pearl et al. (2001) showed no neighborhood association wi th birth weight among whites. They did however find an association between nei ghborhood socioeconomic status and birth weight among Black and Asian residents of California. This study found strongest associations between neighbor hood indicators of material de privation and birth weight and White Non-Hispanic residents of Fl orida when stratifying for race. This inconsistency with findings in the literature may be due to regional confounding and as such, should be researched further. Application to Theory The results of this research unfortunately do not add a clear answer to existing theory on neighborhood deprivation and birt h weight. Rather, it high lights the need for further research into the developmen t of useful metrics of so cioeconomic deprivation in neighborhoods. This research does show a general trend, in which neighborhood or area level measures of deprivation show some association with negative health outcomes. More research needs to be conducted on how best to measure material deprivation and social deprivation and how best to apply this to health outcomes. The multilevel analysis provided a unique framework to examine the role of deprivation and health outcomes, especially low birth weight.

PAGE 76

65Revision to Theory Currently there exists no comprehensive or coherent theo ry regarding the role of group or neighborhood level socioeconomic deprivation. Rather, a milieu of ideas as to potential pathways, and effects of social and material influence on hea lth. This study adds another voice to the research calling for more investigation into the role of contextual effects on health outcomes. Hills Criteria for a Causal Relationship Between Neighborhood Deprivation and Low Birth Weight Outcomes In order to understand the causal relationship between ne ighborhood deprivation and low birth weight outcomes it may be useful to ex amine the results from the multilevel models and determine whether they meet all of Hills criteria of causation. Hills Criteria of Causation is a set of minimum epidemiologic conditions that must be met to establish a causal relationship between an exposure and an outcome. The criteria are temporal relationship, consistency, strength, speci ficity, dose-response re lationship, biologic plausibility coherence and experiment. If all these criteria are met then it can be assumed that there exists a causal relationship between an exposure and an outcome (Hill 1965). Temporal Relationship The first criterion that must be met to es tablish a causal relationship between material deprivation and low birth weight is that of a temporal relationship. For this to be met

PAGE 77

66 material deprivation must precede low birth weight. The aforemen tioned relationship has been the subject of much debate in the li terature. It is comm only agreed upon and assumed that high rates of material depriva tion do, in fact lead to negative health outcomes. The meta-analysis conducted by Pickett and Pearl (2001) provides good examples of this, as does S ubramanian and Kawachi (2004). However, placing socioeconomic status in the causal pathway can be problematic. Subramanian and Kawachi (2004) note the need for longitudinal studies to more properly address this issue. Most of the studies conducted in th is area have been crosssectional/ecological. There are a myriad of confounding factors that may influence low birth weight (and negativ e health outcomes in general). In dividual level factors that were not measured here or in other studies should be considered. In particular, medical history of the mother should be taken into account. Ho wever, due to the private nature of such information it is often difficult to obtain accu rate accounts of previ ous medical history. Additionally, toxic exposures ma y also be linked to low birt h weight. Data regarding doses and duration of toxic exposures are diffi cult if not impossible to determine (Kiely et al. 1994). Strength The second criterion to be met in order to es tablish a causal relationship between material deprivation and low birth weight is the strength of the associ ation. This was not met in this study. In all models there was little, if any, statistically significant correlation between economic deprivation (as measured by the deprivation i ndex) and low birth

PAGE 78

67 weight outcomes. Other re searchers have had mixed re sults. Some have found a significant relationship while others have not (Picket and Pearl 2001; Subramanian and Kawachi 2004). This may be a function of th e type of index being used, rather than the actual strength of associa tion. However, until more rigorous methodologies can be developed to construct deprivation indices in th e United States this problem will persist. Specificity The relationship between material deprivation and low birth weight is not specific, material deprivation may have a number of other harmful effects other than low birth weight. Furthermore, there are a myriad of ot her risk factors for lo w birth weight that were not included in this research (see ch apter three and the literature review for discussion of risk factors). Dose-Response Relationship As of yet, there has been no st udy, this one included, which ca n show an increased risk of low birth weight outcomes with an increase in material deprivation. This too may be due to the inability of the deprivation inde x to properly measure actual neighborhood deprivation. Additionally the lack of any statistically si gnificant relationship between low birth weight variation across neighborhoods precluded any type of dose-response relationship. Coherence The findings of this study are fully compatible with existing epidemiologic theory and knowledge. This study found only small gr oup-level effects on low birth weight

PAGE 79

68 outcomes. Individual-level outcomes explaine d all but a small amount of variance seen in the sample and are most commonly reporte d as having the stronge st association with birth weight. However, as multilevel analysis is a fairly new experimental design in geography and epidemiology little consensus exists in the l iterature as to the effects of group level measures of deprivation on low birth weight. The findings here seem to be in line with some (Pearl et al. 2001; Reij nveld 2001) and at odds with others (Subramanian and Kawachi 2004; Pearl et al. 2001). Biologic Plausibility The research into the pathways in which material deprivation may operate to cause a low birth weight outcome is still very new. Most of the potential pathways have been examined at the individual level. The exte nt to which group-level material deprivation may influence any health outcome, low birt h weight included, has just begun to be examined. There are three proposed pathways in which material deprivation may be linked to negative health outcomes. These are outlined in Subramanian and Kawachi (2004). The first is the so-called structural pathway In this proposed pathway material deprivation could lead to an increase in residential segregation, which in turn, could cause a concentration of povert y and ethnic groups in spatia lly isolated areas. The second pathway is the social cohesion and collec tive social pathway. This pathway uses the concept of social capital, which can be de fined as collective value of social networks (Putnam 2000). In this pathway the presence or absence of collective social pathways,

PAGE 80

69 commonly referred to as social capital, infl uence the behaviors and social environment leading to differences in health outcomes. Th e third pathway is referred to as the policy pathway. In this pathway the implementation of health-related and social policies may further exacerbate the adverse effects of materi al deprivation. Thes e pathways may work together or independently to influence material deprivation and health outcome (Subramanian and Kawachi 2004). Because of the newly formed hypotheses regarding socioeconomic influences on health outcomes, it may be too early to tell if a biologically plausible path way exists. Experiment (Experimental Modification) The study conducted here can, and should be modi fied to attempt to measure the effects material deprivation has on lo w birth weight outcomes. This was an ecologic study aimed at determining if variation in low birth weight in the State of Flor ida could be explained using group-level measures of material depriv ation. An alternat e study design would be to conduct a long-term longitudinal study. Mu ltilevel analysis will allow for the nesting of individuals within groups within different periods of tim e. Additionally, opportunities exist for researchers to study the effects chan ge in socioeconomic status has on health outcomes. For example, an area with a large number of recent lay-offs or areas where certain social programs have seen funding decreases would all se rve as a good starting point in experimentally understanding the role socioeconomic deprivation has on a community. An alternative potential modifica tion of this study would be to adjust for regional confounding by adding another level (3 le vel model) to the current model. The structure of this three level model would be individuals nested within census divisions,

PAGE 81

70 nested within regions. Such studies have been conducted on overall health assessment but none have been conducted on birth weight outcomes (Subramanian and Kawachi 2004). Also, the exploration of alternativ e indices of depriva tion and residential segregation may further shed light on th e group-level effects on birth weight.

