<%BANNER%>

Childhood Disadvantage and Weight Status in Adulthood

Permanent Link: http://ufdc.ufl.edu/UFE0045387/00001

Material Information

Title: Childhood Disadvantage and Weight Status in Adulthood
Physical Description: 1 online resource (226 p.)
Language: english
Creator: Pavela, Gregory M
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2013

Subjects

Subjects / Keywords: aging -- health -- lifecourse -- neighborhoods -- obesity
Sociology and Criminology & Law -- Dissertations, Academic -- UF
Genre: Sociology thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: This research uses a life course approach to examine the association between childhood disadvantage and adult weight.  Two main research questions are posed. First, are childhood conditions associated with Body Mass Index (BMI) as an adult after adjustment for individual sociodemographics? Second, to the extent that childhood conditions are associated with adult BMI, do adult neighborhood characteristics account for this relationship? The association between childhood conditions and adult body mass index is modeled using a two and three-level hierarchical linear model framework. Data on individual childhood and adult characteristics come from waves 2000 through 2008 of the Health and Retirement Study. Data on neighborhood characteristics come from the RAND Center for Population Health and Health Disparities Data Core. Results suggest that among males, childhood conditions are not associated with adult BMI after adjustment for individual characteristics. However, among females,paternal education remains associated with adult BMI after adjustment for individual and neighborhood characteristics.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Gregory M Pavela.
Thesis: Thesis (Ph.D.)--University of Florida, 2013.
Local: Adviser: Peek, Charles W, Iv.
Local: Co-adviser: Henretta, John C.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2015-05-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2013
System ID: UFE0045387:00001

Permanent Link: http://ufdc.ufl.edu/UFE0045387/00001

Material Information

Title: Childhood Disadvantage and Weight Status in Adulthood
Physical Description: 1 online resource (226 p.)
Language: english
Creator: Pavela, Gregory M
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2013

Subjects

Subjects / Keywords: aging -- health -- lifecourse -- neighborhoods -- obesity
Sociology and Criminology & Law -- Dissertations, Academic -- UF
Genre: Sociology thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: This research uses a life course approach to examine the association between childhood disadvantage and adult weight.  Two main research questions are posed. First, are childhood conditions associated with Body Mass Index (BMI) as an adult after adjustment for individual sociodemographics? Second, to the extent that childhood conditions are associated with adult BMI, do adult neighborhood characteristics account for this relationship? The association between childhood conditions and adult body mass index is modeled using a two and three-level hierarchical linear model framework. Data on individual childhood and adult characteristics come from waves 2000 through 2008 of the Health and Retirement Study. Data on neighborhood characteristics come from the RAND Center for Population Health and Health Disparities Data Core. Results suggest that among males, childhood conditions are not associated with adult BMI after adjustment for individual characteristics. However, among females,paternal education remains associated with adult BMI after adjustment for individual and neighborhood characteristics.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Gregory M Pavela.
Thesis: Thesis (Ph.D.)--University of Florida, 2013.
Local: Adviser: Peek, Charles W, Iv.
Local: Co-adviser: Henretta, John C.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2015-05-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2013
System ID: UFE0045387:00001


This item has the following downloads:


Full Text

PAGE 1

1 CHIL D HOOD DISADVANTAGE AND WEIGHT STATUS IN ADULTHOOD By GREGORY PAVELA A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2013

PAGE 2

2 2013 Gregory Pavela

PAGE 3

3 To all those who helped

PAGE 4

4 ACKNOWLEDGMENTS All members of my dissertation committee Dr. Charles Peek, Dr. John Henretta, Dr. Barbara Zsembik, Dr. Robert White, and Dr. Tracey Barnettt, made essential contributions to this dissertation. student, researcher, and person is a model for me and the students I may one day advise. reminder too far beyond the (plural) data is something I hope to remember for the remainder of my career. And Dr. Zsembik did exactly what I hoped she would do when I asked her to be on my committee get to t he heart of the issue during the dissertation proposal de fense. Dr. Barnett was generous with her time and editing, and her good reputation as a committee member ensures she will continue to walk up the hill to serve as the outside committee for sociology students Dr. White development and identification of the life course issues within this project has enriched this and future projects. The support of family and friends, especially Hye Young Nam and Joe Rukus, will hopefully continue well beyond my days in graduate school. Lastly, the editorial staff at the University of Florida has graciously contributed to the professional presentation of this project This material is based upon work supported by the National Science Foundation under Grant No. SES 1129664. Any opinion, findings, and conclusions or recommendations expressed in this mat erial are those of the author and do not necessarily reflect the vi ews of the National Science Foundation.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 7 LIST OF FIGURES ................................ ................................ ................................ ........ 12 LIST OF ABBR EVIATIONS ................................ ................................ ........................... 13 ABSTRACT ................................ ................................ ................................ ................... 14 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 15 The Im portance of Obesity Across the Life Course ................................ ................ 16 Conceptual Links Between Early Life Disadvantage and Later Health Outcomes .. 19 Life Course Perspective ................................ ................................ .......................... 20 Life Course Epidemiology ................................ ................................ ....................... 25 Diss ertation Overview ................................ ................................ ............................. 31 2 DISADVANTAGE AND OBESITY ................................ ................................ ........... 34 Childhood Obesity and Early Disadvantage ................................ ............................ 35 Early Disadvantage and Adult Obesity ................................ ................................ .... 42 Conceptual and Empirical Links between Neighborhood Context and Adult Health ................................ ................................ ................................ .................. 50 Causal Ordering of Health and Socioeconomi c Status ................................ ........... 54 3 DATA AND METHODS ................................ ................................ ........................... 60 Data ................................ ................................ ................................ ........................ 60 Analytical Sample ................................ ................................ ............................. 62 Hierarchical Structure of the Data ................................ ................................ .... 62 Measures ................................ ................................ ................................ .......... 63 Dependent Variable Body Mass Index ................................ ....................... 63 Independent Variables Childhood Measures ................................ ............. 64 Independent Variables Adult Measures ................................ ..................... 65 Independent Variables Neighborhood Measures ................................ ....... 67 Analysis ................................ ................................ ................................ .................. 68 Analyses for Adult BMI in 2000, Individual Characteristics ............................... 69 Analysis for Changes in Adult BMI, Individual Characteristics .......................... 70 Analyses for Adult BMI in 2000, Neighborhood Characteristics ....................... 70 Analysis for Changes in Adult BMI, Neighborhood Characteristics .................. 70

PAGE 6

6 4 CHILDHOOD, ADULT CHARACTERISTICS, AND ADULT WEIGHT .................... 76 Male BMI in 2000 ................................ ................................ ................................ .... 77 Female BMI in 2000 ................................ ................................ ................................ 79 Changes in Male BMI between 2000 and 2008 ................................ ...................... 81 Changes in Female BMI between 2000 and 2008 ................................ .................. 84 Summary ................................ ................................ ................................ ................ 87 5 NEIGHBORHOODS AND ADULT WEIGHT ................................ ......................... 100 Male BMI in 2000 ................................ ................................ ................................ .. 101 Female BMI in 2000 ................................ ................................ .............................. 102 Male BMI, 2000 2008 ................................ ................................ ............................ 103 Female BMI, 2000 2008 ................................ ................................ ....................... 105 Male B MI in 2000 ................................ ................................ ................................ .. 106 Female BMI in 2000 ................................ ................................ .............................. 107 Changes in Male BMI, 2000 2008 ................................ ................................ ........ 109 Changes in Female BMI, 2000 2008 ................................ ................................ .... 110 Summary ................................ ................................ ................................ .............. 111 6 CHILDHOOD CONDITIONS, NEIGHBORHOODS, AND ADULT WEIGHT ......... 133 Female BMI in 2000 ................................ ................................ .............................. 134 Changes in Female BMI ................................ ................................ ....................... 135 Summar y ................................ ................................ ................................ .............. 137 7 DICUSSION AND CONCLUSION ................................ ................................ ........ 152 Synopsis of Childhood Conditions Associated with BMI ................................ ....... 153 Synopsis of Neighborhood Characteristics Associated with BMI .......................... 160 Synopsis of Adult Conditions Associated with BMI ................................ ............... 164 Limitations and Future Research ................................ ................................ .......... 166 APPENDIX A UNADJUSTED BMI ESTIMATES FROM CHAPTER 4 ................................ ......... 171 B UNADJUSTED BMI ESTIMATES FROM CHAPTER 5 ................................ ......... 182 C UNADJUSTED BMI ESTIMATES FROM CHAPTER 6 ................................ ......... 2 01 LIST OF REFERENCES ................................ ................................ ............................. 212 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 226

PAGE 7

7 LIST OF TABLES Table page 3 1 Means, Proportions, and Number of Respondents for Measures of Childhood Disadvantage in the HRS Cohort (1931 1941) ................................ ................... 72 3 2 Means and Proportions for HRS Sample in 2000 2008. ................................ ..... 73 3 3 Descriptive Statistics for Census Tract Characteristics, HRS 2000 .................... 74 3 4 Self reported and Measured BMI, HRS 2006 20010. ................................ ......... 75 4 1 Results from a Hierarchical Linear Model Regressing Male Adjusted BMI on Childhood and Individual Characteristics HRS Cohort 2000. .......................... 90 4 2 Variance Components and Fit Statistics from an HLM Model Estimating Adjusted Male BMI in 2000 ................................ ................................ ................ 91 4 3 Results from a Hierarchical Linear Model Regressing Female Adjusted BMI on Childhood and Individual Characteristics HRS Cohort 2000. ...................... 92 4 4 Variance Components and Fit Statistics from an HLM Model Estimating Adjusted Female BMI in 2000 ................................ ................................ ............ 93 4 5 Results from a Hierarchical Linear Model Regressing Male Adjusted BMI on Childhood and Individual Characteristics HRS Cohort 2000 2008. ................. 94 4 6 Variance Components and Fit Statistics f rom an HLM Model Estimating Unadjusted Male BMI, HRS 2000 2008. ................................ ............................. 96 4 7 Results from a Hierarchical Linear Model Regressing Female Adjusted BMI on Childhood and Individual Characteristics HRS Cohort 2000 2008. ............. 97 4 8 Variance Components and Fit Statistics f rom an HLM Model Estimating Adjusted Female BMI HRS 2000 2008. ................................ ............................ 99 5 1 Results from a Hierarchical Linear Model Regressing Adjusted Male BMI in 2000 on Neighborhood Characteristics HRS Cohort 2000 ............................. 115 5 2 Variance Components and Fit Statistics from an HLM Model Estimating Adjusted Male BMI, in 2000 ................................ ................................ .............. 116 5 3 Results from a Hierarchical Linear Model Regressing Adjusted Female BM I in 2000 on Neighborhood Characteristics HRS Cohort 2000 ......................... 117 5 4 Variance Components and Fit Statistics from an HLM Model Est imating Adjusted Female BMI, in 2000 ................................ ................................ .......... 118

PAGE 8

8 5 5 Results from a Hierarchical Linear Model Regressing Male Adjusted BMI on Neighborhood Characteristics HRS Cohort 2000 20008. .............................. 119 5 6 Variance Components and Fit Statistics from an HLM Model Estimatin g Adjusted Male BMI, in 2000 2008 ................................ ................................ .... 120 5 7 Results from a Hierarchical Linear Model Regressing Female Adjusted BMI on Neighborho od Characteristics HRS Cohort 2000 20008. ......................... 121 5 8 Variance Components and Fit Statistics from an HLM Model Estimating Adjusted Female BMI, in 2000 2008. ................................ ............................... 122 5 9 Results from a Hierarchical Linear Model Regressing Male Adjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000. ................... 123 5 10 Variance Components and Fit Statistics from an HLM Model Estimating Adjusted Male BMI in 2000. ................................ ................................ .............. 124 5 11 Results from a Hierarchical Linear Model Regressing Female Adjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000. .............. 125 5 12 Variance Components and Fit Statistics from an HLM Model Estimating Adjusted Female BMI in 2000. ................................ ................................ .......... 126 5 13 Results from a Hierarchical Linear Model Regressing Male Adjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000 20008. ....... 127 5 14 Variance Components and Fit Statistics from an HLM Model Estimating Changes in Adjusted Male BMI (2000 2008). ................................ ................... 129 5 15 Results from a Hierarchical Linear Model Regressing Female Adjusted BMI on Individual and Neighborhood Characteristics HRS Cohor t 2000 20008. .. 130 5 16 Variance Components and Fit Statistics from an HLM Model Estimating Changes in Adjusted Female BMI (2000 2008). ................................ ............... 132 6 1 Results from a Hierarchical Linear Model Regressing Male Adjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000. ................... 141 6 2 Variance Components and Fit Statistics from an HLM Model Estimating Changes In Adjuste d Male BMI, 2000. ................................ ............................. 142 6 3 Results from a Hierarchical Linear Model Regressing Female Adjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000. .............. 143 6 4 Variance Components and Fit Statistics from an HLM Model Estimating Changes In Adjuste d Female BMI, 2000. ................................ ......................... 144

PAGE 9

9 6 5 Results from a Hierarchical Linear Model Regressing Male Adjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000 20008. ....... 145 6 6 Variance Components and Fit Statistics from an HLM Model Estimating Changes In A djusted Male BMI (2000 2008). ................................ ................... 147 6 7 Results from a Hierarchical Linear Model Regressing Female Adjusted BMI on Individual and N eighborhood Characteristics HRS Cohort 2000 20008. .. 148 6 8 Variance Components and Fit Statistics from an HLM Model Estimating Changes In Adjusted Female BMI (2000 2008). ................................ .............. 150 A 1 Results from a Hierarchical Linear Model Regressing Male Unadjusted BMI on Childhood and Individual Characteristics, HRS Cohort 2000. ...................... 172 A 2 Variance Components and Fit Statistics from an HLM Model Estim ating Unadjusted Male BMI in 2000 ................................ ................................ .......... 173 A 3 Results from a Hierarchical Linear Model Regressing Female Unadjusted BMI on Childhood and Individual Characteristics HRS Cohort 2000. ............. 174 A 4 Variance Components and Fit Statistics from an HLM Model Estimating Un adjusted Female BMI in 2000 ................................ ................................ ..... 175 A 5 Results from a Hierarchical Linear Model Regressing Male Unadjusted BMI on Childhood and Individual Characteristics HRS Cohort 2000 2008. ........... 176 A 6 Variance Components and Fit Statistics from an HLM Model Estimating Ad justed Male BMI, HRS 2000 2008. ................................ ............................... 178 A 7 Results from a Hierarchical Linear Model Regressing Female Adjusted BMI on Childhood and Individual Characteristics HRS Cohort 2000. ..................... 179 A 8 Variance Components and Fit Statistics from an HLM Model Estimating Adjusted Female BMI HRS 2000 2008. ................................ .......................... 181 B 1 Results from a Hierarchical Linear Model Regressing Unadjusted Male BMI in 2000 on Neighborhood Characteristics HRS Cohort 2000 ......................... 183 B 2 Variance Components and Fit Statistics from an HLM Model Estimating Unadju sted Male BMI in 2000. ................................ ................................ .......... 184 B 3 Results from a Hierarchical Linear Model Regressing Unadjusted Female BMI in 2000 on Neighborhood Characteristics HRS Cohort 2000 ................. 185 B 4 Variance Components and Fit Statistics from an HLM Model Estimating Adjusted Female BM I, in 2000 ................................ ................................ .......... 186

PAGE 10

10 B 5 Results from a Hierarchical Linear Model Regressing Male Unadjusted BMI on Neighborhood Characteristics HRS Cohort 2000 20008. ......................... 187 B 6 Variance Components and Fit Statistics from an HLM Model Estimating Unadjusted Male BMI, in 2000 2008 ................................ ............................... 188 B 7 Results from a Hierarchical Linear Model Regressing Female Adjusted BMI on Neighborhood Characteristics HRS Cohort 20 00 20008. ......................... 189 B 8 Variance Components and Fit Statistics from an HLM Model Estimating Unadjusted Female BMI, in 2000 2008. ................................ ........................... 190 B 9 Results from a Hierarchical Linear Model Regressing Male Unadjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000. .............. 191 B 10 Variance Components and Fit Statistics from an HLM Model Estimating Unadjusted Male BMI in 2000. ................................ ................................ .......... 192 B 11 Results from a Hierarchical Linear Model Regressing Female Unadjusted BMI on Individual and Neighborhood Characteristics HRS Coh ort 2000. ...... 193 B 12 Variance Components and Fit Statistics from an HLM Model Estimating Adjusted Female BMI in 2000. ................................ ................................ .......... 194 B 13 Results from a Hierarchical Linear Model Regressing Male Unadjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000 20008. .. 195 B 14 Variance Components and Fit Statistics from an HLM Model Estimating Changes in Unadjusted Male BMI (2000 20 08). ................................ ............... 197 B 15 Results from a Hierarchical Linear Model Regressing Female Adjusted BMI on Individual and Neighborhood Characteris tics HRS Cohort 2000 20008. .. 198 B 16 Variance Components and Fit Statistics from an HLM Model Estimating Changes in Adjusted F emale BMI (2000 2008). ................................ ............... 200 C 1 Results from a Hierarchical Linear Model Regressing Male Unadjusted BMI on Individual and Neighborhood Characteristics, HRS Cohort 2000. ............... 202 C 2 Variance Components and Fit Statistics from an HLM Model Estimating Change s In Unadjusted Male BMI, 2000. ................................ ......................... 203 C 3 Results from a Hierarchical Linear Model Regressing Female Unadjusted BMI on Individual an d Neighborhood Characteristics HRS Cohort 2000. ....... 204 C 4 Variance Components and Fit Statistics from an HLM Model Estimating Chan ges In Unadjusted Female BMI, 2000. ................................ ..................... 205

PAGE 11

11 C 5 Results from a Hierarchical Linear Model Regressing Male Unadjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000 20008. .. 206 C 6 Variance Components and Fit Statistics from an HLM Model Estim ating Changes In Unadjusted Male BMI (2000 2008). ................................ ............... 208 C 7 Results from a Hierarchical Linear Model Regressing Female Unadjusted BM I on Individual and Neighborhood Characteristics, HRS Cohort 2000 20008. ................................ ................................ ................................ .............. 209 C 8 Variance Components and Fit Statistics from an H LM Model Estimating Changes In Unadjusted Female BMI (2000 2008). ................................ .......... 211

PAGE 12

12 LIST OF FIGURES Figure page 1 1 Conceptual Model Linking Childhood Conditions with Adult BMI ........................ 33 6 1 Adjusted Body Mass Index o f Females by Paternal Education ........................ 151

PAGE 13

13 LIST OF ABBREVIATION S BMI Body Mass Index CAD Cumulative Advantage/disadvantage T heory LCE Life Course E pidemiology LCP Life Course P erspective HRS Health and Retirement Study

PAGE 14

14 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy CHIL D HOOD DISADVANTAGE AND WEIGHT STATUS IN ADULTHOOD By Gregory Pavela May 2013 Chair: Charles Peek Cochair: John Henretta Major: Sociology This research uses a life course approach to examine the association between childhood disadvantage and adult weight. Two main research questions are posed. First, are childhood conditions associated with Body Mass Index ( BMI ) as an adult after adjustment for individual sociodemographics? Second, to the extent that childhood conditions are associated with adult BMI, do adult neighborhood characteristics ac count for this relationship? The association between childhood conditions and adult body mass index is modeled using a two and three level hierarc hical linear model framework. Data on individual childhood and adult char acteristics come from waves 2000 through 2008 of the Health and Retirement Study. Data on neighborhood characteristics come from the RAND Center for Population Health and Health Disparities Data Core. Results suggest that among males, childhood conditions are not associated with adult BMI after adjustment for individual characteristics. However, among females, paternal education remains associated with adult BMI after adjustment for individual and neighborhood characteristics.

PAGE 15

15 CHAPTER 1 INTRODUCTION Systems of social stratification have long been known to have myriad effects on both the functioning of society and on the individuals who occupy the various strata. There is substantial evidence that individuals who occupy lower strata, and thus by defini tion are disadvantaged with regards to resources relative to other society members, are more likely to experience a wide variety of adverse events. The childhood, when n umerous circumstances such as family structure, income, and life affected by social disadvantage is health. This research examines the association between disadvanta ge across the life course and a specific dimension of health: body weight. Using data from the Health and Retirement Study, five research questions are posed. First, are childhood conditions associated with BMI as an adult, after adjusting for a range of individual level adult characteristics? Second, are childhood conditions associated with changes in BMI over time as an adult? Third, how do neighborhood characteristics influence adult BMI, after adjusting for individual level characteristics? Fourth, ar e neighborhood characteristics associated with changes in BMI over time as an adult? Finally, to the extent that childhood conditions are associated with adult BMI, do neighborhood characteristics account for this relationship? Results will further the res earch on childhood disadvantage and obesity, as well as contribute to the sociological literature on inequality and the life course epidemiological perspective. The life course epidemiological perspective posits that

PAGE 16

16 early life characteristics influence la ter health. These influences may operate through permanent effects known as scarring, or through a pathway of connected life experiences, with one form of advantage or disadvantage leading to a successive form of advantage or disadvantage. In order to asse ss whether childhood experiences have permanent effects or operate through pathways of connected life experiences, a range of childhood and adulthood measures is necessary. Using a variety of data sources, this research tests whether childhood ex periences influence adult BMI and whether adult neighborhood circumstances retain independent effects on BMI, controlling for other individual level characteristics associated with obesity, including education and income. If adult neighborhood context retains an in dependent effect on adult weight, then the life course perspective should recognize the importance of neighborhood context throughout the life course, especially in relation to obesity. The importance of research on the dynamics of weight and weight relat ed health behaviors across the life course is underscored by the rapid increase in overweight and obese individuals in the US over the past several decades. This introductory chapter lays the foundation for adult weight outcomes as an important area of res earch and then describes the specific research frameworks used in this research the life course perspective. Once the general significance and framing of the research has been established, Chapter 2 reviews the specific conceptual and empirical links betwe en childhood disadvantage neighborhood context, and adult weight The Importance of Obesity Across the Life Course Obesity has been characterized as a national epidemic, and increasing rates of obesity have been predicted to continue for the next several decades (Mokdad, Bowman, and Ford 2001) Obesity is associated with type II diabetes and

PAGE 17

17 cardiovascular disease, the leading cause of death in the United States (Resnick, V alsania and Halter 2000, Poirier Giles, and Bray 2006 ). Using the direct method of age adjustment based on the 2000 census data, estimates of the prevalence of obesity and overweight status for the years of 2007 2008 are 33.8% and 34.2%, respectively. Thu s for the years 2007 2008, 68% of the US population was either overweight or obese (Flegal, Caroll, and Ogden 2010). Increasing rates of obesity are associated with significant increases in the predicted costs, both health and financial, to society. Earlie r estimated the increased costs of health care because they excluded individuals over the age of sixty five, an especially important omission given that increasing proportion of the US populati on that over the age of sixty five (Wang and Beydoun et al. 2008). Wang and Beydoun at al. (2008) estimated the health care costs associated with obesity/overweight would reach $860.7 956.9 billion by 2030, representing 16 18% of total US health care costs Given the social burden obesity represents to the U.S., it is important to understand how risk factors across the life course are associated with becoming either overweight or obese. Results of this dissertation will contribute both to scholarly work on the long term effects of disadvantage and to the public policy discuss ions on how to best address the United States obesity epidemic. Healthy People 2010 objectives regarding weight and weight disparities have not yet been met and the recent overhaul of t he United States health care system make it all the more pressing for public policy makers to have access to information to help design programs to counter trends in United States obesity.

PAGE 18

18 Socioeconomic status throughout the life course is an important ris k factor which has generally been found to be negatively associated with obesity, although the association is usually stronger for women than men (McLaren 2007, Sobal 1989). In the U.S., obesity is stratified along the three standard measures of socioecono mic status: education, income, and occupation. In 2000 26% of individuals with less than a high school education were obese, compared to 15% of those with a college degree (Mokdad et al. 2001). Although measures of socioeconomic status among adults are imp ortant for understanding risk factors for obesity, socioeconomic status early in life may have long lasting consequences for risk of obesity. Indeed, research has shown that childhood disadvantage is associated with increased risk of obesity. Childhood dis advantage measured as neighborhood poverty and low parental education has been found to be associated with an increased risk of obesity for both males and female adolescents, but when socioeconomic disadvantage is measured as either family poverty or welfa re receipt, the association with increased obesity is only found for females in young adulthood (Lee, Harris, and Gordon Larsen 2009) Research has also reported mixed results on a potential interaction between socioeconomic status and race during adolescence, with some research finding that childhood disadvantage leading to an increased risk of being overweight only for non Hispanic whites (Alaimo 2001), and other research finding no such interaction (Scharoun Lee, Kaufman, and Popkin 2009). These findings underscore the importance of treating childho od disadvantage as a multi dimensional concept with different effects for different groups of people in relation to weight outcomes

PAGE 19

19 Conceptual Links Between Early Life Disadvantage and Later Health Outcomes Efforts have been made for half a century to ex plain and predict changing distributions of resources as individuals age. This dissertation, which focuses on health across individual s life course, draws heavily upon these earlier efforts. A particularly important theoretical framework to emerge from th ese efforts is the cumulative advantage disadvantage framework. Although researchers have increasingly relied upon the cumulative advantage/disadvantage (CAD) framework to explain economic inequalities among aging populations theories of CAD were initially popularized by Robert Merton and his research explaining divergent careers of scientists. Merton (1968) introduced an early version of CAD to help explain the fact that proportionately great credit for their contributions to science while relatively unknown scientists tend to get disproportionately little credit for comparable (Merton 1968:57). Multiple processes help explain the Matthew effect in science. Psycho social processes such as name rec ognition and institutional prestige contribute to the unequal distribution of credit in science I nstitutional processes also contribute as, i scientific excellence are allocated far larger resources for investigat ion centers [which] attracts truly promising graduate students [and] retain on their faculties, these scientists of exceptional By linking individual level processes with institutional distributions of reward systems, Merton provi ded sociologists with a useful multilevel framework of processes of inequality. In a similar fashion, this dissertation seeks to explain adult weight status as a function of both the individual experiences of early disadvantage with the institutional proce sses that allocate resources based on individual characteristics, such

PAGE 20

20 as educational attainment. Institutional processes also allocate resources at a scope greater than the individual, including the neighborhood and national. Although national and global processes of resource distribution significantly contribute to global variations in individual health, this dissertation only attempts to capture the effects of neighborhood level resources on US adult weight status. Dannefer (1987) argued that the Matthe w effect occurs in aging, and proposed that aging be conceptualized as a result of social processes operating at multiple levels which regulate intra cohort variation, including population level income inequality, occupational segregation, and individual l evel perceptions of worthiness for future job opportunities continued to develop and explicate the CAD framework as a tool to explain intra cohort variability, giving labels to the resources valued by society (the precious) and to the temp oral distribution of valued resources such that those in early possession of the precious become precocious (O'Rand 1996 ) CAD has increasingly been used to explore intra cohort variation in health outcomes, and life course epidemiology has developed as a f ramework which applies the basic premises of both CAD and the life course perspective to individual health. The l ife course epidemiology perspective will be discussed in greater detail after a general review of the life course perspective. Life Course Pe rspective The life course perspective is a theoretical orientation defining a common domain of inquiry and aids in problem identification (Elder, Johnson, and Crosnoe 2003a ). The broadly defined is the exploration of the processes by which social entities (individuals, couples, families, cohorts etc.) develop along social trajectories through time, are affected by socio historical forces, and affect

PAGE 21

21 As Elder (2003) obse rves, the call to explore the intersections of personal biography, history, and social structure extends to Mills and The Sociological Imagination. It is a perspective that has developed from the work of numerous scholars (Elder 1995; Riley 1987; Ryder 196 5), although this dissertation adopts Elder has set forth several theoretical principles to help guide research operating within a life course perspective (Elder 1995; Elder, J ohnson, and Crosnoe 2003b; Elder and Shanahan 2007). These principles include (1) historical time and place; (2) human development as a lifelong process; (3) linked lives; (4) timing and transitions; and (5) agency. The principle of human development as a lifelong process states simply that human development does not end with childhood. The effects of childhood experiences extend past young adulthood, into middle age and beyond. Many of the longitudinal studies that began in the early 20 th century were fo cused on age specific development the development from childhood to young adulthood. As data collection continued, however, researchers realized the importance of viewing developmental processes as occurring not only during childhood but into adulthood. M any of the same pressures that influence early personality and cognitive development in children, such as housing conditions, parental care, and financial well being readily exert their influence on weight development as well, as will be reviewed in the se ction documenting empirical evidence supporting a life course epidemiology perspective. The life course principle of historical time and place states that the life course of individuals is embedded in both a hi storical and physical context. Elder often poi nts to

PAGE 22

22 powerful effects of national scale events such as the Great Depression and World War Two on the life chances of individuals. The experiences of three different cohorts, the Oakland Growth Study cohort (b.1920 1921); the Berkeley Guidance Study (b.19 28 1920) were each shaped by their experiences with the Great Depression. The Oakland cohort, born in the early 1920 s, experienced the great depression after a childhood during economic boom times but prior t o their entry into the workforce. By contrast, the Berkley cohort experienced the Great Depression during childhood and thus was more likely to experience a fatherless household due to WW2 (Elder 1998). This dissertation examines individual and meso level understood in the context of national trends in obesity. Broad social changes such as suburbanization, advances in the mass production of food, and technology improvements in t he workplace have each contributed to rising obesity rates (Finkelstein, Ruhm, and Kosa 2004; Vandegrift and Yoked 2004). A particularly important change is the mass production of food, with evidence that the increase in caloric consumption by US residents s since 1970 is explained almost entirely by increased snack food consumption for both males and females (Cutler, Glaeser, and Shapiro 2003). Whatever linkages are found between early life disadvantage and later weight status, they must be understood as oc curring during a particular historical period when multiple developments led to rising obesity rates. Indeed, childhood disadvantage in other times and in other places will be linked with very different health outcomes. The principle of linked lives states other, such as parents and children, occupy mutually influential interlocking

PAGE 23

23 development trajectories that extend throughout their lives Elder, Johnson, and Crosnoe 2003). This principle is intuitively tr ue in the relationship between parents and children. The Great Depression presented special difficulties to families, and many families were ultimately forced to separate after being unable to adjust to the exigencies of the time. Children of parents who s troubles in the form of domestic abuse and parental alcoholism (Elder 1998). But the children who experience greater cumulative problems such as chronic disease, emotional problems, difficulty finding a job, and social difficulties are more likely to report fears of stigma, increased care demands, and feelings of failure (Greenfield and Marks 2006 ) Embeddedness in thes e relations can not only affect exposure to hardship, but also the timing of transitions, another central theoretical principle of the life course perspective. succession of (Elder 1998: 3). An age graded life course implies that throughout individual development, there will be many changes in social roles, marked by transitions such as high schoo l graduation, first job, or first marriage. The timing of these transitions can have lifelong effects on individual development, and substantial empirical work has documented associations between timing of events and later outcomes For example, among marr ied couples, timing of birth is associated with divorce (Kravdal 1988). The experiences of the Oakland and Berkeley cohorts demonstrate the importance of timing Transitions experienced at certain periods of the life stage can have either positive or

PAGE 24

24 nega tive consequences, partly through the mechanism of cumulative advantage and disadvantage. Those who entered war service prior to establishing an occupational trajectory experienced trajectories of greater advantage relative to those who entered war service after they had already begun establishing their work careers. Even the gifted members of the Terman study, who encountered war mobilization at an inopportune time in their life, were unable to break from a traj e ctory of greater disadvantage that included greater work instability, lower earned income, and higher rates of divorce (Elder 1998). Human agency is the fifth and final theoretical principle of the life course perspective individuals construct their own life course throu gh choices and actions they take within the opportunities and constraints of history and purely determined by socio historical forces or the experiences of others, but are also affected by individual choices (assuming these are themselves not determined). were more likely to start a successful trajectory of education, work and family (Elder 1999) and thought about the future with a sense of efficacy Belief in the ability to influence future events is readily applicable to issues of weight. Psychological research ind icates that individuals who possess a general tendency to be self regulating and to orient toward contextual factors that promote choice and individual initiative also tend to be more successful in weight loss programs (Williams, Grow, Freedman, Ryan, and Deci 1996) Cutler, Glaeser, and Shapiro (2003), although

PAGE 25

25 not using the terms planful competence or autonomy orientation advocate for the importance of self control for maintaining a healthy weight status. Life Course Epidemiology In the case of health inequalities, life course epidemiology has emerged as a tool to help understand the connections between social ineq ualities and individual health. Life course epidemiology helps to establish the connection between childhood advantage and adult health, and applies many of the principles of the broader life course perspective to the specific experience of health. Indeed, t he fastest growing area of life course research is the examination of health traje ctories across the life course. (Mayer 2009). Research examining heal th trajectories across the life course typically adopts the life course epidemiological perspective whi ch posits that early life events beginning in the pre natal period and extending to adulthood (Ben Schlomo and Kuh 2002:285) have long lasting effects on later health outcomes. Within the life course epidemiological perspective, there are three non mutuall y exclusive explanations linking early life experiences and later health outcomes: (1) the critical period model; (2) accumulation model; and (3) the pathways model. The pathways model can also be seen as a special case of the accumulation model whereby th e experience of health influencing events are correlated with each other and occur in sequence rather than concurrently. Although each of the five principles of the life course perspective (development as a lifelong process, historical time and place, lin ked lives, timi ng, and agency) inform the life course epidemiology approach to disease risk, the principles of development as a lifelong process and timing stand out as central tenets of current LCE literature. The principle of lifelong development states that the development of an individual continues

PAGE 26

26 beyond childhood and into adulthood, and that a comprehensive understanding of the The principle of timing states that the timing of events can have lasting impacts on life course trajectories. The meaning of first marriage at age 30 is significantly different from the meaning of a f ourth marriage at age 30. The predominant LCE models linking early events and later health fo llow from these two principles of the life course perspective. The first explanation linking early life events and later health, known as critical period or latency model, argues that early life exposures can result in biological imprinting, increasing su sceptibility to certain diseases later in life (Galobardes, Lynch and Smith 2004). The Barker or fetal origins hypothesis, for example, conceptualizes poor fetal and early post natal nutrition as shaping metabolic adaptations that lead to increased cardiov ascular risk later in life (Hales and Barker 1992, Pollitt, Rose and Kaufman 2004). In two systematic reviews of the literature, Galobardes, Lynch, and Davey Smith (2004,2008) conclude that all cause mortality risk is higher for those who experienced poor economic circumstances during childhood and that this association is robust between genders and cohorts.(Galobardes, Lynch, and Smith 2008; Galobardes, Lynch, and Davey Smith 2004). In their original review, Galobardes and Lynch et al. (2004) found that f or most causes of mortality, adjusting for adult socioeconomic status attenuated but did not full mediate the relationship between early conditions and later health. Underscoring the complementary nature of the critical period and accumulation model, the d egree of attenuation varies between health outcomes. Independent association between child outcomes and health are especially robust for stomach cancers and stroke mortality, the association with cardiovascular morality is usually

PAGE 27

27 attenuated (but not media ted) by adult circumstances, and non smoking related cancers, prostate cancer, and accidental deaths are best explained by adult conditions (Galobardes and Lynch et al. 2004, Hart, Smith, and Blane 1998) The critical period model can be extended to inclu de a sensitive period model. In a sensitive period, the effects of exposure are magnified relative to effects of the same exposure if it had occurred at another time (Lynch and Smith 2004). Exposures outside of a critical period no effect on later health w hile exposures outside of a sensitive period continue to have effects albeit at a diminished level. Elder (1998) employs a sensitive period model when discussing the differential effects of the Great Depression on different cohorts of individuals. Childre n who had reached adolescence prior to the Great Depression had experienced a relatively secure childhood while children born of life (Elder 1998). Both cohorts were a ffected by the hardships of the Depression, but the effects on life course trajectories were particularly pronounced for the younger cohort. A final permutation of the critical period model allows for the possibility that latent effects only manifest themselves if a third variable is present. This notion underlies the thrifty genotype hypothesis and glucose intolerance, but other evidence of this particular critical period model has been found. Frankel, Elwood, and Sweetnam (1996) found the associatio n between birth weight and risk of coronary heart disease could not be explained by childhood or adult socioeconomic status, however the increased risk of coronary heart disease associated with low birth weight was conditional on having increased BMI in ad ulthood. Koupilova, Leon and Lithell et al.

PAGE 28

28 (1997) found a similar interaction regarding blood pressure. Men with low birth weight but who achieved an above median height in adulthood were at increased risk of high blood pressure relative to those born at normal weight (Koupilova Leon, Lithell, and Berglund 1997; Leon, Lithell, Vger, Koupilov, and Mohsen 1998). The second theoretical explanation, the pathway model, posits that early life events set forth "chains of (Hertzman et al. 2001), which in turn affect health. For example, early childhood disadvantage may lead to poor performance in school, eventually leading to lower income and higher stress jobs, and lower quality housing as an adult. Polit t and Rose et al. (2005) found that the majority of research testing a pathway model found supporting evidence that early life disadvantage was associated with risk factors for CVD, including leisure time, alcohol intake, higher smoking rates, and higher B MI. The accumulation model is the third explanation linking early events and later health. This model hypothesizes that individuals accumulate biological experiences, course. While the critical p eriod model can be conceptualized as an interaction model, the accumulation model is an additive model that sums the accumulation of life experiences (Ben Schlomo and Kuh 2002, Pollitt, Rose et al. 2005). Research that tests the cumulative model of adult h ealth risks usually proceeds by summing the number of times an individual has experienced an adverse event or form of disadvantage in their life (Pollitt, Rose et al. 2005). Forms of disadvantage typically include participation in welfare program s (Kauhane n, Lakka, and Lynch et al. 2006) father working in a manual occupation (Davey Smith 1997) a nd poor housing conditions (Claussen 2003). Support for the cumulative model has been

PAGE 29

29 found for numerous and diverse outcomes, including cardiovascul ar disease (Poll itt and Rose et al. 2005), the association between marital history and mortality (Henretta 2010), hypertension (James and Van Hoewyk et al. 2006), inflammatory markers (Pollitt, Kaufmann and Rose et al. 2007), smoking and drug dependence (Lloyd and Turner 2003) respiratory illnesses (Mann, Wadsworth, and Colley 1992),and cor tisol dysregulation (Gustafsson Anckarsater and Lichtenstein 2010.) Although each of these life course epidemiology models describe distinct processes linking early life events with lat er life outcomes, they do not necessarily contradict each other. Early life events may have a latent effect as well as set an individual along a pathway of cumulative adversity (Power and Hertzman 1997). Because the pathway model emphasizes trajectories of biological and social experiences rather than the physiological imprinting of bio social experiences, the pathway model allows for a greater degree of health modification through interventions and focuses investment s across the life course. Social program s such as school readiness programs can target those on high risk trajectories, changing the social conditions to more closely resemble those of individuals on more advantaged pathways. Policy that favors latent models of childhood disadvantage, however, will tend to shift resource investment earlier in the life course. An important limitation of this re search is the inability to distinguish between the cumulative exposure and pathway models; however, much of the research linking childhood conditions with life course outcomes remains focused on testing whether childhood condition have a direct or indirect association with adult health The primary contribution of this dissertation is to test whether adult neighborhood characteristics explain the residual as sociation between

PAGE 30

30 childhood conditions and adult weight. Future research may use newer techniques to better distinguish between models pathway and cumulative exposure models (Mishra, Nitsch, Black, De Stavola, Kuh, and Hardy 2009) Because of the relative strength of individual effects on health in multilevel studies, it is important to adjust for individual characteristics associated with the outcome of inter obesity in mid late life, this research also tests whether individual adult characteristics, such as educational attainment and marital status, moderate the effects of early life disad vantage on weight gain. Evidence that individual adult circumstances fully mediate the relationship between childhood disadvantage and later life weight would support the pathway model linking early life events with later health outcomes. In contrast, evid ence that the inclusion of adult circumstances, including neighborhood context, does not fully account for the relationship between childhood disadvantage and later life BMI would support the critical period model. Figure 1 1 presents a conceptual model that links early childhood experiences with adult BMI. In the pathway model that links childhood disadvantage to adult BMI, childhood disadvantage leads to adult characteristics (lower education, lower income) that in turn influence health behaviors such a s smoking. These health behaviors contribute directly to adult BMI. Adult characteristics may also influence the type of neighborhood one lives in. Higher individual income may lead to residence in a neighborhood with more food choices and green spaces, di rectly influencing adult BMI. Most previous research which tested for a direct association between childhood disadvantage and adult BMI adjusted only for individual level characteristics. Failure to

PAGE 31

31 include neighborhood characteristics omits an important p athway linking childhood disadvantage to adult BMI. If a direct association remains between childhood conditions and adult BMI after controls for adult and neighborhood socio demographics, the critical period model is supported. Dissertation Overview This dissertation is composed of seven chapters, the first chapter being this general introduction of the life course perspective and life course epidemiology. Chapter 2 reviews the literature linking childhood conditions with adult BMI and neighborhood charac teristics with BMI. The link between childhood conditions and adult weight has been explored since at least the 1960s, but since 1998 there has been a rapid increase in the number of publications on the topic. Between 1965 and 2008, approximately 78 public ations have discussed in some manner the links between childhood social conditions and later weight or fatness into adulthood (Parsons, Powers, Logan, and Summerbell 1999; Senese, Almeida, Fath, Smith, and Loucks 2009) Although not reviewed in as great a depth, many more artic les have focused on the link between childhood conditions and childhood weight. These publications have used a variety of indicators of childhood socioeconomic status, and have conducted analyses on a variety of study populations. This literature is review ed and discussed in relation to the present research. Chapter 3 discusses the data and methods used to answer the primary research questions. Chapters 4, 5, and 6 present results of the analyses. Chapter 4 tests for an association between childhood conditi ons and individual and adult BMI. Chapter 5 tests for an association between neighborhood characteristics and adult BMI. Chapter 6 tests for an association between childhood conditions and adult BMI, including adjustment for neighborhood characteristics. E ach chapter is designed to

PAGE 32

32 motivate the subsequent analysis. Chapter 7 brings together the results of the analytical chapters, and summarizes the main contribution, and offers suggestions for future research.