PAGE 82

71 CHAPTER SIX CONCLUSIONS Consideration of Alternative Explanations There are numerous alternative explanations to the findings of this study that there is only a minor group level influence of deprivation on low birth weight. The first may be that there is actually a difference, however, multilevel models as they exist now, or were executed in this study are not yet sufficien t in explaining group-l evel variation. Additionally, some limitations of the study may have also contribute d to the results (see limitations). The effects of income and de privation on Florida residents seem to contradict some of the findings in the multileve l literature (Pearl et al. 2001). This quite possibly could be due to regiona l differences in population, residential segregation, or other types of regional confounding (more ru ral vs. urban areas, for example). The results here show a moderate effect of gr oup-level variables on birt h weight. However, this study did not adjust for indi vidual income or medical risk s. It is possible that after adjusting for these factors the group level influences may altogether disappear. Consistency with Literature The findings somewhat contrast with publishe d results in the literature. In California Pearl et al. (2001) showed no neighborhood as sociation with birth weight among whites. They did however find an association betw een neighborhood socioeconomic status and

PAGE 83

72 birth weight among Black and Asian residents of Californi a. In this study the strongest association was amongst white residents. This difference may be due to regional confounding. Which is to say that there may be clusters of states within larger regions that are more similar in the distribution a nd effects of contextual variables on health outcomes than other regions. A potential modi fication of this study would be to adjust for regional confounding by adding a nother level (3 level model) to the current model. The structure of this three level model w ould be individuals nested within census divisions, nested within regions. Such st udies have been conduc ted on overall health assessment but none have been conducted on birth weight outcomes (Subramanian and Kawachi 2004). The findings of this study are c onsistent with some of the findings in the literature. There have been mixed results in the use of multilevel modeling of health outcomes in which an index or measure of deprivation is used. See Discussion Chapter 5 and Subramanian and Kawachi (2004) for a complete discussion of income related deprivation and health outcomes. Most studies that have been able to demonstrate an in fluence of group-level deprivation have used State, County or MSA for the group. Smaller scale studies, with the exception of Diez-Roux et al 2001, have had similar re sults. Moreov er, Reijnevled (2001) has proposed that some, if not all, of the studies that have found a significant association between socioeconomic environmen t and health outcomes have incompletely adjusted for individual level socioeconomic factors, whic h could be said about this study. Additionally, of those studies that have shown a pos itive association only show

PAGE 84

73 modest effects on health outcomes and, due to issues of statistical power, are at risk of overestimation of the effects. (Reijnevled, 2001). Limitations The lack of utility of the Deprivation index in predicting geographic variation in low birth weight outcomes may be the resu lt of any of several factors. The deprivation index may have different utility in rural versus urban areas Jordan et al. (2004) report differences in the predictive utility of multiple deprivation in dices in urban versus rural areas. This was not examined in this study. Perhaps the inclusion of a group level variable indicating whether the block group is in a rural area or an urban area. The inability of the individual factors comprising the deprivati on index to predict low birth weight variability may also lend credence to this suggestion. Additionally, the boundaries constr ucted for the purpose of this study were taken directly from the United States Census Bureau. These boundaries are by no means a complete representation of an individuals social and or economic environment. Carefully conceptualizing the idea of social environment, neighbor hood and social capital may provide a better approximation of an individuals social and economic group membership. Such a study was beyond the scope of this research. Moreover, the inclusion of a measure of residential segrega tion may also provide some utility in this area. Perhaps a model of individual-leve l variables and a group level measure of residential segregation or a combination of residential segregation and material/social deprivation may be of more use.

PAGE 85

74 The deprivation index itself may be flawed. The index is part of the social indicators used by the National Association of Planni ng Councils. It was de veloped by Andrulius (2004) and was designed for the MSA level, not the census block group level. This may explain some of the lack of explanatory power of the index. Additionally, crime rates, considered a source of maternal stress by ma ny researchers could not be included in this index while it was in. The reason for the lack of inclusion was that crime data in the US is only available at the MSA level for some areas. Additionally, the factors comprising the deprivation index were measured for 1990 and 2000. The sample was for 1992-1998. Measurements taken for each year would be more accurate representations of the conditions at the time of each birth. Recent literature suggest s that an individuals perceived deprivation may be more important than statistical measures of depriv ation. Studies in wh ich individuals self report negative health outcomes and perceive d social and material deprivation those individuals with a higher fre quency of reported negative he alth outcomes also have a perception of more economic and social deprivation (Bro wn et al.. 2004; Subramanian and Kawachi 2004; Picket and Pearl 2001). Another limitation is a difference in the m odeling or modeling errors. Many models in the literature use different me thods of statistical analysis. For example most studies use marginal models (Subramanian and Kawachi 2004). Marginal models ignore the variance structure when estimating the fixed effect of exposure. This is e rroneous in that the models are specifically ignor ing variability information, which is the purpose of

PAGE 86

75 multilevel modeling. This study was not a marginal model and this may account for differences between the results found here a nd in the literature. Another alternate explanation is that the rando m coefficients associated w ith the area (block groups and census tracts) were not able to be included. This is a common error in multilevel analysis due to software and intensive computation. The random coefficient of interest in this study is the extent of uncondi tional variation of block group -attributable (or census tract) low birth weight outcomes. This will have an e ffect on the overall variance of the model. Additionally, this study utilized MLwiN, a st atistical package desi gned exclusively for developing multilevel models. The software is relatively new and different to existing software packages. Therefore, there is littl e to compare it to as far as ease of use and accuracy of results. The size of the data se t seemed to slow the program and render some of its functions unusable. While this did not actually preclude the statistical assessment of the sample, it put time constraints on those that were run. The large sample size and hierarchical nature of the data created lim itations as to what could be modeled. For example, categorical variables with more th an 3 categories could not be included in the model due to the large size of the data. This was an issue as several of the categorical variables initially considered for the model could not be included and as such were converted to continuous variables. Little in the literature exists regarding multilevel logistic regression analysis. Therefore, more sophisticated models may be necessary to truly model the relationship between the outcome and the variables. Lastly, this model was not able to adjust for individual level income, or medical-risks predating and during pregnancy. This surely would have an effect on the outcomes of

PAGE 87

76 these models. Specifically, with the inclusion of these variab les the effect of group level variables would decrease. Strengths The main strength of this study lies in the f act that it treats low birth weight as an outcome that varies spatially. While the ability of the study to detect significant spatial variation was unsuccessful, the methodology di d not ignore the geographic component of the outcome and at the same time it took into account the individua l level factors that influence the outcome. Diez-Roux (2000) finds that a multilevel methodology is best suited to deal with geographic variability while controlling for individual level factors. Additionally multilevel models are particular ly well suited for avoiding the ecological fallacy common in studies of geographic variab ility. Another streng th of this study was the sample size. Rarely is such a large sample size available to a researcher. With the addition of a more accurate deprivation index the full potential of this sample can be realized. Geographic/Public Health Implications The geographic implications of the findings of this research are the need for geography to adequately address issues of group (n eighborhood) boundaries and membership. Geography is uniquely positioned to examine the boundaries of an individuals physical neighborhood as well as their social group. Th e contextual nature of multilevel analysis makes it a good quantitative methodology to comp lement the qualitative research of contextual effects. Through the examination of the flow of social capital and what makes