PAGE 33

33 Figure 1 1. Conceptual Model Linking Childhood Conditions with Adult BMI

PAGE 34

34 CHAPTER 2 DISADVANTAGE AND OBE SITY The term childhood disadvantage, as used in this research, is a specific example of the general concept social status Social status is a multi dimensional concept including both ascribed and achieved statuses. Ascribed social statuses include race, gender, and age, while achieved social statuses, typically measured as socioeconomic status (SES), include income, education, and occupational status. The use of a global social status measure such as SES is limited, however, because specific indicators of SES may be more important for different outcomes, and the sole use of SES excludes other important measures of status (Alwin and Wray 2005) Despite the regularity with which income, education, and occupation are defined as the core components of socioeconomic status, there is a lack of conceptual clarity on the essential nature of social statu s, nor any theory in the biomedical literature to guide the use of specific SES measures (Oakes and Rossi 2003; Pollitt, Rose, and Kaufman 2005) Other important measures of status have been hypothes ized to include material capital, human capital, social capital, and cultural capital (Oakes and Rossi 2003) There are thus multiple dimensions of social status with which to measure stratification of childhood environments. By themselves, these dimensions of social status, such as experience is formed by the effect of occupying the intersections of multiple social statuses (Williams, Mohammed, Leavell, and Collins 2010) Health and health disadvantage inherent in different ba (Adler and Stewart 2010) This dissertation examines the long term health effects of occupying

PAGE 35

35 disadvantaged social statuses as a child. Because social status is a complex and multi dimensional constru ct, life course epidemiological research should be wary of the ability of single manifest measures of social status to adequately measure childhood disadvantage. This research uses a variety of childhood measures to capture as thoroughly as possible child hood circumstances. There is strong evidence that disadvantaged childhood environments lead to an increased risk of childhood obesity, which in turn is associated with adult obesity (Guo and Chumlea 1999; Singh, Mulder, Twisk, Van Mechelen, and Chinapaw 2008) The remainder of this chapter reviews the link between disadvantaged childhood environments and childhood obesity disadvantage before turning to the link between early disadvantage and later life weight. Because neighborhood characteristics potentially play an important role mediating the relationship between childhood conditions and adult weight, the literature linking neighborhood characteristic s with adult health and adult weight is also reviewed, with an emphasis on the risk of endogeneity when examining neighborhood characteristics across time. Childhood Obesity and Early Disadvantage Before linking childhood disadvantage with childhood obesi ty, causes and trends of childhood obesity should be defined. The lack of a childhood weight measurement is an important limitation of the data used in this project. As the primary contribution of the research is to explain residual associations between ch ildhood conditions and adult weight, any omitted variables in the models that could plausibly explain the link should be considered as well. The link between childhood weight and adult weight is a strong candidate to explain any residual associations betwe en childhood disadvantage and adult weight. If indeed this link explains the residual association, it suggests a critical

PAGE 36

36 period of development insofar as childhood weight exerts a lasting influence on adult weight. The number of adolescents who are overw eight or obese approximately doubled between 1980 and 2000 (Deckelbaum and Williams 2001) The succeeding decade from 2000 2010 also witnessed a significant increase in obesity prevalence for males aged 2 19, although no such increase occurred among females. The current (2009 2010) prevalence of childhood and adolescent obesity in the US is 16.9%, an insignificant increase from the 16.8% estimated prevalence of obesity in 2007 2008 (Flegal, Carroll, Ogden, and Curtin 2010) Alt hough the past thirty years have seen a rapid increase in childhood obesity, the rate of increase may have thus slowed during the past five years (2007 2012). What accounts for the rapid increase in rates of childhood obesity between 1980 through the early 2000s? Current explanations identify calorie rich diets coupled with decreased caloric expenditure as the root cause of obesity. Thus in order to explain the increase in obesity rates among children (and other groups), factors which have disrupted the ba lance between calorie consumption and expenditure must be identified. Endogenous factors, genetics, and a changing social environment are three possible contributors to increased childhood obesity. Endogenous factors include endocrinological or neurologica l syndromes that can lead to increases in weight, however these factors only account for 5% of the obese population (Zakus 1983) Although endogenous endocrinological factors do not account for increasing obesity rates, environmental chemicals that can disrupt normal adipogensis dur ing human development, known as

PAGE 37

37 obesity (Grn and Blumberg 2006) Nicotine exposure in utero is one example of an environmental obeseogen. One study found that prenatal nicotine exposure in rats le d to increased adiposity and changes in modulatory function of perivascular adipose tissue (Gao, Holloway, Zeng, Lim, Petrik, Foster, and Lee 2005) Another possible obesogen is Triflumizole (TFZ), a fungicide not previously identified as toxic or carcinogenic (Li, Pham, Janesick, and Blumberg 2012.) The effects of nic otine and TFZ exposure in mice on obesity demonstrates the potential for environmental chemicals to disrupt the endocrinological regulation of obesity, but it is not yet known to what degree exposure to these chemicals influences human adiposity. Nor is it known if increased exposure to obesogens at the population level occurred during the same time frame that obesity rates in the US began to increase. A second potential source of disruption to the balance between calorie consumption and expenditure is gen etics. Genetics are an important contributor to an studies of twins raised in the same household have estimated the genetic contribution to body mass index to be between 64 to 84 percent (Stunkard, Harris, Pedersen, and McClearn 1990) In studies of adopted children, body mass index class (thin, median, overweight, obese) was significantly associated with biological parents, but not with the adoptee weight class. This association was constant through all weight classes, suggesting that genetics play a strong role in weight across all weight levels (Stunkard, Srensen, Hanis, Teasdale, Chakraborty, Schull, and Schulsinger 1986) However, in order for genetics to contribute to increasing rates of obesity between the 1980s and 2000s, genes themselves should have changed. It is unlik ely that genes themselves have sufficiently changed in the past thirty years to explain

PAGE 38

38 much, if any of the recent increases in obesity. The impact of genes may have changed however, and the nature of this change can provide clues to which changes in the e nvironment have contributed to increasing obesity rates. A key measure of genetic contribution to obesity is the correlation between it suggests that something in t he shared family environment has changed. However, if availability of snacks in schools (Anderson, Butcher, and Schanzenbach 2007) Anderson and Butcher et al. (2007) find that the correlation in BMI between parents and children has increased between the early 1970s and early 2000s, with increases in in increased intergenerational elasticity in BMI is evidence of the increased importance in shared genetic/environmental factors between parents and children. The se shared factors likely include changes in food consumption and energy expenditure such as increases in fast food consumption and sedentary leisure activities that revolve around electronic screens such as television (Anderson and Butcher 2006) Although this genetic, or changing enviro nmental factors) contribute to childhood obesity or adult obesity in later life, the analysis does differentiate between an advantaged and disadvantaged childhood environment. There is strong evidence that disadvantaged childhood environments lead to an in creased risk of childhood obesity, which in turn is associated with adult obesity (Guo

PAGE 39

39 and Chumlea 1999; Singh et al 2008) Earlier research on childhood adiposity in developed countries came to mixed conclusions about the direction of the relationship. State Nutrition Survey of 1968 and 1970 and the National Health and Nutrition Examination Series (NHANES I) of 1970 1974 (Garn, Hopkins, and Ryan 1981; Garn and Clark 1974; Garn and Ryan 1981) The income related reversal of fatness in the female refers to the fact that poor fem ale youth are leaner than their affluent equivalents. The association reverses itself in adulthood however, as poorer women tend to be heavier than more affluent women. More recent research has emphasized an inverse association between childhood socioeconomic status and childhood weight for both sexes in developed countries. In an early review of the literature, an inverse association between family income and childhood adiposity was found in 36% of studie s and a positive association was found in 26% of the studies (Sobal and Stunkard 1 989) In more recent review of research conducted between 1990 and 2005, an inverse association between income and childhood adiposity was found in 42% of studies. No studies found a positive relationship between income and childhood adiposity (Shrewsbury and Wardle 2008) Among children from low income households between 1999 2004, approximately 18% were obese, compared with 15% of children from the overall population in the early 2000s (Anderson, Butcher, and Schanzenbach 2007) .Between 2005 and 2008 among all boys aged 2 19 years, obesity prevalence was 11.9% in households with a poverty income ratio (PIR) 350% or greater (Ogden, Lamb, Carroll, and Flegal 2010) The prevalence of obesity nearly doubles to 21.1% in households with a PIR of less than

PAGE 40

40 130%. The inverse association between PIR and childhood obesity is stronger for non Hispanic whites than other groups. Among non Hispanic blacks and Mexican Americans, no significant association between PIR and obesity was found. Not only are children from lower income families more likely to be obese, but a disproportionate share of the increase in childhood obesity has occurred within low socioeconomic status households. Between 2003 and 2007, obesity prevalence increased by 10% in the overall population, but increased by 23 33% in children from lower status household. Interestingly, although previous research has shown no significant association between income and obesity in the Mexican American population, disproportionate increases in obesity have occurred between racial ethnic gr oups (Singh, Siahpush, and Kogan 2010) Hispanic children and Hawaiian/ Pacific Islanders aged 10 17 in particular have experienced an increase in obesity: in 2003 the prevalence of obesity among Hispanic children was 18.85%. By 2007, the prevalence had increased to 23.42% (Singh, Siahpush, and Kogan 2010) Obesity prevalence among children declined in o ther racial groups. Prevalence of obesity among Asians declin ed from 13.36% to 8.66% decrease. The various patterns of increasing and decreasing prevalence between lower income families and racial ethnic groups demonstrate the multiple competing trends in obesity. Children raised in lower income communities are also more likely to be obese than those raised in more advantaged communities. In one study that compared prevalence of obesity between neighborhoods with varying levels of median household income, researchers found that the prevalence of childhood obesity decreased as median household income increased (Eagle, Sheetz, Gurm, Woodward, Kline Rogers,

PAGE 41

41 Leibowitz, DuRussel Weston, Palma Davis, Aaronson, Fitzgerald, Mitchell, Rogers, Bruenger, Skala, Goldberg, Jackson, Erickson, and Eagle 2012) .The range of prevalence of obesity between neighborhoods in the study was substantial from a low of 9.6% to a high of 42.8%. Such large differences between neighborhoods may be a function of individual circumstances (neighborhood income), as well as neighborhood characteristics. At the family level, children from low income households ate fried food twice as often watched daily television and played video games three times longer, and exercised less frequently. At the neighborhood level, differences in these behaviors may be due to availability of healthy food and green spaces for recreation. Other research has su pported the idea that income differences are associated with differential access to green spaces, which leads to differential levels of physical activity and BMI of children (Evans, Jones Rounds, Belojevic, and Vermeylen 2012) Although this project does not e xplicitly test for a relationship between childhood weight and adult weight, it is an important possible pathway that links childhood disadvantage with adult weight Charney and Goodman (1976) provided early evidence of a link between infant weight and adul t weight. Using data from 366 subjects aged between 20 and 30 years of age, infants whose weight was at the 75th percentile for their height were significantly more likely to be obese. This association was independent of social class, education al level, an d parental weight. A more recent study that investigated the link between childhood weight and weight in adulthood found that about a third of obese preschool children were also obese as adults, and that the risk of obesity was greatest for children at the highest levels of obesity (Serdula, Ivery, Coates, Freedman, Williamson, and Byers 1993) In a 2008 systematic review of the literature on

PAGE 42

42 the tracking of overweight and obesity from childhood to adulthood, all studies reported a consistently positive link between childhood overweight and obesity and adult overweight and obesity status (Singh et al. 2008) Early Disadvantage and Adult Obesity A number of early life conditions are associated with later adiposity outcomes, including genetics, intra uterine growth, social factors, and early physical and dietary behaviors (Barker, Robinson, Osmond, and Barker 1997; Dietz and Gortmaker 1985; Whitaker, Wright, Pepe, Seidel, and Dietz 1997). The relationship between early disadvantage and adult obesity is of special interest because obesity is thought to be t he primary mechanism that links childhood disadvantage with increased risk of cardiovascular disease (Galobardes, Lynch, and Smith 2008; Galobardes, Lynch, and Davey Smith 2004; Senese et al. 2009) Increases in obesity and cardiovascular disease prevalence have helped spur increased research on their risk factors (Heidenreich, Trogdon, Khavjou, Butler, Dracup, Ezekowitz, Finkelstein, Hong, Johnston, Khera, Lloyd Jones, Nelson, Nichol, Orenstein, Wilson, and Woo 2011; Mokdad, Ford, Bowman, Dietz, and Vinicor 2003) Since 1998, the number of articles on the link between childhood socioeconomic position and adult weight has dramatically increased. Prior to 1998, a total of approximatel y thirty p apers examined the relationship between social factors in childhood and later weight (Parsons, Powers, Logan, and Summerbell 1999). Between 1998 and 2008, forty eight publications examined the link between childhood socioeconomic position and adu lt weight (Senese et al. 2009). The majority of research suggests a negative association between childhood socioe conomic status and adult weight children from disadvantaged homes are more likely to be obese adults (Parsons, Powers, Logan, and Summerbell 1999). The

PAGE 43

43 relationship between social factors and adult obesity is comparable in strength to more physiological measures of childhood characteristics. In a 2012 review of over 135 studies that examined early marker s of adult obesity, only seven of 42 variables were found to be either possible or probable makers of later adult obesity. Paternal occupation ( often used as a proxy for family socioeconomic status) was found to be a probable early marker of obesity, along with maternal body mass index, childhood growth patterns, and childhood obesity (Brisbois, Farmer, and McCargar 2012). In all fifteen studies reviewed, there was an association between a lower paternal occupation and adult obesity. Other childhood variab les with more inconsistent associations with later weight included parental diabetes, parental weight and BMI, and early malnutrition exposure. Although a more consistent association with increased adult weight was when both parents had higher BMI, this as sociation was still less consistent than paternal occupation. The relative strength of association between paternal occupation and adult weight suggests that social factors during childhood are at least as influential as genetic factors. Another example o f the relative strength of the association between social factors and adult weight compared to more physiological measures comes from a national UK cohort study designed to test the predictive of chil dhood obesity for adult obesity. C hildhood obesity was f ound to be a relatively poor predictor of adult obesity; however the social class and education of parents were strongly predictive of respondent's later weight. At 36 years of age, 33% of males whose parents were in non manual classes were either overwei ght or obese. In contrast, 46.2% of males whose parents were from manual classes were either overweight or obese. A similar association was found for

PAGE 44

44 females (Braddon, Rodgers, Wadsworth, and Davies 1986). The relative strength of paternal occupation has a lso been found in other study populations. After adjustments for age an ad ult social class, Wannameethee and Whincup et al. (1996) found that having a father in a manual occupation was associated with an increased likelihood of obesity and systolic blood p ressure, and HDL cholesterol. Of all the risk factors, paternal occupation was most strongly associated with obesity. Much of the research on childhood conditions has been conducted in Europe, but there are some early examples of such research in the US. One of the earliest findings linking childhood socioeconomic conditions with adult obesity found an inverse relationship in both men and women (Goldblatt 1965). Goldblatt (1965) also found an inverse association between one's own socioeconomic status and o besity, with remarkably strong findings for its era (earlier research on the relationship between socioeconomic status and obesity tended to have more inconsistent findings). In a sample of 1,660 adults from Manhattan, low status women were six times more likely to be obese than high status women, with similar but less pronounced patterns among men. The Troms Heart Study was another early attempt to the link childhood conditions and adult coron ary disease, including weight. This study collected data on a pproximately 14,652 men and women born between 1925 and 1959 and living in the municipality of Troms. Using data from the Troms study, Arnesen and Forsdahl (1985) examined the link between childhood poverty and coronary risk factors. Poverty during chil dhood was associated with higher total cholesterol in women but not men, and poverty was not associated with body weight for either men or women. Poverty was

PAGE 45

45 associated, however, with lower adult height (Arnesen and Forsdahl 1985). Failure to account for t he lower height of respondents who grew up in poorer condition may explain the lack of association between poverty and weight. Poverty was assessed by asking respondents "What were the living conditions in your family during childhood" and to rate them as "very good", "good", "difficult", or "very difficult." The measurement of childhood conditions using a single measure, poverty, is a limitation of their research. Although research from t he Troms Heart Study failed to find an association between childhood conditions and adult weight, a nother study investigating early childhood conditions and coronary risk factors as an adult did find an association with adult BMI. In a study of 5,645 males from west ern Scotland, Blane and Heart et al. (1996) tested for ass ociations between childhood conditions and cardiovascular risk factors. Social class during childhood was measured using father's main occupation, and social class during adulthood was measured using one's own occupation as an adult. Interestingly, father' s social class was found to be more strongly associated with adult BMI than one's own social class. For all other risk factors (diastolic blood pressure, serum cholesterol, level of physical exercise, smoking, and forced expiratory volume), one's own soci al class was more strongly associated with the risk factor than father's social class (Blane, Hart, Smith, Gillis, Hole, and Hawthorne 1996). Behavioral risk factors thus seemed more strongly associated with present socioeconomic status rather than one's e arlier status; physiological factors (including BMI) were more strongly associated with earlier conditions. The findings of Blane et al. (1996) are consistent with a critical period of development for adult BMI.

PAGE 46

46 The 1958 British Cohort study has proven a rich source of data linking childhood social class with childhood and early adulthood weight Power e t al. ( 1997 ) found evidence that social class differences in weight increased from early childhood to young adulthood in both men and women. Compared to ch ildren from non manual households, children from manual households were more likely to be obese (7% versus 3%), and more likely to remain either overweight or obese into young adultho od (Power, Manor, and Matthews 1999 ). ) research suggests that the effect of childhood social class continues into adulthood, e vidence on the timing of the influence of childhood social class is mixed as others have found the influence of paternal occupation to lessen over time (Lasker and Mas cie Taylor 1989; P ower, Matthews, and Manor 1998) Further work using the 1958 British Cohort study found that the association between social class at birth and weight persisted from age 23 to age 33 (Power, Hertzman, Matthews, and Manor 1997). An invers e association between childhood conditions and adult weight has been found in study populations quite distinct from the United States population. In another large European study linking childhood conditions t o adult weight among males, Ras mussen and Johans son (1998) used data from the national Swedish Medical Birth Registry for men born between 1973 and 1976, which was then linked to the national Military Service Conscription Registry for 1990 1996. Approximately 193,000 children were followed up through th e age of eighteen. Although the primary research goal was to test the relationship between birth weight and birth length and adult body weight, researchers also tested the relationship between maternal education and adult weight. In logistic regression models, men born to women with less than a college degree were

PAGE 47

47 more likely to be overweight. Men born to women with compulsory education only or slightly more than compulsory education only were more likely to be obese relative to men whose mother had a college education. In a large study of 19 year old Dutch ma les, those with fathers in a manual occupation were at an increased risk of obesity (defined as 120% or more of the standard weight for height). What distinguishes Ras mussen and Johansson (1998) study from more recent studies in the United States populati on is that the overall obesity prevalence in the study population was quote low -1.83%, a marked contrast compared to modern obesity rates in the United States. As the above review of childhood disadvantage and life course outcomes research suggests, the m ost common measure of childhood socioeconomic status in studies both prior to 1998 and after 1998 is paternal occupation (Senese et al. 2009) Out of 26 studies linking childhood conditions with adult weight since 1998, approximately 70% have used paternal occupation. In contrast, only 23% (n=6) of studies have included measures of either maternal or paternal education. In studies that use parental education rather than occupation as an indicator of childhood conditions, the evidence for an association with adult BMI is mi xed. Ball and Mishra (2006) found however, this association was partially attenuated once other factors, including social mobility, were included in the analysis. Bielicki a nd Szklarska (2000) did not find an Slightly stronger evidence of a link between parental education and adult weight comes from Laaksonen and Sarl io La hteenkorva (2004), who ex amined the relationship in a mixed sex study population from the Helsinki Health Study. Their results suggested that

PAGE 48

48 socioeconomic position as a child, including paternal education, was inversely associated with obesity among females. This association rem ained after adjustments for adult education and occupation. Among males, however, adjustment for adult socioeconomic position appeared to explain the association between childhood conditions and adult obesity. In one of the few studies to investigate the relationship between maternal education and adult weight, maternal education was inversely associated with young adult BMI among females Chmara 2007) Trotter and Bowen (2009) also found an association between maternal education and adult weight in a sample of adults from Los Angeles. Analyses suggest that a maternal high school diploma was associated with an 8% decrease in median adult weight for whites, 6% decrease among Hispanics, and 11% decrease among Blacks after adjustment for adult socioeconomic position (Trotter, Bowen, and Beresford 2010) Another study indirectly and Retirement Study as part of a larger project that examined life course pathways to adult onset diabetes (Best, Hayward, and Hidajat 2005) Best and Hayward at al. (2005) both education was also found to be significantly associated with lowed prevalence of overweight or obesity among adult women in the HRS cohort. No other measures of childhood con ditions were significant. As Chapter 4 will discuss, the findings of Best and

PAGE 49

49 In studies that link parental socioeconomic with BMI of their children in adulthood, a possible confound is the weight status of the parents. In an attempt to control for the influence of adult weight, Tea s dale and Sorenson et al (1990) identified 2,015 adoptees in Denmark and collected data on their biological parent's social class, adoptee parent's social clas s, and their own current BMI and social class. Social class of the biological parents was available in the adop tion records, which recorded the occupation of the biological father. Occupation of both the father and respondent was measured on a scale from 0 7, with 0 indicating unskilled manual work and 7 indicating professional work. Multivariate analyses found evidence of independent contributions to BMI from the biological father's occupation, adoptee father's occupation, and the respondent's own occupati on. In each case, occupation was inversely associated with BMI. This evidence suggests that the childhood environment contributes to adult weight independently from genetic factors. To summarize the importance of childhood conditions to later weight outcom es, the majority of research has found an inverse association between childhood conditions and adult weight children from disadvantaged environments are more likely to be overweight or obese as adults The further back in time one goes, the less consistent this association is in the literature, but more recent research seems to have reached a consensus on the direction of the relationship. Recent research also generally finds that the association between childhood socioeconomic conditions and adult weight i s stronger in females than in males. Adult socioeconomic status is more likely to explain pattern of previous research, it is more likely that we will find a resid ual association

PAGE 50

50 between childhood conditions and adult BMI among females than males. A final common feature of most previous research is the measurement of childhood conditions with paternal occupation. Fewer studies use parental educational achievement, a nother area where research findings have been mixed. The use of multiple indicators of childhood socioeconomic position, including paternal occupation as well as paternal and maternal education, is an important strength of this research. Conceptual and Em pirical Links between Neighborhood Context and Adult Health The processes producing variation in health range from micro level individual variation of personal resources to large scale structural variation in cultural, economic and social circumstances (Macintyre 1994) The neighborhood exists as an important mezzo level aspect of social structure. The consequences of neighborhood context have been of interes t to researchers since the 1920 s, as growing urbanization increased the visibility of poor urban living conditions ( Sampson 2003b) More recently, the deinstitutionalization of inner city neighborhoods the process by which job opportunities decline, working families move to better neighborhoods, and schools, housing and community organizations deteriorate has been of concern to sociologists (Wacquant and Wilson 1989) .The de institutionalization of neighborhoods has not only influenced the opportunity structures within city neighborhoods, but has also resulted in neighborhoods stratified by social characteristics including socioeconomic status, family structure, and ra ce (Sampson 2003b) Of particular interest in this re search is the role of neighborhood context in the link between early life disadvantage and later health outcomes. Although the definition of a neighborhood itself can be controversial, the use of census tract data, as proposed in this research, can be

PAGE 51

51 usef ul in exploring neighborhood context effects due to their systematic collection and generalizability (Diez Roux 2001) Specific mechanisms linking neighborhood context to i ndividual health have been hypothesized to include the social cohesiveness of the neighborhood, land use patterns, psycho social context, strength of organizational ties within the neighborhood, health selectivity, positive peer interactions, and material resources available, such as parks for recreation (Katz, Kling, and Liebman 2000; Robert and Reither 2004; Sampson 2003a; van Lenthe, Martikainen, and Mackenbach 2007) Regarding obesity, it has been hypothesized that disadvantaged communities may increase the risk of obesity in at least two ways: a lack of resources such as l ighted streets and healthy food options; and a psychosocial context that promotes obesity through mechanisms such as chronic stress (Robert and Reither 2004) .These neighbor hood characteristics intersect with individual level characteristics to help explain social variation in health and obesity. The individual characteristic of interest in this research is the experience of childhood disadvantage with the primary objective of testing whether neighborhood context mediates any association between childhood disadvantage and obesity. Previous research has found consistent yet modest associations between neighborhood disadvantage and negative health outcomes, including mortality, health behaviors, po orer self reported health, low birth weight, and increases in body weight (Pearl, Braveman, and Abrams 2001; Pickett and Pearl 2001; Robert and Reither 2004; Ross and Mi rowsky 2001) Although findings of significant contextual effects are consistent for most health outcomes other than mental health, contextual effects are relatively

PAGE 52

52 modest compared to individual level effects for morbidity and BMI (Pickett and Pearl 2001; Robert and Reither 2004) Despite the relatively modest effect sizes of neighborhood characteristics, a number of studies have found significant associations between neighbor hood characteristics and adult BMI. Do and Dubowitz at al. (2007) found a significant association between economic advantage and lower BMI in a nationally representative sample of men and women. Of note was their finding that ethnic enclaves were not assoc iated improved BMI outcomes, given the association between ethnic enclaves and improved outcomes for other dimensions of health. Robert and Reither (2004), using four waves of data from the A merican Changing Lives Survey, examined the ext ent to which n eighborhood characteristics account for racial disparities in obesity among women. Their results suggested that neighborhood disadvantage measured at baseline wave was associated with B MI, but only marginally explained racial differences in BMI and did not explain differences in BMI over time. Other research has also found an association between lower neighborhood socioeconomic status and increased weight among Black females (Coogan, Cozier, Krishnan, Wise, Adams Campbell, Ros enberg, and Palmer 2010) Unlike Robert and Reither (2004), Coogan and Cozier at al. found neighborhood socioeconomic status to be associated with weight gain not just weight at baseline suggesting that selection into certain neighborhoods by BMI canno t fully explain the relationship. Grafova and Freedman e al. (2008) used data from the 2002 wave of the Health and Retirement Study to test whether economic, built environment, and social characteristics were associated with the likelihood of overweight o r obesity in the HRS

PAGE 53

53 cohort. Their research is particularly relevant because they used the same data and similar measures of neighborhood characteristics used in this study. Rather than use specific measures of neighborhood socioeconomi c status, Graf ova an d Freedman e t al. (2008) use d a scale of economic disadvantage and economic advantage based on percentage of population over 65 years in poverty, percentage of households receiving public assistance income, unemployment rate, percentage of homes without a vehicle, and percentage of population that is black. Their economic advantage scale uses the upper quartile value of owner occupied housing units in the census tract, percentage of families with an annual income over $75,000, and percentage of adults with a college degree. Some of these measures o verlap with measures used in this analysis (unemployment rate, percent population Black, percentage of adults with a college degree). Using these scales, they found that neighborhood economic advantage was associa ted with a reduced likelihood of being obese in men and women. When they substituted the three specific components of economic advantage, each were associated with a reduced likelihood of obesity. To summarize the importance of neighborhood characteristics in this research, it is hypothesized that neighborhood characteristics explain the residual association between childhood conditions and adult weight. Neighborhood characteristics have the potential to influence weight outcomes independent of individual c haracteristics, as the reviewed research suggests. If indeed adult neighborhood characteristics explain the residual association, it suggests that the pathway model/cumulative exposure model is more consistent with the evidence than the critical period mod el, even if the exact

PAGE 54

54 mechanisms through which neighborhood characteristics influence weight cannot be known from the analyses. Causal Ordering of Health and Socioeconomic Status Much of the research linking socioeconomic status with health assumes a part icular causal ordering, with socioeconomic status influencing health. This dissertation makes a similar assumption, but because of its longitudinal approach and relatively comprehensive definition of childhood disadvantage, does not make the assumption na ively. Consideration of t he causal ordering of health and socioeconomic status should not be viewed as the research domain of a small sub discipline of mainstream sociology. Sociology has a long tradition of exploring the production, reproduction, and poss ible fun ctions of class stratification including Marx, Weber, and Durkheim, and classic treatments such as Davis and Moore (1945). Thus insofar as health plays a role in class formation, health has the potential to play a critical role in t he reproduction of social conceptualization of disadvantage allows for the possibility that poor childhood health may influence later adult socioeconomic status, which could in turn affect adult weight status. How might health affect socioeconomic status? Income and wealth are both critical health protective resources greater levels of income and wealth provide greater opportunities to purchase healthier food, access health care services, move to areas more conduciv e to individual health, allow for greater social participation, and (wealth particularly) buffer individuals from the expenses of poor health ( Stansfeld and Marmot 2002, Smith 1999). Yet income and wealth as measures of socioeconomic status also exemplify the effects of individual health on socioeconomic status Individuals who become unhealthy may have to leave

PAGE 55

55 the workforce for a time, affecting not only current income but future income as well if re entry is difficult. Further, individuals in poor health may have to draw on accumulated wealth to meet expenditure requirements for treating their health condition. Smith (1999 ) attempts to isolate the effect of health on economic resources via an examination of the effect s of the onset of new chronic conditio ns (a health shock ) to expenditures. S mith (1999) finds that the economic impacts of severe health shocks are substantial, with a 15% decline in the probability of remaini ng in the workforce (Smith 1999). Further, a forced to take out loans and liquidate assets to meet new expenditure requirements (Smith 1999). Education al attainment is another critical health protective resource that can be affected by poor health the occupational status of individuals, and increases the productive efficiency of individuals such that they ma (Adams 2002; Lynch and Kaplan 2000; Ross and Mirowsky 1999) Kitigawa and Hauser (1968) examined the link between education and mortality, spurring research into the links between socioeconomic statu s and health, and education has become the most commonly used indicators of socioeconomic status in social epidemiological research (Winkleby, Jatulis, Frank, and Fortmann 1992 ) as a measure of socioeconomic status is that individuals usually achieve maximal educational attainment earlier in their lives. This feature helps to shield education from at least one issue of causality that threatened the causal link between income and

PAGE 56

56 wealth with health: health shocks cannot deplete the store of educational resources once maximal educational attainment has been achieved. Educational attainment is not completely shielded from issues of endogeneity, however, as poor health in childhood is strongly associated with reduced educat ional achievement (Behrman 1996) Palloni ( 2006 ) finds that non traditional traits (height and score on a scale of maladjustment) are significantly related to low birth weight and have significant effe c ts on educational attainment -t hus poor childhood health has a measurable effect not just on occupational mobility but on educational attainment through both traditional and non traditional traits. Occupation is another socioeconomic status indicator affec ted by early childhood conditions. Palloni (2006) found that after removing the effect of childhood health, there health on mobility, Palloni estimates that (2006) the proportion of individuals who A second threat to the causal direction of education and health is spuriousness -other characteristics may explain both educational attainment and he alth status. Individuals who are more efficient at accumulating human capital while contemporaneously enrolled in school may already be healthier and have other characteristics that lend them to defer gratification leading to better educational outcomes an d health outcomes (Palloni 2006). Unobserved genetic heterogeneity is another possible source of spuriousness resistance to disease and ability to work (Currie 2008). Researchers have attempted to account for unobs erved variables that explain both education and health outcomes using an instrumental variable an approach (Adams 2002). Using date of birth

PAGE 57

57 measured in quarter years, Adams (2002) found evidence of that education is causally connected to health out comes i n later life, and that the relationship is not entirely spurious. Taking into account the sociological and economic literature which accounts for a dynamic relationship between individual health and socioeconomic status, the current state of research on th e causal relationship suggests that the overall causal flow from socioeconomic status to health appears to be larger than that from health to effect of childhood health in explanations linking better socioecon omic status with better health (Nicolas and Thierry 2011; Palloni 2006). This dissertation, rather than simply controlling for early childhood health, tests for an association between childhood health and adult soci oeconomic and health outcomes. Acknowledging the causal effect of health on socioeconomic status, there remains the complicating issue of intergenerational transmission of health via parental socioeconomic status. Poor childhood health affects later socioe conomic status, which in turn affects the health of the next generation. Thus the issue of individual health becomes an issue of that the socioeconomic status of his mother, father, family, and neighborhood. The consequ ences of neighborhood context on health have been of interest to sociologists since the 1920 s, as growing urbanization increased the visibility of poor urban living conditions (Sampson 2003b). Neighborhood context can affect individual health in numerous w ays, including both the material resources available in a community, the psycho social context of the community, and health selection patterns

PAGE 58

58 (Katz, Kling, and Liebman 2000; Robert and Reither 2004; Sampson 2003a; van Lenthe, Martikainen, and Mackenbach 2 007). Within the framework that the flow of causality is from neighborhood context to individual health, previous research has found associations between neighborhood disadvantage and higher mortality, poorer health behaviors, low birth weight, and increas es in body weight (Pearl, Braveman, and Abrams 2001; Pickett and Pearl 2001; Robert and Reither 2004; Ross and Mirowsky 2001). Yet it is still possible that the mechanism linking neighborhood context and individual health is not purely causal. Selectivity may explain part of the association between neighborhood context and health -people with better health may choose to live in areas with higher socioeconomic status indicators, and those in poorer health drift towards lower socioeconomic neighborhoods. (Du ncan and Raudenbush 1999; Sampson 2003a) Current research suggests that issues of selectivity vary by health outcome of interest. Issues of selectivity are more salient for individuals with two or more chronic diseases, who have an increased probability o f downward migration (van Lenthe, Martikainen, and Mackenbach 2007). However, with regards to BMI, van Lenthe and Martikainen et al. (2007) conclude that health selectivity cannot explain a significant portion of health inequalities between neighborhoods. Further evidence of a causal component in the relationship between neighborhood context and health comes from the Moving to Opportunity (MTO) program (Katz, Kling, and Liebman 2000). The MTO is a randomized housing mobility experiment involving 4,600 low income families who were given a chance to move to a high income neighborhood. Once the families elected to participate, they were assigned by lottery to three groups, allowing researchers to examine effects of neighborhood

PAGE 59

59 context on various outcomes, inc luding health and employment. Those who moved to lower poverty areas reported a general improvement in health status and greater feelings of peace (Katz, Kling, and Liebman 2000).

PAGE 60

60 CHAPTER 3 DATA AND METHODS Research questions within the life course perspective often have intensive data and analytical requirements as they typically cover extended periods of time. A number analysis of data over time within individuals, and the analysis of the influence of neighborhood characteristics require techniques to help account for the dependence of observations. Chapter 3 reviews the data, measures, and analyse s used during the cours e of this research. Data I ndividual data are link ed with census tract level data using three sources ; the Health and Retirement study, RAND HRS dataset (version J), and the RAND Center for Population Health and Health Disparities (CPHHD) Data Core Series. Data on childhood conditions come from the Health and Retirement Study, a national panel study that collects information on the economic, health, marital, and family status of the US population over the age of 50 who were not institutionalized at their first interview (HRS 2008) and sponsored by the National Institute of Aging (grant number NIA U01AG009740). es. It is one of the few nationally representative panel studies that collect information on a variety of dimensions, including health status and childhood disadvantage. Launched in 1992, the HRS has collected nine waves of data on a biennial schedule At the 1992 baseline, the HRS sample size was 12,652 respondents (7,704 households ) with a response rate of 81.7%. The observational unit in the HRS sample is an eligible financial unit, which must include at least one age eligible person, which could include

PAGE 61

61 married couples where only one or both partners were eligible for the survey. The sample is selected using a multi stage probability design with oversamples for blacks, Hispanics, and residents of Florida. At baseline, face to face interviews were conduct ed, with subsequent interviews being conducted over the phone or via mail. The number of participant reinterviews and response rates by wave for the HRS cohort are: Wave 5 (11,762, 85.4% ) ; Wave 6 ( 11,230, 86.6% ) ; Wave 7 ( 10,835, 86.4% ) ; Wave 8 ( 10,026, 88. 6% ) ; and Wave 9 ( 9,587, 88.6%). By 2008, 9.26% of the eligible sample was dropped before deceased. Approximately 2,345 (18.5%) of HRS respondents were deceased by 2008. Additional individual level data come from the RAND HRS dataset, a cleaned and process ed version of the raw data funded by the National Institute on Aging and the Social Security Administration (RAND 2010). Neighborhood characteristics come from the RAND Center for Population Health and Health Disparities ( CPHHD ) Data Core Series. This dataset compiles a range of analytical measures at the census tract, county, and Metropolitan Statistical Area based on the 1990 and 2000 Decennial Census Population and Housing Characteristics Data. Data is available for all census tracts in the United States, Distri ct of Columbia, and Puerto Rico Compiled measured include housing characteristics, population density, racial and ethnic population, as well as demographics and socioeconomic characteristics of an area (Escarce, Lurie, and Jewell 2011). All neighborhood measures used in the analysis are based on the 2000 Decennial Census data. Due to limited geographic data availability, data is only available from waves 2000 through 2008.

PAGE 62

62 Analytical Sample Data used in the analysis draws on the sub sample of HRS respondents born in 1931 through 1941 At baseline, 9,816 respondents (77.6% of the HRS sample) were in this birth cohort. By 2008 (Wave 9), 6,545 respondents from the HRS birth cohort completed core interviews. Information from respondents who completed interviews via proxy were included in the analysis. The proportion of core interviews done by proxy informants by wave are for the HRS cohort are: Wave 5 (8.6%); Wave 6 (9.0%); Wave 7 (7.7%); Wave 8 (5.5%); and Wave 9 (5.4%) For the analysis, data on time varying variables come from Waves 5 9 (2000 2008) due to the limited availability of census tract information. Prior to 2000, geographic data of respondents was linked to 1990 census tract definitions. Time fixed variables a re measured at baseline. Hierarchical Structure of the Data Data is structured hierarchically across four levels in analyses estimating adult BMI at a single point in time (2000) and five levels in analyses estimating changes in adult BMI from 2000 2008. In the four level structure individuals are clustered within financial units, financial units are clustered within census tracts, and census tracts are clustered within counties. In the five level structure, there is an additional level of clustering at the level of individual time points measures of BMI across time are clustered within individuals. Although the full hierarchy of data is four and five levels, analyses suggest that there is insignificant amount of variation in BMI at the county level (Do, Dubowitz, and Bird 2007). Following previous research using data from the Health and Retirement Study, this analysis does not adjust standard errors to account for clustering of individuals within financial units because analyses are stratified by sex (Gra fo va, Freedman, and Kumar 2007). Final analyse s for BMI at a single point in time

PAGE 63

63 are assumed to have a two level structure (individuals within cen s us tracts) and analyses estimating changes in BMI between 2000 and 2 008 are assumed to have a three level st ructure (time points within individuals within census tracts ). Measures The dependent variable in all analyses is BMI, coded as an interval level variable Analyses included childhood, adulthood, and neighborhood sets of variables, described in the following sections. Descriptive statistics of variables used in the analysis are shown in Tables 3 1 through 3 3. Table 3 1 summarizes the means and proportions o f childhood characteristics; Table 3 2 summaries the means and proportions of adulthood characteristics; and Table 3 3 summarizes the means and proportions of neighborhood characteristics. Dependent V ariable Body Mass Index The dependent variable used in a ll analyses is a continuous measure of BMI computed from self reports and direct measurements of height and weight. This measure of BMI has adjusted to correct for reporting bias. The inclusion of both self reported weight and direct measurements of weig ht in the 2004 and 2006 waves allows for the adjustment of self reported weight prior to 2004. Respondents tended to underestimate their weight, resulting in calculations of BMI that were too low. Larger biases are observed among heavier individuals. Self reports of BMI are adjusted using a regression technique similar to Cawley (2000). This method involves regressing the measured value of a variable on the reported value, then multiplying the OLS coefficient by the reported value to create an estimate of the true value (Cawley 2000) In the case reports of weight in 2004 were regressed on direct measurements of weight. Regression analyses were stratified by sex and BMI category

PAGE 64

64 (underweight, normal weight, overweight, or obese). Ba sed on regression results, self reports of BMI were adjusted for Waves 5 9, and these adjusted self reports of BMI were used in all reported analyses. Independent Variables Childhood Measures Measures of childhood circumstances come from HRS waves 1998 2 008. HRS occupation, whether or not their father ever lost their job, and childhood health. Mother grade of manner occupation was recoded into a dichotomous measures (1=laborer; 0 otherwise). O ccupations c lassified as a laborer included farming, forestry, fishing, mechanics and repair, machine/equipment operator, and construction trade. Occupations classified as non laborer included managerial specialty operation, professional specialty operat ion, sales, clerical, and health or food services. Paternal job loss was measured by asking respondents if before the age of 16, there was a time of several months or more when your father had no job. Childhood health was measured by asking respondents to consider their health from birth to age 16 and rate their health as excellent, very good, and mot 9.24 (Table 3 1) Approximately 52% of the sample reported having a father who was a laborer and 18% reported their father was out of wor k for several months or more. Two percent of the sample reported poor childhood health before the age of 16.