PAGE 88

77 a neighborhood, geography can add a much-needed theoretical component to multilevel modeling. Additionally, spatial examination of geographically-attributable variance in health outcomes is the key to a successful multilevel analysis of health and place and as such almost requires a geographic perspective. The public health implications of this research are that more attention should be paid to group level influences and geographic variance of health outcomes. This research has shown some mild effects of context on birt h weight. The differences between this study and others definitely call for more inquiry in to contextual health effects. However, the traditionally individualistic nature of epid emiology should not be fully abandoned. As illustrated by this research, individual-level variables still comprise the majority of negative health outcome risk factors. Moreover, this study has further bolstered th e findings of numerous other public health researchers regarding low birt h weight outcomes. The i ndividual factors of smoking, Black Non-Hispanic ethnicity, primiparity a nd age of the mother were shown to have significant associations with lo w birth weight outcomes that are in agreement with the majority of published literature. This research partially suppor ts the hypothesis that, after ad justing for individual level variables, the odds of having a low birth weight child are high er for a mother living in a neighborhood with a high deprivation index sc ore than a mother living in a neighborhood with a low deprivation index score. The increased odds we re not particularly strong

PAGE 89

78 however, the findings of this research are in the same direction as others in the literature. Additionally, this research was unable to determine if a significant portion of the spatial variation of low birth weight outcomes in the State of Florida is due to neighborhood effects. This research has supported some of the litera ture, particularly con cerning the effect of neighborhood per capita income on birth weight. However, it is partially contradictory to some. Pearl et al. (2001) found a signi ficant relationship between neighborhood socioeconomic variables and birth weight in among Black Non-Hispanics in California. This study of Florida residents found no su ch relationship, in fact, the strongest association between neighborhood-level indica tors and birth weight was among White Non-Hispanics. This suggests more research into the variability of socioeconomic indicators is necessary. Moreover, research into differences in community structures, definitions of communities and social group in different geographic regions will also prove helpful in understandi ng the influence of socio economic context on health outcomes.

PAGE 90

79 REFERENCES Ahern, J., Pickett, K.E., Selvin, S., Abra ms, B., 2003 Preterm birth among African American and white women: a multilevel anal ysis of socioeconomic characteristics and cigarette smoking. J Epidemiol Community Health 57:606 American Academy of Pediatrics, American Co llege of Obstetricians and Gynecologists. 1988. Standard terminology for reporting of repr oductive health statistics in the United States. Public Health Rep 103:464. Anderson, R. T., P. Sorlie, E. Backlund, N. Johnson, and G. A. Kaplan. 1997. Mortality Effects of Community Socioeconomic Status. Epidemiology 8:42. Andrulius, D.P. Reid, H.M., and Duchon, L. 2004. Quality of life in the nations 100 largest cities and their suburbs : New and continuing challenges for improving health and well being. SUNY Downstate Medical Center. Baker D., W., Parker, R.M., Williams, M, V. et al 1998. Health literacy and the risk of hospital admission. J Gen Intern Med. 13:791-798 Backlund, E., Sorlie, P.D., Johnson N.J., 1996. The shape of the relationship between income and mortality in the United States: evidence from the National Longitudinal Mortality Study. Ann Epidemol. 6:12-20 Barker, D.J.P., 1992. Fetal and infant origins of adult onset diseases. British Medical Journal p341 Berendes H.W., Forman MR. Delayed childbear ing: trends and consequences. In: Kiely M, ed. Reproductive and perina tal epidemiology. Boca Raton, Florida: CRC Press, 1991. Brown, A.F., Ettner, S., Piette, J. Weinberg er, M., Gregg, E., et al.. 2004. Socioeconomic position and health among pers ons with Diabetes Mellitus : Conceptual framework and review of the literature. Epidemiologic Reviews 26: 63-77 Bostock L. 2001. Pathways of disadvantage? Walking as a mode of transport among lowincome mothers. Health Soc Care Community 9:11-18.

PAGE 91

80 Carstaris, V., 2000. in Elliott, P., Wakefiled, J., Best, N., and Briggs, D. Eds. 2001. Spatial Epidemiology. Oxford University Press. Cassel, J. 1964. Social sc ience theory as a source of hypotheses in epidemiologic research. Am J Public Health 54: 1482-1488 Couregeau, Daniel in: Courgeau, Daniel (ed). 2003 Me thodology and Epistemology of Multilevel Analysis: Approaches from Different Social Sciences Dordrecht/Boston/London; Kluw er Academic Publishers. Diez Roux, A.V. 2004 The Study of Group-Leve l Factors in Epidemiology: Rethinking Variables, Study Designs, a nd Analytical Approaches Epidemiol Rev 26: 104-111; doi:10.1093/epirev/mxh006 Diez-Roux, A.V., 2003 Potentialities and limita tions of multilevel analysis in public health and epidemiology. In: Courgeau, Daniel (ed). 2003 Methodology and Epistemology of Multilevel Analysis: A pproaches from Different Social Sciences Dordrecht/Boston/London; Kluw er Academic Publishers. Diez-Roux, A.V., Merkin, S.S., Arnett, D ., et al. 2001 Neighborhood of residence and incidence of coronary heart disease. New England Journal of Medicine 345, 99-106. Diez-Roux, A. 2001. Investigating ne ighborhood and area effects on health. Am J Public Health 19 (11):1783-1789 Diez-Roux, A, 2000: Multilevel Analys is in Public Health Research. Ann. Rev. Public Health. 21: 171-92 Diez-Roux 1998 American Jour nal of Public Health Bringing Context back into epidemiology: va riables and fallacies in multilevel analysis Vol 88, Issue2 2-216-222 Diez Roux, A. V., F. J. Nieto, C. Muntaner, H. A. Tyroler, G. W. Comstock, E. Shahar, L. S. Cooper, R. L. Watson, and M. Szklo. 1997. Neighborhood Environments and Coronary Heart Disease: A Multilevel Analysis. American Journal of Epidemiology 146(1):48. Duncan, C., Jones, K., and Moon, G., 1996: Health-related behavi or in context: A multilevel modeling approach. Soc. Sci. Med. 42(6); 817-830.