PAGE 65

65 Independent Variables Adult Measures Adult s ocio demographic variables include age, race, ethnicity, nativity, years of education, household income, and marital status. Age, race, ethnici ty, nativity, and years of education are treated as time fixed variables while household income and marital status are time varying. Race is coded using a three category dummy variable with White (reference), Black or Other as the categories. Hispanic ethn icity is treated separately from race, and is coded dichotomously (1=Hispanic; 0 otherwise). Nativity was measured at baseline and coded dichotomously (1=foreign born; 0 otherwise). Years of education is measured continuously as formal years of education a nd centered at its grand mean for analyses. Marital status is dichotomously married (1=married; 0 otherwise). Total household income is the sum of all income in a household including job earnings, pensions and annuities, social security or disability incom e, or other sources of income. Income is measur ed in nominal dollars and converted into 2008 dollars using the Consumer Price Index (CPI) for analyses. The grand centered mean of total household income is used in all analyses. Analyses also include measur es of a dult health and health behaviors These include functional status, self rated health, smoking status, health care utilization, and insurance status. Functional status is measured as the sum of Activities of Daily Living (ADLs) and Instrumental Acti vities of Daily Living (IADLs) that respondents report any difficulty performing. Higher scores indicate greater functional impairment. Self rated health is measured by asking respondents to rate their health as excellent very good good fair or Higher scores indicate better self rated health. Smoking status of

PAGE 66

66 by asking respondents whether or not they had visited a doctor in the past two years (1=yes; 0 otherwise) or hospital in the past two years (1=yes; 0 otherwise). Health insurance status was modeled using a three category dummy variable and measured by asking respond ents whether they were covered by a private insurance plan (1 if yes; 0 otherwise) or federal government health insurance program (1 if yes; 0 otherwise). Federal plans include Medicare, Medicaid, VA/CHAMPUS or other government insurance program. Responden ts who indicated that that had neither type of insurance coverage were defined as having no insurance (1 if yes; 0 otherwise). In models that estimate BMI in 2000, a time varying indicator for proxy status is included (1=proxy; 0 otherwise). Proxy status comes from the RAND HRS data and is derived directly from the Tracker file. In models that estimate BMI changes between 2000 and 2008, indicators are included for whether or not a respondent attrited, died, or moved between 2000 and 2008 Attrition is coded respondent is thought to be alive but did not respond; 0 if otherwise. Mortality is coded otherwise. 1 Whether or not a respondent mov ed between 2000 and 2008 is constructed from the merged individual and geographic dataset. If a respondent reported living in the same census tract at each is coded as 1 if a respondent moved to a ce nsus tract different from the one they lived in during the 2000 wave. The average age of the analytic sample is 62.3 years (SD=5.84) (Table 3 2). Approximately 74% of the sample is white, 14% black, and 9% Hispanic. 11% of 1 Attrition and mortality come from the Rw1IWSTAT variable in the RAND dataset, which is derived from the HRS Tracker file.

PAGE 67

67 respondents were born outside th e United States. The average number of years of education is 12.34 years and the average total household income is $61,128 in 2008 dollars. Across males and females, the av erage adjusted BMI is 29.19 (Table 3 2). Table 3 4 presents comparisons of adjusted and unadjusted weight and BMI values for respondents in 2004, 2005, and 2006. The average number of ADLs and IADLs is 0.39, smoked, 18% are current smokers, and 44% are for mer smokers. Most of the respondents (93%) had visited the doctor in the past two years, and 21% reported a hospital visit in the past two years. 12% of the sample had no insurance. Independent Variables Neighborhood Measures Data for neighborhood charact eristics comes from the RAND CPHHD data series which is composed of a wide range of measures including census tract characteristics for the year 2000. Measures of neighborhood demographics used in the analysis include proportion over the age of 65, propo rtion rural, proportion black, proportion Hispanic, proportion foreign born. The average percent of individuals over the age of 65 within a census tract is 14% and the average percent rural is 23%. The average percent community black percent Hispanic and percent foreign born are 15%, 12%, and 9%, respectively. Measures of neighborhood socioeconomic status include proportion unemployed, proportion of females with a college degree, proportion of males with a college degree, and proportion of families below the poverty line. One measure of the built environment, walkability, is also included in analyses. The measure of walkability used in analyses is street loops given the number of intersections in a census tract A higher score indicates greater street

PAGE 68

68 connectivity (i.e. a greater number of connections between one loc ation and another) The average percent of individuals unemployed within a census tract is 6%. The average percent of individuals with a college degree is 5% for both males and females and the average proportion of families below the poverty line is 9%. Analysis In order to test for the long term influence of childhood co nditions on adult weight status independent of adu lt and neighborhood socio demographics, analyses were conducted in three phases. Chapt er 4 examines the association between childhood conditions and adult BMI independent of individual adult characteristics. Chapter 5 tests for an association between neigh borhood socio demographics and adult weight, independent of individual characteristics. Chapter 6 brings together the analyses of Chapters 4 and 5 by testing for an association between childhood conditions and adult weight independent of both individual an d neighborhood socio demographics. All a nalyses are stratified by sex Analyses are unweighted because many of the attributes used in HRS selection weights are included in the models (Botoseneanu and Liang) When sampling weights are a function of independent variables included in a model (as is the cas unweighted ordinary least squares estimates are less biased and have smaller standard errors than weighted estimates (Winship and Radbill 1994 Hierarchical linear models are used to account for the dependence of observations within individuals and individuals within census tracts. F ailing to account for correlations between observations due to clustering can have a negative impact on regression models including biased estimates of the slope and standard error, often resulting in a standard error that is too small (Bryk and Raudenbush 2002).

PAGE 69

69 To account for the non independence of the errors and adjust standard errors, a t wo level hierarchical model is used to estimate adult BMI in 2000 2 Level 1 variables consist of individual level observations for all individuals. Level 2 variables include the characteristics of the census tract (j) from which individual (i) was sampled Cases with missing data at level 2 are excluded from the analysis. All continuous variables are grand mean cente red. Time is centered such that the estimated intercept reflects estimated BMI in 2000. The hierarchical models for estimated adult BMI in 2000 are modeled as follows: Level ( 3 1) Level Xkjt + Uoi ( 3 2) Level ( 3 3) In Equation 3 1, Y it represents BMI for individual i in census tract j oi is the person 1i represents the slope for the i th individual in census tract j X kit represents the k th time varying covariate for individual i in tract j ij represents residuals from person i s linear trajectory in tract j after controlling for the effects of X 1 k Equations 3 2 and 3 3 depict inter individual differences as functions of an intercept and non time varying covariates (e.g. gender, age, race, and childhood variables) and a level 2 residual ( U oi or U 1i ) Analyses for Adult BMI in 2000 Individual Characteristics The dependent variable in this model is a cross sectional measure of adjusted adult BMI in 2000, continuously coded. In Chapter 4, Model 1 a djusts for proxy status (time varying ) and childh ood conditions. Model 2 adds adult demographics without childhood 2 In some cases, standard errors were not available at the time tables were prepared. All tables report standard errors when available.

PAGE 70

70 c onditions. Model 3 adds childhood conditions and Model 4 adds adult socioeconomic status. Model 5 adds adult health behaviors. Analysis for Changes in Adult BMI, Individual Characteristics The dependent variable in this model is a longitudinal measure of adult BMI from 2000 2008. Variables are entered into the model in a fashion similar to the model estimating adult BMI in 2000. The model for change is differentiated from cross sectional estimated of adult BMI by an additional level of hierarchy (time poin ts within individuals), thus these models include an estimated random effect for the slope of BMI between 2000 and 2008. Variables entered into the model are used to predict both the intercept (BMI in 2000) and slope (changes in BMI between 2000 and 2008). In Chapter 4, Model 1 adjusts for mortality, attrition, and proxy status (time varying) and includes measures of childhood conditions Model 2 adds race, gender, and age. Model 3 adds socioeconomic status Model 4 adds health behaviors. Analyses for Ad ult BMI in 2000 Neighborhood Characteristics The dependent variable in this model is a cross sectional measure of adult BMI in 2000, continuously coded. In Chapter 5, Model 1 includes an indicator for proxy status and Model 2 adds neighborhood demographics. Model 3 adds neighborhood socioeconomic status and Model 4 adds a measure of walkability. Analysis for Changes in Adult BMI, Neighborhood Characteristics The dependent variable in this model is a longitudinal measure of adult B MI from 2000 2008. As in the models with just individual level characteristics, variables entered into the model are used to predict both the intercept (BMI in 2000) and slope (changes in BMI between 2000 and 2008). In Chapter 5, Model 1 includes indicato rs for proxy status (time varying), mortality, attrition, and whether or not a respondent moved

PAGE 71

71 between 2000 and 2008. Model 2 adds neighborhood demographics. Model 3 adds neighborhood socioeconomic status and Model 4 adds a measure of walkability.

PAGE 72

72 T able 3 1. Means, Proportions, and Number of Respondents for Measures of Childhood Disadvantage in the HRS Cohort (1931 1941) Variable N Mean Initial Wave of Measurement (Year) Missing (DK, NA) Missing due to Mortality Father's Education 11607 8.99 RAND 1875 0 Mother's Education 12070 9.24 RAND 1412 0 Father Laborer 10584 0.52 Wave 4 (1998) 855 630 Father Lost Job 11403 0.18 Wave 4 (1998) 131 630 Poor Childhood Health 10584 0.02 Wave 4 (1998) 19 630 1 Total does not include live drop outs. 9.26% of the original HRS cohort sample has been dropped before death

PAGE 73

73 Table 3 2. Means and Proportions for HRS Sample in 2000 2008 Covariates Mean (SD) Model Controls Died 0.19(0.39) Attrit 0.31(0.46) Proxy Status 0.08(0.27) Moved Since 2000 0.45( 0.49 ) Adult Demographics Age(years) 62.27(5.84) White (ref) 0.74(0.43) Black 0.14(0.35) Other 0.03(0.15) Hispanic 0.09(0.29) Foreign Born 0.11(0.31) Years of Education 12.34(3.18) Household Income 61,128(10,1596) Married 0.77(3.18) Health BMI (adjusted) 29.19 Functional Status 0.39(1.26) Self Rated Health 3.30(1.13) Non Smoker 0.39(0.49) Current Smoker 0.18(0.38) Former Smoker 0.44(0.49) Visited Doctor 0.93(0.24) Visited Hospital 0.21(0.41) No Insurance 0.12(0.33)

PAGE 74

74 Table 3 3. Descriptive Statistics for Census Tract Characteristics, HRS 2000 Mean (SD) Neighborhood Demographics Proportion Over Age 65 0.14(0.07) Proportion Rural Population 0.23(0.36) Black Population 0.15(0.25) Hispanic Population 0.12(0.20) Foreign Born Population 0.09(0.12) Neighborhood SES Proportion Unemployed 0.06(0.05) Proportion Females with a BA 0.05(0.03) Proportion Males with a BA 0.05(0.03) Walkability 0.16( 0.36 ) Proportion Families Below Poverty Line 0.09(0.09)

PAGE 75

75 Table 3 4. Self reported and Measured BMI HRS 2006 20010. Year Self Reported BMI Measured BMI Difference 2006 27.93 29.17 1.24 2008 28.2 0 29.5 0 1.3 0 2010 28.22 29.51 1.29

PAGE 76

76 CHAPTER 4 CHILDHOOD, ADULT CHA RACTERISTICS, AND AD ULT WEIGHT Chapter 4 presents results from analyses testing whether individual adult characteristics such as educational attainment and health behaviors can account for residual associations between childhood conditions and adult weight. Childhood socioeconomic posi tion and health can have lifelong influence s on adult health including weight related outcomes Whether childhood conditions have a direct association with adult weight or operate indirectly through adult characteristics remains an area of research. In a review of literature between 1998 and 2008, childhood socioeconomic position was inversely associated with adulthood obesity in 70% of studies in females and 27% of studies in males. After measures of adult socioeconomic position were in included in models inverse associations were found in 47% of studies with females and 14% of studies in males (Senese et al. 2009) Other r esearch based on the HRS has also found a persistent association between childhood socioeconomic position and likelihood of being overweight or obese among females (Best, Hayward, and Hidajat 2005) Their research did not include a full array of adult health behaviors, however, as the primary aim of that study w as t o determine whether obesity is a pathway linking life course socioeconomic status with diabetes. Analyses in Chapter 4 extend research using the HRS research by the inclusion of additional measures of adult health behaviors testing for an association betw een childhood conditions and changes in BMI in older adults, and accounting for dependence of observations. Persistent associations between childhood conditions and adult weight are consistent with th e critical period model of life course epidemiology. I f childhood conditions have a direct effect, then the results would also be consistent with a

PAGE 77

77 cumulative exposure model. However, if the association between childhood characteristics and adult weight is mediated by adult characteristics, the pathways model of life course epidemiology is supported The analyses of chapter 4 and subsequent chapters do not disentangle a cumulative exposure model versus pathways model. Rather, analyses test for direct and indirect associations between childhood conditions and a dult weight. A framework for disentangling critical, cumulative and pathways models has been suggested by other researchers (Mishra et al. 2009) A lthough analyses presented in Chapters 4 thro ugh 6 do not disentangle between models, they do provide evidence of long term effects of childhood conditions with regards to adult weight that can motivate more precise tests of cumulative vs. pathway models. Male BMI in 2000 Table 4 1 presents results from two level hierarchical linear model s (individuals nested within census tracts) predicting adjusted adult BMI among males in 2000. Full regression analyses adjust for socio demographic characteristics, health status and health behaviors. Model 1 inclu des controls for proxy status and childhood conditions. The average adjusted BMI among males who did not report poor childhood health, a father who lost a job, or a father who was a laborer, and whose parents were at the grand mean of educational attainmen t, is estimated to be 28.33 i n 2000 (b=28.33, p<.01). Among the measures of childhood conditions, only having a father as a laborer was significantly associated with BMI (b=0.44, p<.01). Having a father who was a laborer is thus associated with higher BMI among men before additional adult controls are included. Model 2 includes controls for proxy status and adult socio demographics without measures of childhood conditions. Being older is associated with a lower BMI

PAGE 78

78 (b= 0.13, p<.01). White males do not signi categories in BMI. Hispanic males do not significantly differ from non Hispanic males. Being foreign born is associated with a lower BMI (b= 0.99), as is having higher educational attainment (b= 0.10, p<.01). Being married is associated with a significantly higher BMI (b=1.16, p<.01) The positive association between being married and BMI among males is in line with previous research (Jeffery and Rick 2002; Sobal, Rauschenbach, and Frongillo Jr 1992) Model 3 reintroduces childhood conditions and includes controls for demographic characteristics. Having a father who was a la borer remains significantly associated with adult BMI (b=0.44, p<.01). Age, race, nativity, and marital status do not appear to mediate the relationship between paternal occupation and adult BMI. Model 4 additionally adjusts model 3 with controls for years of education and household income. Higher educational attainment is significantly associated with a lower BMI (b= 0.08, b<.01). Household income is not associated with BMI. After controls for socioeconomic status are included, paternal occupation is no lo nger associated with adult BMI. Model 5 additionally includes measures for health and health behaviors. Better self rated health is associated with lower BMI (b= 0.41, p<.01), as is being a current smoker (b= 1.89, p<.01). Visiting a doctor in the past two years is associated with a higher BMI (b=0.89, p<.01). Educational attainment continues to be independently associated with adult BMI (b=0.08, p<.01) Across all models, there is significant variation in BMI between individuals (Table 4 2) However, ther e is no significant variation between census tracts. The lack of significant variation between census tracts among males is evidence that neighborhood characteristics, insofar as the exert an influence on weight, cannot explain differences in

PAGE 79

79 BMI between c ensus tracts. According to all three measures of model fit, (deviance, AIC, and BIC) the best fitting model is model 5, the full model which includes measures of childhood characteristics, adult socio demographics, and adult health and health behaviors. F emale BMI in 2000 Table 4 3 presents results from two level hierarchical linear models predicting adjusted adult BMI among females in 2000. As with males, full regression analyses adjust for socio demographic characteristics, health status, and health beha viors. Model 1 includes controls for proxy status and childhood conditions. The average adjusted BMI among females who did not report poor childhood health, a father who lost a job, or a father who was a laborer, and whose parents were at the grand mean of educational attainment, is estimated to be 28.32 in 2000 (b=28.32, p<.01). Among the measures of significantly associated with BMI (b=0.45, p<.05; b= 0.17, p<.01). Having a fat her who was a laborer is thus associated with higher BMI and having a father with greater educational attainment is associated with lower adult BMI among females before additional adult controls are included. Each additional year of paternal education abov e the grand mean of paternal education is associated with a 0.17 decrease in predicted BMI. Model 2 includes controls for proxy status and adult socio demographics without measures of childhood conditions. Being older is associated with a lower BMI (b= 0. 08, p<.01). Black females are predicted to have a significantly higher BMI than white females (b=3.17, p<.01), although there is no estimated difference between white Hispanics, Hispanic females are

PAGE 80

80 predicted t o have a significantly higher BMI (b=1.37, p<.01). As with males, being foreign born is associated with a lower BMI (b= 1.41), as is having higher educational attainment (b= 0.26, p<.01). Higher household income among females (but not males) is associated with a lower BMI (b= 0.02, p<.05). Model 3 reintroduces childhood conditions and includes controls for demographic characteristics. Having a father who was a laborer is no longer associated with BMI among females However, paternal education retains an i nverse association with BMI. Age, race, nativity, and marital status do not appear to mediate the relationship between paternal education and adult BMI. Model 4 additionally adjusts model 3 with controls for years of education and household income. Higher educational attainment and income attainment are significantly associated with a lower BMI (b= 0.17 b<.01 ; b= 0.02, p<.05 ). Model 5 additionally includes measures for health and health behaviors. Number of ADLs and IADLs is positively associated with adul t BMI (b=0.34, p<.01). Individuals who reported better self rated health (b= 1.08, p<.01) and current smokers (b= 2.70, p<.01) are predicted to have a lower BMI in 2000. Females who visited a hospital in the past two years are predicted to have a higher BM I (b=0.64 p<.01). Educational attainment continues to be independently a ssociated with adult BMI (b=0.11 p<.01) however his association is attenuated after the inclusion of health behaviors. Across all models, there is significant variation in BMI bet ween individuals and between census tracts (Table 4 4 ). Approximately 3.5% of variation in BMI among females is associated with the census tract level. Unlike males, this represents a statistically significant amount of variation, even if substantively, va riation in BMI predominantly occurs within between individuals who live within the same census tract

PAGE 81

81 (96.5%). The relatively small amount of variation at the census tract level suggests that neighborhood characteristics (as measured at the census tract lev el) can at most, only explain a small amount of variation in female BMI. According to all three measures of model fit, (deviance, AIC, and BIC) the best fitting model is model 5, the full model which includes measures of childhood characteristics, adult socio demographics, and adult health and health behaviors. Changes in Mal e BMI between 2000 and 2008 Table 4 5 presents results from three level hierarchical linear models (waves within individuals within census tracts) predicting adjusted adult BMI among males between 2000 and 2008. As with the two level hierarchical linear mo dels with predicted BMI in 2000, the full regression analysis using three level models adjust for socio demographic characteristics, health status, and health behaviors. The key difference between the two and three level models is that the three level mod els test whether childhood conditions are associated with changes in BMI between 2000 and 2008. As with two level models, a random effect is estimated for the intercept to account for the dependence of observations within census tracts. Time is coded such that the intercept in each model represents the predicted BMI in 2000. An additional random effect is estimated for the slope which indicates the predicted rate of change between each wave of the HRS. The inclusion of an estimated random effect helps to a ccount for the dependence of observations within individuals. Because the HRS is conducted on a two year schedule, the slope is the predicted rate of chan ge across two year intervals. In the written summary of the models, coefficients of covariates are int erpreted for both their association with the predicted intercept (estimated BMI in 2000) and slope (rate of change between two year intervals). In most cases, coefficients for the predicted BMI in

PAGE 82

82 2000 should be similar to the previous models which only es timated BMI in 2000. However, coefficients that are not identical to each other do not necessarily indicate inconsistencies in model specification as the coefficients come from different regression estimations and should be expected to vary somewhat. Model 1 includes controls for proxy status whether or not a respondent died during the eight year observation period, attrition due to non mortality, and whether or not a respondent moved during the eight year observation period. Model 1 also includes childhoo d conditions. The average estimated BMI of males in 2000 who did not report poor childhood health, a father who lost a job, or a father who was a laborer, and whose parents were at the grand mean of educational attainment, is estimated to be 27.98 in in 20 00 (b=27.98, p<.01). The average rate of change in male BMI was positive (b=0.09, p<.01) indicating that BMI is expected to increase by 0.09 points every two years. Having a father who was a laborer is associated with a higher BMI (b=0.54, p<.01). Regardi ng rate of change (Table 4 5 cont), male BMI increased an average of .09 points every two years (b=0.09, p<.01). Respondents who died at any point during the observation period are predicted to have a lower rate of BMI increase (b= 0.44, p<.01). None of th e measures of childhood conditions were associated with change in BMI. Model 2 additionally adjusts for demographic characteristics. Having a father who was a laborer retains a significant association with BMI in 2000 (b=0.49, p<.01). Being older is assoc iated with a lower BMI in 2000 (b= 0.07), as is being foreign born (b= 0.94, p<.01). Being married is associated with a higher BMI (b=0.51, p<.01). Respondents who died during the observation period are predicted to have a lower initial BMI (b= 0.71, p<.01 ). Regarding rate of change, BMI is predicted to increase by

PAGE 83

83 an average of 0.14 points every two years (b=0.14, p<.01). Demographic variables associated with rate of change include age and being black. Being older is associated with a decreased rate of BMI change (b= 0.01, p<.01). Being older is thus associated with both lower initial BMI, as well as lower average rate of BMI change. Blacks, although they do not significantly differ from whites in their predicted BMI in 2000, are predicted to have a slower rate of BMI change between 2000 and 2008 (b= 0.17). Model 3 additionally adjust for adult socioeconomic characteristics. Greater educational attainment is associated with lower BMI in 2000 (b= 0.11, p<.01), however household income is not associated with BMI. Once adult socioeconomic characteristics Regarding rate of change, BMI is predicted increase by an average of 0.12 points every two years. Neither educa tion nor i ncome are associated with rate of change. The association between being black and lower rate of BMI change is not mediated by adult socioeconomic characteristics. Model 4, the full model, includes measures of health and health behaviors. Males who report better health are predicted to have lower BMI in 2000 (b= 0.16, p<.05). Males who smoke are also predicted to have lower BMI (b= 0.99, p<.01). Being older retains i ts inverse association with BMI (b= 0.08, p<.01). Being foreign born and having greater educational attainment also continue to be associated with lower BMI (b= 1.02, p<.01; b= 0.12, p<.01). Regarding rate of change, the only health measure associated with BMI was self rated health males who report better health are predicted to have an increased rate of BMI change (b=0.03, p<.01). Other than mortality, the only other

PAGE 84

84 variable in the full model for males associated with rate of change is age being older is associated with a decreased rate of change (b= 0.01, p<.01). Across all models for men, th ere is significant variation in the BMI slope (Table 4 6). Although the variation is significant, it represents only a very small proportion of total variation in BM I (1.05%). If this were a much younger sample, such as adolescents still developing, we would expect much more of the total variation in BMI to be associated with BMI changes over time. However, given the age of the HRS cohort, it is expected that changes in BMI constitute only a tiny fraction of total variation in BMI. As in the two level models for men, most of the variation in BMI occurs between people within a census tract. An insignificant amount of variation in BMI is associated with the census tract level. According to all three measures of model fit the best fitting model is model 4, the full model which includes measures of childhood characteristics, adult socio demographics, and adult health and health behaviors. Changes in Female BMI between 2000 and 2008 Table 4 7 presents results from three level hierarchical linear models predicting adjusted adult BMI among females between 2000 and 2008. As with the two level hierarchical linear models with predicted female BMI in 2000, the full regression anal ysis using three level models adjust for socio demographic characteristics, health status, and health behaviors. Model 1 includes childhood conditions, controls for proxy status, whether or not a respondent died during the eight year observation period, a ttrition due to non mortality, and whether or not a respondent moved during the eight year observation period. The average estimated BMI of fe males in 2000 who did not report poor childhood health, a father who lost a job, or a father who was a laborer, an d whose parents were at the grand mean of educational attainment, is estimated to be

PAGE 85

85 27. 76 in in 2000 (b=27.76 p<.01). Of the measures of childhood conditions, only 0.19, p<.01). Females with more highly educated fathers are predicted to have a lower BMI in 2000. Regarding rate of change, the average rate of change in female BMI was positive (b=0.16, p<.01) indicating that BMI is expected to increase by 0.16 points every two years. Respo ndents who died at any point during the observation period are predicted to have a lower rate of BMI increase (b= 0.43 p<.01). with a slightly increased rate of BMI change (b=0.01, p<.01). No other measure of childhood co nditions was associated with change in BMI. Model 2 additionally adjusts for demographic characteristics. Being older is associated with a lower BMI in 2000 ( b= 0.02). Being black is associated with a higher BMI in 2000 (b=3.22, p< .01). Relative to non Hispanics, Hispanics are also predicted to have a higher BMI (b=1.25, p<.01). Being married is associated with a higher BMI in 2000 (b=0.50, p<.01). Having a father who was a laborer retains a significant association with BMI in 2000 (b= 0.19, p<.01). Regarding rate of change, BMI is predicted to increase by an average of 0.22 points every two years (b=0.22 p<.01). The only d emogr aphic variable associated with rate of change is age. Being older is associated with a decreased rate of B MI change (b= 0.01, p<.01). As was found for males, being older is associated with both lower initial BMI, as well as lower average rate of BMI change. Black females although they are predicted to have significantly higher BMI than white females, do not d iffer in predicted changes in BMI over time. Model 3 additionally adjust for adult socioeconomic characteristics. Greater educational attainment is associated with lower BMI in 2000 (b= 0. 18 p<.01. Adult

PAGE 86

86 socioeconomic characteristics do not appear to att enuate the association between paternal education and female adult BMI Regarding predicted rate of change in model 3 BMI is predicte d increase by an average of 0.2 2 points every two years. Adult socioeconomic status is not associated with rate of change in BMI Only age is associated with BMI being older is associated with a decreased rate of BMI change (b= 0.02, p<.01). Model 4, the full model, includes measures of health and health behaviors. Fem ales who report better health are predicted to have lower BMI in 2000 (b= 0.28 p<.05). Fem ales who smoke are also pred icted to have lower BMI (b= 1.25 p<.01). Being married continues to be associated with a higher BMI (b=0.46, p<.01). G re ater educational attainment continue s to be associated with lower BMI ( b= 0.19 p<.01; b= 0.12, p<.01). Adult health status and health behaviors do not appear to attenuate the association between paternal education and female adult BMI (b= 0.14, p<.01) Regarding rate of change, no health measure is associated with BMI. Females with more highly educated fathers are predicted to have slightly increased rates of BMI change (b=0.01, p<,01). Other than mortality, the only other variable in the full model for fe males associated with rate of change is age being older is associated wit h a decreased rate of change (b= 0.01, p<.01). Across all models for females there is significant variation in BMI slope (Table 4 8 ). Although the variation is significant, as it did for men, variation in BMI over time represents only a very small propor tion of total variation in BMI ( 0.98 %) As in the two level models for females most of the variation in BMI occurs between individuals within a census tract. Unlike men, however, there is significant variation in BMI at the census

PAGE 87

87 tract level. 4.6% of tota l variation in BMI occurs between census tracts. The higher proportion of BMI variation at the census tract level among females suggests that neighborhood characteristics may explain a higher proportion of BMI differences between females in different censu s tracts. Nonetheless, it is important to put contextual effects into context the vast majority of BMI variation is still associated with individual characteristics among females ( 9 4.6%). The lack of significant variation in BMI at the census tract for mal es, and relatively small proportion among females, should not be neighborhoods at the census tr act level. According to all three measures of model fit the best fitting model is model 4, the full model which includes measures of childhood characteristics, adult socio demographics, and adult health and health behaviors. Summary Results from these analyses help demonstrate the influence of childhood conditions on adult weight, as well as the gender specific associations between early conditions and adult weight. Results also suggest that childhood conditions have little associatio n with changes in BMI over an eight year observation period among older adults. Among males, having a father who was a laborer was associated was higher initial BMI. This association remained after controls for age, race, and nativity were included in the model. However, a fter controls for adult socioeconomic position were included the association between paternal occupation and adult BMI was no longer statistically significant. T he finding that adult socioeconomic status explains the link between childhoo d conditions and adult weight is consistent with a pathway model linking childhood conditions with adult weight among males. This finding is also

PAGE 88

88 consistent with previous research that suggests among men, adult socioeconomic position is more likely to medi ate associations between childhood conditions and adult weight (Senese et al. 2009). Among females, paternal education was inversely associated with initial BMI in 2000 This association persisted after controls for demographics, adult socioeconomic posit ion, adult health status and health behaviors. The finding that childhood conditions retain an independent association with adult weight after a full range of controls for adult conditions is consistent with a critical period model linking childhood condi tions with adult weight among females. Within the critical period model, the direct association between childhood conditions and adult weight may be explained by a number of factors including biological programming or the development of lifelong health beh aviors not accounted for in the model. Given that the association between childhood conditions and adult BMI seem s to operate differently between men and women, interpretations of the persistent association among women should consider the influence of gend er norms regarding weight management among women. It may be the case that women raised in a higher socioeconomic status household may be exposed to different or more stringent ideas about weight management, and exposure to these norms inculcates the develo pment of lifelong weight related health behaviors (Sobal and Stunkard 1989) If permanent habits were formed in childhood, is this a case of a critical period of development? In order for these health behaviors to be truly consistent with a critical period model, they should begin at an early age and persist across the life course. also possible that processes of biological scarring differ between men and women.

PAGE 89

89 Among both males and females, the amount of variation in BMI at the census tract level was relatively small compared to the proportion of variation in BMI between ind ividuals. Male BMI, in fact, did not significantly vary between census tracts. Previous research using the Americans Changing Lives study has in contrast to these findings found significant variation in male BMI at the census tract level. The ACL study u sed significantly older census tract information; however, and relied on 1980 census tract definitions. The lack of significant variation among males at the census tract level does not mean that neighborhood characteristics exert insignificant effects on a dult male weight, however it does mean that neighborhood characteristics measured at the census tract level likely do not explain BMI differences between males. Female BMI did vary significantly between census tracts, and so neighborhood characteristics ma y play some role in explaining BMI differences between females. The residual association between paternal education and female adult BMI may indeed be due to neighborhood characteristics. The next chapter provides modest evidence that neighborhood characte ristics are associated with adult weight among females net of individual characteristics; Chapter 6 brings the results from Chapter 4 and Chapter 5 together in a final analysis.

PAGE 90

90 Table 4 1. Results from a Hierarchical Linear Model Regressing Male Adjusted BMI on Childhood and Individual Characteristics HRS Cohort 2000. Covariates Model 1 Model 2 Model 3 Model 4 Model 5 Intercept 28.33** 27.94** 27.38** 27.52** 28.54** Model Controls Proxy Status 0.19 0.45 0.38 0.45 0.44 Childhood Conditions Poor Health 0.16 0.04 0.09 0.19 Father Lost Job 0.15 0.17 0.18 0.14 Father Laborer 0.44* 0.44* 0.37 0.39 0.03 0.00 0.02 0.02 0.05 0.04 0.02 0.02 Adult Demographics Age 0.13** 0.14** 0.14** 0.16** White (ref) Black 0.36 0.41 0.31 0.23 Other 0.58 0.54 0.48 0.48 Hispanic (ref=not Hispanic) 0.47 0.58 0.36 0.31 Foreign Born 0.99** 0.79* 0.82* 0.85* Married 1.16** 1.31** 1.32** 1.12** Years of Education 0.10** 0.08** 0.08** Household Income 0.01 0.01 0.01 Health Functional Status 0.16 Self Rated Health 0.41** Current Smoker 1.89** Former Smoker 0.04 Visited Doctor 0.89** Visited Hospital 0.34 No Insurance 0.30 Notes: *p < .05; **p < .01.

PAGE 91

91 Table 4 2. Variance Components and Fit Statistics from an HLM Model Estimating Adjusted Male BMI in 2000 Variance Components Model 1 Model 2 Model 3 Model 4 Model 5 In rate of Change Within group 23.23** 22.26** 22.49** 22.45** 21.78** Between Group Level 2 < 0.00 < 0.00 < 0.00 < 0.00 < 0.00 Between Group Level 3 N, Level 1 3083 3191 2962 2962 2950 N, Level 2 1879 1927 1814 1814 1807 N, Level 3 Goodness of fit Deviance 18455.0 18973.7 17637.8 17643.8 17487.1 AIC 18457.0 18975.7 17639.8 17645.8 17489.1 BIC 18462.2 18981.3 17645.3 17651.3 17494.6 Notes: *p < .05; **p < .01.

PAGE 92

92 Table 4 3. Results from a Hierarchical Linear Model Regressing Female Adjusted BMI on Childhood and Individual Characteristics HRS Cohort 2000. Covariates Model 1 Model 2 Model 3 Model 4 Model 5 Intercept 28.32** 28.46** 27.93** 28.12** 31.69** Model Controls Proxy Status 1.73** 2.20** 1.63* 1.76** 2.71** Childhood Conditions Poor Health 0.08 0.00 0.13 0.84 Father Lost Job 0.16 0.02 0.03 0.06 Father Laborer 0.45* 0.44 0.28 0.27 0.06 0.04 0.00 0.03 0.17** 0.14** 0.11** 0.10** Adult Demographics Age 0.08** 0.08** 0.09** 0.13** White (ref) Black 3.17** 3.14** 3.09** 2.46** Other 0.16 0.49 0.28 0.48 Hispanic (ref=not Hispanic) 1.37** 1.50** 1.12* 0.75 Foreign Born 1.41** 0.89** 1.03** 1.30** Married 0.23 0.23 0.14 0.23 Years of Education 0.26** 0.17** 0.11* Household Income 0.02* 0.02* 0.02* Health Functional Status 0.34** Self Rated Health 1.08** Current Smoker 2.70** Former Smoker 0.31 Visited Doctor 0.51 Visited Hospital 0.64** No Insurance 0.50 Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01

PAGE 93

93 Table 4 4. Variance Components and Fit Statistics from an HLM Model Estimating Adjusted Female BMI in 2000 Variance Components Model 1 Model 2 Model 3 Model 4 Model 5 In rate of Change Within group 36.63** 36.53** 35.92** 35.89** 32.37** Between Group Level 2 1.32* 0.35 0.69 0.50 0.99* Between Group Level 3 N, Level 1 3843 4039 3723 3723 3710 N, Level 2 2221 2268 2140 2140 2133 N, Level 3 Goodness of fit Deviance 24880.0 26045.4 23975.0 23963.0 23558.2 AIC 24884.0 26049.4 23979.0 23967.0 23562.2 BIC 24895.4 26060.9 23990.3 23978.3 23573.5 Notes: *p < .05; **p < .01.

PAGE 94

94 Table 4 5. Results from a Hierarchical Linear Model Regressing Male Adjusted BMI on Childhood and Individual Characteristics HRS Cohort 2000 2008. Covariates Model 1 Model 2 Model 3 Model 4 Intercept 27.98** 27.74** 27.92** 29.64** For Intercept (Initial Status) Model Controls Proxy Status 0.23 0.16 0.10 0.11 Died 0.58 0.71* 0.64* 0.79* Attrit 0.06 0.20 0.21 0.28 Moved Since 2000 Childhood Conditions Poor Health 0.02 0.27 0.36 0.35 Father Lost Job 0.13 0.12 0.13 0.12 Father Laborer 0.54* 0.49* 0.39 0.33 0.00 0.01 0.00 0.01 0.01 0.02 0.01 0.00 Adult Demographics Age 0.07** 0.08** 0.08** White (ref) Black 0.51 0.38 0.35 Other 0.13 0.04 0.12 Hispanic (ref=not Hispanic) 0.64 0.36 0.33 Foreign Born 0.94* 0.99* 1.02** Married 0.51* 0.52* 0.48* Years of Education 0.11** 0.12** Household Income 0.00 0.00 Health Functional Status 0.01 Self Rated Health 0.16* Current Smoker 0.99** Former Smoker 0.40 Visited Doctor 0.46 Visited Hospital 0.02 No Insurance 0.06 Notes: *p < .05; **p < .01.

PAGE 95

95 Table 4 5 Continued Covariates Model 1 Model 2 Model 3 Model 4 Interval 0.09** 0.14** 0.12* 0.11 For Slope (Change) Model Controls Proxy Status 0.04 0.02 0.01 0.01 Died 0.44** 0.41** 0.40** 0.42** Attrit 0.00 0.01 0.02 0.02 Moved Since 2000 0.01 0.00 0.01 0.01 Childhood Conditions Poor Health 0.06 0.04 0.03 0.03 Father Lost Job 0.02 0.02 0.02 0.02 Father Laborer 0.03 0.02 0.01 0.00 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 Adult Demographics Age 0.01** 0.01** 0.01** White (ref) Black 0.12* 0.11* 0.09 Other 0.17 0.18 0.17 Hispanic (ref=not Hispanic) 0.09 0.07 0.06 Foreign Born 0.07 0.08 0.08 Married 0.03 0.03 0.03 Years of Education 0.01 0.01 Household Income 0.00 0.00 Health Functional Status 0.00 Self Rated Health 0.03* Current Smoker 0.06 Former Smoker 0.06 Visited Doctor 0.09 Visited Hospital 0.02 No Insurance 0.01 Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 96

96 Table 4 6. Variance Components and Fit Statistic s from an HLM Model Estimating Una djusted Male BMI, HRS 2000 2008. Variance Components Model 1 Model 2 Model 3 Model 4 In rate of Change 0.31** 0.29** 0.29** 0.29** Within group 2.71** 2.68** 2.68** 2.67** Between Group Level 2 26.22** 25.36** 25.23** 25.06** Between Group Level 3 0.24 0.45 0.52 0.49 N, Level 1 14276 13760 13760 13613 N, Level 2 3570 3428 3428 3424 N, Level 3 1929 1865 1865 1863 Goodness of fit Deviance 68422.9 65710.4 65736.0 65074.2 AIC 68436.9 65724.4 65750.0 65088.2 BIC 68475.9 65763.1 65788.7 65126.9 Notes: *p < .05; **p < .01.

PAGE 97

97 Table 4 7. Results from a Hierarchical Linear Model Regressing Female Adjusted BMI on Childhood and Individual Characteristics HRS Cohort 2000 2008 Covariates Model 1 Model 2 Model 3 Model 4 Intercept 27.76** 26.90** 27.16** 28.71** For Intercept (Initial Status) Model Controls Proxy Status 0.49 0.66 0.62 0.95 Died 2.02** 2.01** 1.85** 1.96** Attrit 0.43 0.51 0.54 0.50 Moved Since 2000 0.01 0.00 0.01 0.04 Childhood Conditions Poor Health 0.02 0.12 0.23 0.43 Father Lost Job 0.09 0.14 0.09 0.09 Father Laborer 0.25 0.27 0.14 0.11 0.08 0.04 0.00 0.01 0.19** 0.17** 0.14** 0.14** Adult Demographics Age 0.02 0.02 0.03 White (ref) Black 3.22** 3.22** 3.06** Other 0.21 0.01 0.19 Hispanic (ref=not Hispanic) 1.25* 0.91 0.70 Foreign Born 0.47 0.65 0.81 Married 0.50* 0.51* 0.46* Years of Education 0.18** 0.19** Household Income 0.00 0.00 Health Functional Status 0.10 Self Rated Health 0.28** Current Smoker 1.25** Former Smoker 0.11 Visited Doctor 0.24 Visited Hospital 0.17 No Insurance 0.01 Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 98

98 Table 4 7 Continued Covariates Model 1 Model 2 Model 3 Model 4 Interval 0.16** 0.22** 0.22** 0.05 For Slope (Change) Model Controls Proxy Status 0.10 0.15 0.14 0.20 Died 0.43** 0.43** 0.42** 0.46** Attrit 0.02 0.01 0.01 0.01 Moved Since 2000 0.02 0.01 0.01 0.01 Childhood Conditions Poor Health 0.07 0.09 0.10 0.13 Father Lost Job 0.04 0.06 0.06 0.06 Father Laborer 0.06 0.04 0.05 0.05 0.00 0.00 0.01 0.07 0.01* 0.01* 0.01* 0.01* Adult Demographics Age 0.01** 0.02** 0.01** White (ref) Black 0.06 0.06 0.03 Other 0.15 0.15 0.26 Hispanic (ref=not Hispanic) 0.03 0.02 0.00 Foreign Born 0.11 0.11 0.10 Married 0.07 0.07 0.06 Years of Education 0.00 0.00 Household Income 0.00 0.00 Health Functional Status 0.01 Self Rated Health 0.03 Current Smoker 0.04 Former Smoker 0.04 Visited Doctor 0.06 Visited Hospital 0.01 No Insurance 0.02 Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 99

99 Table 4 8. Variance Components and Fit Statistics from an HLM Model Estimating Adjusted Female BMI HRS 2000 2008. Variance Components Model 1 Model 2 Model 3 Model 4 In rate of Change 0.49** 0.47** 0.47** 0.47** Within group 3.95** 3.91** 3.91** 3.92** Between Group Level 2 43.17** 41.88** 41.82** 40.11** Between Group Level 3 2.32** 1.81* 1.65* 1.77* N, Level 1 18345 17793 17793 17589 N, Level 2 4322 4183 4183 4182 N, Level 3 2248 2168 2168 2168 Goodness of fit Deviance 95383.9 92171.9 92183.9 91108.4 AIC 95395.9 92183.9 92195.9 91120.4 BIC 95430.3 92218.0 91130.0 91154.5 Notes: *p < .05; **p < .01.

PAGE 100

100 CHAPTER 5 NEIGHBORHOODS AND AD ULT WEIGHT Chapter 5 presents results from two and three level hierarchical linear models that test for associations between neighborhood demographics and socioeconomic status with adult weight. After presenting results with just neighborhood characteristics, individ ual characteristics are introduced into the models. Neighborhood socio demographics, such as percent black, are by definition aggregates of individual characteristics. Yet the influence of aggregate characteristics may exist above and beyond individual cha racteristics. For example, a higher proportion of residents with a college degree or higher incomes may create sufficient demand for amenities such as health food stores and green spaces that individuals could utilize. The availability of amenities such as health food stores is referred to as the built environment in the literature on neighborhood characteristics Only one measures of the built environment walkability, is included in these analyses. All other neighborhood measures are demographic and socioe conomic characteristics. Previous research has found these characteristics to be associated with adult BMI. A higher proportion of black residents and a higher percentage of residents in poverty has been found to be associated with higher BMI (Boardman, Saint Onge, Haberstick, Timberl ake, and Hewitt 2008) Other research has found gender differences in the association between neighborhood socioeconomic disadvantage and weight, with females showing more consistent associations than males (Robert and Reither 2004) By presenting models with neighborhood characteristics and sans individual characteristics, potentially influential neighborhood characteristics are identified. If these characteristics mai ntain their association with adult weight after adjustment for individual

PAGE 101

101 characteristics, then there is evidence that aggregate characteristics have an influence independent of individual characteristics. Male BMI in 2000 Table 5 1 presents results from a two level hierarchical linear model regressing adjusted male BMI in 2000 on neighborhood characteristics. All models do contain one individual level variable proxy status. In each model, proxy status was associated with a significantly lower BMI in 2000. Model 1 includes only proxy status as a predictor of BMI. Among males who did not have a proxy, the average estimated BMI in 2000 is 28.55 (b=28.55, p<.01). Model 2 introduces neighborhood demographics. Percent Hispanic and percent foreign born are signi ficantly associated with male BMI. A higher percentage of Hispanics is associated with a higher BMI (b=0.12, p<.05) while a higher percentage of foreign born residents is associated with a lower BMI (b= 0.25, p<.05). Percent rural, percent over the age o f 65, and percent black are not associated with BMI. Model 3 additionally includes neighborhood socioeconomic characteristics. Only percent of males with a college degree or higher is associated with BMI (b= 1.31, p<.01). Males who live in census tracts wi th a higher percentage of males who attended college are thus predicted to have lower BMI. Percent of families in poverty, percent of vacant homes, percen t unemployed, percent females with a BA degree and median household value of a census tract are not a ssociated with BMI. Model 4 additionally includes a measure of walkability, which is not associated with BMI. Table 5 2 presents estimated random effects and measures of model fit for each model in Table 5 1. In model 1, there is significant variation in male BMI at both the individual and census tract level. Approximately 3.5% of variation in male BMI is associated at the census tract level, with the rest of the variation occurring between

PAGE 102

102 individuals. Once neighborhood demographics are included in model 2, there is not significant variation in male BMI. This suggests that the racial, age, and nativity characteristics of a census tract explain observed variation between census tracts in male BMI. According the all three measures of model fit, model 4 prov ides the best fit to the data. Female BMI in 2000 Table 5 3 presents results from a two level hierarchical linear model regressing adjusted male BMI in 2000 on neighborhood characteristics. As with males, a ll models include a control for proxy status and i n each model, proxy status was associated with a significantly lower BMI in 2000. From model 1, a mong males who did not have a proxy, the average estimated BMI in 2000 is 28. 50 (b=28.50 p<.01). Model 2 introduces neighborhood demographics. Percent rural, p ercent black, percent Hispanic and percent foreign born are significantly associated with fe male BMI. Females living in census tract that is more rural are predicted to have a higher BMI (b=0.12, p<.01). Those living in a census tract with a higher perc entage of blacks and a hig h er percentage of Hispanics are also predicted to have a higher BMI (b=0.40, p<.01; b=0.35, p<.01). A higher percentage of foreign born residents is associated w ith a lower BMI (b= 0.30, p<.01). P ercent over the age of 65 is not a ssociated with BMI. Model 3 additionally includes neighborhood socioeconomic characteristics. Only percent of males with a college degree or hig her is associated with BMI (b= 2.2 1, p<. 01). Like males, fem ales who live in census tracts with a higher percent age of males who attended college are predicted to have lower BMI. After adjustment for neighborhood socioeconomic characteristics, only percent black is associated with female BMI (b=0.19, p<.01). However, this association appears to be attenuated by soci oeconomic

PAGE 103

103 characteristics. Model 4 additionally includes a measure of walkability, which is not associated with BMI. Table 5 4 presents estimated random effects and measures of model fit for each model in Table 5 3. In model 1, there is significant varia tion in female BMI at both the individual and census tract level. Approximately 4.6% of variation in female BMI is associated at the census tract level, with the rest of the variation occurring between individuals. Once neighborhood demographics are includ ed in model 2, there is not significant variation in female BMI. This suggests that, as they did for males, the racial, age, and nativity characteristics of a census tract explain observed variation between census tracts in female BMI in 2000. According th e all three measures of model fit, model 4 provides the best fit to the data. Male BMI, 2000 2008 Table 5 5 presents results from a three level hierarchical linear model regressing adjusted male BMI between 2000 and 2008 on neighborhood characteristics. As with the two level hierarchical linear models with predicted BMI in 2000, the full regression analysis using three level models adjust neighborhood demographics and neighborhood socioeconomic status. The three level models, in addition to testing for an association between neighborhood characteristics and BMI in 2000, also test whether neighborhood characteristics are associated with changes in BMI between 2000 and 2008. Time is coded such that the intercept in each model represents the predicted BMI in 2 000. An additional random effect is estimated for the slope which indicates the predicted rate of change between each wave of the HRS. In the written summary of the models to follow, coefficients of neighborhood covariates are interpreted for both their a ssociation with the predicted intercept (estimated BMI in 2000) and slope (rate of change between two

PAGE 104

104 year intervals). Model 1 includes controls for proxy status, whether or not a respondent died during the eight year observation period, attrition due to n on mortality, and whether or not a respondent moved during the eight year observation period. None of these controls were associated with BMI in 2000, however respondents who died between 2000 and 2008 are estimated to have a rate of BMI change (b= 0.41, p <.01). Male BMI is predicted to increase by 0.07 points over two year intervals (b=0.07, p<.05). Model 2 additionally adjusts for neighborhood demographics. Percent black and percent Hispanic are associated with an increased BMI in 2000 (b=0.11, p<.01; b= 0.21, p<.05). Regarding rate of change, living in a census tract with a higher percentage of adults over the age of 65 is associated with an increased rate of change in BMI (b=0.05, p<.05). A higher percentage of blacks is associated with a reduced rate of change (b= 0.02, p<.05). Model 3 additionally adjusts for neighborhood socioeconomic characteristics. No measure of neighborhood socioeconomic status is associated with BMI in 2000 or rate of change in BMI. Once neighborhood socioeconomic measures are inc luded in the model, percent black and percent Hispanic are not associated with BMI in 2000. Percent black is no longer associated with rate of change in BMI. Model 4 additionally adjusts for walkability, which is not associated with initial levels or rate of change in BMI. Table 5 6 presents estimated random effects and measures of model fit for each model in Table 5 4. There is significant variation in rate of change, within an individual, and between individuals in all four models. Most of the variation in BMI (91.5%) occurs between individuals. Approximately 8 4 % of the variation in male BMI is due to changes

PAGE 105

105 within an individual. There is no significant variation between census tracts in male BMI. According to all three measures of model fit, model 4 is the best fitting model. Female BMI, 2000 2008 Table 5 7 presents results from a three level hierarchical linear model regressing adjusted female BMI between 2000 and 2008 on neighborhood characteristics. Model 1 includes controls for proxy status, whether or not a respondent died during the eight year observation period, attrition due to non mortality, and whether or not a respondent moved during the eight year observation period. Individuals who died between 2000 and 2008 are predicted to have higher BMI in 2000 (b=2.66, p<.01 but a reduced rate of change (b= 0.41, p<.01).Female BMI is predicted to increase by 0.20 points over two year intervals (b=0.20, p<.05). Model 2 additionally adjusts for neighborhood demographics. Percent rural, per cent black and percent Hispanic are associated with BMI in 2000. Females who live in a more rural census tract are predicted to have a higher BMI in 2000 *b=0.11, p<.01). Females who live in census tracts with a higher percentage of blacks, and females who live in areas with a higher percentage of Hispanics are also predicted to have a higher BMI in 2000 (b=0.42, p<.01; b=0.26, p<.01 respectively ). Neighborhood demographics are not associated with rate of change in BMI Model 3 additionally adjusts for nei ghborhood socioeconomic characteristics. Females who live in census tract with a higher percentage of males with a college degree are predicted to have a lower BMI (b= 3.01, p<.01). No measure of neighborhood socioeconomic status is associated rate of chan ge in BMI Model 4 additionally adjusts for walkability, which is not associated with initial levels or rate of change in BMI.

PAGE 106

106 Table 5 8 presents estimated random effects and measures of model fit for each model in Table 5 7. There is significant variation in rate of change, within an individual, between individuals and between census tracts in model 1. Most of the variation in BMI ( 87.8 %) occurs between individuals. Approximately 4.3 % of the variation in fe male BMI is due to changes within an in dividual. 5.2% of variation in female BMI is associated with the census tract level. Once models adjust for census tract level demographics, an insignificant amount of variation in female BMI is associated with the census tract level. According to all thre e measures of model fit, model 4 is the best fitting model. Male BMI in 2000 Table 5 9 presents results from a hierarchical linear model regressing adjusted male BMI in 2000 on individual and neighborhood characteristics. Chapter 4 presented results from models that incrementally adjusted for sets of covariates that did not include neighborhood characteristics. Thus the first model in Table 5 9 presents results from a model that fully adjusts for individual characteristics as well as the set of neighborhoo d demographic variables. Of neighborhood demographics, only percent black is associated with BMI in 2000 individuals who live in a census tract with a higher proportion of black respondents are predicted to have a higher BMI (b=0.09, p<.01). Of individual demographic characteristics, being older is associated with a lower BMI in 2000 (b= 0.13, p<.01); foreign born respondents are predicted to have a lower BMI (b= 1.08, p<.01); and individuals with more years of education are predicted to have a lower BMI ( b= 0.11, p<.01). Married males are predicted to have a higher BMI (b=1.03, p<.01). Of the health and health behavior measures, functional status, self rated health, smoking, and visiting a doctor are associated with male BMI. Those with

PAGE 107

107 greater number of I ADLs and ADLs, higher self rated health, and who report being current smokers are predicted to have a lower BMI in 2000. Respondents who report they visited a doctor in the past two years are predicted to have a higher BMI. Model 2 additionally adjusts fo r neighborhood socioeconomic characteristics. Males who live in a census tract with a higher percentage of college educated males are predicted to have a lower BMI in 2000 (b= 1.16, p<.05). Once neighborhood socioeconomic controls are adjusted for, percent age of black residents living in a census tract is no longer associated with BMI. Model 3 additionally adjusts for walkability, which is not associated with BMI. Although walkability is not significantly associated with BMI, percentage of males with a coll ege degree is no longer associated with BMI. Table 5 10 presents estimated random effects and measures of model fit for each model in Table 5 9. Ninety eight percent of the variation in BMI occurs within census tracts, and there is an insignificant amount of variation between census tracts. According to all three measures of model fit, model 3 is the best fitting model. Female BMI in 2000 Table 5 11 presents results from a hierarchical linear model regressing adjusted female BMI in 2000 on individual and neighborhood characteristics. The first model in Table 5 11 presents results from a model that fully adjusts for individual characteristics as well as the set of neighborhood demographic variables. Of neighborhood demographics, only percent rural is associ ated with BMI in 2000 females who live in a more rural census tract are predicted to have a higher BMI (b=0.07, p<.01). Of individual demographic characteristics, being older is associated with a lower BMI in 2000 (b= 0.10, p<.01); foreign born respondents are predicted to have a lower BMI (b=

PAGE 108

108 0.93, p<.01); and individuals with more years of education are predicted to have a lower BMI (b= 0.17, p<.01). Compared to white females, black females are predicted to have higher BMI (b=2.35, p<.01). Of the health a nd health behavior measures status, self rated health and smoking are associated with female BMI. Females who report higher self rated health and who report being current smokers are predicted to have a lower BMI in 2000. Females who report a greater numbe r IADLs and ADLs are predicted to have a higher BMI (b= .34, p<.01). Model 2 additionally adjusts for neighborhood socioeconomic characteris tics. Fem ales who live in a census tract with a higher percentage of college educated males are predicted to h ave a lower BMI in 2000 (b= 1.55, p<.01 ). Once neighborhood socioeconomic controls are adjusted for, a higher percentage of black residents living in a census tract is associated with a lower BMI (b= 0.12, p<.01) Percent rural is no longer associated with BMI. Of the individual characteristics, being a former smoker is associated with a higher BMI (b=0.39, p<.05). Model 3 additionally adjusts for walkability, which is not associated with BMI. A higher percentage of college educated males remains associated with a lower female BMI (b= 1.35, p<.01). Of the individual characteristics, being older, having more years of education, having better self rated health, and being a current smoker remain associated with a lower BMI. Black females are still predicted to have a higher BMI relative to white (b=2.38, p<.01), as are females who report more IADLs and ADLs (b=0.36, p<.01). Table 5 12 presents estimated random effects and measures of model fit for each model in Table 5 11 About 98 of the variation in BMI occurs wit hin census tracts, and

PAGE 109

109 there is an insignificant amount of variation between census tracts. According to all three measures of model fit, model 3 is the best fitting model. Changes in Male BMI, 2000 2008 Table 5 13 presents results from three level hierarchical linear models predicting adjusted adult BMI among males between 2000 and 2008. The three level models, in addition to testing for an association between neighborhood characteristics and BMI in 2000, also test whether individual and neighborhood characteristics are associated with changes in BMI between 2000 and 2008. Time is coded the same as in previous models. Model 1 includes controls for proxy status, whether or not a respondent died during the e ight year observation period, attrition due to non mortality, and whether or not a respondent moved during the eight year observation period. Chapter 4 presented results from models that incrementally adjusted for sets of covariates that did not include ne ighborhood characteristics, thus the first model in Table 5 13 presents results from a model that fully adjusts for individual characteristics as well as the set of neighborhood demographic variables. After controls for individual demographics, socioecono mic status, and health status, neighborhood demographics are not associated with initial or rate of change in BMI among males. Of the individual characteristics, being older, having more years of education, and being a current or former smoker are associat ed with adult BMI in 2000. Older males are predicted to have lower BMI (b= 0.11, p<.01), as are males with more years of education (b= 0.13, p<.01). Males who currently smoke, or report being a former smoker, are also predicted to have a lower BMI (b= 1.41 p<.01; b= 0.62, p<.01 respectively ). Regarding rate of change, the only measure associated with change in BMI was being a former smoker. Males who reported being a former smoker are

PAGE 110

110 expected to have an increased rate of BMI change relative to those who never smoked (b=0.09, p<.01). Model 2 additionally includes neighborhood socioeconomic status. No measure of socioeconomic status is associated with either BMI in 2000 or changes in BMI between 2000 and 2008. Model 3 additionally includes walkability, whi ch is not associated with either BMI in 2000 or changes in BMI. In the full model, model 3, being older, having more years of education, and being either a current or former smoker are associated with a lower BMI. Other than respondents who died between 20 00 and 2008, no measure at either the individual or neighborhood level is associated with change in BMI. Table 5 14 presents estimated random effects and measures of model fit for each model in Table 5 13. There is significant variation in rate of change, within an individual, and between individuals in all four models. Most of the variation in BMI (91.3%) occurs between individuals. Approximately 8.7% of the variation in male BMI is due to changes within an individual. There is no significant variation be tween census tracts in male BMI. According to all three measures of model fit, model 4 is the best fitting model. Changes in Female BMI 2000 2008 Table 5 15 presents results from three level hierarchical linear models predicting adjusted adult BMI among fe males between 2000 and 2008. After controls for individual demographics, socioeconomic status, and health status, neighborhood demographics are not associated with initial or rate of change in BMI among males. Of the individual demographics, being black or Hispanic, and being older, are associated with female BMI in 2000. Both black and Hispanic females are predicted to have a higher BMI than white females (b=3.18, p<.01; b=1.68, p<.01 respectively ). Of the individual socioeconomic characteristics, havin g more years of education is associated with a

PAGE 111

111 lower BMI in 2000 (b= 0.19, p<.01). Of the measures of health and health behaviors, females who currently smoke are predic ted to have a lower BMI (b= 1.15, p<.01 ) Being older was the only measure associated w ith change in BMI Older fem ales are expected to have an de creased rate of BMI change (b= 0.01 p<.01). Model 2 additionally includes neighborhood socioeconomic status. Females who live in a census tract with a higher percentage of males with at least a c ollege degree are predicted to have a lower BMI (b= 2.77, p<.01). No measure of neighborhood socioeconomic status is associated with rate of change in BMI. Model 3 additionally includes walkability, which is not associated with either BMI in 2000 or change s in BMI. In the full model, model 3, being black or Hispanic remains associated with a higher initial BMI. Being older, having more years of education, being a current smoker, and having better self rated health are associated with a lower initial BMI. O ther than respondents who died between 2000 and 2008 and age no measure at either the individual or neighborhood level is associated with change in BMI. Table 5 16 presents estimated random effects and measures of model fit for ea ch model in Table 5 15 T here is significant variation in rate of change, within an individual, between individuals and between census tracts in all four models. Mo st of the variation in BMI (88.4 %) occurs betwee n individuals. Approximately 8.6 % of the variation in fe male BMI is due to changes within an individual. According to all three measures of model fit, model 3 is the best fitting model. Summary The primary objective of analyses in Chapter 5 was to test for associations between neighborhood socio demographic characteristic s and adult weight, net of individual characteristics. A parallel objective was to test whether individual socio

PAGE 112

112 demographic characteristics are associated with adult weight net of neighborhood characteristics. A third objective was to test whether individ ual and neighborhood socio demographic characteristics were associated with changes in adult BMI between 2000 and 2008. Among males, neighborhood demographics that predicted BMI in 2000 were percent Hispanic and percent foreign born however these associat ions became insignificant once models controlled for neighborhood socio economic status. A higher percentage of Hispanics and foreign born residents were also associated with increased BMI among females, and these associations were again explained by neighb orhood socioeconomic characteristics. A particularly important measure of neighborhood socioeconomic status for both males and females was percent of males with at least a college degree the higher the percentage of males with a BA the lower the predicted BMI. Once individual characteristics were included in the model for men, percent Hispanic was no longer associated with BMI in 2000 even before adjustment for neighborhood socioeconomic characteristics. In the full model for males which included all individual measures and a measure of neighborhood walkability, neighborhood socioeconomic status was no longer associated with weight. Once individual characteristics were included in the model for women, demographic characteristics had inconsistent associ ations with BMI. The one neighborhood measure consistently associated with female BMI in 2000 was percentage of males with at least a college degree.

PAGE 113

113 The most consistent individual measures associated with a lower BMI in 2000 for both men and women were more years of education, better self rated health, being foreign born, and being a current smoker. These associations persisted through adjustments for all neighborhood socio demographics. Individual characteristics associated with a higher BMI in fully ad justed models were different for men and women. Among men, being married and having visited a doctor in the past two years was associated with a higher BMI, whereas among women being black was associated with a higher BMI. Functional status seems to influe nce BMI differently between men and women. Number of IADLs and ADLs is inversely associated with BMI among men, but positively associated among women. Although there is only modest evidence of an association between neighborhood characteristics and BMI among older females, there is even less evidence of an association with changes in BMI. For both males and females, no measure of neighborhood demographic s or socioeconomic status is associated with change in BMI between 2000 and 2008. The lack of a significant association between neighborhood socio demographics and changes in BMI is likely due to the relatively small amount of variation in BMI that occurs within individuals over an eight year period. If variation in BMI over ti m e was observed across a longer period of time, or duri ng a different part of the life course, neighborhood characteristics may be more likely to have an association with changes in B MI. Nonetheless percent of males with a BA is associated with a lower initial BMI among females. This association may explain the residual association

PAGE 114

114 Chapter 6 includes the primary individual characteristic of interest in models predicting adult BMI childhood conditions.

PAGE 115

115 Table 5 1. Results from a Hierarchical Linear Model Regressing Adjusted Male BMI in 2000 on Neighborhood Characteristics HRS Cohort 2000 Covariates Model 1 Model 2 Model 3 Model 4 Fixed Effects Intercept 28.55 (.08) ** 28.56 (.08) ** 28.58 (.08) ** 28.57 (.09)** Interval Model Controls Proxy Status 0.52 (.22) ** 0.56 (.06) ** 0.57 ( .56) ** 0.50 (.22) Neighborhood Demographics Percent Rural 0.01 (.01) 0.01 (.00) 0.02 (.02) Percent Over 65 0.06 (.06) 0.07 (.04) 0.06 (.06) Percent Black 0.06 (.06) 0.01 (.01) 0.00 (.04) Percent Hispanic 0.12 (.12)* 0.03 (.06) 0.01 (.01) Percent Foreign Born 0.25 (.10) 0.12 (.10) 0.11 (.12) Neighborhood SES Street Connectivity 0.33 (1.27) Percent Families Poverty 0.14 (.16) 0.09 (.16) Percent Vacant Houses 0.10 (.12) 0.10 (.12) Percent Unemployed 0.14 (.27) 0.16 (.28) Percent Males w/BA 1.31 (.60) 1.30 (.61) Percent Females w/BA 0.53 (.61) 0.59 (.62) Median Household Value 0.02 (.01) 0.02(.01) Notes: BMI = body mass index; SRH = self rated health; Median Household Value in in Tens of Thousands of Dollars *p < .05; **p < .01.

PAGE 116

116 Table 5 2. Variance Components and Fit Statistics from an HLM Model Estimating Adjusted Male BMI, in 2000 Variance Components Model 1 Model 2 Model 3 Model 4 In rate of Change Within group 21.06** 21.03** 21.03** 37.0.** Between Group Level 2 0.73* 0.70 0.65 0.27 Between Group Level 3 N, Level 1 3689 3683 3683 3615 N, Level 2 1992 1988 1988 1941 N, Level 3 Goodness of fit Deviance 21834.5 21810.7 21802.5 21392.4 AIC 21838.5 21814.7 21806.5 21396.4 BIC 21849.7 21814.7 21817.7 21407.5 Notes: *p < .05; **p < .01.

PAGE 117

117 Table 5 3. Results from a Hierarchical Linear Model Regressing Adjusted Female BMI in 2000 on Neighborhood Characteristics HRS Cohort 2000 Covariates Model 1 Model 2 Model 3 Model 4 Fixed Effects Intercept 28.50 ( .15 ) ** 28.53 ( .14 ) ** 28.54 ( .14 ) ** 28.54 ( .14 ) ** Interval Model Controls Proxy Status 1.73 ( .59 ) ** 1.68 ( .58 ) ** 1.54 ( .58 ) ** 1.55 ( .58 ) ** Neighborhood Demo s Percent Rural 0.12 ( .03 ) ** 0.01 ( .03 ) 0.02 ( .03 ) Percent Over 65 0.11 ( .14 ) 0.24 ( .14 ) 0.24 ( .14 ) Percent Black 0.40 ( .04 ) ** 0.19 ( .05 ) ** 0.19 ( .05 ) ** Percent Hispanic 0.35 ( .09 ) ** 0.09 ( .11 ) 0.01 ( .11 ) Percent Foreign Born 0.30 ( .14 ) ** 0.06 ( .15 ) 0.04 ( .16 ) Neighborhood SES Street Connectivity 1 .22 ( 1.52 ) Pct. Families Poverty 0.07 ( .19 ) 0.14 ( .19 ) Pct. Vacant Houses 0.17 ( .14 ) 0.13 ( .15 ) Pct. Unemployed 0.31 ( .35 ) 0.37 ( .35 Pct. Males w/BA 2.21 ( .72 ) ** 1.99 ( .72 ) ** Pct. Females w/BA 0.42 ( .73 ) 0.58 ( .74 ) Median Household Value 0.01 ( .01 ) 0.01 ( .01 ) Notes: BMI = body mass index; SRH = self rated health; Median Household Value in in Tens of Thousands of Dollars *p < .05; **p < .01.

PAGE 118

118 Table 5 4. Variance Components and Fit Statistics from an HLM Model Estimating Adjusted Female BMI, in 2000 Variance Components Model 1 Model 2 Model 3 Model 4 In rate of Change Within group 37.47** 37.28** 37.04** 37.0.** Between Group Level 2 1.72** 0.70 0.67 0.27 Between Group Level 3 N, Level 1 4516 4510 4510 4417 N, Level 2 2351 2347 2347 2288 N, Level 3 Goodness of fit Deviance 29371.9 29215.0 29187.3 28534.4 AIC 29375.9 29219.0 29191.3 28538.4 BIC 29387.4 29230.6 29202.8 28549.9 Notes: *p < .05; **p < .01.

PAGE 119

119 Table 5 5. Results from a Hierarchical Linear Model Regressing Male Adjusted BMI on Neighborhood Characteristics HRS Cohort 2000 20008. Covariates Model 1 Model 2 Model 3 Model 4 For Initial Status (BMI 2000) Fixed Effects Intercept 28.39 (.18) ** 28.40 (.18) ** 28.40 (.18) ** 28.40 (.18) ** Proxy Status 0.18 (.35) 0.12 (.35) 0.12 (.35) 0.16 (.36) Died 0.73 (.48) 0.64 (.48) 0.61 (.48) 0.64 (.48) Attrit 0.25 (.24) 0.18 (.35) 0.21 (.36) 0.18 (.36) Moved Since 2000 0.03 (.25) 0.00 (.25) 0.02 (.26) 0.00 (.26) Neighborhood Demographics Percent Rural 0.03 (.04) 0.01 (.04) 0.01(.04) Percent Over 65 0.09 (.18) 0.19 (.19) 0.17 (.19) Percent Black 0.11 (.05) ** 0.10 (.07) 0.09 (.07) Percent Hispanic 0.21 (.09) 0.12 (.11) 0.13 (.11) Percent Foreign Born 0.27 (.16) 0.15 (.17) 0.24 (.19) Neighborhood SES Street Connectivity 1.52 (1.97) Percent Families Poverty 0.18 (.25) 0.03 (.26) Percent Vacant Houses 0.17 (.19) 0.15 (.19) Percent Unemployed 0.12 (.47) 0.06 (.48) Percent Males w/BA 1.03 (.93) 1.17 (.94) Percent Females w/BA 0.15 (.95) 0.33 (.96) For Rate of Change (2000 2008) Interval 0.07 (.06) 0.07 (.03) 0.07 (.03) 0.07 (.03) Proxy Status 0.03 (.18) 0.02 (.06) 0.03 (.35) 0.03 (.35) Died 0.41 (.09) ** 0.40 (.09) ** 0.40 (.09) ** 0.42 (.49) ** Attrit 0.04 (.06) 0.03 (.06) 0.03 (.06) 0.03 (.36) Moved Since 2000 0.01 (.25) 0.02 (.04) 0.03 (.04) 0.03 (.26) Neighborhood Demographics Percent Rural 0.01 (.01) 0.00 (.01) 0.03 (.01) Percent Over 65 0.05 (.03) 0.06 (.03) 0.06 (.03) Percent Black 0.02 (.01) 0.02 (.01) 0.02 (.01) Percent Hispanic 0.02 (.01) 0.02 (.01) 0.02 (.02) Percent Foreign Born 0.02 (.02) 0.01 (.02) 0.02 (.03) Neighborhood SES Street Connectivity 0.10 (.09) Percent Families Poverty 0.02 (.04) 0.03 (.04) Percent Vacant Houses 0.01 (.03) 0.03 (.03) Percent Unemployed 0.02 (.07) 0.06 (.07) Percent Males w/BA 0.15 (.14) 0.03 (.14) Percent Females w/BA 0.03 (.14) 0.03 (.14) Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 120

120 Table 5 6. Variance Components and Fit Statistics from an HLM Model Estimating Adjusted Male BMI, in 2000 2008 Variance Components Model 1 Model 2 Model 3 Model 4 In rate of Change 0.41** 0.42** 0.42** 0.42** Within group 2.68** 2.67** 2.66** 2.69** Between Group Level 2 33.38** 33.43** 33.46** 33.09** Between Group Level 3 0.01 0.01 0.01 0.01 N, Level 1 11834 11815 11815 11610 N, Level 2 3702 3696 3696 3628 N, Level 3 1992 1988 1988 1941 Goodness of fit Deviance 58004.9 57903.3 57918.6 56941.1 AIC 58016.9 57915.3 57930.6 56953.1 BIC 58050.4 57948.9 57964.2 56986.5 Notes: *p < .05; **p < .01.

PAGE 121

121 Table 5 7. Results from a Hierarchical Linear Model Regressing Female Adjusted BMI on Neighborhood Characteristics HRS Cohort 2000 20008. Covariates Model 1 Model 2 Model 3 Model 4 For Initial Status (BMI 2000) Fixed Effects Intercept 27.97** 27.90** 27.86** 27.88** Proxy Status 0.98 0.97 1.01 1.08 Died 2.66** 2.39** 2.21** 1.91** Attrit 0.27 0.41 0.44 0.56 Moved Since 2000 0.05 0.12 0.26 0.29 Neighborhood Demographics Percent Rural 0.11** 0.01 0.01 Percent Over 65 0.26 0.37 0.42* Percent Black 0.42** 0.17* 0.16* Percent Hispanic 0.26** 0.01 -0.08 Percent Foreign Born 0.30 0.09 0.14 Neighborhood SES Street Connectivity 1.59 Percent Families Poverty 0.05 0.15 Percent Vacant Houses 0.03 0.01 Percent Unemployed 0.61 0.67 Percent Males w/BA 3.01* 2.61** Percent Females w/BA 0.47 1.06 For Rate of Change (2000 2008) Interval 0.20** 0.20** 0.20** 0.19** Proxy Status 0.21 0.34 0.20 0.22 Died 0.41** 0.41* 0.41* 0.36* Attrit 0.27 0.41 0.01 0.03 Moved Since 2000 0.03 0.03 0.02 0.03 Neighborhood Demographics Percent Rural 0.01 0.01 0.01 Percent Over 65 0.03 0.04 0.04 Percent Black 0.01 0.00 0.00 Percent Hispanic 0.00 0.01 0.01 Percent Foreign Born 0.01 0.01 0.20 Neighborhood SES Street Connectivity 0.24 Percent Families Poverty 0.00 0.02 Percent Vacant Houses 0.01 0.00 Percent Unemployed 0.02 0.03 Percent Males w/BA 0.15 0.16 Percent Females w/BA 0.03 0.03 Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 122

122 Table 5 8. Variance Components and Fit Statistics from an HLM Model Estimating Adjusted Female BMI, in 2000 2008 Variance Components Model 1 Model 2 Model 3 Model 4 In rate of Change 0.65** 0.65** 0.65** 0.65** Within group 4.02** 4.03** 4.03** 4.04** Between Group Level 2 53.08** 53.11** 52.54** 52.34** Between Group Level 3 2.76** 1.29 0.79 0.79 N, Level 1 15552 15532 15532 15224 N, Level 2 4624 4618 4618 4522 N, Level 3 2351 2347 2347 2288 Goodness of fit Deviance 83079.2 82886.2 82875.8 81233.2 AIC 83091.2 82898.2 82887.8 81245.2 BIC 83125.8 82932.7 82922.4 81279.6 Notes: *p < .05; **p < .01.

PAGE 123

123 Table 5 9. Results from a Hierarchical Linear Model Regressing Male Adjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000. Covariates Model 1 Model 2 Model 3 For Initial Status (BMI 2000) Fixed Effects Intercept 28.81** 28.67** 28.81** Model Controls Proxy Status 0.69** 0.69** 0.63** Adult Demographics Age 0.13** 0.12** 0.13** White (ref) Black 0.39 0.30 0.26 Other 0.89 0.87 0.72 Hispanic (ref=not Hispanic) 0.09 0.18 0.08 Foreign Born 1.08** 1.04** 1.12** Years of Education 0.11** 0.09** 0.09** Household Income 0.00 0.00 0.00 Married 1.03** 1.04** 1.05** Health Functional Status 0.15* 0.15* 0.16* Self Rated Health 0.33** 0.30** 0.33** Current Smoker 1.83** 1.85** 1.89** Former Smoker 0.08 0.11 0.11 Visited Doctor 0.82** 0.87** 0.82* Visited Hospital 0.07 0.08 0.06 No Insurance 0.02 0.03 0.04 Neighborhood Demographics Percent Rural 0.01 0.03 0.03 Percent Over 65 0.14 0.13 0.11 Percent Black 0.09* 0.03 0.02 Percent Hispanic 0.09 0.01 0.02 Percent Foreign Born 0.13 0.01 0.02 Neighborhood SES Street Connectivity 0.11 Percent Families Poverty 0.13 0.07 Percent Vacant Houses 0.07 0.07 Percent Unemployed 0.29 0.32 Percent Males w/BA 1.16* 1.10 Percent Females w/BA 0.68 0.69 Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 124

124 Table 5 10. Variance Components and Fit Statistics from an HLM Model Estimating Adjusted Male BMI in 2000. Variance Components Model 1 Model 2 Model 3 In rate of Change Within group 20.02** 20.04** 19.93** Between Group Level 2 0.40 0.34 0.37 Between Group Level 3 N, Level 1 3512 3512 3446 N, Level 2 1914 1914 1868 N, Level 3 Goodness of fit Deviance 20599.1 20591.6 20192.9 AIC 20603.1 20595.6 20196.9 BIC 20614.2 20606.7 20208.0 Notes: *p < .05; **p < .01.

PAGE 125

125 Table 5 11. Results from a Hierarchical Linear Model Regressing Female Adjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000. Covariates Model 1 Model 2 Model 3 Fixed Effects Intercept 32.05** 31.85** 31.89** Model Controls Proxy Status 3.08** 2.93** 2.96** Adult Demographics Age 0.10** 0.10** 0.10** White (ref) Black 2.35** 2.48** 2.38** Other 0.66 0.68 0.69 Hispanic (ref=not Hispanic) 0.45 0.74 0.77 Foreign Born 0.93** 0.85* 0.78* Years of Education 0.17** 0.13** 0.11** Household Income 0.01 0.01 0.01 Married 0.38 0.38 0.34 Health Functional Status 0.34** 0.35** 0.36** Self Rated Health 1.07** 1.03** 1.05** Current Smoker 2.70** 2.73** 2.72** Former Smoker 0.37 0.39* 0.39 Visited Doctor 0.34 0.34 0.35 Visited Hospital 0.28 0.25 0.25 No Insurance 0.49 0.46 0.49 Neighborhood Demographics Percent Rural 0.07* 0.01 0.00 Percent Over 65 0.04 0.19 0.19 Percent Black 0.03 0.12* 0.12 Percent Hispanic 0.11 0.13 0.17 Percent Foreign Born 0.21 0.04 0.01 Neighborhood SES Street Connectivity 0.39 Percent Families Poverty 0.12 0.05 Percent Vacant Houses 0.23 0.19 Percent Unemployed 0.40 0.49 Percent Males w/BA 1.55** 1.35* Percent Females w/BA 0.70 0.89 Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 126

126 Table 5 12. Variance Components and Fit Statistics from an HLM Model Estimating Adjusted Female BMI in 2000. Variance Components Model 1 Model 2 Model 3 In rate of Change Within group 33.18** 33.05** 33.03** Between Group Level 2 0.61 0.42 0.33 Between Group Level 3 N, Level 1 4349 4349 4260 N, Level 2 2256 2256 2240 N, Level 3 Goodness of fit Deviance 27679.8 27638.6 27060.4 AIC 27683.8 27642.6 27064.4 BIC 27695.2 27654.0 27075.8 Notes: *p < .05; **p < .01.

PAGE 127

127 Table 5 13. Results from a Hierarchical Linear Model Regressing Male Adjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000 20008. Covariates Model 1 Model 2 Model 3 Fixed Effects Intercept 29.74** 29.85** 29.99** Model Controls Died 0.83 0.78 0.84** Attrit 0.02 0.01 0.04 Proxy Status 0.07 0.14 0.09 Moved Since 2000 0.07 0.03 0.07 Adult Demographics Age 0.11** 0.11** 0.11** White (ref) Black 0.71 0.41 0.57 Other 0.14 0.10 0.20 Hispanic (ref=not Hispanic) 0.00 0.15 0.11 Foreign Born 0.61 0.72 0.74 Years of Education 0.13** 0.12** 0.12** Household Income 0.00 0.00 0.00 Married 0.51 0.39 0.46 Health Functional Status 0.06 0.06 0.08 Self Rated Health 0.20 0.19 0.22 Current Smoker 1.41** 1.39** 1.42** Former Smoker 0.62* 0.62* 0.65* Visited Doctor 0.48 0.31 0.36 Visited Hospital 0.12 0.12 0.12 No Insurance 0.19 0.08 0.05 Neighborhood Demographics 0.02 Percent Rural 0.02 0.03 0.04 Percent Over 65 0.07 0.04 0.02 Percent Black 0.05 0.06 0.00 Percent Hispanic 0.01 0.04 0.07 Percent Foreign Born 0.00 0.61 0.22 Neighborhood SES Street Connectivity 1.72 Percent Families Poverty 0.16 0.00 Percent Vacant Houses 0.10 0.12 Percent Unemployed 0.28 0.31 Percent Males w/BA 0.62 0.75 Percent Females w/BA 0.18 0.21 Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 128

128 Table 5 13 Continued Covariates Model 1 Model 2 Model 3 For Rate of Change (2000 2008) Fixed Effects Interval 0.21 0.18 0.19 Model Controls Died 0.37** 0.36** 0.38** Attrit 0.00 0.00 0.00 Proxy Status 0.01 0.02 0.02 Moved Since 2000 0.01 0.01 0.02 Adult Demographics Age 0.01 0.01 0.00 White (ref) Black 0.13 0.05 0.06 Other 0.17 0.15 0.17 Hispanic (ref=Not Hispanic) 0.01 0.03 0.06 Foreign Born 0.00 0.00 0.02 Years of Education 0.01 0.01 0.01 Household Income 0.00 0.00 0.00 Married 0.03 0.01 0.02 Health Functional Status 0.01 0.00 0.00 Self Rated Health 0.04 0.04 0.04 Current Smoker 0.13 0.12 0.01 Former Smoker 0.09* 0.08 0.08 Visited Doctor 0.11 0.08 0.09 Visited Hospital 0.01 0.01 0.00 No Insurance 0.00 0.04 0.01 Neighborhood Demographics Percent Rural 0.00 0.00 0.00 Percent Over 65 0.04 0.04 0.04 Percent Black 0.01 0.01 0.01 Percent Hispanic 0.02 0.02 0.02 Percent Foreign Born 0.02 0.02 0.02 Neighborhood SES Street Connectivity 0.12 Percent Families Poverty 0.01 0.01 Percent Vacant Houses 0.03 0.03 Percent Unemployed 0.03 0.01 Percent Males w/BA 0.12 0.12 Percent Females w/BA 0.08 0.07 Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 129

129 Table 5 14. Variance Components and Fit Statistics from an HLM Model Estimating Changes in Adjusted Male BMI (2000 2008). Variance Components Model 1 Model 2 Model 3 In rate of Change 0.40** 0.41** 0.41** Within group 2.67** 2.61** 2.63** Between Group Level 2 32.30** 32.46** 32.02** Between Group Level 3 0.01 0.01 0.01 N, Level 1 10512 10512 11123 N, Level 2 N, Level 3 1811 1811 1768 Goodness of fit Deviance 51488.7 51491.5 50612.2 AIC 51500.7 51503.9 50624.2 BIC 51533.7 51536.6 50657.1 Notes: *p < .05; **p < .01.