PAGE 92

81 Dovey, K. 1985. An ecology of place and pl acemaking. Structures process, knots of meaning. In Dovey, K., Downton, P., and Missingham, G., eds Place and Placemaking, Proceedings of the PAPER 85 Conference. Melbourne, 19-22 June 1985. Faculty of Architecture and Building. Royal Melbour ne Institute of Technology. Melbourne Australia. Epstein, A.M., Taylor, W., C., Seage, G., R ., 1985. Effects of patients socioeconomic status and physicians tr aining and practice on patie nt-doctor communication. Am J Med. 7:101-106. Fiscella, K., Goodwin. M., A., Stange, K.,C. 2002. Does pa tient educational level affect office visits to family physicians? J Natl Medical Assoc. 94:157-165. Gazamararian, J., A., Parker, R., M., Willia ms. M.,W., et al. 1999 Reading Skills and family planning knowledge and practices in a low-income managed-care population. Obstet Gynecol. 93:239-244 Geronimus AT, Korenman S. 1993. Maternal yo uth or family background? On the health disadvantages of infants with teenag e mothers. Am J Epidemiol. 137:213. Giddens, A. 1984. The Constitution of Society: Outline of the Theory of Structuration Berkeley, University of California Press. Goldstein H., Rasbash J., 1996. Improved approximations for multilevel models with binary responses J R Stat Soc 159: 505. Greenland, S., 2001: Ecologic ve rsus individual-level sour ces of bias in ecologic estimates of contextual health effects Int. J. Epidemiol 30: 1343-1350 Haan M., Kaplan G.A., Camacho, T., 1987. P overty and health: perspectives evidence from the Alameda County Study. Am J of Epidemiol. 125:989-998 Hack, M., Taylor, G., Klein, N., et al.. 1994 School-age outcomes in children with birth weights under 750 g. New England Journal of Medicine 331:753-59. Harding, J.E. 2001. The nutritional basis of the fetal origins of adult disease. Int J Epidemiology 30(1) 15-23. Hawe, P., 1998. Making sense of context-le vel influences on h ealth. Editorial in Am J Epidemiol. 146:1-4. Hox, J.J., 1995 Applied Multilevel Analysis TT-Publikaties-Amsterdam. Institute of Medicine, Comm ittee to Study the Prevention of Low Birth Weight. 1985. Preventing low birth weight Washington, DC: Na tional Academy Press

PAGE 93

82 Jarman, B. 1984. Underprivilege d areas: validation and distribution of scores. Brit Med J 289: 1587-1592. Jones, K. and Duncan, C. 1998. Modeling co ntext and heterogeneity : Applying multilevel models. In E. Scarbrough and E. Tanenbaum (eds.), Research Strategies in the Social Sciences Oxford University Press Krieger, N., D. L. Rowley, A. A. Herman, B. Avery, and M. T. Phillips. 1993. Racism, Sexism and Social Class: Implications for Studies of Health, Disease, and Well-Being. American Journal of Preventive Medicine 9(suppl 2):82. Kessel SS, Kleinman JC. Koontz AM, Hogue CJ R, Berendes HW. Racial differences in pregnancy outcomes. Clin Perinatol 1988;15:745. Kiely, J.L., Yu, S., Rowley, D.L., 1994. Lo w birth weight and intrauterine growth retardation. In: From Data to Action; CDCs Public Health Surveillance for Women, Infants and Children. Atlanta. Kline J, Stein Z, Susser M. Conception to bi rth: epidemiology of prenatal development. New York: Oxford University Press, 1989. Kleinman, J., and Kessel, S. 1987.Racial differences in low birth weight. New England Journal of Medicine 317:749-53. Kramer MS. Determinants of low birth we ight: methodological assessment and metaanalysis. Bull World Hea lth Organ 1987;65:663. Kreiger, N., Williams, D.R., Moss, N.E. 1997 M easuring social class in US public health research: concepts, met hodologies, and guidelines, Annu Rev Public Health 18:341-378. Kuh D, and Ben-Shlomo, Y. 1997., A Life course approach to chronic disease epidemiology. New York: Oxford University Press p.317 Kuhn, T., 1962 The Structure of Scientific Revolutions. University of Chicago Press.pp 211 Larzfeld, P.F. and Menzel, H., 1961. On the relation between individual and collective properties. In: A. Etzioni (ed.) Complex Organizations: A Sociological Reader. New York: Holt, Reinhardt & Winston. LeClere, F. B., R. G. Roge rs, and K. D. Peters. 1998. Neighborhood Social Context and Racial Differences in Womens Heart Disease Mortality. Journal of Health Social Behavior 39:91

PAGE 94

83 Lynch, J. W., and G. A. Kaplan 2000. Socioeconomic Position. In Social Epidemiology edited by L. F. Berkman and I. Kawachi, 13. New York: Oxford University Press Lynch, J.W., Kaplan, G.A., Shema, S.J. 1997 Cumulative impact of sustained economic hardship on physical, cognitive, psychological and social functioning. N Engl J Med 1997; 333:1889-1895 Marmot, M. 2002. The influence of income on health: views of an epidemiologist. Does money really matter? Or is it a marker for something else? Health Aff ( Millwood) 21:31. Macintyre, S., Maciver, S., and Sooman, A ., 1993. Area class and health: should we be focusing on places or people? Journal of Social Policy 22:213-234. Melnikow, J., Paliescheskey, M., Stewart, G., K. 1997. Effect of a transportation incentive on compliance with the first pr enatal appointment: a randomized trial. Obstet Gynecol. 89:1023-1027. Morgan M, Chinn S. 1983. ACORN group, social class, and child health. J Epidemiol Community Health;37:196. NCHS. Vital statistics of the United States, 1988 Vol. I, natality. Hyattsville, Maryland: US Department of Health and Human Services, Public Health Service, CDC, 1990. NCHS. Vital statistics of the United States, 2001 Vol. I, natality. Hyattsville, Maryland: US Department of Health and Human Services, Public Health Service, CDC, 2002. OCampo, P., X. Xue, M.-C. Wang, a nd M. O. Caughy. 1997. Neighborhood Risk Factors for Low Birth we ight in Baltimore: A Multilevel Analysis. American Journal of Public Health 87(7):1113. ODonnell, C.R. ,Tharp, R., G., and Wilson, K., 1993. Activity settings as the unity of analysis a theoretical basis for comm unity intervention and development. American Journal of Public Health. 86: 678-683 Pearl, M., Braveman, P., and Abrams, B. 2001 The relationship of neighborhood socioeconomic characteristics to birthw eight among 5 ethnic groups in California. Am J Public Health 19 (11) 1808-1814