PAGE 130

130 Table 5 15. Results from a Hierarchical Linear Model Regressing Female Adjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000 20008. Covariates Model 1 Model 2 Model 3 Fixed Effects Intercept 29.08** 29.08** 28.97** Model Controls Died 2.27** 2.35** 2.07** Attrit 0.28 0.37 0.50 Proxy Status 0.83 0.78 0.85 Moved Since 2000 0.11 0.18 0.23 Adult Demographics Age 0.03 0.02 0.02 White (ref) Black 3.18** 3.37** 3.31** Other 0.40 0.24 0.33 Hispanic (ref=not Hispanic) 1.68** 2.12** 2.07** Foreign Born 0.55 1.04* 1.02* Years of Education 0.19** 0.23** 0.25** Household Income 0.01 0.01 0.01 Married 0.24 0.14 0.29 Health Functional Status 0.12 0.05 0.06 Self Rated Health 0.39** 0.39** 0.37** Current Smoker 1.15** 1.26** 1.20** Former Smoker 0.00 0.02 0.04 Visited Doctor 0.29 0.05 0.01 Visited Hospital 0.37 0.36 0.36 No Insurance 0.27 0.38 0.39 Neighborhood Demographics Percent Rural 0.05 0.05 0.03 Percent Over 65 0.27 0.43* 0.48* Percent Black 0.02 0.21 0.18 Percent Hispanic 0.01 0.32* 0.39** Percent Foreign Born 0.23 0.01 0.05 Neighborhood SES Street Connectivity 2.31 Percent Families Poverty 0.03 0.08 Percent Vacant Houses 0.08 0.06 Percent Unemployed 0.68 0.73 Percent Males w/BA 2.77** 2.33* Percent Females w/BA 0.46 0.88 Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 131

131 Table 5 15 Continued Covariates Model 1 Model 2 Model 3 For Rate of Change (2000 2008) Fixed Effects Interval 0.02 0.03 0.05 Model Controls Died 0.47** 0.49** 0.44** Attrit 0.06 0.03 0.01 Proxy Status 0.22 0.21 0.23 Moved Since 2000 0.01 0.01 0.01 Adult Demographics Age 0.01** 0.01** 0.01** White (ref) Black 0.06 0.07 0.07 Other 0.20 0.15 0.18 Hispanic (ref=Not Hispanic) 0.12 0.09 0.09 Foreign Born 0.12 0.06 0.05 Years of Education 0.00 0.01 0.01 Household Income 0.00 0.00 0.00 Married 0.04 0.03 0.05 Health Functional Status 0.01 0.00 0.00 Self Rated Health 0.04 0.04 0.04 Current Smoker 0.02 0.03 0.04 Former Smoker 0.05 0.04 0.03 Visited Doctor 0.06 0.02 0.02 Visited Hospital 0.02 0.05 0.03 No Insurance 0.08 0.08 0.09 Neighborhood Demographics Percent Rural 0.00 0.00 0.01 Percent Over 65 0.04 0.04 0.05 Percent Black 0.00 0.01 0.01 Percent Hispanic 0.01 0.03 0.02 Percent Foreign Born 0.03 0.01 0.02 Neighborhood SES Street Connectivity 0.28 Percent Families Poverty 0.01 0.01 Percent Vacant Houses 0.02 0.02 Percent Unemployed 0.03 0.01 Percent Males w/BA 0.21 0.15 Percent Females w/BA 0.01 0.04 Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 132

132 Table 5 16. Variance Components and Fit Statistics from an HLM Model Estimating Changes in Adjusted Female BMI (2000 2008). Variance Components Model 1 Model 2 Model 3 In rate of Change 0.59** 0.59** 0.59** Within group 3.96** 3.96** 3.97** Between Group Level 2 46.93** 46.61** 46.23** Between Group Level 3 1.59* 1.15 1.19 N, Level 1 13767 13767 13523 N, Level 2 4107 4107 4029 N, Level 3 2136 2136 2083 Goodness of fit Deviance 73015.3 72982.3 71685.3 AIC 73027.3 72994.3 71707.3 BIC 73061.3 73028.3 71741.1 Notes: *p < .05; **p < .01.

PAGE 133

133 CHAPTER 6 CHILDHOOD CONDITIONS NEIGHBORHOODS, AND ADULT WEIGHT Chapter 4 presented evidence of a persistent association between childhood conditions and adult BMI among females. Higher paternal education was associated with lower adult BMI in 2000 after model adjustments for individual demographics, socioeconomic status, health status, and health behaviors. Chapter 5 presented evidence that neighborhood socioeconomic status, as me asured by percentage of males with at least a college degree, is associated with adult BMI among females. This association exists independent ly of controls for individual socioeconomic status. Chapter 6 presents results from an analysis that tests whether the residual association between childhood conditions and female adult BMI is explained by neighborhood socio demographics. tests whether neighborhood characteristics explain the residual asso ciation between childhood conditions and adult weight in females in the United States population. Among males, there was no residual association between childhood conditions and adult BMI, indicating that adult socioeconomic position may be play a more imp ortant role in the link between childhood conditions and adult weight in males than females. The results for males are presented in Table 4 1, 4 2, 4 5, and 4 6. These results are only briefly discussed because of the lack of residual association between c hildhood conditions and adult weight in males. The lack of a residual association after controls for adult socioeconomic position among males is consistent with what most other studies have found (Senese et al. 2009)

PAGE 134

134 Female BMI in 2000 Table 6 3 presents results from a two level hierarchical linear model regressing female BMI in 2000 on childhood conditions, adult characteristics, and neighborhood characteristics. Model 1 includes controls for proxy status childhood conditions, adult socio demographics, health status, and health behaviors. This is the same model presented in Chapter 4 that regressed female BMI in 2000 on individual characteristics but with an additional adjustment for neighborhood demographics. Females residing in more rural census tracts are predicted to have a higher BMI in 2000 than those in less rural census tracts (b=0.06, p<.05). After controls for neighborhood demographics, paternal education remains associated with a lower adult BMI (b= 0.09, p<.01). Individual adult demographics associated with BMI are being black and foreign born. Black females are predicted to have a higher BMI relative to white females (b=2.56, p<.01), while foreign born females are predicted to have a lower BMI (b= 0.97, p<.01). Individual educational attainment is associated with a lower BMI (b= 0.10, p<.05). Individual health characteristics associated with BMI are functional status, self rated health, and smoking. Females who report a greater number of IADLs and ADLs are predicted to have a higher BMI in 2000. Females who report better self rated health or c urrently smoke are predicted to have a lower BMI ( b= 1.07, p<.01; b= 2.67, p<.01 respectively ). Females who visited a hospital in the past two years are predicted to have a higher BMI (b=0.66, p<.05) while females who have no insurance are predicted to ha ve a lower BMI (b= 0.56). Model 2 additionally adjusts for neighborhood socioeconomic characteristics. Percent unemployed and percent of males with a college degree or more are associated with female BMI in 2000. Living in a census tract with higher unem ployment

PAGE 135

135 is associated with a higher BMI (b=0.69, p<.05) and living in a census tract with a higher percent of males with a college degree is associated with a lower BMI (b= 1.53, p<.01). Of the childhood conditions, paternal education remains associated w ith female BMI in 2000 (b= 0.09, p<.01). Being black or Hispanic remains associated with a higher BMI, while being foreign born and having more years of education is associated with a lower BMI. After controls for socioeconomic status, percent rural is no longer associated with female BMI. However, percent black is associated with a lower BMI (b= 0.14, p<.01). Model 3, additionally adjusts for walkability, which is not associated with female BMI in 2000. In the full model, paternal education remains associ ated with adult female BMI. Being black or Hispanic remains associated with a higher BMI, while being foreign born and having more years of education is associated with a lower BMI. Table 6 4 presents estimated random effects and measures of model fit for each model in Table 6 3 There is significant variation in BMI between individuals in each model, however an insignificant amount of variation is associated with the census tract level. Approximately 97 % of the variation in BMI occurs within census tracts and there is an insignificant amount of variation between census tracts. According to all three measures of model fit, model 3 is the best fitting model. Changes in Female BMI Table 6 7 presents results from three level hierarchical linear models predic ting adjusted adult BMI among females between 2000 and 2008. After controls for individual demographics, socioeconomic status, and health status, neighborhood demographics are not associated with initial or rate of change in BMI among males. The independen t variable of interest the individual demographics, being black or Hispanic are associated with female BMI in

PAGE 136

136 2000. As in the two level models, both black and Hispanic females are predicted to have a higher BMI than white females (b=3.09, p<.01; b=1.67, p<.01 respectively ). Of the individual socioeconomic characteristics, having more years of education is associated with a lower BMI in 2000 (b= 0.17, p<.01). Of the measures of health and he alth behaviors, females who currently smoke are predicted to have a lower BMI (b= 1.15, p<.01), as are females who report better self rated health (b= 0.39, p<.01). Regarding rate of change, being older was associated with a lower rate of BMI change (b= 0. 01, p<.01) and reporting better health was associated with a higher rate of BMI change. Model 2 additionally includes neighborhood socioeconomic status. Females who live in a census tract with a higher percentage of males with at least a college degree ar e predicted to have a lower BMI (b= 2.80, p<.01). After adjusting for neighborhood socioeconomic status, paternal education continues to be associated with lower female BMI in 2000. Females living in a census tract with a higher percentage of adults over t he age of 65 are predicted to have a lower BMI in 2000 (b= 0.48, p<.05) and females living in a census tract with a higher percentage of blacks are expected to have a lower BMI (b= 0.21, p<.05). No measure of neighborhood socioeconomic status is associated with rate of change in BMI. Model 3 additionally includes walkability, which is not associated with either BMI in 2000 or changes in BMI. be associated with a lower female BMI. Being black or Hispanic remains associated wi th a higher initial BMI. Having more years of education, being a current smoker, and having better self rated health remain associated with a lower initial BMI. Other than

PAGE 137

137 respondents who died between 2000 and 2008 and age, no measure at either the indivi dual or neighborhood level is associated with change in BMI. Table 6 8 presents estimated random effects and measures of model fit for ea ch model in Table 6 7 There is significant variation in rate of change, within an individual, between individuals, and between census tracts in all four models. Most of the variation in BMI (88.4%) occurs between individuals. Approximately 8.6% of the variation in female BMI is due to changes within an individual. According to all thre e measures of model fit, model 2 (the full model without a measure of walkability) is the best fitting model. Summary Neighborhood socio demographics do not explain the residual association between childhood conditions and adult weight among females. Females from households with more educated fathers are consistently predicted to have a lower BMI than females with less educated fathers. This association persists across a full range of individual and neighborhood socio demographic controls. These results are consistent with the critical period model of life course epidemiology or an additive cumulative model and suggest that experiences before the age of sixteen have lasting associations with adult weight. It is possible that additional adult characteristics not included in these models may expl ain the association (the relevant pathway may not be modeled) ; however, the analyses in Chapter 6, by including neighborhood socio demographics, include a broader range of measures than is often included in other research. In the fully adjusted models pred icting BMI and BMI change, percent of males with at least a college degree is associated with a lower initial female BMI. Figure 6 1 presents a simple, unadjusted bivariate relationship between paternal education and adjusted

PAGE 138

138 female BMI observed in the Hea lth and Retirement Study. Although BMI differences between categories do no indicate statistical significance, a clear trend is observed in the data. After adjustment for individual and neighborhood socioeconomic characteristics, community percent black a nd Hispanic were associated with a lower initial BMI. Such a finding is interesting because it suggests that the aggregate effect of living in an area with a higher concent ration of blacks or Hispanics may be opposite that of the individual characteristics of being black or Hispanic (among females). Previous research has found that community percent black is not associated with female BMI once individual socioeconomic status is taken into account (Robert and Reither 2004) Such a seemingly paradoxical pattern of individual characteristics having the opposite association as the aggregates of individual characteristics suggests that neighborhood level mechanisms influencing adu lt female weight are distinct from individual level mechanisms. A similar pattern has been observed in arrest rates although being black is associated with an increased risk of an arrest; some studies show that living in an area with a higher community per cent black is associated with a lower arrest rate. In the case of arrest rates, the mechanism is thought to be social control insofar as higher community percent black is an indicator of segregation, it is segregation that operates as a form of social cont rol rather than policing (Parker, Stults, and Rice 2005) If future research replicates the finding that community percent black and Hispanic are associated with lower BMI (and currently the research suggests no association) in other study populations, it will be useful to consider the mechanisms linking these ch aracteristics to lower BMI. Parker and Stults et al. (2005) also find

PAGE 139

139 evidence that community income inequality is positively associated with female BMI. Although income inequality did not substantially reduce racial inequalities in the work of Parker and Stults et al. (2005), other forms of inequality may still play a role. A higher concentration of blacks and Hispanics may be an indicator of reduced levels of some other forms of inequality. In addition to paternal education, a number of individual chara cteristics are associated with female BMI after adjustment for neighborhood socio demographics. In models predicting BMI changes, black and Hispanic females were consistently predicted to have a higher BMI than white females. Educational attainment maintai ned an inverse association with BMI even after controls for health status and health behaviors. And of the health and health behavior set of covariates, reporting better health and being a smoker were consistently associated with a lower initial BMI. As d iscussed in Chapter 5, individual and neighborhood socio demographics do not exert an influence on changes in BMI between 2000 and 2008. The only variables associated with changes in BMI was age and mortality older adults and those who died sometime betwee n 2000 and 2008 experience a lower rate of change in BMI. Better self rated health was only tenuously associated with a lower weight, but lost significance after model adjustment for neighborhood socioeconomic characteristics. The lack of an association between any variable and change in BMI is not surprising given the study population of older adults, whose body weight is not likely to fluctuate much over an eight year period. Previous research using the Health and Retirement Study has found heterogeneou s trajectories of body weight among older adults; however, these studies were able to utilize all waves of the HRS. Analyses in Chapters

PAGE 140

140 4 6 which modeled changes in BMI over time were limited to four waves of data because of the limitations in geographic data. As the HRS continues to collect data and more waves are made available, it may be more feasible to test for the influence of neighborhood socio demographics on changes in BMI. Nonetheless these analyses will continue to face difficulties with regard s to changing census tract def initions, increased attrition, and issues of residential stability.

PAGE 141

141 Table 6 1. Results from a Hierarchical Linear Model Regressing Male Adjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000 Covariates Model 1 Model 2 Model 3 Fixed Effects Intercept 28.55** 28.49** 28.63** Model Controls Proxy Status 0.44 0.44 0.42 Childhood Conditions Poor Health 0.20 0.24 0.46 Father Lost Job 0.15 0.12 0.10 Father a Laborer 0.40 0.31 0.30 0.02 0.01 0.01 0.02 0.02 0.02 Adult Demographics Age 0.16** 0.16** 0.17** White (ref) Black 0.11 0.19 0.30 Other 0.47 0.42 0.20 Hispanic 0.13 0.24 0.01 Foreign Born 0.80* 0.78* 0.87* Years of Education 0.08* 0.07* 0.07* Household Income 0.01 0.00 0.01 Married 1.13** 1.13** 1.12** Health Functional Status 0.17* 0.16 0.16 Self Rated Health 0.41** 0.38** 0.41** Current Smoker 1.89** 1.90** 1.92** Former Smoker 0.04 0.06 0.09 Visited Doctor 0.89** 0.91** 0.89** Visited Hospital 0.33 0.34 0.35 No Insurance 0.31 0.34 0.38 Neighborhood Demographics Percent Rural 0.00 0.02 0.03 Percent Over 65 0.19 0.17 0.16 Percent Black 0.04 0.01 0.03 Percent Hispanic 0.08 0.01 0.02 Percent Foreign Born 0.07 0.01 0.08 Neighborhood SES Street Connectivity 0.44 Percent Families Poverty 0.21 0.11 Percent Vacant Houses 0.00 0.00 Percent Unemployed 0.29 0.28 Percent Males w/BA 0.94 0.96 Percent Females w/BA 0.17 0.95 Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 142

142 Table 6 2. Variance Components and Fit Statistics from an HLM Model Estimating Changes In Adjusted Male BMI, 2000. Variance Components Model 1 Model 2 Model 3 In rate of Change Within group 21.80** 21.73** 21.73** Between Group Level 2 < 0.00 < 0.00 < 0.00 Between Group Level 3 N, Level 1 2944 2944 2894 N, Level 2 1803 1803 1760 N, Level 3 Goodness of fit Deviance 17467.3 17460.1 17156.2 AIC 17469.3 17462.1 17158.2 BIC 17474.8 17467.6 17163.7 Notes: *p < .05; **p < .01.

PAGE 143

143 Table 6 3 Results from a Hierarchical Linear Model Regressing Female Adjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000 Covariates Model 1 Model 2 Model 3 Fixed Effects Intercept 31.65 31.57** 31.60** Model Controls Proxy Status 2.64** 2.50** 2.52** Childhood Conditions Poor Health 0.81 0.82 0.87 Father Lost Job 0.02 0.07 0.08 Father a Laborer 0.20 0.08 0.09 0.03 0.04 0.04 0.09** 0.09** 0.09** Adult Demographics Age 0.12** 0.12** 0.12** White (ref) Black 2.56** 2.70** 2.61** Other 0.44 0.45 0.41 Hispanic 0.93 1.26* 1.28** Foreign Born 0.97* 0.92* 0.93* Years of Education 0.10 0.07 0.07 Household Income 0.01 0.01 0.01 Married 0.28 0.28 0.26 Health Functional Status 0.34** 0.36** 0.38** Self Rated Health 1.07** 1.04** 1.06** Current Smoker 2.67** 2.70** 2.71** Former Smoker 0.37 0.39 0.42 Visited Doctor 0.51 0.50 0.55 Visited Hospital 0.66* 0.60* 0.58* No Insurance 0.56* 0.50 0.53 Neighborhood Demographics Percent Rural 0.06* 0.01 0.00 Percent Over 65 0.07 0.20 0.23 Percent Black 0.00 0.14* 0.14* Percent Hispanic 0.07 0.16 0.21* Percent Foreign Born 0.26 0.10 0.06 Neighborhood SES Street Connectivity 1.54 Percent Families Poverty 0.25 0.02 Percent Vacant Houses 0.16 0.14 Percent Unemployed 0.69* 0.79* Percent Males w/BA 1.53* 1.34 Percent Females w/BA 0.78 0.95 Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 144

144 Table 6 4. Variance Components and Fit Statistics from an HLM Model Estimating Changes In Adjusted Female BMI, 2000. Variance Components Model 1 Model 2 Model 3 In rate of Change Within group 32.37 ** 35.78 ** 32.32** Between Group Level 2 0.94 0.62 0.59 Between Group Level 3 N, Level 1 3705 3705 3640 N, Level 2 2129 2129 2077 N, Level 3 Goodness of fit Deviance 23532.3 23494.2 23071.5 AIC 23536.3 23498.2 23075.5 BIC 23547.6 23509.5 23086.8 Notes: *p < .05; **p < .01.

PAGE 145

145 Table 6 5. Results from a Hierarchical Linear Mode l Regressing Male Adjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000 20008. Covariates Model 1 Model 2 Model 3 For Initial Status (BMI 2000) Fixed Effects Intercept 29.74 (.78) ** 29.77 ( .78 ) ** 29.89 ( .78 ) ** Model Controls Died 0.83 (.52) 0.83 ( .52 ) 0.87 ( .53 ) Attrit 0.02 (.41) 0.01 ( .41 ) 0.04 ( .42 ) Proxy Status 0.07 (.39) 0.07 ( .39 ) 0.02 ( .40 ) Moved Since 2000 0.07 (.27) 0.07 ( .27 ) 0.12 ( .27 ) Childhood Conditions Poor Health 0.32 (1.12) 0.27 ( 1.12 ) 0.04 ( 1.13 ) Father Lost Job 0.01 (.31) 0.01 ( .31 ) 0.01 ( .31 ) Father a Laborer 0.39 (.30) 0.34 ( .30 ) 0.38 ( .31 ) 0.04 (.05) 0.04 ( .05 ) 0.01 ( .05 ) 0.01 (.05) 0.01 ( .05 ) 0.01 ( .05 ) Adult Demographics Age 0.12 (.03) 0.11 ( .03 ) 0.13 ( .03 ) White (ref) Black 0.71 (.62) 0.76 ( .62 ) 0.93 ( .62 ) Other 0.14 (1.0 0 ) 0.19 ( 1.0 0 ) 0.46 ( 1.02 ) Hispanic 0.00 (.73) 0.06 ( .74 ) 0.35 ( .76 ) Foreign Born 0.61 (.53) 0.61 ( .53 ) 0.63 ( .54 ) Years of Education 0.13 (.05) ** 0.12 ( .05 ) ** 0.12 ( .05 ) ** Household Income 0.00 (.01) 0.00 ( .01 ) 0.00 ( .01 ) Married 0.51 (.35) 0.53 ( .34 ) 0.57 ( .35 ) Health Functional Status 0.06 (.12) 0.06 ( .12 ) 0.09 ( .13 ) Self Rated Health 0.19 (.12) 0.21 ( .12 ) 0.23 ( .12 ) Current Smoker 1.41 (.39) ** 1.42 ( .39 ) ** 1.46 ( .40 ) ** Former Smoker 0.63 (.29) 0.65 ( .29 ) 0.69 ( .30 ) Visited Doctor 0.48 (.44) 0.49 ( .44 ) 0.55 ( .45 ) Visited Hospital 0.13 (.27) 0.14 ( .27 ) 0.14 ( .27 ) No Insurance 0.19 (.49) 0.21 ( .49 ) 0.17 ( .49 ) Neighborhood Demographics Percent Rural 0.02 (.04) 0.00 ( .04 ) 0.02 ( .05 ) Percent Over 65 0.07 (.19) 0.01 ( .20 ) 0.03 ( .21 ) Percent Black 0.05 (.08) 0.00 ( .10 ) 0.02 ( .10 ) Percent Hispanic 0.17 (.01) 0.11 ( .14 ) 0.16 ( .15 ) Percent Foreign Born 0.21 (.18) 0.09 ( .19 ) 0.23 ( .21 ) Neighborhood SES Street Connectivity 1.84 ( 2.09 ) Percent Families Poverty 0.16 ( .27 ) 0.02 ( .28 ) Percent Vacant Houses 0.09 ( .20 ) 0.09 ( .20 ) Percent Unemployed 0.33 ( .50 ) 0.21 ( .51 ) Percent Males w/BA 0.56 ( .96 ) 0.71 ( .98 ) Percent Females w/BA 0.53 ( 1.00 ) 0.75 ( 1.01 ) Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 146

146 Table 6 5 Continued Covariates Model 1 Model 2 Model 3 For Rate of Change (2000 2008) Fixed Effects Interval 0.21 ( .13 ) 0.21 ( .13 ) 0.21 ( .13 ) Model Controls Died 0.37 ( .10 ) ** 0.37 ( .10 ) ** 0.39 ( .10 ) ** Attrit 0.00 ( .07 ) 0.01 ( .01 ) 0.01 ( .07 ) Proxy Status 0.01 ( .07 ) 0.01 ( .07 ) 0.00 ( .07 ) Moved Since 2000 0.01 ( .04 ) 0.02 ( .04 ) 0.02 ( .04 ) Childhood Conditions Poor Health 0.15 ( .17 ) 0.16 ( .17 ) 0.15 ( .17 ) Father Lost Job 0.04 ( .05 ) 0.04 ( .05 ) 0.03 ( .05 ) Father a Laborer 0.01 ( .05 ) 0.02 ( .05 ) 0.03 ( .05 ) 0.00 ( .01 ) 0.00 ( .01 ) 0.00 ( .01 ) 0.00 ( .01 ) 0.00 ( .01 ) 0.00 ( .01 ) Adult Demographics Age 0.01 ( .00 ) 0.01 ( .00 ) 0.01 ( .00 ) White (ref) Black 0.13 ( .09 ) 0.11 ( .10 ) 0.12 ( .10 ) Hispanic 0.01 ( .11 ) 0.01 ( .11 ) 0.06 ( .12 ) Foreign Born 0.00 ( .08 ) 0.01 ( .08 ) 0.01 ( .08 ) Years of Education 0.00 ( .01 ) 0.01 ( .01 ) 0.01 ( .01 ) Household Income 0.00 ( .00 ) 0.00 ( .00 ) 0.00 ( .00 ) Married 0.04 ( .06 ) 0.03 ( .06 ) 0.03 ( .06 ) Health Functional Status 0.01 ( .02 ) 0.01 ( .02 ) 0.01 ( .02 ) Self Rated Health 0.03 ( .02 ) 0.04 ( .02 ) 0.04 ( .02 ) Current Smoker 0.13 ( .07 ) 0.12 ( .07 ) 0.13 ( .07 ) Former Smoker 0.09 ( .04 ) 0.09 ( .04 ) 0.09 () .05 Visited Doctor 0.11 ( .08 ) 0.11 ( .08 ) 0.12 ( .08 ) Visited Hospital 0.01 ( .05 ) 0.01 ( .05 ) 0.01 ( .05 ) No Insurance 0.08 ( .10 ) 0.01 ( .10 ) 0.00 ( .10 ) Neighborhood Demographics Percent Rural 0.00 ( .01 ) 0.01 ( .01 ) 0.00 ( .01 ) Percent Over 65 0.03 ( .03 ) 0.04 ( .02 ) 0.04 ( .03 ) Percent Black 0.01 ( .01 ) 0.01 ( .02 ) 0.01 ( .02 ) Percent Hispanic 0.01 ( .02 ) 0.03 ( .02 ) 0.04 ( .03 ) Percent Foreign Born 0.02 ( .03 ) 0.02 ( .03 ) 0.05 ( .03 ) Neighborhood SES Street Connectivity 0.19 ( .32 ) Percent Families Poverty 0.02 ( .04 ) 0.03 ( .04 ) Percent Vacant Houses 0.03 ( .03 ) 0.03 ( .03 ) Percent Unemployed 0.04 ( .08 ) 0.05 ( .08 ) Percent Males w/BA 0.18 ( .15 ) 0.12 ( .15 ) Percent Females w/BA 0.04 ( .15 ) 0.02 ( .15 ) Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 147

147 Table 6 6. Variance Components and Fit Statistics from an HLM Model Estimating Changes In Adjusted Male BMI (2000 2008). Variance Components Model 1 Model 2 Model 3 In rate of Change 0.40** 0.40** 0.39** Within group 2.67** 2.67** 2.69** Between Group Level 2 32.30** 32.38 ** 31.89** Between Group Level 3 0.01 0.01 0.01 N, Level 1 10512 10512 10328 N, Level 2 N, Level 3 1811 1811 1768 Goodness of fit Deviance 51488.7 51503.9 50625.2 AIC 51500.7 51515.9 50637.2 BIC 51533.7 51515.9 50670.1 Notes: *p < .05; **p < .01.

PAGE 148

148 Table 6 7. Results from a Hierarchical Linear Model Regressing Female Adjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000 20008. Covariates Model 1 Model 2 Model 3 For Initial Status (BMI 2000) Fixed Effects Intercept 29.08** 29.26** 28.85** Model Controls Died 2.28** 1.93* 1.98* Attrit 0.28 0.32 0.38 Proxy Status 0.83 1.68 1.01 Moved Since 2000 0.11 0.15 0.24 Childhood Conditions Poor Health 0.96 0.79 1.00 Father Lost Job 0.12 0.16 0.11 Father a Laborer 0.07 0.08 0.08 0.06 0.07 0.07 0.17** 0.16** 0.16** Adult Demographics Age 0.03 0.03 0.03 White (ref) Black 3.09** 3.27** 3.37** Other 0.40 0.40 0.41 Hispanic 1.67* 2.04** 2.07** Foreign Born 0.55 1.04 0.99 Years of Education 0.19** 0.15* 0.15* Household Income 0.00 0.01 0.01 Married 0.24 0.29 0.28 Health Functional Status 0.12 0.12 0.13 Self Rated Health 0.39** 0.34** 0.34** Current Smoker 1.15** 1.13** 1.13** Former Smoker 0.00 0.00 0.06 Visited Doctor 0.29 0.27 0.25 Visited Hospital 0.37 0.49 0.47 No Insurance 0.27 0.27 0.27 Neighborhood Demographics Percent Rural 0.05 0.07 0.06 Percent Over 65 0.28 0.48* 0.49* Percent Black 0.02 0.21* 0.23* Percent Hispanic 0.01 0.31 0.43** Percent Foreign Born 0.23 0.13 0.05 Neighborhood SES Street Connectivity 1.74 Percent Families Poverty 0.02 0.01 Percent Vacant Houses 0.26 0.16 Percent Unemployed 0.43 0.64 Percent Males w/BA 2.80** 2.74** Percent Females w/BA 0.72 0.73 Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 149

149 Table 6 7 Continued Covariates Model 1 Model 2 Model 3 For Rate of Change (2000 2008) Fixed Effects Interval 0.02 0.04 0.05 Model Controls Died 0.47** 0.44* 0.44* Attrit 0.06 0.07 0.04 Proxy Status 0.21 0.24 0.25 Moved Since 2000 0.01 0.01 0.01 Childhood Conditions Poor Health 0.19 0.18 0.18 Father Lost Job 0.06 0.06 0.06 Father a Laborer 0.05 0.05 0.05 0.02 0.02 0.01 Fat 0.02 0.02 0.02 Adult Demographics Age 0.01** 0.01** 0.01** White (ref) Black 0.06 0.07 0.07 Hispanic 0.12 0.12 0.12 Foreign Born 0.02 0.10 0.11 Years of Education 0.00 0.00 0.00 Household Income 0.00 0.00 0.00 Married 0.04 0.06 0.05 Health Functional Status 0.01 0.01 0.01 Self Rated Health 0.04* 0.04 0.03 Current Smoker 0.04 0.01 0.01 Former Smoker 0.03 0.05 0.05 Visited Doctor 0.06 0.06 0.06 Visited Hospital 0.02 0.02 0.02 No Insurance 0.08 0.09 0.09 Neighborhood Demographics Percent Rural 0.00 0.01 0.01 Percent Over 65 0.04 0.06 0.06 Percent Black 0.01 0.01 0.01 Percent Hispanic 0.01 0.03 0.03 Percent Foreign Born 0.01 0.02 0.02 Neighborhood SES Street Connectivity 0.00 Percent Families Poverty 0.02 0.02 Percent Vacant Houses 0.03 0.02 Percent Unemployed 0.04 0.05 Percent Males w/BA 0.18 0.22 Percent Females w/BA 0.04 0.04 Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01

PAGE 150

150 Table 6 8. Variance Components and Fit Statistics from an HLM Model Estimating Changes In Adjusted Female BMI (2000 2008). Variance Components Model 1 Model 2 Model 3 In rate of Change 0.59** 0.59** 0.59** Within group 3.96** 3.97** 3.97** Between Group Level 2 46.93** 46.30 ** 46.25** Between Group Level 3 1.58 1.25 1.19 N, Level 1 13767 13523 13523 N, Level 2 N, Level 3 2136 2083 2083 Goodness of fit Deviance 73015.3 68285.0 71709.8 AIC 73027.3 68297.0 71721.8 BIC 73027.3 68330.7 71721.8 Notes: *p < .05; **p < .01.

PAGE 151

151 Figure 6 1. Adjusted Body Mass Index of Females by Paternal Education 25 26 27 28 29 30 31 32 Less than High School High School Some College College BMI of Females by Paternal Education

PAGE 152

152 CHAPTER 7 DICUSSION AND CONCLU SION This project examined the link between childhood conditions and adult BMI. In doing so, the project falls within an active research area that links early life events to health outcomes across the life course. Overweight and obesity are particularly important health outcomes to consider given the increased prevalence of both weigh t categories over the past several decades. Although this project used BMI as an outcome rather than selected BMI cut offs indicating overweight or obesity, identifying the correlates of increased BMI can be of use to future research that focuses specifically on BMI ca tegories. A number of previous studies have already linked childhood socioeconomic status with increased risk of overweight or obesity. This project sought to extend prior research by testing whether adult neighborhood characteristics explain residual asso ciations between childhood conditions and adult BMI. At the outset of this project, five key research questions were asked. First, are childhood conditions associated with BMI as an adult, after adju sting for a range of individual level adult characterist ics ? Second, are childhood conditions associated with changes in BMI over time as an adult? Third are neighborhood characteristics associated with adult BMI, after adjusting for individual level characteristics? Fourth, are neighborhood characteristics as sociated with changes in BMI over time as an adult? Finally, to the extent that childhood conditions are associated with adult BMI, do neighborhood characteristics account for this relationship? Chapter 4 presented results from analyses that attempt to ans wer the f irst two questions; Chapter 5 the second two questions; and Chapter 6 the final research question. Chapter 7 summarizes the overall

PAGE 153

153 findings, discusses their implications, and identifies limitations and potential for future research. Synopsis of Childhood Conditions Associated with BMI To answer the first two research questions: childhood conditions are associated with BMI as an adult only among females, and childhood conditions are not associated with changes in BMI over an eight year period in e ither males o r females. Childhood conditions considered in the analyses in clude d paternal occupation As the review of the literature made clear, t he most common indictor of childhood socioeconomic sta tus is paternal occupation. Paternal occupation is usually dichotomized into an indicator for whether or not the father was a laborer (as it was in this analysis) and the majority of research has found an inverse association between paternal occupation an d adult weight. Past research has often been limited in its measurement of childhood conditions by its reliance on paternal occupation, and this project sought to address this limitation by the inclusion of a broader range of measures. Past research has al so been limited in its cross sectional measurement of BMI few studies have tested whether childhood conditions are associated with changes in BMI in the adult population. This project sought to address this gap by including longitudinal measurements of BMI in t he adult population. Although little research has examined BMI longitudinally past research has consistently documented an inverse association between childhood conditions and adult weight. The explanation for this inverse association is an active area of research why do childhood conditions influence adult weight so many decades later? Life course epidemiology offers two general answers. First, childhood conditions can result in last a lifetime,

PAGE 154

154 regardless of later life conditions. The second answer is that childhood conditions set individuals on certain trajectories, and that as individuals move along these trajectories, they encounter experiences that affect their health. Unders tanding which answer best explains the link between childhood conditions and adult weight has policy implications if indeed childhood conditions retain permanent influences on adult weight, then intervention efforts must either focus their efforts at earli er stages of the life course or develop more intensive strategies later in life Results presented in Chapter 4 suggest that among males, having a father who was a laborer was associated with a higher BMI in 2000. Once educational attainment and income were included in the analysis having a father who was a laborer was no longer associated with adult BMI among males. Among females, however, childhood conditions retained their association with a dult BMI in 2000. Specifically, paternal education was inversely associated with adult BMI in females. This association persisted after a full range of controls were included, including adult health status and health behaviors. The se findings are consiste nt with previous research which suggests that p rior to controls for adult socioeconomic status childhood conditions are inversely associated with adult BMI in both men and women. O nce adult socioeconomic status is adjusted for in models the association i s much weaker among men but retains i ts association with female BMI, suggesting sex differences in how childhood conditions influence weight. What might explain these sex differences? Specifically, why might d by paternal education, influence adult weight outcomes in females?

PAGE 155

155 Insofar as paternal education is a valid marker of social class, t here is evidence that social class is associated with body dissatisfaction and greater valuation of thinness among adoles cent girls. If this is the case, the mechanism linking education and body weight among females may not be entirely positive, as educationa l attainment would then operate as a marker for membership in a social class which stigmatizes weight rather than a s a marker for increased access to resources. In a study of adolescent girls, Ogden (1999) found that girls from a higher class reported greater body dissatisfaction, concern about weight, and restrained eating behaviors. Associations between social class, bo dy dissatisfaction and weight concern were explained by a greater valuation of female thinness and p lacing a greater importance on physical appearance. In relation to the results from this analysis, it may be that girls raised in households with greater pa ternal education are exposed to certain values relating to body aesthetics, and that these values influence eating and exercise behaviors. The notion that early exposure to cultural values can have a lasting influence on health parallels the logic of the critical period model. Kuh and Ben Schlomo et as a limited time window in which an exposure can have adverse or protective effects on development and subsequent disease outcome. Outside this developmental window there is no excess disease risk associated with exposure. biological scarring is a commonly identified mechanism linking critical period to subsequent disease outcomes, Kuh and Ben Schlomo et al. (2003) do not limit critical period effects to su ch mechanisms allowing for the possible existence of developmental periods during which acquisition of norms or behavior are optimal Language serves as an excellent example of an outcome that

PAGE 156

156 does not rely on biological scarring as a mechanism. Language acquisition has long been hypothesized to have a critical period of development whereby it is easier to learn a language between the ages of five and puberty (Lenneberg 1967) It is possible that there a critical period of development exists for the acquisition of certain beliefs about notions regar ding ideal body size and beauty across the life course. But like language, exposure to cultural beliefs about the body early in life may engender a cultural fluency with lifelong implications for weight related health outcomes. Values relating to body aest hetic may also vary by race and ethnicity. There is some evidence that African Americans perceive a larger body size as ideal relative to Americans of European descent. (Antin and Hunt 2011; Harris, Walters, and Waschull 1991) Harris an d Walters et al. (1991) found that Latino American college students had similar levels of body satisfaction as European Americans. Preference for a larger female body type may serve to protect women from body dissatisfaction, although there is no research consensus on whether prevalence of eating disorder symptoms differs between racial and ethnic groups (Dounchis, Hayden, and Wilfley 2001) Despite the lack of a consensus on prevalence of eating disorder symptoms, differing valuations of body thinness may lead to differing behaviors of diet and exercise. Among males, there is some evidence that Blacks in the U.S., Hispanic Americans, and Asians are at greater risk of developing eating problems relative to whites (Ricciardelli, McCabe, William s, and Thompson 2007) Future research may wish to examine the competing influences of class and ethnicity on beliefs about the ideal body size.

PAGE 157

157 Exposure and internalization of c ertain values pertaining to body image can be see n as a form of cultural ca pital. Cultural capital, as conceived by Bourdieu, includes a range of resources that includes verbal ability, cultural awareness, and aesthetic preferences (Swartz 1997). Bourdieu originally formulated his conception of cultural capital in an attempt to explain differences in academic success among students in the French educational system, and considered higher education to be a site of systems of thought and values wer e maintained and legitimated (Swartz 1997). Bourdieu believed that cultural capital takes three forms: embodied, objectified, and institutionalized cultural capital (Swartz 1997) Embodied capital refers to long lasting dispositions of the mind and body, i ncluding language; objectified cultural capital includes pictures, books, machines and other cultural artifacts; and institutionalized capital refers to objects such as college degree (Bourdieu 1986). Beliefs about body be seen as a form of embodied cultural capital. If these beliefs and attitudes are formed early in adulthood, and stick with one (females) throughout the life course, then it is also consistent with the c ritical period model of life course epidemiology. Taken together, possible explanation for gender differences in the relationship between early life conditions and later adult weight. This explanation hinges on a number of things which First, there is no evidence in these analyse s that having a father who is more highly educated is associated with the intergenerational transmission of certain bodily aesthetics. Previous research has found

PAGE 158

158 interest educat ional attainment (the same association was not found in men) (Flouri 2006) It education, leading to greater educational attainment and improved health outcomes. A second explanation for gender specific asso ciations between socioeconomic The notion of Intersectionality is important for understanding why, at any given socioeconomic status a woma s experience and ma tended to dichotomize individuals into different grou ps (black or white; poor or not poor), According to Collins, this type of mentality hid es the fact that individuals who occupy multiple groups have life experiences unique to their social location defined by the intersections of their group membership. Ot her scholars for exploring inequalities among people of different characteristics. Zinn and Dill (1996) cial location within these systems. For example, i n addition to the burdens of lower material resources, women must also contend with gender norms that structure individual behaviors. Individual behaviors at the household level are particularly important, as it is at the household level that the allocation of resources between partners and across generations occurs ( Moss 200 2). For households that struggle with limited resources, the allocation of material goods (like money) and immaterial goods (like free time) may

PAGE 159

159 be distributed especially unequally between men and women. Women are more likely than men to hold themselves responsible for child care, and women in lower income households will have fewer resources to assist them in their child care responsibil ities. Some research suggests that food is one outlet for women whose lives are defined by responsibilities over which they have little control (Davis 2010). with lower wei ght among women is the changing nature and availability of higher education in the US. Children born between 1931 and 1941 were born during an era of significantly lower levels of educational attainment. In 1940, only 25% of the populati on had completed hi gh school. But by 2009, 87% had completed high school, representing a 335% increase in the percent of the population with a high school degree (US Bureau of the Census 2007 ). There was a similarly dramatic increase in the percent of US adults with a colleg e degree in 1940 only 5% of the population had a college degree, but by 2009 the percentage had increased to 30% (higher than the percentage of adults with a high school degree in 1940). As low as the prevalence of college education was in 1940, and as ra pid the increase in national educational attainment between 1940 and 2009 has been the period between 1900 and 1941 was an era of even lower average educational attainment and no rapid gains in the number of adults with a high school or college degree. In 1910, only 13.5% of the US adult population had four years of high school degree and 2.7% had a college degree (US Bureau of the Census 2003 ). Such low levels of educational attainment during this era is an important conside ration given that the parent s of the HRS cohort would have attended high school or, if they were among

PAGE 160

160 the lucky few, college, during this era The meaning of education has surely changed since an era when college attendance was so rare, and periodical writers wrote many articles try ing to convince a skeptical nation that a woman could simultaneously be attractive and go to college (Gordon 1987) Insofar as college represented a rare and unique experience for this cohort, having a father wi th a college education may set that father apart for reasons other than what they learned at college. Interpretations about the influence of parental education must take into consideration the changing nature and increasing availability of higher educatio n, and future research should explore if the changing nature of education is associated with a change in its association with weight related health outcomes. T he validity of any explanation which relies on early life socialization to explain the childhood conditions and adult BMI link depends on whether adult characteristics explain the residual association between paternal education and adult female weight. Results from Chapter 5 and 6 do provide some evidence that neighborhood characteristics, although t hey are associated with adult weight among females, do not explain the residual association between childhood conditions and adult weight. Indeed, the pattern of association between neighborhood socioeconomic characteristics and female BMI is consistent wi th the view that cultural beliefs about the ideal female body weight help to explain sex differences in the association between socioeconomic status and BMI. This is explained further in the follow section. Synopsis of Neighborhood Characteristics Associat ed with BMI To answer the third key research questions of this project: neighborhood demographics are not associated BMI among men (other than percent over the age of 65 in cross sectional models), and are inconsistently associated with BMI among

PAGE 161

161 women. Ne ighborhood socioeconomic status is not associated with BMI among men, but is associated with BMI among women. To answer the fourth key research question: n eighborhood demogra phics are not associated rate of change in BMI Results from Chapter 5 (individual and neighborhood characteristics without childhood predictors) suggest that neighborhood demographics explain little variation in BMI at the census tract level. For both males and females, neighborhood socioeconomic characteristics explained the association between most neighborhood demographics and adult BMI. In females, one demogra phic characteristic did remain associated with BMI in fully adjusted models estimating BMI change -percent Hispanic was associated a lower initial BMI. Analyses from Chapter 6 answer the fifth key research question: neighborhood characteristics do no explain the residual association between childhood conditions and adult BMI among females. Chapter 6, which included childhood conditions, also tell s a somewhat different story of demographic differences among females. In these analyses, community percent black and community percent Hispanic were associated with a lower initial BMI after adjustment for individual and neighborhood socioeconomic characteristics. The inverse a ssociation between these demographics and BMI in females was found in both cross sectional and longitudinal analyses of BMI why the inclusion of childhood conditions should reveal an association between neighborhood demographics and adult bod y weight. Previous research has not found an association between the racial and ethnic demographics of a census tract and BMI. Boardman and Saint Onge (2005) found that living in a census tract where at least a quarter of the residents are black is associ ated

PAGE 162

162 with an increased in the likelihood of being overweight or obese but that this association was completely attenuated by neighborhood socioeconomic status. Similarly, Robert and Reither (2004) found that community percent black was not associated with individual BMI (regardless of adjustment for community socioeconomic status). Grafova and Freedman (2008) used community percent black as one of several items in a scale measuring economic disadvantage, which they found to be associated with increased risk of overweight or obesity in older adults. Because Grafova and Freedman (2008) included a number of other socioeconomic indicators in their scale of economic advantage, it is unclear if their finding is truly inconsistent with other research that finds no association between community black and BMI when community percent black is modeled separately from other indicators. Although neighborhood socioeconomic status seems to explain demographic associations with BMI, only one neighborhood characteristic was a ssociated with BMI, and only among females neighborhood educational attainment as defined by percentage of males with a BA or more living in the census tract. It was unexpected that community percent of males with a college degree would be associated with female BMI but not male BMI. One possible explanation for this finding is selection related to husband centered migration. Wives are more likely to defer to husbands in the decision to move, and higher annual family income is associated with husband cente red migration (Shihadeh 1991) It may be that males with greater educational attainment tend to select wives with lower BMI, and then move to areas with jobs that match their educational qualifications. In considering this explanation, it is imp ortant to remember that t he study population consisted of older adults who lived during an era when single

PAGE 163

163 wage earners were more common, and the wage earner was more likely to be male. As couple migration becomes less male centered over time, this explana tion will be less applicable in later cohorts. A second explanation for the association between male educational attainment and female BMI may be related to the previous discussion of socioeconomic status and body image. Living in a neighborhood with a h igher percentage of males with a BMI may indicate living in a neighborhood that is more likely place greater value on female thinness. Previous research that examined the relationship between socioeconomic status and beliefs about body image, satisfaction management has f ound an inverse association between body satisfaction and socioeconomic status among adolescent girls (Ogden 1999) Among adult women, greater educational attainment is associated with an increased likelihood of perceiving (Paeratakul, White, Williamson, Ryan, and Bray 2002) social clas s of her neighborhood may be associated with beliefs about weight. The association between male educational attainment and female BMI may also indicate living in an area with greater access to resources that can be used for healthy weight management, but if that were the case, why is community percent of males with a college degree not associated with male BMI? Any explanation linking n eighborhood characteristics with adult BMI must contend with sex differences in the association, especially as they are consistent sex differences in the association between socioeconomic status and females are also seen at the individual level.

PAGE 164

164 Synopsis of Adult Conditions Associated with BMI The most consistent individual measures associated with a lower BMI in 2000 for both men and women were more years of education, better self rated health, being foreign born, and being a current smoker. Although resu lts suggest that for females, early health y weight intervention programs may be worthwhile, the adult measures associated with BMI suggest a number of other possibilities for intervention programs insofar as their associations are causal. Smoking status is associated with lower BMI, and indeed campaigns to reduce smoking may have had the unintended effect of increasing the prevalence of overweight and obesity in the US (Rashad, Grossman, and Chou 2005) will be liberalized in an effort to reduce obesity prevalence, even if being foreign born is associated with lower BMI. Thus of the adult characteristics associated with adult BMI, perhaps the most useful for health interventions is education al attainment The persistent association between educat ional attainment and BMI has been found in numerous other studies (Baum Ii and Ruhm 2009; McLaren 2007; Mokdad Bowman, and Ford 2001; Rashad, Grossman, and Chou 2005) Proposed mechanisms linking higher educational attainment to improved health include b etter work and economic conditions, social and psychological resources, and greater productive or allocative efficiency (Adams 2002; Ross and Wu 1995) In the context of obesity, higher educational attainment may be indicative of higher paying jobs and a concomitant ability to afford healthier food. Educational attainment is stro ngly associated with employment (Ross and Wu 1995) In February 2013, those with a college degree or higher had an unemployment rate of 3.8% compared to 11.2% for those with less than a high school

PAGE 165

165 degree. Those with a high school degree had an unemployment rate of 7.9% (Bureau of Labor Statistics 2013). I ndividuals with less education are less likely to have a job, and may not have the resources to afford the increased costs of healthier food. C aloric density is inversely associated with cost of food thus high fat, high sugar meals are more affordable than meals with lower caloric density but greater nutritional value (Drewnowski and Specter 2004) Cohort Study found that the healthiest diet cost over $1,000/year more than the least healthy diet (Cade, Upmeier, Calvert, and Greenwood 1999) Some of the largest differences in expenditures were in meat, fruit, and vegetables Not only are energy dense foods cheaper, but they are also associated with diminished satiety and increased palatability (Drewnowski 1998) Individuals with reduced income are t hus at ri sk of eating less healthy foods that are dense in calories, and are more likely to eat these foods in greater quantity. In addition to improved access to resources, higher educational attainment may develop the s ocial and psychological resources necessary to improve health generally and maintain a healthy weight specifically Having a college degree is associated with a stronger social support system and those with better social support systems have better health outcomes including mortality (Ross and Mirowsky 1999) There is some evidence to suggest that individuals living in areas with greater social capital are less likely to be overweight or obese (Kim, Subramanian, Gortmaker, and Kawachi 2006) Insofar as educational attainment leads people to networks with greater social activity and engagement with the community generally, it may also serve to low er likelihood of obesity. Of the psychological resources associated with education and

PAGE 166

166 obesity, e of control, and having an inner locus of control is associated with a reduced likelihood of obesity (Gale, Batty, and Deary 2008; Strudler Wallston and Wallston 1978) The notion that higher educational attainment leads to great er resources both material and immaterial, and that these resources can then be used by individuals to maintain a healthy weight, complements the previous discussion on the possible role of cultural capital in explaining gender differences in the association between childhood conditions and weight. Beliefs about ideal weight help to explain the motivation to maintain a certain body weight, while resources such as income and an inner locus of control make it possible for an individual to act on their motivations. Limitations and Future Research There are a number of limitations to this research. One of the most important limitations is the use of census tracts as an indicator for neighborhood residence Although this is standard practice in much of the literature on neighborhood characteristics and health, census tracts may be too rough a proxy for neighbor hoods to capture sufficient variation in BMI. Indeed, results from analyses suggest there is little to no variation in male BMI at the census tract level. And among females, variation in BMI at the census tract level is quite small relative to variation in BMI between individuals. This is not to say that neighborhood characteristics do not matter for adult BMI they may matter very much. However, neighborhood characteristics at the census tract level do no explain much variation in BMI between census tracts. Future research may wish to follow the lead of the Project on Human Development in Chicago Neighborhoods, which identified 343 neighborhood clustered within the city of Chicago, providing a much more fine grained analysis of neighborhoods than census trac ts (Sampson,

PAGE 167

167 Raudenbush, and Earls 1997) Smaller neighborhoods may vary more in BMI and in resources related to healthy weight management, making it easier to identify important neighborhood characteristics. Related to lack of significant variation in BMI between census tracks is the lack of much variation in BMI over time. T his lack of variation is primarily due to two factors: the age of the study population and the limited ( eight year) observation period. The BMI of older adults may not vary much over an eight year period, making it difficult to identify factors associated with changes in BMI. Given the relatively small amount of variation in BMI observed in this analysis, any factor associated with BMI changes in older adults would have to have a rather large effect size. Analyses were limited to eight years of data due to issues related to the coding of census tracts. Although an additional eight years of data is available on the BMI of older adults, census tract information is not available that is consistent across all sixteen years. Given the lack of an association with childhood conditions and changes in BMI, we are left with a question: at what point do childhood conditions exert a n influence on BMI trajectories? Results from this point along the life course, and that this influence can still be observed among older adults. Future research should more precisely identify the timing of the influence childhood conditions And i f beliefs about body thinness and ideal body image do play a role in the link between early childhood and later adult BMI, at what point do these beliefs begin to manifest themselves, and is there a point at which they become

PAGE 168

168 Research that more precisely identifies the timing of childhood influences may help overcome another limitation of this research: the inability to distinguish between cumulative exposure and pathway models that link childhood conditions with adult weight outcomes. Although the knowledge that neighborhood conditions do not expla in the residual association between childhood conditions and adult BMI among females is a significant contribution to the literature, it tells us relatively little about how and why childhood conditions have a long term influence. Among males, the associat ion seems to be explained by adult socioeconomic status is this because childhood disadvantage sets males on a trajectory of reduced adult socioeconomic status, or is it the sum of disadvantaged experiences that matters more for health outcomes. Mishra and Nitsch (2009) provide a statistical framework to disentangle these processes and answer these questions which remain unanswered by this dissertation. Another possible limitation of this research is the use of retrospective measures of childhood conditions Some research has cast doubt on the validity of recall measures with regards to adverse childhood experiences, including sexual or physical abuse (Hardt and Rutter 2004) Other resear ch is more optimistic about the use of retrospective measures of childhood socioeconomic position (Krieger, Okamoto, and Selby 1998) K riger and Okamoto et al. (1998), using a cross section study of adul t l evel and childhood soci al class, found a high level of concordance among twins about their childhood socioeconomic status. and 81% agreed on childhood social class, and recall concordance did not vary by education or race. Regarding measures of childhood health, research has found that

PAGE 169

169 recall measures of childhood health in the Health and Retirement Study to be reliable, although there is no method to determine the validity of this measure because the HRS did not directly m easure childhood health (Haas 2007) Another limitation is the lack of a measure of childhood weight. The lack of such a measure precludes the ability to test for an important pathway through which childhood conditions may influence adult weight. U.S. children from disadvantaged homes are more likely to be either overweight or obese, and weight status in childhood is known to track into adulthood. (Power and Matthews 1997) Thus what may appear to be unexplained residual associations between childhood conditions and adult weight may be due to the failure of controlling for childhood weight. Nonetheless, the policy conclusion would be the same of childhood weight explained the association between childhood socioeconomic status and adult weight: policy interventions should continue to target early life. Lastly, mortality prior to the year 2000 has the potential to affect the observed as sociation between childhood conditions and weight. In a sample of middle aged men, lowest mortality was observed in men who weight 20% less than the US average for me n (Lee and Manson et al. 1993) E xcess mortality has been observed in both underweight populations and obese populations, with greater exce ss observed in higher degrees of obesity (Flegal and Graubard et al. 2005) Consequently, the largest population on which this study was based those surviving to middle age likely has a low representation of those who were underweight or obese compared to earlier ages. Less clear is the mortality risk associated wit h being overweight relative to those classified as having a healthy BMI. Some research has found that being overweight is

PAGE 170

170 associated with reduced mortality risk (Orpana and Berthelot et al. 2010) If childhood conditions are associated with an increased risk of obesity, then analyses may underestimate the association because obese i ndividuals will not be included in the analysis. However, if childhood conditions are associated with being overweight, mortality selection is less of an issue insofar as overweight confers a lower risk of morality.

PAGE 171

171 APPENDIX A UNADJUSTED BMI ESTIMATES FROM C HAPTER 4 This appendix presents results from models that estimate unadjusted BMI using the same models as in Chapter 4.

PAGE 172

172 Table A 1. Results from a Hierarchical Linear Model Regressing Male U nadjusted BMI on Childhood and Individual Characteristics, HRS Cohort 2000. Covariates Model 1 Model 2 Model 3 Model 4 Model 5 Intercept 27.59 (.15) ** 27.19 (.23) ** 26.66 (.27) ** 26.79 (.28) ** 27.78 ( .55 ) ** Model Controls Proxy Status 0.11 (.25) 0.34 (.16) 0.27 (.26) 0.33 (.26) 0.33 ( .26 ) Childhood Conditions Poor Health 0.04 (.04) 0.07 (.20) 0.03 (.73) 0.07 ( .73 ) Father Lost Job 0.23 (.23) 0.27 (.27) 0.27 (.20) 0.24 ( .20 ) Father Laborer 0.40 (.19) 0.40 (.19) 0.33 (.19) 0.36 ( .19 ) 0.02 (.03) 0.01 (.03) 0.00 (.03) 0.01 ( .03 ) 0.03 (.03) 0.03 (.03) 0.01 (.03) 0.01 ( .03 ) Adult Demographics Age 0.13 (.02) ** 0.14 (.02) ** 0.14 (.02) ** 0.16 ( .02 ) ** White (ref) Black 0.28 (.26) 0.32 (.29) 0.23 (.29) 0.15 ( .29 ) Other 0.58 (.29) 0.47 (.64) 0.42 (.64) 0.42 ( .64 ) Hispanic (ref=not Hispanic) 0.44 (.35) 0.59 (.37) 0.39 (.33) 0.31 ( .38 ) Foreign Born 1.01 (.30) ** 0.84 (.33) 0.87 (.25) ** 0.88 ( .32 ) ** Married 1.16 (.24) ** 1.31 (.25) ** 1.32 ( .25 ) ** 1.14 ( .25 ) ** Years of Education 0.10 (.03) ** 0.08 ( .03 ) 0.08 ( .03 ) ** Household Income 0.01 (.01) 0.01 ( .01 ) 0.35 ( .01 ) Health Functional Status 0.13 ( .08 ) Self Rated Health 0.39 ( .09 ) ** Current Smoker 1.86 ( .26 ) ** Former Smoker 0.04 ( .19 ) Visited Doctor 0.82 ( .33 ) Visited Hospital 0.34 ( .20 ) No Insurance 0.24 ( .30 ) Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 173

173 Table A 2. Variance Components and Fit Statistics from an HLM Model Estimating Unadjusted Male BMI in 2000 Variance Components Model 1 Model 2 Model 3 Model 4 Model 5 In rate of Change Within group 21.36** 20.43** 20.65** 20.62** 20.01** Between Group Level 2 0.00 0.00 0.00 0.00 0.00 Between Group Level 3 N, Level 1 3083 3191 2962 2962 2950 N, Level 2 1879 1927 1814 1814 1807 N, Level 3 Goodness of fit Deviance 18197.1 18700.3 17385.7 17392.8 17239.0 AIC 18199.1 18702.3 17387.7 17394.8 17241.0 BIC 18204.6 18707.9 17393.2 17400.4 17246.5 Notes: *p < .05; **p < .01.

PAGE 174

174 Table A 3. Results from a Hierarchical Linear Model Regressing Female U nadjusted BMI on Childhood and Individual Characteristics HRS Cohort 2000. Covariates Model 1 Model 2 Model 3 Model 4 Model 5 Intercept 27.45 ( .18 ) ** 27.59 ( .19 ) ** 27.12 ( .24 ) ** 27.28 ( .25 ) ** 30.46 ( .61 ) ** Model Controls Proxy Status 1.73 ( .62 ) ** 2.10 ( .60 ) ** 1.58 ( .63 ) 1.71 ( .63 ) ** 2.69 ( .63 ) ** Child Cond s Poor Health 0.11 ( .84 ) 0.05 ( .84 ) 0.07 ( .83 ) 0.73 ( .80 ) Father Lost Job 0.18 ( .24 ) 0.03 ( .24 ) 0.08 ( .24 ) 0.09 ( .23 ) Father Laborer 0.43 ( .21 ) 0.04 ( .21 ) 0.27 ( .21 ) 0.26 ( .21 ) c. 0.06 ( .04 ) 0.04 ( .04 ) 0.00 ( .04 ) 0.02 ( .04 ) 0.15 ( .03 ) ** 0.13 ( .03 ) ** 0.10 ( .03 ) ** 0.09 ( .03 ) ** Adult Dem s Age 0.07 ( .02 ) ** 0.08 ( .02 ) ** 0.08 ( .02 ) ** 0.12 ( .02 ) ** White (ref) Black 3.03 ( .27 ) ** 2.99 ( .30 ) ** 2.94 ( .29 ) ** 2.34 ( .29 ) ** Other 0.07 ( .35 ) 0.42 ( .66 ) 0.25 ( .66 ) 0.41 ( .66 ) Hispanic (ref=not Hispanic) 1.13 ( .40 ) ** 1.24 ( .43 ) ** 0.88 ( .43 ) 0.54 ( .44 ) Foreign Born 1.30 ( .34 ) ** 0.82 ( .36 ) ** 0.94 ( .36 ) ** 1.18 ( .36 ) ** Married 0.20 ( .21 ) 0.25 ( .21 ) 0.14 ( .22 ) 0.14 ( .22 ) Yrs. Education 0.24 ( .04 ) ** 0.15 ( .04 ) ** 0.10 ( .04 ) Household Income 0.03 ( .01 ) 0.02 ( .01 ) 0.01 ( .01 ) Health Func. Status 0.39 ( .09 ) ** Self Rated Health 0.97 ( .10 ) ** Cr nt Smoker 2.61 ( .26 ) ** Fmr. Smoker 0.34 ( .20 ) Vst d Doctor 0.45 ( .41 ) Vst d Hospital 0.62 ( .25 ) ** No Insurance 0.54 ( .27 ) Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 175

175 Table A 4. Variance Components and Fit Statistics from an HLM Model Estimating Unadjusted Female BMI in 2000 Variance Components Model 1 Model 2 Model 3 Model 4 Model 5 In rate of Change Within group 32.95** 32.94** 32.37** 32.35** 29.16** Between Group Level 2 1.36* 0.50 0.80 0.63 1.04* Between Group Level 3 N, Level 1 3843 4039 3723 3723 3710 N, Level 2 2221 2268 2140 2140 2133 N, Level 3 Goodness of fit Deviance 24491.2 25650.9 23607.3 23569.2 23188.3 AIC 24495.2 25654.9 23611.3 23600.2 23192.3 BIC 24506.6 25666.3 23622.6 23611.6 23203.7 Notes: *p < .05; **p < .01.

PAGE 176

176 Table A 5. Results from a Hierarchical Linear Model Regressing Male U nadjusted BMI on Childhood and Individual Characteristics HRS Cohort 2000 2008. Covariates Model 1 Model 2 Model 3 Model 4 Intercept 27.23 ( .16 ) ** 27.17 ( .19 ) ** 27.27 ( .19 ) ** 27.92 ( 0.35 ) ** For Intercept Model Controls Proxy Status 0.11 ( .09 ) 0.08 ( .09 ) 0.08 ( .09 ) 0.12 ( .17 ) Died 0.09 ( .18 ) 0.06 ( .18 ) 0.03 ( .18 ) 0.42 ( .23 ) Attrit 0.14 ( .18 ) 0.31 ( .19 ) 0.3 1 ( .19 ) 0.26 ( .23 ) Moved Since 2000 Childhood Conditions Poor Health 0.23 ( .60 ) 0.20 ( .60 ) 0.10 ( .60 ) 0.24 ( .68 ) Father Lost Job 0.25 ( .17 ) 0.21 ( .17 ) 0.21 ( .17 ) 0.21 ( .21 ) Father Laborer 0.27 ( .16 ) 0.20 ( 16 ) 0.16 ( 0.16 ) 0.19(0.19) 0.01 ( .03 ) 0.01 ( .03 ) 0.00 ( .03 ) 0.00 ( .03 ) 0.04 ( .03 ) 0.02 ( .03 ) 0.02 ( .03 ) 0. 0 1 ( .03 ) Adult Demographics Age 0.07 ( .01 ) ** 0.08 ( .01 ) ** 0.08 ( .02 ) ** White (ref) Black 0.28 ( .24 ) 0.23 ( 0.23 ) 0.29 ( .28 ) Other 0.83 ( .52 ) 0.80(.53 ) 0.66 ( .62 ) Hispanic (ref=not Hispanic) 0.48(.32) 0.38 ( .32 ) 0.18 ( .36 ) Foreign Born 0.85 ( .28 ) 0.87 ( .28 ) 0.92 ( .31 ) ** Married 0. 31(.10 ) 0.31 ( .10 ) 0.29 ( .16 ) Years of Education 0.03 ( .03 ) 0.07 ( .02 ) ** Household Income 0.00 ( .00 ) 0.00 ( .00 ) Health Functional Status 0.04 ( 0.05 ) Self Rated Health 0.05 ( 0.05 Current Smoker 0.64 ( .21 ) ** Former Smoker 0.28 ( .18 ) Visited Doctor 0.01 ( .15 ) Visited Hospital 0.03 ( .10 ) No Insurance 0.17 ( .15 ) Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 177

177 Table A 5 Continued Covariates Model 1 Model 2 Model 3 Model 4 Interval 0.10 ( .02 ) ** 0.11 ( .02 ) ** 0.12 ( .03 ) 0.02 () For Slope (Change) Model Controls Proxy Status 0.01 ( .09 ) 0.02 ( .02 ) 0.01 ( .09 ) 0 .01 ( 0.04 ) Died 0.29 ( .18 ) ** 0.25 ( .03 ) ** 0.25 ( .03 ) ** 0.36 ( .04 ) ** Attrit 0.15 ( .18 ) 0.05 ( .03 ) 0.06 ( .03 ) 0.03 ( .04 ) Moved Since 2000 0.05 ( .15 ) 0.00 ( .02 ) 0.01 ( .03 ) 0.02 ( 0.02 ) Childhood Conditions Poor Health 0.06 ( .60 ) 0.06 ( .07 ) 0.06 ( 0.07 ) 0.00 ( .01 ) Father Lost Job 0.01 ( .17 ) 0.02 ( .02 ) 0.02 ( .02 ) 0.02 ( .03 ) Father Laborer 0.00 ( .16 ) 0.01 ( .02 ) 0.01 ( .02 ) 0.01 ( .03 ) 0.01 ( .03 ) 0.00 ( .00 ) 0.00 ( .00 ) 0.00 ( .00 ) 0.00 ( .03 ) 0.00 ( .00 ) 0.00 ( .00 ) 0.00 ( .00 ) Adult Demographics Age 0.01 ( .00 ) ** 0.01 ( .00 ) ** 0.01 ( .00 ) ** White (ref) Black 0.08 ( .03 ) 0.09 ( .03 ) 0.09 ( .04 ) Other 0.04 ( .06 ) 0.04 ( .07 ) 0.05 ( .09 ) Hispanic (ref=not Hispanic) 0.09 ( .04 ) 0.05 ( .04 ) 0.01 ( .05 ) Foreign Born 0.07 ( .03 ) 0.06 ( .03 ) 0.06 ( .04 ) Married 0.00 ( .02 ) 0.00 ( .00 ) 0.00 ( .03 ) Years of Education 0.00 ( .00 ) 0.01 ( .00 ) Household Income 0.00 ( .00 ) 0.00 ( .00 ) Health Functional Status 0.00 ( .01 ) Self Rated Health 0.01 ( .01 ) Current Smoker 0.02 ( .03 ) Former Smoker 0.03 ( .03 ) Visited Doctor 0.00 ( .03 ) Visited Hospital 0.00 ( .02 ) No Insurance 0.03 ( .04 ) Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 178

178 Table A 6. Variance Components and Fit Statistics from an HLM Model Estimating Adjusted Male BMI, HRS 2000 2008. Variance Components Model 1 Model 2 Model 3 Model 4 In rate of Change 0.31** 0.29** 0.29** 0.29** Within group 2.71** 2.68** 2.68** 2.67** Between Group Level 2 26.22** 25.36** 25.23** 25.06** Between Group Level 3 0.24 0.45 0.52 0.49 N, Level 1 14276 13760 13760 13613 N, Level 2 3570 3428 3428 3424 N, Level 3 1929 1865 1865 1863 Goodness of fit Deviance 68422.9 65710.4 65736.0 65074.2 AIC 68436.9 65724.4 65750.0 65088.2 BIC 68475.9 65763.1 65788.7 65126.9 Notes: *p < .05; **p < .01.

PAGE 179

179 Table A 7. Results from a Hierarchical Linear Model Regressing Female Adjusted BMI on Childhood and Individual Characteristics HRS Cohort 2000. Covariates Model 1 Model 2 Model 3 Model 4 Intercept 26.78 ( .19 ) ** 26.17 ( .21 ) ** 26.38 ( .21 ) ** 27.42 ( .39 ) ** For Intercept (Initial Status) Model Controls Proxy Status 0.19 ( .22 ) 0.22 ( .22 ) 0.21 ( .02 ) 0.76 ( .41 ) Died 1.38 ( .29 ) ** 1.24 ( .29 ) ** 1.12 ( .29 ) ** 1.31 ( .34 ) ** Attrit 0.42 ( .23 ) 0.52 ( .24 ) 0.55 ( .23* ) 0.45 ( .27 ) Moved Since 2000 0.04 ( .18 ) 0.02 ( .18 ) 0.02 ( .18 ) 0.00 ( .20 ) Childhood Conditions Poor Health 0.84 ( .77 ) 0.82 ( .76 ) 0.74 ( .76 ) 0.86 ( .84 ) Father Lost Job 0.24 ( .21 ) 0.07 ( .21 ) 0.12 ( .21 ) 0.17 ( .24 ) Father Laborer 0.36 ( .19 ) 0.33 ( .19 ) 0.22 ( .19 ) 0.27 ( .21 ) 0.06 ( .03 ) 0.04 ( .03 ) 0.00 ( .03 ) 0.00 ( .04 ) 0.16 ( .03 ) ** 0.13 ( .03 ) ** 0.10 ( .03 ) ** 0.11 ( .03 ) ** Adult Demographics Age 0.01 ( .02 ) 0.02 ( .02 ) 0.02 ( .02 ) White (ref) Black 3.09 ( .26 ) ** 3.09 ( .26 ) ** 2.97 ( .29 ) ** Other 0.07 ( .60 ) 0.12 ( .60 ) 0.11 ( .67 ) Hispanic (ref=not Hispanic) 1.21 ( .38 ) 0.92 ( .39 ) 0.47 ( .43 ) Foreign Born 0.78 ( .32 ) 0.94 ( .33 ) 0.73 ( .37* ) Married .25 ( .09 ) 0.28 ( .09 ) 0.29 ( .14 ) Years of Education 0.16 ( .04 ) ** 0.16 ( .04 ) ** Household Income 0.01 ( .00 ) 0.16 ( .04 ) ** Health Functional Status 0.15 ( .05 ) ** Self Rated Health 0.15 ( .05 ) ** Current Smoker 1.04 ( .22 ) ** Former Smoker 0.29 ( .20 ) Visited Doctor 0.17 ( .20 ) Visited Hospital 0.07 ( .12 ) No Insurance 0.23 ( .14 ) Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 180

180 Table A 7 Continued Covariates Model 1 Model 2 Model 3 Model 4 Interval 0.19 ( .02 ) ** 0.21 ( .02 ) ** 0.22 ( .02 ) ** 0.10 ( .07 ) For Slope (Change) Model Controls Proxy Status 0.04 ( .05 ) 0.06 ( .05 ) 0.06 ( .05 ) 0.15 ( .09 ) Died 0.30 ( .04 ) ** 0.27 ( .04 ) ** 0.27 ( .04 ) ** 0.35 ( .06 ) ** Attrit 0.01(.03 ) 0. 02 ( .03 ) 0.01 ( .03 ) 0.02 ( .04 ) Moved Since 2000 0.01 ( .02 ) 0.00 ( .02 ) 0.00 ( .02 ) 0.01 ( .02 ) Childhood Conditions Poor Health 0.05 ( .08 ) 0.04 ( .08 ) 0.04 ( .08 ) 0.08 ( .11 ) Father Lost Job 0.02 ( .02 ) 0.01 ( .02 ) 0.02 ( .02 ) 0.01 ( .03 ) Father Laborer 0.03 ( .02 ) 0.00 ( .02 ) 0.02 ( .02 ) 0.01 ( .03 ) 0.00 ( .00 ) 0.00 ( .00 ) 0.00 ( .00 ) 0.00 ( .00 ) 0.01 ( .00 ) 0.01 ( .00 ) 0.00 ( .00 ) 0.01 ( .00 ) Adult Demographics Age 0.01 ( .00 ) ** 0.02 ( .00 ) ** 0.02 ( .00 ) ** White (ref) Black 0.06 ( .03 ) 0.05 ( .03 ) 0.02 ( .04 ) Other 0.15 ( .06 ) 0.12 ( .06 ) 0.14 ( .09 ) Hispanic (ref=not Hispanic) 0.05 ( .03 ) 0.05 ( .04 ) 0.03 ( .06 ) Foreign Born 0.05 ( .03 ) 0.05 ( .03 ) 0.11 ( .05 ) Married 0.03 ( .02 ) 0.04 ( .02 ) 0.05 ( .03 ) Years of Education 0.00 ( .00 ) 0.00 ( .00 ) Household Income 0.00 ( .00 ) 0.00 ( .00 ) Health Functional Status 0.02 ( .01 ) Self Rated Health 0.01 ( .01 ) Current Smoker 0.03 ( .03 ) Former Smoker 0.00 ( .03 ) Visited Doctor 0.04 ( .04 ) Visited Hospital 0.02(.03 No Insurance 0.07 ( .03 ) Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 181

181 Table A 8. Variance Components and Fit Statistics from an HLM Model Estimating Adjusted Female BMI HRS 2000 2008. Variance Components Model 1 Model 2 Model 3 Model 4 In rate of Change 0.49** 0.47** 0.47** 0.47** Within group 3.95** 3.91** 3.91** 3.92** Between Group Level 2 43.17** 41.88** 41.82** 40.11** Between Group Level 3 2.32** 1.81* 1.65* 1.77* N, Level 1 18345 17793 17793 17589 N, Level 2 4322 4183 4183 4182 N, Level 3 2248 2168 2168 2168 Goodness of fit Deviance 95383.9 92171.9 92183.9 91108.4 AIC 95395.9 92183.9 92195.9 91120.4 BIC 95430.3 92218.0 91130.0 91154.5 Notes: *p < .05; **p < .01.

PAGE 182

182 APPENDIX B UNADJUSTED BMI ESTIM ATES FROM CHAPTER 5 This appendix presents results from models that estimate unadjusted BMI using the same models as in Chapter 5.

PAGE 183

183 Table B 1. Results from a Hierarchical Linear Model Regressing Unadjusted Male BMI in 2000 on Neighborhood Characteristics HRS Cohort 2000 Covariates Model 1 Model 2 Model 3 Model 4 Fixed Effects Intercept 27.81 ( .08 ) ** 27.82 ( .08 ) ** 27.84 ( .08 ) ** 27.84 ( .03 ) ** Interval Model Controls Proxy Status 0.50 ( .21 ) 0.54 ( .21 ) 0.56 ( .21 ) ** 0.48 ( .21 ) Neighborhood Demo s Pct. Rural 0.00 ( .02 ) 0.02 ( .03 ) 0.02 ( .03 ) Pct. Over 65 0.06 ( .10 ) 0.07 ( .11 ) 0.05 ( .12 ) Pct. Black 0.05 ( .03 ) 0.01 ( .04 ) 0.01 ( .05 ) Pct. Hispanic 0.12 ( .05 ) 0.05 ( .07 ) 0.01 ( .07 ) Pct. Foreign Born 0.23 ( .09 ) 0.17 ( .09 ) 0.11 ( .12 ) Neighborhood SES Street Connectivity 0.11 ( 1.21 ) Pct. Families Poverty 0.15 ( .15 ) 0.10 ( .16 ) Pct. Vacant Houses 0.08 ( .11 ) 0.09 ( .11 ) Pct. Unemployed 0.22 ( .26 ) 0.26 ( .26 ) Pct. Males w/BA 1.27 ( .57 ) 1.25 ( .58 ) Pct Females w/BA 0.19 ( .55 ) 0.59 ( .59 ) Notes: BMI = body mass index; SRH = self rated health; Median Household Value in in Tens of Thousands of Dollars *p < .05; **p < .01.

PAGE 184

184 Table B 2. Variance Components and Fit Statistics from an HLM Model Estimating Unadjusted Male BMI in 2000. Variance Components Model 1 Model 2 Model 3 Model 4 In rate of Change Within group 19.29** 19.28** 19.31** 19.26** Between Group Level 2 0.54 0.54 0.44 0.48 Between Group Level 3 N, Level 1 3689 3683 3683 3615 N, Level 2 1992 1988 1988 1941 N, Level 3 Goodness of fit Deviance 21489.7 21469.3 21462.3 21064.1 AIC 21493.7 21493.3 21462.3 21068.1 BIC 21504.9 21484.5 21473.5 21079.3 Notes: *p < .05; **p < .01.

PAGE 185

185 Table B 3. Results from a Hierarchical Linear Model Regressing Unadjusted Female BMI in 2000 on Neighborhood Characteristics HRS Cohort 2000 Covariates Model 1 Model 2 Model 3 Model 4 Fixed Effects Intercept 27.71 (.09) ** 27.66 (.09) ** 27.67(.09) ** 27.67 (.09) ** Interval Model Controls Proxy Status 1.69 (.53) ** 1.65(.52) ** 1.53(.52) ** 1.52 (.52) ** Neighborhood Demo s Percent Rural 0.09 (.03) ** 0.00 (.03) 0.01 (.03) Percent Over 65 0.11 (.13) 0.23 (.13) 0.22 (.14) Percent Black 0.37 (.04) ** 0.17 (.05) ** 0.16 (.05) ** Percent Hispanic 0.32 (.07) ** 0.00 (.08) 0.02 (.08) Percent Foreign Born 0.30 (.11) ** 0.09(.12) 0.05 (.13) Neighborhood SES Street Connectivity 0.18 (1.41) Percent Families Poverty 0.08 (.17) 0.14 (.18) P ct. Vacant Houses 0.13 (.13) 0.08 (.14) P ct. Unemployed 0.33 (.32) 0.41 ( .32 ) P ct. Males w/BA 2.25(.67) ** 2.07 ( .69 ) ** P ct. Females w/BA 0.33 (.66) 0.33 ( .69 ) Notes: BMI = body mass index; SRH = self rated health; Median Household Value in in Tens of Thousands of Dollars *p < .05; **p < .01.

PAGE 186

186 Table B 4. Variance Components and Fit Statistics from an HLM Model Estimating Adjusted Female BMI, in 2000 Variance Components Model 1 Model 2 Model 3 Model 4 In rate of Change Within group 33.65** 33.56** 33.42** 33.37** Between Group Level 2 1.88** 0.94* 0.56 0.49 Between Group Level 3 N, Level 1 4516 4510 4510 4417 N, Level 2 2351 2347 2347 2288 N, Level 3 Goodness of fit Deviance 28924.1 28779.5 28714.2 28107.5 AIC 28928.1 28783.5 28718.2 28111.5 BIC 28939.6 28795.1 28729.7 28123.0 Notes: *p < .05; **p < .01.

PAGE 187

187 Table B 5. Results from a Hierarchical Linear Model Regressing Male Unadjusted BMI on Neighborhood Characteristics HRS Cohort 2000 20008. Covariates Model 1 Model 2 Model 3 Model 4 For Initial Status (BMI 2000) Fixed Effects Intercept 27.69(.17) ** 27.77 (.24) ** 27.70 ( .17 ) ** 27.70 ( .17 ) ** Proxy Status 0.46 (.31) 0.14 (.56) 0.41 ( .32 ) 0.46 ( .32 ) Died 1.25 (.43) ** 0.07(.78) 1.11 ( .43 ) 1.11 ( .33 ) Attrit 0.32(.32) 0.81 (.49) 0.28 ( .33 ) 0.27 ( .33 ) Moved Since 2000 0.18 (.23) 0.22 (.33) 0.14 ( .23 ) 0.16 ( .23 ) Neighborhood Dems. Percent Rural 0.03 (.05) 0.01 ( .04 ) 0.02 ( .04 ) Percent Over 65 0.26 (.24) 0.16 ( .17 ) 0.16 ( .18 ) Percent Black 0.14 (.07) 0.10 ( .07 ) 0.09 ( .07 ) Percent Hispanic 0.17(.13) 0.12 ( .10 ) 0.11 ( .10 ) Percent Foreign Born 0.28 ( .22 ) 0.21 ( .15 ) 0.23 ( .18 ) Neighborhood SES Street Connectivity 0.82 ( 1.79 ) Percent Families Poverty 0.06 ( .23 ) 0.03 ( .23 ) Percent Vacant Houses 0.14 ( .17 ) 0.13 ( .17 ) Percent Unemployed 0.06 ( .42 ) 0.09 ( .43 ) Percent Males w/BA 1.09 ( .84 ) 1.11 ( .85 ) Percent Females w/BA 0.19 ( .82 ) 0.45 ( .87 ) For Rate of Change (2000 2008) Interval 0.06 (.02) 0.05 ( .04 ) 0.06 ( .02 ) 0.06 ( .03 ) Proxy Status 0.08 (.31) 0.04 ( .11 ) 0.08 ( .06 ) 0.08 ( .06 ) Died 0.53 (.43) ** 0.20 ( .16 ) 0.52 ( .08 ) ** 0.53 ( .08 ) ** Attrit 0.05 (.32) 0.18 ( .09 ) 0.04 ( .04 ) 0.04 ( .05 ) Moved Since 2000 0.04 (.23) 0.07 ( .06 ) 0.05 ( .03 ) 0.05 ( .04 ) Neighborhood Demographics Percent Rural 0.00 ( .01 ) 0.00 ( .00 ) 0.00 ( .00 ) Percent Over 65 0.09 ( .04 ) 0.07 ( .03 ) 0.06 ( .03 ) Percent Black 0.02 ( .01 ) 0.02 ( .01 ) 0.02 ( .01 ) Percent Hispanic 0.01 ( .02 ) 0.02 ( .02 ) 0.02 ( .02 ) Percent Foreign Born 0.00 ( .04 ) 0.01 ( .02 ) 0.02 ( .03 ) Neighborhood SES Street Connectivity 0.03 ( .28 ) Percent Families Poverty 0.04 ( .04 ) 0.04 ( .04 ) Percent Vacant Houses 0.03 ( .03 ) 0.03 ( 0.03 ) Percent Unemployed 0.07 ( .07 ) 0.10 ( .07 ) Percent Males w/BA 0.02 ( .13 ) 0.02 ( .13 ) Percent Females w/BA 0.05 ( .12 ) 0.02 ( .13 ) Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 188

188 Table B 6. Variance Components and Fit Statistics from an HLM Model Estimating Unadjusted Male BMI, in 2000 2008 Variance Components Model 1 Model 2 Model 3 Model 4 In rate of Change 0.38** 0.37** Within group 2.06** 2.08** Between Group Level 2 28.12** 27.70** Between Group Level 3 0.00 0.00 N, Level 1 11834 11815 11815 11610 N, Level 2 3702 3696 3696 3628 N, Level 3 1992 1988 1988 1941 Goodness of fit Deviance 55767.1 54857.0 AIC 55779.1 54869.0 BIC 55812.6 54902.4 Notes: *p < .05; **p < .01. Some estimates are unavailable due to improper specification of clustering in the models.