PAGE 95

84 Pickett, K., E. and Pearl, M. 2001 Multile vel analyses of neighbourhood socioeconomic context and health outcomes: a critical review. J Epidemiol Community Health. 55:111122 Putnam R., D., 2000. Bowling Alone: The Collapse and R evival of American Community New York: Simon & Schuster. Raudenbush, Stephen W. and Sampson, Robert J. (1999). Ecometrics: Toward a Science of Assessing Ecological Settings, with Application to the Systematic Social Observation of Neighborhoods. Sociological Methodology 29 1-41. Raugh., V.A., Andrews, H.F., and Garfinkel, R.S., 2001. The contribution of maternal age to racial disparities in birthweitght: A multilevel perspective. Am J Public Health 19 (11) 1815-1823 Ravelli, G.P., Stein, Z.A., Susser, M.W. 1976. Obesity in young men after famine exposure in utero and early infancy. New England Journal of Medicine 295: 349-353 Reader, S. 2001. Detecting and analyzing clus ters of low-birth weight incidence using exploratory spatial data analysis. Geo Journal 53: 149-159. Reijneveled, S., A., 2001.Explanations for differences in health outcomes between neighborhoods of varying socioeconomic level. J Epidemiol Community Health. 55:847-848 Reijneveled, S., A., 1998. The impact of i ndividual and area char acteristics on urban socioeconomic differences in health and smoking. Int J Epidemiol. 27:33-40 Rice, N., 2001 Binomial Regression in Leyl and, A.H., and Goldstein, H., (eds) Multilevel Modeling of Health Statistics John Wiley and Sons, Ltd New York. Roberts, E. M. 1997. Neighbor hood Social Environments and the Distribution of Low Birthweight in Chicago. American Journal of Public Health 87(4):5970150603 Schur, C., L., Albers, L., A., 1996, Language, sociodemographics and health care us of Hispanic adults. J Health Care Poor Underserved 7:140-158 Shiono, P.H., Klebanoff, M.A., Graubard, B ., et al.. 1986. Birth we ight among women of different ethnic groups. Journal of the American Medical Association 255:48-52.

PAGE 96

85 Shiono, P.H., Klebanoff, M.A ., Nugent, R.P., et al.. 1995. The impact of cocaine and marijuana use on low birth weight a nd preterm birth: A multicenter study. American Journal of Obstetrics and Gynecology. 172:19-27. Smith G., D., Hart, C., Blane, D., et al 1998. Adverse socioeconomic conditions in childhood and cause specific adult mortality: perspective observational study. BMJ. 316:1631-1635. Stein ZA, Susser M. Intrauterine growth reta rdation: Epidemiological issues and public health significance. Semin Perinatol 1984; 8:5. Subramanian S.V., Kawachi, I. 2004. Income inequality and Health: What have we learned so far? Epidemiologic Reviews 26: 78-91 Swanborn, P.G., 1981. Methods of Social Research. Amsterdam/Meppel: Boom. Szklo, M., and Nieto, F. J., 2000. Epidemiology: Beyond the Basics. Aspen Publishers, Gaithersburg. Md. Teaching Resources and Materials for So cial Scientists (TRAMSS webpage) 1999 http://tramss.data-archive.au.uk Takano, T., Nakamura, K., 2001. An analysis of health levels and va rious indicators of urban environments for Healthy Cities projects. J Epidemiol Comm Health. 55:263-270. Terry, M.B., and Susser, E., 2001. Commentary: The impact of fetal and infant exposures along the life course Int J Epidemiology 30 (1):95-96. Townsend, P. Phillimore, P. and Beattie, A. (1988). Health and deprivation: inequality and the North. London Croom Helm. Vassen, N., Joop, A.J., Heutinik, P., Hofman, A., Lamberts, S.W.J., Oostra, B.A., Pols, H.A.P., van Duijn, C.M. 2002. Association be tween genetic variation in the gene for insulin-like growth facto r-I and low birth weight. Lancet 359: 1036-1037 Wilcox A, Russell I. Why small black infants have a lower mortality rate than small white infants: the case for populationspecific standards for birth weight J Pediatr 1990;116:7. Williamson, D., Fast, J., E., 1989 Poverty a nd medical treatment: when public policy compromises accessibility. Can J. Public Health. 89:120-124. Williams, M., V., Baker, D., W., Parker, R ., M., et al. 1998. Relationship of functional health literacy to patients knowledge of their chronic dis ease: a study of patients with hypertension and diabetes. Arch Intern Med. 158:166-172

PAGE 97

86 Ylpp A. 1919 Zur physiologie, klinik, zum schicksal der frhgeborenen. Zeitschrift fr kinderheilkunde 24:1.

PAGE 98

87Appendices

PAGE 99

Appendix A: Multilevel Model Results Tables Table A.1 Results of Complete Multilevel Model For Year 2 000 Census Block Groups Showing Odds Ratios For Individual and Group Level Variables Full Model Deprivation Index Poverty Vehicle Ownership Educational Attainment Per Capita Income Unemployment Linguistic Isolation Variables Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Smoking 2.222.172.222.182.162.182.22 Not Married 1.291.291.401.291.171.291.29 Female 1.181.181.171.181.181.181.18 Black Non-Hispanic 2.021.872.022.021.871.912.02 Hispanic 1.171.161.171.161.161.161.16 Younger than 18 yrs 1.131.131.121.131.131.131.13 Older than 34 yrs 1.381.381.381.381.381.381.38 Primiparous 1.521.521.521.521.521.521.52 Low Gain 1.271.271.271.271.271.271.27 Excess Gain 0.360.380.360.380.380.360.39 Group Variable 2000 Deprivation Index 1.08*** *** *** *** *** *** 2000 Poverty *** 1.21*** *** *** *** *** 2000 No Vehicle *** *** 1.07*** *** *** *** 2000 Low Educational Attainment *** *** *** 1.15*** *** *** 2000 Per Capita Income *** *** *** *** 1.27*** *** 2000 Unemployment *** *** *** *** *** 1.11*** 2000 Linguistic Isolation *** *** *** *** *** *** 1.04

PAGE 100

Appendix A (Continued) Table A.2 Results of Complete Multilevel Model For Year 1 990 Census Block Groups Showing Odds Ratios For Individual and Group Level Variables. Full Model Deprivation Index Poverty Vehicle Ownership Educational Attainment Per Capita Income Unemployment Linguistic Isolation Variables Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Smoking 2.172.202.222.172.162.172.22 Not Married 1.281.291.301.521.291.291.29 Female 1.171.161.171.111.181.171.18 Black Non-Hispanic 1.921.851.962.022.021.922.02 Hispanic 1.131.121.131.161.161.131.16 Younger than 18 yrs 1.131.121.131.131.131.131.13 Older than 34 yrs 1.381.361.391.381.381.381.38 Primiparous 1.601.611.611.521.521.601.52 Low Gain 1.261.261.261.271.271.261.27 Excess Gain 0.380.390.380.380.380.380.38 Group Variable 1990 Deprivation Index 1.06*** *** *** *** *** *** 1990 Poverty *** 1.05*** *** *** *** *** 1990 No Vehicle *** *** 1.03*** *** *** *** 1990 Low Educational Attainment *** *** *** 1.08*** *** *** 1990 Per Capita Income *** *** *** *** 1.13*** *** 1990 Unemployment *** *** *** *** *** 1.05*** 1990 Linguistic Isolation *** *** *** *** *** *** 1.02