PAGE 189

189 Table B 7. Results from a Hierarchical Linear Model Regressing Female Adjusted BMI on Neighborhood Characteristics HRS Cohort 2000 20008. Covariates Model 1 Model 2 Model 3 Model 4 For Initial Status (BMI 2000) Fixed Effects Intercept 27.19 ( .18 ) ** 27.05 ( .27 ) ** 27.11 ( .18 ) ** 27.11 ( .18 ) ** Proxy Status 0.92 ( .73 ) 0.11 ( 1.39 ) 0.87 ( .73 ) 0.91 ( .74 ) Died 2.57 ( .67 ) ** 1.19 ( 1.27 ) 2.05 ( .66 ) ** 1.73 ( .68 ) Attrit 0.41 ( .38 ) 0.75 ( .60 ) 0.59 ( .37 ) 0.72 ( .38 ) Moved Since 2000 0.03 ( .25 ) 0.08 ( .38 ) 0.16 ( .26 ) 0.23 ( .26 ) Neighborhood Dem os Percent Rural 0.13 ( .06 ) 0.00 ( .04 ) 0.01 ( .05 ) Percent Over 65 0.20 ( .27 ) 0.38 ( .19 ) 0.43 ( .19 ) Percent Black 0.42 ( .08 ) ** 0.19 ( .07 ) ** 0.18 ( .07 ) Percent Hispanic 0.29 ( .14 ) 0.01 ( .11 ) 0.05 ( .12 ) Percent Foreign Born 0.19 ( .25 ) 0.01 ( .17 ) 0.11 ( .19 ) Neighborhood SES Street Connectivity 2.36 ( 2.00 ) Percent Families Poverty 0.15 ( .25 ) 0.05 ( .19 ) P ct. Vacant Houses 0.19 ( .19 ) 0.18 ( .19 ) P ct. Unemployed 0.45 ( .46 ) 0.41 ( .47 ) P ct. Males w/BA 2.87 ( .95 ) ** 2.55 ( .97 ) ** P ct. Females w/BA 0.26 ( .92 ) 0.98 ( .98 ) For Rate of Change (2000 2008) Interval 0.18 ( .03 ) ** 0.18 ( .04 ) ** 0.18 ( .03 ) ** 0.18 ( .03 ) ** Proxy Status 0.22 ( .13 ) 0.28 ( .26 ) 0.20 ( .13 ) 0.21 ( .13 ) Died 0.40 ( .12 ) ** 0.21 ( .27 ) 0.39 ( .13 ) 0.34 ( .13 ) ** Attrit 0.27 ( .06 ) 0.08 ( .11 ) 0.04 ( .06 ) 0.05 ( .06 ) Moved Since 2000 0.03 ( .04 ) 0.03 ( .07 ) 0.02 ( .04 ) 0.02 ( .04 ) Neighborhood Demo s Percent Rural 0.00 ( .01 ) 0.00 ( .00 ) 0.01 ( .01 ) Percent Over 65 0.02 ( .05 ) 0.03 ( .03 ) 0.04 ( .03 ) Percent Black 0.00 ( .01 ) 0.00 ( .01 ) 0.00 ( .01 ) Percent Hispanic 0.00 ( .03 ) 0.00 ( .02 ) 0.01 ( .02 ) Percent Foreign Born 0.02 ( .04 ) 0.03 ( .02 ) 0.00 ( .03 ) Neighborhood SES Street Connectivity 0.38 ( .29 ) Percent Families Poverty 0.05 ( .04 ) 0.04 ( .04 ) Percent Vacant Houses 0.03 ( .03 ) 0.00 ( .03 ) Percent Unemployed 0.00 ( .07 ) 0.02 ( .07 ) Percent Males w/BA 0.18 ( .14 ) 0.15 ( .14 ) Percent Females w/BA 0.06 ( .13 ) 0.07 ( .14 ) Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 190

190 Table B 8. Variance Components and Fit Statistics from an HLM Model Estimating Unadjusted Female BMI, in 2000 2008. Variance Components Model 1 Model 2 Model 3 Model 4 In rate of Change 0.57** 0.57** Within group 3.19** 3.21** Between Group Level 2 44.75** 44.59** Between Group Level 3 1.13 1.08 N, Level 1 15552 15532 15532 15224 N, Level 2 4624 4618 4618 4522 N, Level 3 2351 2347 2347 2288 Goodness of fit Deviance 80085.5 78553.6 AIC 80099.5 78565.6 BIC 80139.8 78600.0 Notes: *p < .05; **p < .01. Some estimates are unavailable due to improper specification of clustering in the models.

PAGE 191

191 Table B 9. Results from a Hierarchical Linear Model Regressing Male Unadjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000. Covariates Model 1 Model 2 Model 3 For Initial Status (BMI 2000) Fixed Effects Intercept 28.03 (.46) ** 27.90 ( .46 ) ** 28.04 ( .47 ) ** Model Controls Proxy Status 0.66 (.23) ** 0.66 ( .23 ) ** 0.60 ( .23 ) ** Adult Demographics Age 0.11 (.01) ** 0.11 ( .01 ) ** 0.11 ( .02 ) ** White (ref) Black 0.41 (.33) 0.36 ( .33 ) 0.30 ( .34 ) Other 0.84 (.53) 0.82 ( .53 ) 0.70 ( .53 ) Hispanic (ref=not Hispanic) 0.24 (.38) 0.33 ( .38 ) 0.24 ( .39 ) Foreign Born 0.99 (.29) ** 0.96 ( .29 ) ** 1.02 ( .30 ) ** Years of Education 0.10 (.03) ** 0.08 ( .03 ) ** 0.09 ( .03 ) ** Household Income 0.00 (.00) 0.01 ( .00 ) 0.01 ( .00 ) Married 0.94 (.21) ** 0.94 ( .21 ) ** 0.95 ( .22 ) ** Health Functional Status 0.14 (.06) ** 0.14 ( .06 ) 0.15 ( .06 ) Self Rated Health 0.34 (.08) ** 0.31 ( .08 ) ** 0.34 ( .08 ) ** Current Smoker 1.80 (.23) ** 1.81 ( .23 ) ** 1.86 ( .23 ) ** Former Smoker 0.00 (.17) 0.03 ( .17 ) 0.05 ( .17 ) Visited Doctor 0.89 (.29) ** 0.93 ( .29 ) ** 0.89 ( .29 ) Visited Hospital 0.06 (.18) 0.08 ( .18 ) 0.06 ( .06 ) No Insurance 0.05 (.26) 0.07 ( .26 ) 0.05 ( .27 ) Neighborhood Demographics Percent Rural 0.01 (.02) 0.03 ( .03 ) 0.03 ( .12 ) Percent Over 65 0.13 ( .10 ) 0.12 ( .11 ) 0.09 ( .12 ) Percent Black 0.09 ( .05 ) 0.04 ( .05 ) 0.01 ( .06 ) Percent Hispanic 0.08 ( .07 ) 0.01 ( .07 ) 0.03 ( .08 ) Percent Foreign Born 0.13 ( .10 ) 0.07 ( .10 ) 0.02 ( .12 ) Neighborhood SES Street Connectivity 0.12 ( 1.19 ) Percent Families Poverty 0.15 ( .15 ) 0.10 ( .16 ) Percent Vacant Houses 0.05 ( .11 ) 0.05 ( .11 ) Percent Unemployed 0.38 ( .28 ) 0.43 ( .28 ) Percent Males w/BA 1.14 ( .56 ) 1.04 ( .57 ) Percent Females w/BA 0.27 ( .54 ) 0.69 ( .57 ) Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 192

192 Table B 10. Variance Components and Fit Statistics from an HLM Model Estimating Unadjusted Male BMI in 2000. Variance Components Model 1 Model 2 Model 3 In rate of Change Within group 18.25** 18.29** 18.22** Between Group Level 2 0.22 0.15 0.18 Between Group Level 3 N, Level 1 3512 3512 3446 N, Level 2 1914 1914 1868 N, Level 3 Goodness of fit Deviance 20250.0 20242.3 19857.3 AIC 20254.0 20246.3 19861.3 BIC 20265.1 20257.4 19872.3 Notes: *p < .05; **p < .01.

PAGE 193

193 Table B 11. Results from a Hierarchical Linear Model Regressing Female Unadjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000. Covariates Model 1 Model 2 Model 3 Fixed Effects Intercept 30.82 ( .55 ) ** 30.63 ( .55 ) ** 30.65 ( .55 ) ** Model Controls Proxy Status 3.06 ( .54 ) ** 2.92 ( .54 ) ** 2.96 ( .54 ) ** Adult Demographics Age 0.10 ( .02 ) ** 0.10 ( .02 ) ** 0.10 ( .02 ) ** White (ref) Black 2.29 ( .36 ) ** 2.42 ( .36 ) ** 2.30 ( .36 ) ** Other 0.43 ( .60 ) 0.45 ( .59 ) 0.45 ( .60 ) Hispanic (ref=not Hispanic) 0.39 ( .40 ) 0.66 ( .44 ) 0.66 ( .45 ) Foreign Born 0.93 ( .33 ) ** 0.86 ( .33 ) ** 0.77 ( .34 ) Years of Education 0.16 ( .03 ) ** 0.12 ( .04 ) ** 0.11 ( .04 ) ** Household Income 0.02 ( .01 ) 0.01 ( .01 ) 0.01 ( .01 ) Married 0.25 ( .19 ) 0.25 ( .19 ) 0.21 ( .20 ) Health Functional Status 0.38 ( .08 ) ** 0.39 ( .08 ) ** 0.40 ( .08 ) ** Self Rated Health 1.00 ( .09 ) ** 0.97 ( .09 ) ** 0.98 ( .09 ) ** Current Smoker 2.62 ( .24 ) ** 2.65 ( .24 ) ** 2.65 ( .24 ) ** Former Smoker 0.35 ( .19 ) 0.37 ( .19 ) 0.37 ( .19 ) Visited Doctor 0.38 ( .38 ) 0.38 ( .38 ) 0.42 ( .39 ) Visited Hospital 0.23 ( .23 ) 0.20 ( .33 ) 0.18 ( .23 ) No Insurance 0.54 ( .25 ) 0.51 ( .25 ) 0.52 ( .25 ) Neighborhood Demographics Percent Rural 0.05 ( .03 ) 0.02 ( .03 ) 0.02 ( .03 ) Percent Over 65 0.04 ( .12 ) 0.17 ( .13 ) 0.17 ( .13 ) Percent Black 0.03 ( .05 ) 0.14 ( .06 ) 0.14 ( .06 ) Percent Hispanic 0.09 ( .08 ) 0.14 ( .09 ) 0.17 ( .09 ) Percent Foreign Born 0.20 ( .11 ) 0.04 ( .12 ) 0.01 ( .13 ) Neighborhood SES Street Connectivity 0.02 ( 1.35 ) Percent Families Poverty 0.11 ( .17 ) 0.06 ( .17 ) Percent Vacant Houses 0.18 ( .13 ) 0.15 ( .13 ) Percent Unemployed 0.45 ( .30 ) 0.54 ( 02 ) Percent Males w/BA 1.55 ( .65 ) 1.41 ( .66 ) Percent Females w/BA 0.56 ( .63 ) 0.67 ( .67 ) Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 194

194 Table B 12. Variance Components and Fit Statistics from an HLM Model Estimating Adjusted Female BMI in 2000. Variance Components Model 1 Model 2 Model 3 In rate of Change Within group 29.79** 29.70** 29.64** Between Group Level 2 0.78* 0.59 0.51 Between Group Level 3 N, Level 1 4349 4349 4260 N, Level 2 2256 2256 2240 N, Level 3 Goodness of fit Deviance 27244.3 27204.3 26632.2 AIC 27248.3 27208.3 26636.2 BIC 27259.7 27219.8 26647.6 Notes: *p < .05; **p < .01.

PAGE 195

195 Table B 13. Results from a Hierarchical Linear Model Regressing Male Unadjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000 20008. Covariates Model 1 Model 2 Model 3 Fixed Effects Intercept 28.94 ( .63 ) ** 28.93 ( .63 ) ** 28.99 ( .64 ) ** Model Controls Died 1.23 ( .45 ) ** 1.22 ( .45 ) ** 1.24 ( .45 ) ** Attrit 0.01 ( .35 ) 0.03 ( .36 ) 0.01 ( .36 ) Proxy Status 0.15 ( .34 ) 0.16 ( .34 ) 0.21 ( .23 ) Moved Since 2000 0.19 ( .23 ) 0.18 ( .24 ) 0.21 ( .45 ) Adult Demographics Age 0.09 ( .02 ) ** 0.09 ( .02 ) ** 0.09 ( .02 ) ** White (ref) Black 0.26 ( .52 ) 0.31 ( .51 ) 0.41 ( .52 ) Other 0.40 ( .85 ) 0.38 ( .85 ) 0.20 ( .87 ) Hispanic (ref=not Hispanic) 0.11 ( .60 ) 0.18 ( .61 ) 0.04 ( .62 ) Foreign Born 0.79 ( .45 ) 0.78 ( .45 ) 0.79 ( .46 ) Years of Education 0.15 ( .04 ) ** 0.13 ( .04 ) ** 0.14 ( .04 ) ** Household Income 0.00 ( .00 ) 0.00 ( .01 ) 0.00 ( .04 ) Married 0.21 ( .29 ) 0.23 ( .30 ) 0.28 ( .30 ) Health Functional Status 0.02 ( .10 ) 0.01 ( .10 ) 0.01 ( .10 ) Self Rated Health 0.10 ( .10 ) 0.11 ( .10 ) 0.13 ( .10 ) Current Smoker 1.12 ( .34 ) ** 1.15 ( .34 ) ** 1.19 ( .34 ) ** Former Smoker 0.54 ( .26 ) 0.56 () .26 0.59 ( .26 ) Visited Doctor 0.21 ( .37 ) 0.22 ( .37 ) 0.24 ( .38 ) Visited Hospital 0.04 ( .22 ) 0.05 ( .23 ) 0.05 ( .23 ) No Insurance 0.46 ( .42 ) 0.49 ( .42 ) 0.45 ( .42 ) Neighborhood Demographics Percent Rural 0.01 ( .03 ) 0.01 ( .04 ) 0.03 ( .04 ) Percent Over 65 0.02 ( .16 ) 0.04 ( .18 ) 0.04 ( .18 ) Percent Black 0.09 ( .07 ) 0.05 ( .08 ) 0.03 ( .08 ) Percent Hispanic 0.12 ( .10 ) 0.05 ( .12 ) 0.06 ( .12 ) Percent Foreign Born 0.18 ( .15 ) 0.09 ( .17 ) 0.17 ( .18 ) Neighborhood SES Street Connectivity 0.96 ( 1.81 ) Percent Families Poverty 0.06 ( .23 ) 0.04 ( .24 ) Percent Vacant Houses 0.09 ( .17 ) 0.09 ( .17 ) Percent Unemployed 0.13 ( .43 ) 0.06 ( .44 ) Percent Males w/BA 0.61 ( .85 ) 0.67 ( .86 ) Percent Females w/BA 0.27 ( .86 ) 0.35 ( .88 )

PAGE 196

196 Table B 13 Continued Covariates Model 1 Model 2 Model 3 For Rate of Change (2000 2008) Fixed Effects Interval 0.12 ( .11 ) 0.13 ( .11 ) 0.12 ( .11 ) Model Controls Died 0.47 ( .08 ) ** 0.48 ( .08 ) ** 0.49 ( .09 ) ** Attrit 0.00 ( .06 ) 0.00 ( .06 ) 0.00 ( .06 ) Proxy Status 0.03 ( .06 ) 0.02 ( .06 ) 0.03 ( .06 ) Moved Since 2000 0.03 ( .04 ) 0.01 ( .03 ) 0.04 ( .04 ) Adult Demographics Age 0.01 ( .00 ) 0.01 ( .00 ) 0.01 ( .00 ) White (ref) Black 0.06 ( .08 ) 0.05 ( .08 ) 0.05 ( .08 ) Other 0.09 ( .13 ) 0.09 ( .13 ) 0.09 ( .13 ) Hispanic (ref=Not Hispanic) 0.04 ( .09 ) 0.05 ( .09 ) 0.05 ( .09 ) Foreign Born 0.00 ( .07 ) 0.00 ( .07 ) 0.01 ( .07 ) Years of Education 0.01 ( .01 ) 0.01 ( .07 ) 0.01 ( .01 ) Household Income 0.00 ( .00 ) 0.00 ( .00 ) 0.00 ( .00 ) Married 0.00 ( .05 ) 0.01 () .05 0.00 ( .05 ) Health Functional Status 0.01 ( .02 ) 0.00 ( .02 ) 0.00 ( .02 ) Self Rated Health 0.02 ( .02 ) 0.02 ( .02 ) 0.02 ( .02 ) Current Smoker 0.09 ( .06 ) 0.09 ( .06 ) 0.01 ( .05 ) Former Smoker 0.06 (.04) 0.06 ( .04 ) 0.06 ( .04 ) Visited Doctor 0.06 ( .07 ) 0.06 ( .07 ) 0.06 ( .07 ) Visited Hospital 0.01 ( .07 ) 0.01 ( .04 ) 0.01 ( .07 ) No Insurance 0.07 ( .08 ) 0.07 ( .08 ) 0.06 ( .08 ) Neighborhood Demographics Percent Rural 0.00 ( .01 ) 0.00 ( .00 ) 0.00 ( .01 ) Percent Over 65 0.04 ( .03 ) 0.05 ( .02 ) 0.05 ( .03 ) Percent Black 0.01 ( .01 ) 0.01 ( .01 ) 0.01 ( .01 ) Percent Hispanic 0.01 ( .02 ) 0.02 ( .02 ) 0.02 ( .02 ) Percent Foreign Born 0.01 ( .02 ) 0.02 ( .03 ) 0.03 ( .03 ) Neighborhood SES Street Connectivity 0.02 ( .28 ) Percent Families Poverty 0.03 ( .04 ) 0.04 ( .04 ) Percent Vacant Houses 0.02 ( .03 ) 0.02 ( .03 ) Percent Unemployed 0.06 ( .07 ) 0.09 ( .07 ) Percent Males w/BA 0.11 ( .13 ) 0.11 ( .13 ) Percent Females w/BA 0.02 ( .13 ) 0.03 ( .14 )

PAGE 197

197 Table B 14. Variance Components and Fit Statistics from an HLM Model Estimating Changes in Unadjusted Male BMI (2000 2008). Variance Components Model 1 Model 2 Model 3 In rate of Change 0.36** 0.36** 0.36** Within group 2.03** 2.03** 2.04** Between Group Level 2 27.34** 27.42** 27.23** Between Group Level 3 0.00 0.00 0.01 N, Level 1 11324 11324 11123 N, Level 2 N, Level 3 1811 1811 1768 Goodness of fit Deviance 53251.9 53269.5 52369.6 AIC 53263.9 53281.5 52381.6 BIC 53297.3 53314.8 52414.9 Notes: *p < .05; **p < .01.

PAGE 198

198 Table B 15. Results from a Hierarchical Linear Model Regressing Female Adjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000 20008. Covariates Model 1 Model 2 Model 3 Fixed Effects Intercept 28.13 ( .69 ) ** 28.00 ( .69 ) ** 27.92 ( .71 ) ** Model Controls Died 2.20 ( .66 ) ** 2.13 ( .67 ) ** 1.80 ( .68 ) ** Attrit 0.42 ( .37 ) 0.46 ( .40 ) 0.58 ( .40 ) Proxy Status 0.74 ( .78 ) 0.76 ( .78 ) 0.79 ( .79 ) Moved Since 2000 0.04 ( .26 ) 0.13 ( .26 ) 0.19 ( .26 ) Adult Demographics Age 0.01 ( .02 ) 0.01 ( .02 ) 0.01 ( .02 ) White (ref) Black 3.53 ( .53 ) ** 3.67 ( .05 ) ** 3.57 ( .05 ) ** Other 0.37 ( .89 ) 0.27 ( .88 ) 0.32 ( .89 ) Hispanic (ref=not Hispanic) 1.16 ( .65 ) 1.51 ( .65 ) 1.46 ( .67 ) Foreign Born 0.87 ( .48 ) 0.79 ( .48 ) 0.74 ( .49 ) Years of Education 0.25 ( .05 ) ** 0.19 ( .05 ) ** 0.21 ( .05 ) ** Household Income 0.01 ( .01 ) 0.01 ( .01 ) 0.01 ( .01 ) Married 0.07 ( .24 ) 0.07 ( .24 ) 0.10 ( .25 ) Health Functional Status 0.07 ( .10 ) 0.08 ( .10 ) 0.11 ( .10 ) Self Rated Health 0.30 ( .11 ) ** 0.29 ( .11 ) ** 0.28 ( .12 ) Current Smoker 0.89 ( .37 ) ** 0.92 ( .34 ) ** 0.88 ( .34 ) ** Former Smoker 0.07 ( .27 ) 0.06 ( .27 ) 0.06 ( .27 ) Visited Doctor 0.08 ( .48 ) 0.10 ( .47 ) 0.08 ( .48 ) Visited Hospital 0.24 ( .26 ) 0.22 ( .26 ) 0.25 ( .26 ) No Insurance 0.66 ( .33 ) 0.65 ( .33 ) 0.65 ( .34 ) Neighborhood Demographics Percent Rural 0.06 ( .04 ) 0.05 ( .04 ) 0.03 ( .05 ) Percent Over 65 0.25 ( .18 ) 0.41 ( .19 ) 0.47 ( .19 ) Percent Black 0.02 ( .07 ) 0.21 ( .09 ) 0.21 ( .09 ) Percent Hispanic 0.06 ( .11 ) 0.21 ( .13 ) 0.28 ( .13 ) Percent Foreign Born 0.14 ( .17 ) 0.02 ( .18 ) 0.05 ( .20 ) Neighborhood SES Street Connectivity 2.88 ( 1.99 ) Percent Families Poverty 0.24 ( .25 ) 0.14 ( .25 ) Percent Vacant Houses 0.27 ( .19 ) 0.25 ( .19 ) Percent Unemployed 0.46 ( .46 ) 0.46 ( .46 ) Percent Males w/BA 2.71 ( .94 ) ** 2.31 ( .96 ) Percent Females w/BA 0.64 ( .97 ) 1.03 ( .98 ) Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 199

199 Table B 15 Continued Covariates Model 1 Model 2 Model 3 For Rate of Change (2000 2008) Fixed Effects Interval 0.05 ( .12 ) 0.05 ( .12 ) 0.06 ( .12 ) Model Controls Died 0.45 ( .13 ) ** 0.46 ( .13 ) ** 0.40 ( .13 ) ** Attrit 0.03 ( .06 ) 0.02 ( .06 ) 0.00 ( .07 ) Proxy Status 0.22 ( .14 ) 0.22 ( .14 ) 0.23 ( .14 ) Moved Since 2000 0.01 ( .04 ) 0.01 ( .04 ) 0.01 ( .04 ) Adult Demographics Age 0.01 ( .00 ) ** 0.01 ( .00 ) ** 0.01 ( .00 ) ** White (ref) Black 0.16 ( .08 ) 0.16 ( .08 ) 0.15 ( .08 ) Other 0.14 ( .13 ) 0.14 ( .13 ) 0.16 ( .14 ) Hispanic (ref=Not Hispanic) 0.01 ( .10 ) 0.01 ( .10 ) 0.00 ( .10 ) Foreign Born 0.10 ( .07 ) 0.10 ( .07 ) 0.09 ( .07 ) Years of Education 0.00 ( .01 ) 0.00 ( .01 ) 0.00 ( .01 ) Household Income 0.00 ( .00 ) 0.00 ( .00 ) 0.00 ( .00 ) Married 0.02 ( .04 ) 0.03 ( .04 ) 0.03 ( .04 ) Health Functional Status 0.00 ( .02 ) 0.01 ( .02 ) 0.01 ( .02 ) Self Rated Health 0.03 ( .02 ) 0.03 ( .02 ) 0.03 ( .02 ) Current Smoker 0.02 ( .06 ) 0.01 ( .06 ) 0.02 ( .06 ) Former Smoker 0.02 ( .04 ) 0.02 ( .04 ) 0.02 ( .04 ) Visited Doctor 0.02 ( .09 ) 0.03 ( .09 ) 0.03 ( .09 ) Visited Hospital 0.00 ( .05 ) 0.01 ( .05 ) 0.01 ( .05 ) No Insurance 0.15 ( .06 ) 0.14 ( .06 ) 0.14 ( .06 ) Neighborhood Demographics Percent Rural 0.00 ( .01 ) 0.01 ( .01 ) 0.01 ( .01 ) Percent Over 65 0.02 ( .03 ) 0.04 ( .03 ) 0.05 ( .03 ) Percent Black 0.01 ( .01 ) 0.01 ( .01 ) 0.01 ( .01 ) Percent Hispanic 0.01 ( .02 ) 0.01 ( .02 ) 0.01 ( .02 ) Percent Foreign Born 0.01 ( .02 ) 0.01 ( .03 ) 0.01 ( .03 ) Neighborhood SES Street Connectivity 0.39 ( .29 ) Percent Families Poverty 0.01 ( .04 ) 0.04 ( .04 ) Percent Vacant Houses 0.02 ( .03 ) 0.05 ( .03 ) Percent Unemployed 0.01 ( .07 ) 0.03 ( .07 ) Percent Males w/BA 0.15 ( .13 ) 0.15 ( .14 ) Percent Females w/BA 0.04 ( .14 ) 0.09 ( .14 )

PAGE 200

200 Table B 16. Variance Components and Fit Statistics from an HLM Model Estimating Changes in Adjusted Female BMI (2000 2008). Variance Components Model 1 Model 2 Model 3 In rate of Change 0.54** 0.55** 0.55** Within group 3.13** 3.13** 3.14** Between Group Level 2 41.69** 41.61** 41.48** Between Group Level 3 1.72* 1.22 1.16 N, Level 1 13767 13767 13523 N, Level 2 4107 4107 4029 N, Level 3 2136 2136 2083 Goodness of fit Deviance 76718.2 76698.8 75221.5 AIC 76732.2 76712.8 75235.5 BIC 76772.3 76752.8 75275.4 Notes: *p < .05; **p < .01.

PAGE 201

201 APPENDIX C UNADJUSTED BMI ESTIM ATES FROM CHAPTER 6 This appendix presents results from models that estimate unadjusted BMI using the same models as in Chapter 6.

PAGE 202

202 Table C 1. Results from a Hierarchical Linear Model Regressing Male Unadjusted BMI on Individual a nd Neighborhood Characteristics HRS Cohort 2000. Covariates Model 1 Model 2 Model 3 Fixed Effects Intercept 27.78 (.55) ** 27.72 (.55) ** 27.86 ( .56 ) ** Model Controls Proxy Status 0.35 (.26) 0.34 (.26) 0.32 ( .27 ) Childhood Conditions Poor Health 0.09 (.73) 0.12 (.73) 0.27 ( .73 ) Father Lost Job 0.25 (.20) 0.22 (.20) 0.19 ( .20 ) Father a Laborer 0.35 (.19) 0.28 (.19) 0.27 ( .19 ) 0.01 (.03) 0.00 (.03) 0.00 ( .03 ) 0.01 (.03) 0.00 (.03) 0.00 ( .03 ) Adult Demographics Age 0.16 (.02) ** 0.16 (.02) ** 0.16 ( .02 ) ** White (ref) Black 0.02 (.04) 0.09 (.40) 0.18 ( .40 ) Other 0.02 (.64) 0.35 (.64) 0.16 ( .48 ) Hispanic 0.21 (.47) 0.31 (.47) 0.12 ( .48 ) Foreign Born 0.83 (.34) 0.51 (.34) 0.90 ( .34 ) ** Years of Education 0.08 (.03) 0.06 (.03) 0.06 ( .03 ) Household Income 0.01 (.01) 0.00 (.01) 0.01 ( .01 ) Married 1.15 (.25) ** 1.15 (.25) ** 1.14 ( .25 ) ** Health Functional Status 0.13 (.08) 0.11 (.08) 0.12 ( .08 ) Self Rated Health 0.39 (.09) ** 0.37 (.09) ** 0.39 () .09 Current Smoker 1.84 (.26) ** 1.86 (.26) ** 1.90 ( .27 ) ** Former Smoker 0.03 (.19) 0.06 (.19) 0.09 ( .19 ) Visited Doctor 0.83 (.32) 0.86 (.33) ** 0.84 ( .33 ) ** Visited Hospital 0.33 (.21) 0.33 (.27) 0.34 ( .21 ) No Insurance 0.25 (.29) 0.27 (.30) 0.31 ( .30 ) Neighborhood Demographics Percent Rural 0.01 (.02) 0.02 (.03) 0.02 ( .03 ) Percent Over 65 0.22 (.22) 0.19 (.13) 0.18 ( .13 ) Percent Black 0.04 (.05) 0.01 (.06) 0.02 ( .06 ) Percent Hispanic 0.06 (.08) 0.01 (.09) 0.01 ( .09 ) Percent Foreign Born 0.06 (.11) 0.01 (.11) 0.05 ( .13 ) Neighborhood SES Street Connectivity 0.21 ( 1.34 ) Percent Families Poverty 0.19 (.18) 0.01 ( .18 ) Percent Vacant Houses 0.01 (.12) 0.00 ( .12 ) Percent Unemployed 0.25 (.32) 0.25 ( .03 ) Percent Males w/BA 0.86 (.62) 0.86 ( .06 ) Percent Females w/BA 0.13 (.62) 0.03 ( .06 ) Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 203

203 Table C 2. Variance Components and Fit Statistics from an HLM Model Estimating Changes In Unadjusted Male BMI, 2000. Variance Components Model 1 Model 2 Model 3 In rate of Change Within group 20.02** 19.99** 20.01** Between Group Level 2 0.00 0.00 0.00 Between Group Level 3 N, Level 1 2944 2944 2894 N, Level 2 1803 1803 1760 N, Level 3 Goodness of fit Deviance 17219.4 17214.3 16921.0 AIC 17221.4 17216.3 16923.0 BIC 17226.9 17221.8 16928.5 Notes: *p < .05; **p < .01.

PAGE 204

2 04 Table C 3. Results from a Hierarchical Linear Model Regressing Female Unadjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000. Covariates Model 1 Model 2 Model 3 Fixed Effects Intercept 30.41 (.61) ** 30.35 (.61) ** 30.39 ( .62 ) ** Model Controls Proxy Status 2.65 (.63) ** 2.52 (.63) ** 2.54 ( .63 ) ** Childhood Conditions Poor Health 0.70 (.80) 0.71 (.80) 0.76 ( .80 ) Father Lost Job 0.08 (.23) 0.11 (.23) 0.12 ( .23 ) Father a Laborer 0.19 (.21) 0.08 (.21) 0.08 ( .21 ) 0.02 (.04) 0.03 (.04) 0.04 ( .04 ) Education (years) 0.08 (.03) 0.08 (.03) ** 0.08 (.03) Adult Demographics Age 0.12 (.02) ** 0.11 (.02) ** 0.12 ( .02 ) ** White (ref) Black 2.55 (.40) ** 2.68 (.40) ** 2.58 ( .41 ) ** Other 0.34 (.64) 0.34 (.64) 0.30 ( .64 ) Hispanic 0.76 (.50) 1.06 (.50) 1.07 ( .51 ) ** Foreign Born 0.89 (.37) 0.85 (.37) 0.86 ( .37 ) Years of Education 0.09 (.04) 0.06 (.04) 0.06 ( .04 ) Household Income 0.01 (.01) 0.01 (.01) 0.01 ( .01 ) Married 0.27 (.21) 0.27 (.21) 0.26 ( .21 ) Health Functional Status 0.40 (.09) ** 0.42 (.09) ** 0.43 ( .09 ) ** Self Rated Health 0.95 (.10) ** 0.92 (.10) ** 0.95 ( .10 ) ** Current Smoker 2.58 (.26) ** 2.60 (.26) ** 2.62 ( .26 ) ** Former Smoker 0.38 (.20) 0.40 (.20)* 0.43 ( .21 ) Visited Doctor 0.45 (.41) 0.44 (.41) 0.50 ( .41 ) Visited Hospital 0.63 (.26) 0.58 (.25) 0.58 ( .25 ) No Insurance 0.59 (.27) 0.54 (.27) 0.57 ( .27 ) Neighborhood Demographics Percent Rural 0.06 (.03) 0.01 (.03) 0.00 ( .03 ) Percent Over 65 0.05 (.13) 0.17 (.14) 0.20 ( .14 ) Percent Black 0.01 (.06) 0.14 (.07) 0.14 ( .07 ) Percent Hispanic 0.05 (.08) 0.14 (.10) 0.19 ( .10 ) Percent Foreign Born 0.23 (.13) 0.08 (.13) 0.04 ( .14 ) Neighborhood SES Street Connectivity 1.53 ( 1.45 ) Percent Families Poverty 0.27 (.19) 0.24 ( .19 ) Percent Vacant Houses 0.16 (.14) 0.14 ( .14 ) Percent Unemployed 0.53 (.33) 0.62 ( .34 ) Percent Males w/BA 1.41 (.69) 1.24 ( .70 ) Percent Females w/BA 0.79 (.67) 0.93 ( .68 ) Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 205

205 Table C 4. Variance Components and Fit Statistics from an HLM Model Estimating Changes In Unadjusted Female BMI, 2000. Variance Components Model 1 Model 2 Model 3 In rate of Change Within group 29.17** 29.21** 29.19** Between Group Level 2 0.98* 0.65 0.62 Between Group Level 3 N, Level 1 3705 3705 3640 N, Level 2 2129 2129 2077 N, Level 3 Goodness of fit Deviance 23164.1 23129.8 22715.4 AIC 23168.1 23133.8 22719.4 BIC 23179.4 23145.2 22730.7 Notes: *p < .05; **p < .01.

PAGE 206

206 Table C 5. Results from a Hierarchical Linear Model Regressing Male Unadjusted BMI on Individual and Neighborhood Characteristics HRS Cohort 2000 20008. Covariates Model 1 Model 2 Model 3 For Initial Status (BMI 2000) Fixed Effects Intercept 28.69 ( .70 ) ** 28.72 ( .70 ) ** 28.76 ( .70 ) ** Model Controls Died 1.24 ( .47 ) ** 1.23 ( .47 ) ** 1.24 ( .47 ) ** Attrit 0.01 ( .37 ) 0.01 ( .38 ) 0.02 ( .38 ) Proxy Status 0.17 ( .35 ) 0.19 ( .35 ) 0.23 ( .35 ) Moved Since 2000 0.22 ( .24 ) 0.21 ( .24 ) 0.25 ( .25 ) Childhood Conditions Poor Health 0.15 ( 1.02 ) 0.12 ( 1.02 ) 0.02 ( 1.03 ) Father Lost Job 0.18 ( .28 ) 0.18 ( .28 ) 0.16 ( .29 ) Father a Laborer 0.30 ( .27 ) 0.28 ( .28 ) 0.38 ( .28 ) 0.01 ( .05 ) 0.01 ( .05 ) 0.00 ( .05 ) 0.00 ( .04 ) 0.00 ( .04 ) 0.00 ( .04 ) Adult Demographics Age 0.10 ( .03 ) ** 0.11 ( .03 ) ** 0.11 ( .03 ) ** White (ref) Black 0.53 ( .56 ) 0.56 ( .56 ) 0.66 ( .57 ) Other 0.02 ( .92 ) 0.00 ( .92 ) 0.16 ( .93 ) Hispanic 0.12 ( .66 ) 0.16 ( .67 ) 0.09 ( .69 ) Foreign Born 0.73 ( .48 ) 0.73 ( .48 ) 0.77 ( .49 ) Years of Education 0.14 ( .04 ) ** 0.13 ( .05 ) ** 0.13 ( .05 ) ** Household Income 0.00 ( .01 ) 0.00 ( .01 ) 0.00 ( .01 ) Married 0.34 ( .31 ) 0.36 ( .31 ) 0.36 ( .31 ) Health Functional Status 0.02 ( .11 ) 0.02 ( .11 ) 0.02 ( .11 ) Self Rated Health 0.11 ( .11 ) 0.12 ( .11 ) 0.12 ( .11 ) Current Smoker 1.14 ( .36 ) ** 1.15 ( .36 ) ** 1.20 ( .36 ) ** Former Smoker 0.56 ( .27 ) 0.58 ( .27 ) 0.60 ( .27 ) Visited Doctor 0.27 ( .39 ) 0.28 ( .39 ) 0.28 ( .40 ) Visited Hospital 0.08 ( .24 ) 0.28 ( .24 ) 0.09 ( .24 ) No Insurance 0.54 ( .43 ) 0.56 ( .43 ) 0.57 ( .44 ) Neighborhood Demographics Percent Rural 0.01 ( .04 ) 0.00 ( .04 ) 0.00 ( .04 ) Percent Over 65 0.04 ( .17 ) 0.01 ( .18 ) 0.10 ( .02 ) Percent Black 0.05 ( .07 ) 0.03 ( .08 ) 0.56 ( .08 ) Percent Hispanic 0.15 ( .11 ) 0.12 ( .12 ) 0.15 ( .13 ) Percent Foreign Born 0.20 ( .16 ) 0.18 ( 1 8 ) 0.73 ( .18 ) Neighborhood SES Street Connectivity 0.67 ( 1.90 ) Percent Families Poverty 0.02 ( .25 ) 0.10 ( .26 ) Percent Vacant Houses 0.07 ( .18 ) 0.06 ( .18 ) Percent Unemployed 0.02 ( .45 ) 0.12 ( .46 ) Percent Males w/BA 0.67 ( .88 ) 0.71 ( .89 ) Percent Females w/BA 0.34 ( .87 ) 0.46 ( .88 ) Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 207

207 Table C 5 Continued Covariates Model 1 Model 2 Model 3 For Rate of Change (2000 2008) Fixed Effects Interval 0.12 ( .12 ) 0.12 ( .69 ) 0.11 ( .12 ) Model Controls Died 0.37 ( .09 ) ** 0.48 ( ..09 ) ** -0.49 ( .09 ) ** Attrit 0.00 ( .06 ) 0.02 ( .06 ) 0.02 ( .06 ) Proxy Status 0.03 ( .07 ) 0.04 ( .07 ) 0.04 ( .07 ) Moved Since 2000 0.04 ( .04 ) 0.04 ( .04 ) 0.04 ( .04 ) Childhood Conditions Poor Health 0.10 ( .15 ) 0.11 ( .15 ) 0.11 ( .15 ) Father Lost Job 0.01 ( .04 ) 0.00 ( .04 ) 0.00 ( .04 ) Father a Laborer 0.01 ( .04 ) 0.01 ( .04 ) 0.01 ( .04 ) 0.00 ( .01 ) 0.00 ( .01 ) 0.01 ( .01 ) 0.00 ( .01 ) 0.00 ( .01 ) 0.00 ( .01 ) Adult Demographics Age 0.01 ( .00 ) 0.01 ( .00 ) 0.01 ( .00 ) White (ref) Black 0.11 ( .09 ) 0.09 ( .09 ) 0.09 ( .09 ) Hispanic 0.01 ( .10 ) 0.02 ( .10 ) 0.03 ( .10 ) Foreign Born 0.02 ( .02 ) 0.02 ( .07 ) 0.01 ( .08 ) Years of Education 0.01 ( .01 ) 0.01 ( .01 ) 0.02 ( .01 ) Household Income 0.00 ( .00 ) 0.00 ( .00 ) 0.00 ( .00 ) Married 0.01 ( .05 ) 0.01 ( .05 ) 0.02 ( .05 ) Health Functional Status 0.01 ( .02 ) 0.00 ( .02 ) 0.01 ( .02 ) Self Rated Health 0.02 ( .02 ) 0.02 ( .02 ) 0.02 ( .02 ) Current Smoker 0.13 ( .06 ) 0.09 ( .06 ) 0.09 ( .06 ) Former Smoker 0.06 ( .04 ) 0.06 ( .04 ) 0.07 ( .04 ) Visited Doctor 0.07 ( .07 ) 0.06 ( .07 ) 0.06 ( .07 ) Visited Hospital 0.06 ( .04 ) 0.00 ( .04 ) 0.00 ( .04 ) No Insurance 0.07 ( .09 ) 0.07 ( .09 ) 0.07 ( .08 ) Neighborhood Demographics Percent Rural 0.00 ( .01 ) 0.00 ( .01 ) 0.00 ( .01 ) Percent Over 65 0.04 ( .02 ) 0.05 ( .03 ) 0.05 ( .03 ) Percent Black 0.00 ( .01 ) 0.01 ( .01 ) 0.01 ( .01 ) Percent Hispanic 0.02 ( .02 ) 0.03 ( .02 ) 0.03 ( .02 ) Percent Foreign Born 0.02 ( .03 ) 0.03 ( .03 ) 0.04 ( .03 ) Neighborhood SES Street Connectivity 0.01 ( .29 ) Percent Families Poverty 0.04 ( .04 ) 0.04 ( .04 ) Percent Vacant Houses 0.02 ( .03 ) 0.02 ( .03 ) Percent Unemployed 0.08 ( .07 ) 0.11 ( .07 ) Percent Males w/BA 0.10 ( .14 ) 0.10 ( .14 ) Percent Females w/BA 0.05 ( .13 ) 0.05 ( .14 ) Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 208

208 Table C 6. Variance Components and Fit Statistics from an HLM Model Estimating Changes In Unadjusted Male BMI (2000 2008). Variance Components Model 1 Model 2 Model 3 In rate of Change 0.34** 0.35** 0.35** Within group 2.07** 2.07** 2.08** Between Group Level 2 27.37** 27.47 ** 27.25** Between Group Level 3 0.00 0.01 0.00 N, Level 1 10512 10512 10328 N, Level 2 N, Level 3 1811 1811 1768 Goodness of fit Deviance 49604.8 49607.5 48780.2 AIC 49616.8 49619.5 48792.2 BIC 49649.8 49652.5 48825.0 Notes: *p < .05; **p < .01.

PAGE 209

209 Table C 7. Results from a Hierarchical Linear Model Regressing Female Unadjusted BMI on Individual a nd Neighborhood Characteristics HRS Cohort 2000 20008. Covariates Model 1 Model 2 Model 3 For Initial Status (BMI 2000) Fixed Effects Intercept 27.85 (.75) ** 27.82 ( .75 ) ** 27.74 ( .77 ) ** Model Controls Died 1.92 ( .70 ) ** 1.87 ( .70 ) ** 1.64 ( .71 ) Attrit 0.37 ( .41 ) 0.38 ( .41 ) 0.47 ( .42 ) Proxy Status 0.92 ( .82 ) 0.98 ( .82 ) 1.03 ( .83 ) Moved Since 2000 0.10 ( .26 ) 0.19 ( .26 ) 0.22 ( .27 ) Childhood Conditions Poor Health 0.65 ( 1.16 ) 0.69 ( 1.15 ) 0.72 ( 1.15 ) Father Lost Job 0.04 ( .32 ) 0.09 ( .32 ) 0.04 ( .32 ) Father a Laborer 0.13 ( .29 ) 0.03 ( .29 ) 0.02 ( .29 ) 0.04 ( .05 ) 0.04 ( .05 ) 0.04 ( .05 ) 0.14 ( .04 ) ** 0.13 ( .05 ) ** 0.13 ( .05 ) ** Adult Demographics Age 0.03 ( .02 ) 0.01 ( .02 ) 0.02 ( .02 ) White (ref) Black 3.52 ( .57 ) ** 3.73 ( .56 ) ** 3.70 ( .57 ) ** Other 0.22 ( .91 ) 0.21 ( .91 ) 0.41 ( .91 ) Hispanic 0.89 ( .70 ) 1.31 ( .70 ) 1.27 ( .71 ) Foreign Born 0.20 ( .52 ) 0.13 ( .52 ) 0.12 ( .52 ) Years of Education 0.15 ( .06 ) ** 0.10 ( .06 ) 0.11 ( .06 ) Household Income 0.00 ( .01 ) 0.00 ( .01 ) 0.01 ( .01 ) Married 0.15 ( .25 ) 0.15 ( .25 ) 0.15 ( .26 ) Health Functional Status 0.12 ( .11 ) 0.13 ( .11 ) 0.14 ( .11 ) Self Rated Health 0.29 ( .11 ) ** 0.27 ( .11 ) 0.26 ( .12 ) Current Smoker 0.86 ( .35 ) 0.89 ( .34 ) ** 0.86 ( .35 ) ** Former Smoker 0.11 ( .28 ) 0.09 ( .28 ) 0.07 ( .28 ) Visited Doctor 0.18 ( .49 ) 0.19 ( .49 ) 0.18 ( .50 ) Visited Hospital 0.27 ( .27 ) 0.24 ( .27 ) 0.26 ( .27 ) No Insurance 0.61 ( .34 ) 0.59 ( .35 ) 0.58 ( .35 ) Neighborhood Demographics Percent Rural 0.06 ( .04 ) 0.06 ( .04 ) 0.04 ( .05 ) Percent Over 65 0.23 ( .19 ) 0.43 ( .20 ) 0.48 ( .20 ) Percent Black 0.05 ( .08 ) 0.24 ( .09 ) ** 0.24 ( .10 ) ** Percent Hispanic 0.08 ( .12 ) 0.23 ( .13 ) 0.29 ( .14 ) Percent Foreign Born 0.22 ( .18 ) 0.01 ( .18 ) 0.01 ( .20 ) Neighborhood SES Street Connectivity 2.31 ( 2.06 ) Percent Families Poverty 0.25 ( .26 ) 0.20 ( .20 ) Percent Vacant Houses 0.36 ( .20 ) 0.33 ( .48 ) Percent Unemployed 0.31 ( .48 ) 0.40 ( .48 ) Percent Males w/BA 2.89 ( .96 ) ** 2.51 ( .98 ) ** Percent Females w/BA 0.41 ( .94 ) 0.70 ( .95 ) Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 210

210 Table C 7 Continued Covariates Model 1 Model 2 Model 3 For Rate of Change (2000 2008) Fixed Effects Interval 0.07 ( .13 ) 0.07 ( .13 ) 0.08 ( .13 ) Model Controls Died 0.42 ( .13 ) ** 0.42 ( .13 ) 0.38 ( .14 ) ** Attrit 0.05 ( .07 ) 0.07 ( .07 ) 0.04 ( .07 ) Proxy Status 0.24 ( .15 ) 0.24 ( .15 ) 0.26 ( .15 ) Moved Since 2000 0.01 ( .04 ) 0.01 ( .04 ) 0.01 ( .01 ) Childhood Conditions Poor Health 0.11 ( .18 ) 0.11 ( .18 ) 0.11 ( .18 ) Father Lost Job 0.02 ( .05 ) 0.02 ( .05 ) 0.03 ( .05 ) Father a Laborer 0.03 ( .04 ) 0.03 ( .04 ) 0.03 ( .01 ) 0.01 ( .01 ) 0.01 ( .01 ) 0.01 ( .01 ) 0.01 ( .01 ) 0.01 ( .01 ) 0.01 ( .01 ) Adult Demographics Age 0.01 ( .00 ) ** 0.01 ( .00 ) ** 0.01 ( .00 ) ** White (ref) Black 0.01 ( .08 ) 0.16 ( .08 ) 0.15 ( .09 ) Hispanic 0.02 ( .10 ) 0.12 ( .10 ) 0.01 ( .11 ) Foreign Born 0.17 ( .08 ) 0.17 ( .08 ) 0.17 ( .08 ) Years of Education 0.00 ( .01 ) 0.00 ( .01 ) 0.00 ( .01 ) Household Income 0.00 ( .00 ) 0.00 ( .00 ) 0.00 ( .00 ) Married 0.04 ( .04 ) 0.04 ( .04 ) 0.04 ( .04 ) Health Functional Status 0.01 ( .02 ) 0.01 ( .02 ) 0.01 ( .02 ) Self Rated Health 0.03 ( .02 ) 0.02 ( .02 ) 0.02 ( .02 ) Current Smoker 0.01 ( .06 ) 0.01 ( .06 ) 0.01 ( .06 ) Former Smoker 0.02 ( .04 ) 0.02 ( .04 ) 0.02 ( .04 ) Visited Doctor 0.04 ( .09 ) 0.04 ( .09 ) 0.04 ( .09 ) Visited Hospital 0.00 ( .05 ) 0.00 ( .05 ) 0.00 ( .05 ) No Insurance 0.14 ( .07 ) 0.14 ( .07 ) 0.14 ( .07 ) Neighborhood Demographics Percent Rural 0.00 ( .01 ) 0.01 ( .01 ) 0.01 ( .01 ) Percent Over 65 0.03 ( .03 ) 0.05 ( .03 ) 0.02 ( .03 ) Percent Black 0.01 ( .01 ) 0.02 ( .01 ) 0.02 ( .01 ) Percent Hispanic 0.00 ( .02 ) 0.01 ( .02 ) 0.01 ( .02 ) Percent Foreign Born 0.00 ( .03 ) 0.00 ( .03 ) 0.02 ( .03 ) Neighborhood SES Street Connectivity 0.08 ( .30 ) Percent Families Poverty 0.03 ( .04 ) 0.02 ( .04 ) Percent Vacant Houses 0.05 ( .03 ) 0.05 ( .03 ) Percent Unemployed 0.03 ( .07 ) 0.02 ( .07 ) Percent Males w/BA 0.20 ( .14 ) 0.17 ( .14 ) Percent Females w/BA 0.01 ( .14 ) 0.01 ( .14 ) Notes: BMI = body mass index; SRH = self rated health; *p < .05; **p < .01.

PAGE 211

211 Table C 8. Variance Components and Fit Statistics from an HLM Model Estimating Changes In Unadjusted Female BMI (2000 2008). Variance Components Model 1 Model 2 Model 3 In rate of Change 0.51** 0.51** 0.52** Within group 3.16** 3.16** 3.18** Between Group Level 2 39.76** 39.69 ** 39.54** Between Group Level 3 2.03* 1.43 1.42 N, Level 1 13767 13523 13523 N, Level 2 N, Level 3 2136 2083 2083 Goodness of fit Deviance 70632.5 70600.8 69403.5 AIC 70646.5 70614.8 69417.5 BIC 70686.2 70654.4 69457.0 Notes: *p < .05; **p < .01.

PAGE 212

212 LIST OF REFERENCES Adams, Scott J. 2002. "Educational Attainment and Health: Evidence from a Sample of Older Adults." Education Economics 10:97 109 Adler, Nancy E. and Judith Stewart. 2010. "Preface to The Biology of Disadvantage: Socioeconomic Status and Health." Annals of the New York Academy of Sciences 1186:1 4. Alaimo, Katherine, Christine M. Olson, and Edward A. Frongillo, Jr. 2001. "Low Family Income and Food Insufficiency in Relation to Overweight in US Children: Is There a Paradox?" Archives of Pediatrics Adolescent Medicine 155:1161 1167. Alwin, Duane F. an d Linda A. Wray. 2005. "A Life Span Developmental Perspective on Social Status and H ealth." The Journals of Gerontology Series B: Psychological Sciences and Social Sciences 60:s7 s14. Anderson, Patricia M. and Kri stin F. Butcher. 2006. "Childhood Obesity: Trends and Potential Causes." The Future of Children 16:19 45. Anderson, Patricia., Kristin. Butcher, and Diane. Schanzenbach. 2007. "Childhood Disadvantage and Obesity: Is Nurture Trumping Nature?" NBER Working P aper Series. Antin, Tamar M. J. and Geoffrey Hunt. 2011. "Embodying both stigma and satisfaction: an interview study of African American women." Critical Public Health 23:17 31. Arnesen, E. and A. Forsdahl. 1985. "The Tromso Heart Study: Coronary Risk Fact ors And Their Association With Living Conditions During Childhood ." J ournal of Epidemio logical & Community Health 39:210 214. Ball, Kylie and Gita D. Mishra. 2006. "Whose Socioeconomic Status Influences A Woman's Obesity Risk: Her Mother's, Her Father's, O r Her Own?" International Journal of Epidemiology 35:131 138. Baum Ii, Charles L. and Christopher J. Ruhm. 2009. "Age Socioeconomic Status And Obesity Growth ." Journal of Health Economics 28:635 648. Behrman, Jere R. 1996. "THE IMPACT OF HEALTH AND NUTRITION ON EDUCATION." The World Bank Research Observer 11:23 37. Ben Shlomo, Y. and D. Kuh. 2002. A Life Course Approach To Chronic Disease Epidemiology: Conceptual Models, Empirical Challenges And Interdisciplinary Perspectives ." Int ernational J ournal of Epidemiol ogy 31:285 293. Best, Latrica E., Mark D. Hayward, and Mira M. Hidajat. 2005. "Life Course Pathways t o Adult Onset Diabetes ." Biodemography and Social Biology 52:94 111.

PAGE 213

213 Bielicki, T., A. Sz klarska, Z. Welon, And R. M. Malina. 2000. "Variation In The Body Mass Index Among Young Adult Polish Males Between 1965 a nd 1995." International Journal Of Obesity a nd Related Metabolic Disorders : Journal of t he International Association f or The Study o f Obesity 24:658 662. Blane, D., C. L. Hart, G. Davey Smith, C. R. Gillis, D. J. Hole, and V. M. Hawthorne. 1996. "Association o f Cardi ovascular Disease Risk Factors w ith Socioecon omic Position During Childhood a nd During Adulthood ." British Medical Journal 313 :1434 Braddon, Fiona E. M., Bryan Rodgers, Michael E. J. Wadsworth, and John M. C. Davies. 1986. "Onset Of Obesity In A 36 Year Birth Cohort Study." British Medical Journal (Clinical Research Edition) 293:299 303. Bryk, A. S. and S. W. Raudenbush. 199 2. Hierarchical Linear Models: Applications and Data Analysis Methods. Boardman, J. D., J. M. Saint Onge, B. C. Haberstick, D. S. Timberlake, and J. K. Hewitt. 2008. Do Schools Moderate the Genetic Determinants of Smoking? Behav ioral Genet ics 38:234 246 Bourdieu, Pierre. 1986. The forms of capital. In J. Richardson (Ed.) Handbook of Theory and Research for the Sociology of Education (New York, Greenwood), 241 258. Botoseneanu, Anda and Jersey Liang. 2011. "Social Stratification of Body Weight Trajectory in Middle Age and Older Americans: Results From a 14 Year Longitudinal Study." Journal of Aging and Health 23:454 480 Brisbois, T. D., A. P. Farmer, an d L. J. McCargar. 2012. "Early M arkers o f Adult Obesity: A R eview." Obesity Reviews 13:347 367. Cade, J., H. Upmeier, C. Calvert, and D. Greenwood. 1999. "Costs of a healthy diet: analysis from the UK Women's Cohort Study." Public H ealt h N utrition 2:505 512. Cawley, John. 2000. "Body Weight and Women's Labor Market Outcomes." National Bureau of Econo mic Research Working Paper Series No. 7841. Claussen, B., G. Davey Smith, and D. Thelle. 2003. "Impact of childhood and adulthood socioeconomic position on cause specific mortality: the Oslo Mortality Study." J ournal of Epidemio ogical & Community Health 57 :40 45. Coogan, Patricia F., Yvette C. Cozier, Supriya Krishnan, Lauren A. Wise, Lucile L. Adams Campbell, Lynn Rosenberg, and Julie R. Palmer. 2010. "Neighborhood Socioeconomic Status in Relation to 10 Year Weight Gain in the Black Women's Health Study." Obesity 18:2064 2065.

PAGE 214

214 Currie, Janet. 2008. "Healthy, Wealthy, and Wise: Socioeconomic Status, Poor Health in Childhood, and Human Capital Development." National Bureau of Economic Research Working Paper Series No. 13987. Cutler, David, Edward Glaeser, and Jesse Shapiro. 2003. "Why Have Americans Become More Obese?" National Bureau of Economic Research Working Paper Series No. 9446. D annefer, Dale. 1987. "Aging as Intracohort D ifferentiat ion: Accentuation, the Matthew Effect, and the Life C ourse." Sociological Forum 2:211 236. Davis, Kathy. 2010. "Women and Obesity: What Feminists Do (and Don't) Have to Say About It." Pp 33 41 in Whose Weight Is it Anyway:Essays on Ethics and Eating edited by S. Vandamme, S. van de Vathorst and I. de Beaufort. Acco Davis, Kingsley and Wilbert E. Moore. 1945. "Some Principles of Stratification." American Sociological Review 10:242 249. Deckelbaum, Richard J. and Christine L. Williams. 2001. "Childhood Obesity: The Health Issue." Obesity Research 9:239S 243S. Diez Roux, Ana V. 2001. "Investigating Neighborhood and Area Effects on Health." American Journal of Public Health 91:1783 1789. Do, D. Phuong, Tamara Dubowitz, Chloe E. Bird, Nicole Lurie, Jos J. Escarce, and Brian K. Finch. 2007. "Neighborhood Context and E thnicity Differences in Body Bass Index: A M ultilevel Analysis Using the NHANES III S urvey (1988 1994)." Economics & Human Biology 5:179 203. Dounchis, Jennifer Zoler, Helen A. Hayden, and Denise E. Wilfley. 2001. "Obesity, Body Image, and Eating Disorders in Ethnically Diverse Children a nd Adolescents ." Pp. 67 98 in Body Image, Eating Disorders, and Obesity in Y ou th: Assessment, Prevention, andT treatment E dited by J. K. T. L. Smolak. Washington, DC, US: American Psychological Association. Drewnowski, Adam 1998. "Energy Density, Palatability, and Satiety: Implications for Weight Control." Nutrition Reviews 56:347 353. Drewnowski, Adam and S. E. Specter. 2004. "Poverty and Obesity: the Role of E nergy density and energy costs." The American Journal of Clinic al Nutrition 79:6 16. Duncan, Greg J. and Stephen W. Raudenbush. 1999. "Assesing the Effects of Context in Studies of Child a nd Youth Development." Educational Psychologist 34:29 41.

PAGE 215

215 Eagle, Taylor F., Anne Sheetz, Roopa Gurm, Alan C. Woodward, Eva Kline Rogers, Robert Leibowitz, Jean DuRussel Weston, LaVaughn Palma Davis, Susan Aaronson, Catherine M. Fitzgerald, Lindsey R. Mitchell, Bruce Rogers, Patricia Bruenger, Katherine A. Skala, Caren Goldberg, Elizabeth A. Jackson, Steven R. Erickson, and Kim A. Eagle. 2012. "Understanding Childhood Obesity i n America : Linkages Between Househol d Income, Community Resources, a nd Children's Behaviors ." American Heart Journal 163:836 843. Elder, G. H. 1995. "The Life Course Paradigm: Social Change a nd Individual Development. Examining Lives in Context : P erspe ctives on the Ecology of Human De velopment 1st:xx, 708. Elder, Glen H., Jr. 1998. "The Life Course as Developmental Theory." Child Development 69:1 12. Elder, Glen H., Monica Kirkpatrick Johnson, and Robert Crosnoe. 2003. "The Emergence and Development of Life Course Theory." Pp. 3 19 19 in Handbook of the Life Course Handbooks of Sociology and Social Research edited by J. T. Mortimer and M. J Shanahan: Springer US. Elder, Glen H. and Michael J. Shanahan. 2007. The Life Course and Human Development : John Wiley & Sons, Inc. Escarce, Jose J., Nicole Lurie, and Adria Jewell. RAND Center for Population Health and Health Disparities (CPHHD) Data Co re Series: Decennial Census Abridged, 1990 2010 [United States] [Computer file]. ICPSR27866 v1. Ann Arbor, MI: Inter university Consortium for Political and Social Research [distributor], 2011 10 21. Evans, Gary W., McKenzie L. Jones Rounds, Goran Belojevi c, and Francoise Vermeylen. 2012. "Family Income and Childhood Obesity i n Eight European Cities: The Mediating Roles o f Neighborhood Characteristics a nd Physical Activity ." Social Science & Medicine 75:477 481. Finkelstein, Eric A., Christopher J. Ruhm, an d Katherine M. Kosa. 2004. Economic Causes and Consequences of Obesity ." Annual Review of Public Health 26:239 257. Flegal, Katherine M., Margaret D. Carroll, Cynthia L. Ogden, and Lester R. Curtin. 2010. "Prevalence and Trends in Obesi ty Among US Adults, 1999 2008." The Journal of the American Medical Association 303:235 241. Flegal Km, Graubard B. I. Willi amson D. F. Gail M. H. 2005. "Ex cess deaths associated with underweight, overweight, and obesity." The Journal of the American Medical Association 293: 1861 1867. Flouri, Eirini. 2006. "Parental interest in children's education, children's self esteem and locus of control, and later educational attainment: Twenty six year follow up of the 1970 British Birth Cohort." British Journal of Educational Psycholo gy 76:41 55.

PAGE 216

216 Frankel, Stephen, P. Elwood, G. Davey Smith, S. Frankel, Peter Sweetnam, and John Yarnell. 1996. "Birthweight, Body Mass Index In Middle Age, a nd Incident Coronary Heart Disease ." The Lancet 348:1478 1480. Freedman, Vicki A., Irina B. Grafova, Robert F. Schoeni, and Jeannette Rogo wski. 2008. "Neighborhoods and Disability in Later L ife." Social Science & Medicine 66:2253 2267. Gale, Catharine R., G. David Batty, and Ian J. Deary. 2008. "Locus of Control at Age 10 Years and Health Outcomes and Be haviors at Age 30 Years: The 1970 British Cohort Study." Psychosomatic Medicine 70:397 403. Galobardes, B., J. W. Lynch, and G. Davey Smith. 2008. "Is t he Association Between Childho od Socioeconomic Circumstances a nd Cause Specific Mortality Established? U pdate of a Systematic Review ." Journal of Epidemiology and Community Health 62:387 390. Galobardes, Bruna, John W. Lynch, and George Davey Smith. 2004. "Childhood Socioeconomic Circumstances and Cause specific Mortality in Adulthood: Systematic Review and Interpretation." Epidemiologic Reviews 26:7 21. Gao, Y. J., A. C. Holloway, Z. H. Zeng, G. E. Lim, J. J. Petrik, W. G. Foster, and R. M. Lee. 2005. "Prenatal Exposure t o Nic otine Causes Postnatal Obesity a nd Altered Perivascular Adipose Tissue Function ." Obe ity s Res earch 13:687 692. Garn, S. M., P. J. Hopkins, and A. S. Ryan. 1981. "Different ial fatness gain of low income boys and girls." The American Journal of Clinical N utrition 34:1465 1468. Garn, SM. and DC. Clark. 1974. "Economics of fatness." Ecology of Food and Nutrition 3:19 20. Garn, SM. and A S. Ryan. 1981. "Replicating the income Rev ersal of F atness." Ecology of Food and Nutrition 10:237 239. Goldbla tt P Moore M. E. Stunkard A. J. 1965. "Social Factors in Obesity." The Journal of the American Medical Association 192:1039 1044. Gordon, Lynn D. 1987. "The Gibson Girl Goes to College: P opular Culture and Women's Higher Education in the Progressive Era, 1890 1920." American Quarterly 39:211 230. Bei ng." Journal of Marriage and Family 68:442 454. Grn, Felix and Bruce Blumberg. 2006. "Environmental Obesogens: Organotins and Endocrine Disruption via Nuclear Receptor Signaling." Endocrinology 147:s50 s55.

PAGE 217

217 Guo, Shumei S. and William Camero n Chumlea. 1999 "Tracking of Body Mass Index in Children in Relation to Overweight in A dulthood." The American Journal of Clinical Nutrition 70:145s 148s. Gustafsson, Per E., Henrik Anckarster, Paul Lichtenstein, Nina Nelson, and Per A. Gustafsson. 2010. "Does Quantity Have a Quality All Its Own? Cumulative Adversity And Up a nd Down Regulation Of Circ adian Salivary Cortisol Levels i n Healthy Children ." Psychoneuroendocrinology 35:1410 1415. Haas, Steven 2007. "The Long Term Effects o f Poor Childhood Health: An Assessment and Application o f Retrospective Reports ." Demography 44:113 135. Hales, CN. and DJP. Barker. 1992. "Type 2 (Non Insulin Dependent) Diabetes Mellitus: The Thrifty Phenotype Hypothesis ." Diabetol ogia 35:595 601. Hardt, Jochen And Michael Rutter. 2004. "Validity of Adult Retrospective Reports o f Adverse Childhood Experiences: Review of t he Evidence ." Journal of Child Psychology and Psychiatry 45:260 273. Harris, Mary B., Laurie C. Walters, and Stefanie Waschull. 1991. "Gender and Ethnic Differences in Obesity Related Behaviors and Attitudes In a College Sample." Journal of Applied Social Psychology 21:1545 1566. Hart, C. L., G. D. Smith, and D. Blane. 1998. "Inequalities in mortality by social class measured at 3 stages of the lifecourse." Am erican J ournal of Public Health 88:471 474. Health and Retirement Study,(Sample Sizes and Response Rates) public use dataset. Produced and distributed by the University of Michigan with funding from the National Institute on Aging (grant number NIA U01AG009740). Ann Arbor, MI, (2011). Heidenreich, Paul A., Justin G. Trogdon, Olga A. Khavjou, Javed Butler, Kathleen Dracup, Michael D. Ezekowitz, Eric Andrew Finkelstein, Yuling Hong, S. Claiborne Johnston, Amit Khera, Donald M. Lloyd Jones, Sue A. Nelson, Graham Nic hol, Diane Orenstein, Peter W. F. Wilson, and Y. Joseph Woo. 2011. "Forecasting the Future of Cardiovascular Disease in the United States: A Policy Statement From the American Heart Association." Circulation 123:933 944. Henretta, John C. 2010. "Lifetime M arital History and Mortality After Age 50." Journal of Aging and Health 22:1198 1212 Hertzman, C., C. Power, S. Matthews, and O. Manor. 2001. "Using an Interactive Framework of Society and Lifecourse to Explain Self Rated Health i n Early Adulthood ." Soc ial Sci ence of Med icine 53:1575 1585. Hill, C. P. 2000 Black Feminist Thought: Knowledge Consciousness, And The Politics Of Empowerment New York: Routledge.

PAGE 218

218 James, Sherman A., John Van Hoewyk, Robert F. Belli, David S. Strogatz, David R. Williams, and Trev illore E. Raghunathan. 2006. "Life Course Socioeconomic Position and Hypertension in African American Men: The Pitt County Study." American Journal of Public Health 96:812 817. Jeffery, Robert W. and Allison M. Rick. 2002. "Cross Sectional and Longitudinal Associations between Body Mass Index and Marriage Related Factors." Obesity Research 10:809 815. Katz, Lawrence F., Jeffrey R. Kling, and Jeffrey B. Liebman. 2000. "Moving to O pportunity in Boston: Early Results of a Randomized Mobility Experiment." National Bureau of Economic Research Working Paper Series No. 7973. Kauhanen, Laura, Hanna Maaria Lakka, John W. Lynch, and Jussi Kauhanen. 2006. "Social Disadvantages in Childhood a nd Risk Of All Cause Death and Cardiovascular Disease in Later Life: A Comparison of Historical a nd Retrospective Childhood Information ." International Journal of Epidemiology 35:962 968. Kim, Daniel, S. V. Subramanian, Steven L. Gortmaker, and Ichiro Kawa chi. 2006. "US State A nd County Level Social Capital in Relation to Obesity a nd Physical Inactivity: A Multilevel, Multivariable Analysis ." Social Science & Medicine 63:1045 1059. Kitagawa, Evelyn M. and Philip M. Hauser. 1968. "Education Differentials in Mortality by Cause of Death: United States, 1960." Demography 5:318 353. Koupilova Ilona, David A. Leon, Hans O. Lithell, and Lars Berglund. 1997. "Size at Birth and Hypertension in Longitudinally Followed 50 70 Year Old Men." Blood Pressure 6:223 228. Krieger, N., A. Okamoto, and J. V. Selby. 1998. call Of Childhood Social Class a Validation Study f or Public Health Research" Am erican J ournal of Epidemiol ogy 147:704. Kuh, D., Y. Ben Shlomo, J. Lynch, J. Hal lqv ist, and C. Power. 2003. "Life Course E pidemiology." Journal of Epidemiology and Community Health 57:778 783. Laaksonen, Mikko, Sirpa Sarlio Lhteenkorva, and Eero Lahelma. 2004. "Multiple Dimensions of Socioeconomic Position and Obesity among Employees : The Helsinki Health Study." Obesity Research 12:1851 1858. Lee, Hedwig, Kathleen Harris, and Penny Gordon Larsen. 2009. "Life Course Perspectives on the Links Between Poverty and Obesity During the Transition to Young Adulthood." Population Research and Policy Review 28:505 532.

PAGE 219

219 Lee I, Manson J. E. Hennekens C. H. Paffe nbarger R. S., Jr. 1993. "Body W eigh t and Mortality: A 27 year Follow up of Middle Aged M en." The Journal of the American Medical Association 270:2823 2828. Lenneberg, E.H. 1967. Biological Foundations of Language New York: Wiley. Leon, D. A., H. O. Lithell, D. Vger, I. Koupilov, and R. Mohsen. 1998. Reduced Fetal Growth Rate a nd Increased Risk of Death f rom Ischaem ic Heart Disease: Cohort Study o f 15 000 Swedishmen a nd Women Born 1915 29 Br itish Med ical J ournal 317:241 245 Li, Xia., Hang T. Pham, Amanda. Janesick, and Bruce. Blumberg. 2012. "Triflumizole is an Obesogen in Mice that Acts through Peroxisome Proliferator Activated Recptor Gamma. Environmental Health Perspectives 120: 17 20 1726 Lynch, JW. and GA. Kaplan. 2000. "Socioeconomic Position." Pp. 13 25 in Social Epidemiology edited by L. Berkman and I. Kawachi. New York: Oxford University Press. Macintyre, S ally. 1994. "Understanding the Social Patterning of Health: The Role of the Social S ciences." Journal of Public Health 16:53 59. Mann, S. L., M. E. Wadsworth, and J. R. Colley. 1992. "Accumulation of Factors Influencing Res piratory Illness In Members of A National Birth Cohort a nd Their Offspring ." Journal of Epidemiology and Community Health 46:286 292. Mayer, K. U. 2009. "New Directions in Life Course Research." Annual Review of Sociology 35:413 433. McLaren, Lindsay. 2007. "Socioeconomic Status and Obesity." Epidemiologic Reviews 29:29 48. Merton, Robert K. 1968. "The Ma tthew Effect in Science." Science 159:56 63. Mishra, Gita, Dorothea Nitsch, Stephanie Black, Bianca De Stavola, Diana Kuh, and Rebecca Hardy. 2009. "A Structured Approach to Modelling the Effects o f Binary Exposure Variables Over t he Life Course ." Internat ional Journal of Epidemiology 38:528 537. Mokdad, A. H., E. S. Ford, B. A. Bowman, W. H. Dietz, and F. Vinicor. 2003. Prevelance of Obesity, Diabetes, and Obesity Related Health Risk Factors. Journal of the American Medical Association 289:76 79 Mokdad, A li H., Barbara A. Bowman and Earl S. Ford. 2001. "The C ont inuing Epidemics of Obesity and D iabetes in the United States." Journal of the American Medical Association 10:1195 1200.

PAGE 220

220 Moss, Nancy E. 2002. "Gender Equity a nd Socioec onomic Inequality: A Framework for the Patterning o f Women's Health." Social Science & Medicine 54:649 661. Nicolas, Sirven and Debrand Thierry. 2011. "Social Capital and Hea lth of Older Europeans." IRDES Institut for Research and Information in Health E conomics. Oakes, J. Mic hael and Peter H. Ross i. 2003. "The Measurement of SES in Health Research: Current pPactice and Steps Toward a New A pproach." Social Science & Medicine 56:769 784. Ogden, Cynthia., Molly. Lamb, Margaret. Carroll, and Katherine. Flegal. 2010. "Obesity and S ocioeconomic Status in Children and Adolescents: United States, 2005 2008." NCHS Data Brief no 51. Hyattesville, MD: National Center for Health Statistics. Ogden, JaneThomas Debra. 1999. "The Role of Familial Values in Understanding the Impact of Social Cl ass on Weight Concern." International Journal of Eating Disorders 25:273 279. O'Rand, Angela M. 1996. "The Precious and the Precocious: Understanding Cumulative Disadvantage and Cumulative Advantage Over the Life Course." The Gerontologist 36:230 238. Orpa na, Heather M., Jean Marie Berthelot, Mark S. Kaplan, David H. Feeny, Bentson McFarland, and Nancy A. Ross. 2010. "BMI and Mortality: Results From a National Longitudinal Study of Canadian Adults." Obesity 18:214 218. Palloni, Alberto. 2006. "Reproducing I nequalities: Luck, Wallets, and the Enduring Effects of Childhood Health." Demography 43:587 615. Paeratakul, Sahasporn, Marney A. White, Donald A. Williamson, Donna H. Ryan, and George A. Bray. 2002. "Sex, Race/Ethnicity, Socioeconomic Status, and BMI in Relation to Self Perception of Overweight." Obesity Research 10:345 350. Parker, Karen F., Brian J. Stults, and Stephen K. Rice. 2005. Racial Threat, Concentrated Disadvantage And Social Control: Considering The Macro Level Sources Of Variation In Arrests ." Criminology 43:1111 1134. Parsons, T. J., C. Powers, S. Logan, and C. D Summerbell. 1999. "Childhood Predictors of Adult Obesity: A Systematic R eview." International Journal of Obesity & Related Metabolic Disorders 23:S1 S107. Pearl, Michelle, Paula Br aveman, and Barbara Abrams. 2001. "The Relationship of Neighborhood Socioeconomic Characteristics to Birthweight Among 5 Ethnic Groups in California." American Journal of Public Health 91:1808 1814.

PAGE 221

221 Pickett, K. E. and M. Pearl. 2001. "Multilevel analyses o f neighbourhood socioeconomic context and health outcomes: a critical review." Journal of Epidemiology and Community Health 55:111 122. Poirier, Paul, Thomas D. Giles, George A. Bray, Yuling Hong, Judith S. Stern, F. Xavier Pi Sunyer, and Robert H. Eckel. 2006. "Obesity and Cardiovascular Disease: Pathophysiology, Evaluation, and Effect of Weight Loss." Circulation 113:898 918. Pollitt, Ricardo, Kathryn Rose, and Jay Kaufman. 2005. "Evaluating the Evidence for Models o f Lif e Course Socioeconomic Factors a nd Cardiovascular Outcomes: A Systematic Review ." BMC Public Health 5:7. Power, C., O. Manor, and S. Matthews. 1999 "Child t o Adult Socioeconomic Conditions and Obesity I n A National Cohort ." In ternational J ournal of Obes ity Relat ed Metab olic Disord ers 27:1081 1086. Power, Chris and Sharon Matthews. 1997. "Origins of Health Inequa lities in a National Population Sample ." The Lancet 350:1584 1589. RAND HRS Data, Version L. Produced by the RAND Center for the Study of Aging, with funding from the National Institute on Aging and the Social Security Administration. Santa Monica, CA (December 2011). Rashad, Inas, Michael Grossman, and Shin Yi Chou. 2005. "The Super Size of America: An Economic Estimation of Body Mass Index and Obesity in Adults." National Bureau of Economic Research Working Paper Series No. 11584. Rasmussen, Finn and Malin Johansson. 1998. "The Relation of Weight, Length and Ponderal Index at Birth to Body Mass Index and Overweight among 18 Year Old Males in Sweden." European Journal of Epi demiology 14:373 380. Resnick, Helaine E., Paola Valsania, Jeffrey B. Halter, and Xihong Lin. 2000. "Relation of Weight Gain a nd Weight Lo ss on Subsequent Diabetes Risk i n Overweight Adults ." Journal of Epidemiology and Community Health 54:596 602. Ricciar delli, Lina A., Marita P. McCabe, Robert J. Williams, and J. Kevin Thompson. 2007. "The Role Of Ethnicity and Culture in Body Image a nd Disordered Eating A mong Males ." Clinical Psychology Review 27:582 606. Riley, Matilda White. 1987. "On the Significance of Age in Sociology." American Sociological Review 52:1 14. Robert, Stephanie A. and Eric N. Reither. 2004. A Multilevel Analys is o f Race, Community Disadvantage, a nd Body Mass Index Among Adults In The US." Social Science & Medicine 59:2421 2434. Ross, C atherine E. and Chia ling Wu. 1995. "The Links Between Education and Health." American Sociological Review 60:719 745.

PAGE 222

222 Ross, Catherine E. and John Mirowsky. 1999. "Refining the Assocation between Education and Health: The Effects of Quantity, Credential, a nd Selectivity." Demography 36:445 460. 2001. "Neighborhood Disadvantage, Disorder, and Health." Journal of Health and Social Behavior 42:258 276. Ryder, Norman B. 1965. "The Cohort as a Concept in the Study of Social Change." American Sociological Revi ew 30:843 861. Sampson, Robert J. 2003a. "Neighborhood level Context and Health: Lessons from S ociology." Pp. 132 146 in Neighborhoods and H ealth edited by I. Kawachi and L. F. Berkman. New York: Oxford University Press. 2003b. "The Neighborhood Contex t of Well Being." Perspectives in Biology and Medicine 46:S53 S64. Sampson, Robert J., Stephen W. Raudenbush, and Felton Earls. 1997. "Neighborhoods and Violent Crime: A Multilevel Study of Collective Efficacy." Science 277:918 924. Scharoun Lee, M., J. S. Kaufman, B. M. Popkin, and P. Gordon Larsen. 2009. "Obesity, Race/Ethnicity a nd Life Cour se Socioeconomic Status Across the Transition from Adolescence t o Adulthood ." Journal of Epidemiology and Community Health 63:133 139. Senese, Laura C., Nisha D. Almeida, Anne Kittler Fath, Brendan T. Smith, and Eric B. Loucks. 2009. "Associations Between Childhood Socioeconomic Position and Adulthood Obesity." Epidemiologic Reviews 31:21 51. Serdula, M. K., D. Ivery, R. J. Coates, D. S. Freedman D. F. Williamson, and T. Byers. 1993. "Do Obese Children Become Obese Adults? A Review of the Literature." Preventive Medicine 22:167 177. Shihadeh, Edward S. 1991. "The Prevalence of Husband Centered Migration: Employment Consequences for Married Mother s." Journal of Marriage and Family 53:432 444. Shrewsbury, Vanessa and Jane Wardle. 2008. "Socioeconomic Status and Adiposity in Childhood: A Systematic Review of Cross sectional Studies 1990 2005." Obesity 16:275 284. Singh, A. S., C. Mulder, J. W. R. Twi sk, W. Van Mechelen, and M. J. M. Chinapaw. 2008. "Tracking of Childhood Overweight Into Adulthood: A Systematic Review of t he Literature ." Obesity Reviews 9:474 488. Singh, Gopal K., Mohammad Siahpush, and Michael D. Kogan. 2010. "Rising Social Inequaliti es in US Childhood Obesity, 2003 2007." Annals of Epidemiology 20:40 52.

PAGE 223

223 Smith, James P. 1999. "Healthy Bodies And Thick Wallets: The Dual Relation Between Health And Economic Status." The Journal Of Economic Perspectives 13:145 166. Smith, George Davey, C arole Hart, David Blane, Charles Gillis, and Victor Hawthorne. 1997. "Lifetime Soc ioeconomic Position a nd Mortality: Prospective Observational Study ." British Medical Journal 314:547. Sobal, Jeffery and Albert J. Stunkard. 1989. "Socioeconomic Status and Obesity: A Review of the Literature." Psychological Bulletin 105:260 275. Sobal, Jeffrey, Barbara S. Rauschenbach, and Edward A. Frongillo Jr. 1992. "Marital status, fatness and obesity ." Social Science & Medicine 35:915 923. Stansfeld, S. A. and M. G. Marmot. 2002. "Stress and the Heart : P sychosocial Pathways t o Coronary Heart Disease."xi, 304. Strudler Wallston, Barbara and Kenneth A. Wallston. 1978. "Locus of Control and Health: A Re view of the Literature." Health Education & Behavior 6:107 117. Stunkard, Albert J., Jennifer R. Harris, Nancy L. Pedersen, and Gerald E. McClearn. 1990. "The Body Mass Index of Twins Who Have Been Reared Apart." New England Journal of Medicine 322:1483 14 87. Stunkard, Albert J., Thorkild I. A. Srensen, Craig Hanis, Thomas W. Teasdale, Ranajit Chakraborty, William J. Schull, and Fini Schulsinger. 1986. "An Adoption Study of Human Obesity." New England Journal of Medicine 314:193 198. Swartz, David. 1997. Culture & Power: The Sociology of Pierre Bourdieu University of Chicago Press Teasdale, T. W., Thorkild I. A. Sorensen, and Albert J. Stunkard. 1990. "Genetic And Early Environmental Components In Sociodemographic Influen ces On Adult Body Fatness." Britis h Medical Journal 300:1615 1618. Trotter, LaTonya J., Deborah J. Bowen, and Shirley A. A. Beresford. 2010. "Testing for Racial/Ethnic Differences in the Association Between Childhood Socioeconomic Position and Adult Adiposity." American Journal of Public Health 100:1088 1094. Turner, R. Jay and Donald A. Lloyd. 2003. "Cumulative adversity and drug dependence in young adults: racial/ethnic contrasts." Addiction 98:305 315. U.S. Bureau of the Censu s. 2003. "Statistical Abstract of the United States: Table No HS 22. Educational Attainment by Sex: 1910 to 2002." Retrieved April, 2013 ( http://www.census.gov/statab/hist/HS 22.pdf ). __ 2009. Educational Attainment in the United States: 2007. Retrieved April 2013 ( http://www.census.gov/prod/2009pubs/p20 560.pd f )

PAGE 224

224 U.S. Bureau of Labor Statistics. 2013. "Table A 4 Employment Status Of The Civilian Population 25 Years And Over By Educational Attainment." Retrieved April 2013 ( http://www.bls.gov/news.release/empsit.t04.htm ) van Lenthe, Frank J., Pekka Martikainen, and Johan P. Mackenbach. 2007. "Neighbourhood Inequalities In Health a nd Health Related Behaviour: Results o f Selective Migration ?" Health & Place 13:123 137. Vandegrift, Donald and Tom mer Yoked. 2004. "Obesity Rate s, Income, and S ubu rban Sprawl: An Analysis of US S tates." Health & Place 10:221 229. Wacquant, LoC J. D. and William Julius Wilson. 1989. "The Cost of Racial and Class Exclusion in the Inner City." The Annals of the American Academy of Political and Soci al Science 501:8 25. Wannamethee, S. G., P. H. Whincup, G. Shaper, and M. Walker. 1996. "Influence of Fathers' Social Class o n Cardiovascular Disease In Middle Aged Men. Lancet 348:1259 1263. Wang, Youfa, May A. Beydoun, Lan Liang, Benjamin Caballero, and Shiriki K. Kumanyika. 2008. "Will All Americans Become Overweight or Obese? Estimating the Progression and Cost of the US Obesity Epidemic." Obesity 16:2323 2330. Williams, David R., Selina A. Mohammed, Jacinta Leavell, and C hiquita Collins. 2010. "Race, Socioeconomic Status, and Health: Complexities, Ongoing Challenges, and Research O pportunities." Annals of the New York Academy of Sciences 1186:69 101. Williams, Geoffrey C., Virginia M. Grow, Zachary R. Freedman, Richard M. Ryan, and Edward L. Deci. 1996. "Motivational predictors of weight loss and weight loss maintenance." Journal of Personality and Social Psychology 70:115 126. Winkleby, M. A., D. E. Jatulis, E. Frank, and S. P. Fortmann. 19 92. "Socioeconomic Status a nd Health: How Education, Income, and Occupation Contribute to Risk Factors f or Cardiovascular Disease." American Journal of Public Health 82:816 820. Winship, Christopher and Larry Radbill. 1994. "Sampling Weights and Regression Analysis." Sociological Methods & Research 23:230 257. Chmara. 2007. "Childcare, height and BMI among female Polish university students, 2005." Economics & Human Biology 5:435 442. Zakus, Gloria E. 1983. "Obesity in Chil dren and Adolescents." Social Work in Health Care 8:11 29.

PAGE 225

225 Zinn, Maxine Baca and Dill Bonnie Thornton. 1996. "Theorizing Difference from Multiracial Feminism." Feminist Studies 22:321 331.

PAGE 226

226 BIOGRAPHICAL SKETCH Greg Pavela weight related health outcomes and social relationships. In 2006 Pavela graduated from the University of Virginia with a Bachelor of Arts in sociology Pavela received his Mas ter of Arts i n sociology in 2009 from the University of Florida. He received his Ph.D. from the University of Florida in the spring of 2013 He will continue his work as a post doctoral research fellow at the Nutrition and Obesity Research Cente r at the University of A labama, Birmingham.