PAGE 101

Appendix A (Continued) Table A.3 Results of Multilevel Model For White Non-Hispa nic Mothers Year 2000 Census Block Groups Showing Odds Ratios For Individual and Group Level Variables Deprivation IndexPoverty Vehicle Ownership Educational Attainment Per Capita Income Unemployment Linguistic Isolation Individual Variable Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Smoking 2.242.232.242.24 2.222.242.24 Not Married 1.371.381.381.38 1.381.381.38 Female 1.151.151.151.15 1.151.151.15 Younger than 18 yrs 1.321.321.331.33 1.321.321.33 Older than 34 yrs 1.391.381.381.38 1.401.381.38 Primiparous 1.771.771.771.77 1.771.771.77 Low Gain 1.801.771.801.80 1.801.801.80 Excess Gain 0.380.380.380.38 0.380.380.38 Group Variable 2000 Deprivation Index 1.16*** *** *** *** *** *** 2000 Poverty *** 1.08*** *** *** *** *** 2000 No Vehicle *** *** 1.08*** *** *** *** 2000 Low Educational Attainment *** *** *** 1.13 *** *** *** 2000 Per Capita Income *** *** *** *** 1.12*** *** 2000 Unemployment *** *** *** *** *** 1.10*** 2000 Linguistic Isolation *** *** *** *** *** *** 1.10

PAGE 102

Appendix A (Continued) Table A.4 Results of Multilevel Model For White Non-Hispa nic Mothers Year 1990 Census Block Groups Showing Odds Ratios For Individual and Group Level Variables. Full Model Deprivation Index Poverty Vehicle Ownership Educational Attainment Per Capita Income Unemployment Linguistic Isolation Variables Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Smoking 2.232.232.242.232.232.242.25 Not Married 1.371.381.391.381.381.391.38 Female 1.151.151.151.151.151.151.15 Younger than 18 yrs 1.321.321.331.321.321.331.33 Older than 34 yrs 1.391.391.381.391.391.381.38 Primiparous 1.771.771.771.771.771.761.77 Low Gain 1.801.801.811.801.801.811.80 Excess Gain 0.380.380.380.380.380.380.38 Group Variable 1990 Deprivation Index 1.09*** *** *** *** *** *** 1990 Poverty *** 1.07*** *** *** *** *** 1990 No Vehicle *** *** 1.27*** *** *** *** 1990 Low Educational Attainment *** *** *** 1.04*** *** *** 1990 Per Capita Income *** *** *** *** 1.07 *** *** 1990 Unemployment *** *** *** *** *** 1.08 *** 1990 Linguistic Isolation *** *** *** *** *** *** 1.18

PAGE 103

Appendix A (Continued) Table A.5 Results of Multilevel Model For Black Non-Hispa nic Mothers Year 2000 Census Block Groups Showing Odds Ratios For Individual and Group Level Variables. Full Model Deprivation Index Poverty Vehicle Ownership Educational Attainment Per Capita Income Unemployment Linguistic Isolation Variables Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Smoking 2.092.092.082.092.092.092.08 Not Married 1.221.221.231.231.231.221.22 Female 1.211.211.211.211.211.211.21 Younger than 18 yrs 1.131.131.131.131.131.131.13 Older than 34 yrs 1.271.151.281.281.281.281.28 Primiparous 1.411.411.411.411.411.411.41 Low Gain 1.921.920.191.941.921.931.93 Excess Gain 0.430.430.430.430.430.430.43 Group Variable 2000 Deprivation Index 1.04*** *** *** *** *** *** 2000 Poverty *** 1.12*** *** *** *** *** 2000 No Vehicle *** *** 1.04*** *** *** *** 2000 Low Educational Attainment *** *** *** 1.02*** *** *** 2000 Per Capita Income *** *** *** *** 1.10*** *** 2000 Unemployment *** *** *** *** *** 1.03*** 2000 Linguistic Isolation *** *** *** *** *** *** 1.13

PAGE 104

Appendix A (Continued) Table A.6 Results of Multilevel Model For Black Non-Hispa nic Mothers Year 1990 Census Block Groups Showing Odds Ratios For Individual and Group Level Variables Full Model Deprivation Index Poverty Vehicle Ownership Educational Attainment Per Capita Income Unemployment Linguistic Isolation Variables Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Smoking 2.092.082.09 2.092.082.092.09 Not Married 1.231.231.22 1.221.221.221.23 Female 1.211.211.21 1.211.221.221.22 Younger than 18 yrs 1.131.131.13 1.131.131.131.13 Older than 34 yrs 1.281.281.28 1.281.281.281.28 Primiparous 1.411.411.41 1.411.411.411.41 Low Gain 1.931.931.93 1.921.921.921.93 Excess Gain 0.430.430.43 0.430.430.430.43 Group Variable 1990 Deprivation Index 1.02*** *** *** *** *** *** 1990 Poverty *** 1.01*** *** *** *** *** 1990 No Vehicle *** *** 1.05 *** *** *** *** 1990 Low Educational Attainment *** *** *** 1.04*** *** *** 1990 Per Capita Income *** *** *** *** 1.04*** *** 1990 Unemployment *** *** *** *** *** 1.04*** 1990 Linguistic Isolation *** *** *** *** *** *** 1.04

PAGE 105

Appendix A (Continued) Table A.7 Results of Multilevel Model For Hispanic Mothers Year 2000 Census Block Groups Showing Odds Ratios For Individual and Group Level Variables. Full Model Deprivation Index Poverty Vehicle Ownership Educational Attainment Per Capita Income Unemployment Linguistic Isolation Variables Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Smoking 2.162.152.15 2.162.152.152.15 Not Married 1.271.271.27 1.271.281.291.27 Female 1.111.111.11 1.181.181.181.18 Younger than 18 yrs 1.351.351.35 1.131.131.131.13 Older than 34 yrs 1.411.151.41 1.381.381.381.38 Primiparous 1.641.641.64 1.641.641.631.64 Low Gain 1.321.321.32 1.321.321.331.32 Excess Gain 0.390.390.39 0.390.390.390.39 Group Variable 2000 Deprivation Index 1.12*** *** *** *** *** *** 2000 Poverty *** 1.09*** *** *** *** *** 2000 No Vehicle *** *** 1.12 *** *** *** *** 2000 Low Educational Attainment *** *** *** 1.06*** *** *** 2000 Per Capita Income *** *** *** *** 1.10*** *** 2000 Unemployment *** *** *** *** *** 1.28*** 2000 Linguistic Isolation *** *** *** *** *** *** 1.11

PAGE 106

Appendix A (Continued) Table A.8 Results of Multilevel Model For Hispanic Mothers Ye ar 1990 Census Block Groups Showing Odds Ratios For Individual and Group Level Variables. Full Model Deprivation Index Poverty Vehicle Ownership Educational Attainment Per Capita Income Unemployment Linguistic Isolation Variables Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Smoking 2.152.162.15 2.162.162.152.15 Not Married 1.271.271.27 1.271.271.271.26 Female 1.111.111.11 1.111.111.111.11 Younger than 18 yrs 1.351.351.35 1.351.351.351.35 Older than 34 yrs 1.411.411.41 1.411.411.411.41 Primiparous 1.641.641.63 1.641.641.641.64 Low Gain 1.321.321.32 1.321.321.321.32 Excess Gain 0.390.390.39 0.390.390.390.39 Group Variable 1990 Deprivation Index 1.06*** *** *** *** *** *** 1990 Poverty *** 1.05*** *** *** *** *** 1990 No Vehicle *** *** 1.07 *** *** *** *** 1990 Low Educational Attainment *** *** *** 1.03*** *** *** 1990 Per Capita Income *** *** *** *** 1.05*** *** 1990 Unemployment *** *** *** *** *** 1.07*** 1990 Linguistic Isolation *** *** *** *** *** *** 1.08

PAGE 107

Appendix A (Continued) Table A.9 Results of Complete Multilevel Model For Year 2 000 Census Tracts Showing Odds Ratios For Individual and Group Level Variables. Full Model Deprivation Index Poverty Vehicle Ownership Educational Attainment Per Capita Income Unemployment Linguistic Isolation Variables Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Smoking 2.182.182.182.182.182.182.18 Not Married 1.281.281.281.281.281.281.28 Female 1.171.171.171.171.171.171.17 Black Non-Hispanic 1.901.901.901.901.901.901.90 Hispanic 1.131.131.131.131.131.131.13 Younger than 18 yrs 1.121.121.121.121.121.121.12 Older than 34 yrs 1.361.361.361.361.361.361.36 Primiparous 1.611.611.611.611.611.611.61 Low Gain 1.271.271.271.271.271.271.27 Excess Gain 0.380.380.380.380.380.380.38 Group Variable 2000 Deprivation Index 1.07*** *** *** *** *** *** 2000 Poverty *** 1.19*** *** *** *** *** 2000 No Vehicle *** *** 1.05*** *** *** *** 2000 Low Educational Attainment *** *** *** 1.13*** *** *** 2000 Per Capita Income *** *** *** *** 1.02 *** 2000 Unemployment *** *** *** *** *** 1.10*** 2000 Linguistic Isolation *** *** *** *** *** *** 1.03

PAGE 108

Appendix A (Continued) Table A.10 Results of Complete Multilevel Models Year 1990 Census Tracts Showing Odds Ratios For Individual and Group Level Variables. Full Model Deprivation Index Poverty Vehicle Ownership Educational Attainment Per Capita Income Unemployment Linguistic Isolation Variables Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Smoking 2.182.182.182.182.182.182.18 Not Married 1.281.281.281.281.281.281.28 Female 1.171.171.171.171.171.171.17 Black Non-Hispanic 1.901.901.901.901.901.901.90 Hispanic 1.131.131.131.131.131.131.13 Younger than 18 yrs 1.121.121.121.121.121.121.12 Older than 34 yrs 1.361.361.361.361.361.361.36 Primiparous 1.611.611.611.611.611.611.61 Low Gain 1.271.271.271.271.271.271.27 Excess Gain 0.380.380.380.380.380.380.38 Group Variable 1990 Deprivation Index 1.07 *** *** *** *** *** *** 1990 Poverty *** 1.19*** *** *** *** *** 1990 No Vehicle *** *** 1.05*** *** *** *** 1990 Low Educational Attainment *** *** *** 1.13 *** *** *** 1990 Per Capita Income *** *** *** *** 1.02*** *** 1990 Unemployment *** *** *** *** *** 1.10*** 1990 Linguistic Isolation *** *** *** *** *** *** 1.03

PAGE 109

Appendix A (Continued) Table A.11 Results of Multileve l Model For White Non-Hispanic Mothers Year 2000 Census Tracts Showing Odds Ratios For Individual and Group Level Variables. Full Model Deprivation Index Poverty Vehicle Ownership Educational Attainment Per Capita Income Unemployment Linguistic Isolation Variables Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Smoking 2.202.192.20 2.202.182.202.20 Not Married 1.361.361.36 1.361.361.361.36 Female 1.151.151.15 1.151.151.151.15 Younger than 18 yrs 1.241.241.24 1.241.241.241.24 Older than 34 yrs 1.361.361.36 1.361.371.361.36 Primiparous 1.741.741.74 1.741.741.741.74 Low Gain 1.941.911.95 1.951.941.951.95 Excess Gain 0.800.800.80 0.800.800.800.80 Group Variable 2000 Deprivation Index 1.11*** *** *** *** *** *** 2000 Poverty *** 1.05*** *** *** *** *** 2000 No Vehicle *** *** 1.06 *** *** *** *** 2000 Low Educational Attainment *** *** *** 1.09 *** *** 2000 Per Capita Income *** *** *** *** 1.08*** *** 2000 Unemployment *** *** *** *** *** 1.06 2000 Linguistic Isolation *** *** *** *** *** *** 1.07

PAGE 110

Appendix A (Continued) Table A.12 Results of Multileve l Model For White Non-Hispanic Mothers Year 1990 Census Tracts Showing Odds Ratios For Individual and Group Level Variables Full Model Deprivation Index Poverty Vehicle Ownership Educational Attainment Per Capita Income Unemployment Linguistic Isolation Variables Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Smoking 2.192.192.202.192.202.202.21 Not Married 1.361.361.371.361.371.371.36 Female 1.151.151.151.151.151.151.15 Younger than 18 yrs 1.241.241.241.231.241.241.24 Older than 34 yrs 1.361.361.361.361.351.351.36 Primiparous 1.741.741.741.741.741.741.74 Low Gain 1.941.941.951.941.951.951.95 Excess Gain 0.800.800.800.800.800.800.80 Group Variable 2000 Deprivation Index 1.09*** *** *** *** *** *** 2000 Poverty *** 1.04*** *** *** *** *** 2000 No Vehicle *** *** 1.17*** *** *** *** 2000 Low Educational Attainment *** *** *** 1.07*** *** *** 2000 Per Capita Income *** *** *** *** 1.05*** *** 2000 Unemployment *** *** *** *** *** 1.05*** 2000 Linguistic Isolation *** *** *** *** *** *** 1.11

PAGE 111

Appendix A (Continued) Table A.13 Results of Multileve l Model For Black Non-Hispanic Mothers Year 2000 Census Tracts Showing Odds Ratios For Individual and Group Level Variables Full Model Deprivation Index Poverty Vehicle Ownership Educational Attainment Per Capita Income Unemployment Linguistic Isolation Variables Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Smoking 2.062.052.052.052.062.052.05 Not Married 1.211.211.211.221.221.211.22 Female 1.201.201.201.201.201.201.20 Black Non-Hispanic 1.101.101.101.101.101.101.10 Hispanic 1.261.261.261.261.261.261.27 Younger than 18 yrs 1.401.401.401.401.401.401.40 Older than 34 yrs 2.092.092.092.122.102.102.11 Primiparous 0.820.140.820.820.820.820.82 Low Gain 1.941.911.951.951.941.951.95 Excess Gain 0.800.800.800.800.800.800.80 Group Variable 2000 Deprivation Index 1.03 *** *** *** *** *** *** 2000 Poverty *** 1.05 *** *** *** *** *** 2000 No Vehicle *** *** 1.05 *** *** *** *** 2000 Low Educational Attainment *** *** *** 1.01 *** *** *** 2000 Per Capita Income *** *** *** *** 1.04 *** *** 2000 Unemployment *** *** *** *** *** 1.02 *** 2000 Linguistic Isolation *** *** *** *** *** *** 1.08

PAGE 112

Appendix A (Continued) Table A.14 Results of Multileve l Model For Black Non-Hispanic Mothers Year 1990 Census Tracts Showing Odds Ratios For Individual and Group Level Variables. Full Model Deprivation Index Poverty Vehicle Ownership Educational Attainment Per Capita Income Unemployment Linguistic Isolation Variables Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Smoking 2.092.052.05 2.052.052.052.05 Not Married 1.231.221.21 1.211.211.211.22 Female 1.211.201.20 1.201.201.201.20 Younger than 18 yrs 1.131.101.10 1.101.101.101.10 Older than 34 yrs 1.281.261.26 1.261.261.261.26 Primiparous 1.411.401.40 1.401.401.401.39 Low Gain 1.931.941.94 1.941.941.941.94 Excess Gain 0.800.820.82 0.820.820.820.82 Group Variable 1990 Deprivation Index 1.01*** *** *** *** *** *** 1990 Poverty *** 1.00*** *** *** *** *** 1990 No Vehicle *** *** 1.03 *** *** *** *** 1990 Low Educational Attainment *** *** *** 1.03*** *** *** 1990 Per Capita Income *** *** *** *** 1.03*** *** 1990 Unemployment *** *** *** *** *** 1.02*** 1990 Linguistic Isolation *** *** *** *** *** *** 1.03

PAGE 113

Appendix A (Continued) Table A.15 Results of Multileve l Model For Hispanic Mothers Year 2000 Census Tracts Showing Odds Ratios For Individual and Group Level Variables. Full Model Deprivation Index Poverty Vehicle Ownership Educational Attainment Per Capita Income Unemployment Linguistic Isolation Variables Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Smoking 2.122.112.112.122.122.112.11 Not Married 1.261.261.261.261.261.271.26 Female 1.101.101.101.101.101.101.10 Younger than 18 yrs 1.261.261.261.261.261.261.26 Older than 34 yrs 1.391.391.391.391.391.391.39 Primiparous 1.621.621.621.621.621.611.62 Low Gain 1.371.371.371.371.371.381.37 Excess Gain 0.800.800.800.800.800.800.80 Group Variable 2000 Deprivation Index 1.08*** *** *** *** *** *** 2000 Poverty *** 1.06*** *** *** *** *** 2000 No Vehicle *** *** 1.08*** *** *** *** 2000 Low Educational Attainment *** *** *** 1.04*** *** *** 2000 Per Capita Income *** *** *** *** 1.07*** *** 2000 Unemployment *** *** *** *** *** 1.18*** 2000 Linguistic Isolation *** *** *** *** *** *** 1.07

PAGE 114

Appendix A (Continued) Table A.16 Results of Multileve l Model For Hispanic Mothers Year 1990 Census Tracts Showing Odds Ratios For Individual and Group Level Variables. Full Model Deprivation Index Poverty Vehicle Ownership Educational Attainment Per Capita Income Unemployment Linguistic Isolation Variables Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Odds Ratio Smoking 2.112.122.11 2.122.122.122.11 Not Married 1.261.261.26 1.261.261.251.25 Hispanic 1.101.101.10 1.101.101.101.10 Younger than 18 yrs 1.261.261.26 1.261.261.261.26 Older than 34 yrs 1.391.391.38 1.391.391.391.38 Primiparous 1.621.621.61 1.621.621.621.62 Low Gain 1.371.371.37 1.371.371.371.37 Excess Gain 0.800.800.80 0.800.800.800.80 Group Variable 1990 Deprivation Index 1.04*** *** *** *** *** *** 1990 Poverty *** 1.03*** *** *** *** *** 1990 No Vehicle *** *** 1.05 *** *** *** *** 1990 Low Educational Attainment *** *** *** 1.02*** *** *** 1990 Per Capita Income *** *** *** *** 1.03*** *** 1990 Unemployment *** *** *** *** *** 1.05*** 1990 Linguistic Isolation *** *** *** *** *** *** 1.05

PAGE 115

Appendix B: Low Birth Weight and Deprivation Maps Map B.1 Percent Live Births <2500g; State of Florida for Years 1992-1997

PAGE 116

Appendix B (Continued) Map B.2 Miami, Dade County; Depr ivation Index and Low Birth Weight

PAGE 117

Appendix B (Continued) Map B.3 Orange County / Orlando Florida; Deprivation Index and Low Birth Weight

PAGE 118

Appendix B (Continued) Map B.4 Hillsborough County; Depriv ation Index and Low Birth Weight

PAGE 119

Appendix B (Continued) Map B.5 Jacksonville Florida; Depr ivation Index and Low Birth Weight


xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 001498261
003 fts
006 m||||e|||d||||||||
007 cr mnu|||uuuuu
008 041209s2004 flua sbm s000|0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0000499
035
(OCoLC)57710697
9
AJU6866
b SE
SFE0000499
040
FHM
c FHM
090
G116 (ONLINE)
1 100
Green, Joseph William.
2 245
A multilevel modeling analysis of the geographic variability of low birth weight occurrence in Florida
h [electronic resource] /
by Joseph William Green Jr.
260
[Tampa, Fla.] :
University of South Florida,
2004.
502
Thesis (M.A.)--University of South Florida, 2004.
504
Includes bibliographical references.
516
Text (Electronic thesis) in PDF format.
538
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
500
Title from PDF of title page.
Document formatted into pages; contains 119 pages.
520
ABSTRACT: The associations of neighborhood level socioeconomic deprivation and low birth weight were investigated among 1,030,443 singleton live births in the State of Florida between the years 1992 and 1997. Census data for per capita income, unemployment, percent of individuals living below the poverty line, vehicle ownership and educational attainment were used as neighborhood level indicators of socioeconomic status. Additionally, these variables were combined into a deprivation index to measure relative deprivation of neighborhoods across Florida. Birth data were linked to census block groups and tracts, which were used as proxies for low birth weight. Multilevel models were used to model the relationship between the deprivation index and each of the indicators and low birth weight, while adjusting for individual level risk factors.After adjusting for individual level factors no consistent relationship between neighborhood socioeconomic measures and low birth weight could be established. The relationship between neighborhood socioeconomic factors and low birth weight varied across ethnic categories. Among White Non-Hispanics and Hispanics measures of socioeconomic deprivation had a small association with low birth weight. However, for Black Non-Hispanics neighborhood measures had little consistency in predicting the occurrence of low birth weight
590
Adviser: Reader, Steven.
653
medical geography.
deprivation.
multilevel modeling.
low birth weight.
690
Dissertations, Academic
z USF
x Geography
Masters.
773
t USF Electronic Theses and Dissertations.
4 0 856
u http://digital.lib.usf.edu/?e14.499