USF Libraries
USF Digital Collections

Schools as moderators of neighborhood influences on adolescent academic achievement and risk of obesity

MISSING IMAGE

Material Information

Title:
Schools as moderators of neighborhood influences on adolescent academic achievement and risk of obesity a cross-classified multilevel investigation
Physical Description:
Book
Language:
English
Creator:
Bell-Ellison, Bethany A
Publisher:
University of South Florida
Place of Publication:
Tampa, Fla
Publication Date:

Subjects

Subjects / Keywords:
Social influences
School effects
Adolescent development
Hierarchical linear modeling
Ecological systems theory
Dissertations, Academic -- Measurement and Evaluation -- Doctoral -- USF   ( lcsh )
Genre:
non-fiction   ( marcgt )

Notes

Summary:
ABSTRACT: Grounded in Bronfenbrenner's (1979) Ecological Systems Theory and through the application of cross-classified random effects models, the goal of this study was to examine simultaneously neighborhood and school influences on adolescent academic achievement and risk of obesity, as well as the moderating effects of schools on these outcomes. By examining concurrently neighborhood and school influences on achievement and risk of obesity, this study aimed to fill gaps in the social determinants literature. For example, it is unclear if where an adolescent lives or where she/he attends school has a stronger influence on academic achievement. We also do not know if schools can moderate neighborhood influences on adolescent achievement, nor do we know much about the relationships among schools, neighborhoods, and adolescent risk for obesity.Using data from the National Longitudinal Study of Adolescent Health and the Adolescent Health and Academic Achievement study, four research questions were investigated: (1) To what extent are neighborhood influences on U.S. middle and high school students' academic achievement moderated by school environments? (2) What are the relative influences of neighborhood and school environments on U.S. middle and high school students' academic achievement? (3) To what extent are neighborhood influences on U.S. middle and high school students' risk of obesity moderated by school environments? (4) What are the relative influences of neighborhood and school environments on U.S. middle and high school students' risk of obesity? Findings did not suggest a moderating relationship between neighborhood and school factors examined in this study.In terms of relative relationships with academic achievement, three neighborhood factors (affluence, racial composition, and urbanicity) and two school characteristics (student body racial composition and school socioeconomic status) appeared to have the strongest relationships with adolescent achievement after controlling for individual and other neighborhood and school characteristics. For adolescent risk of obesity, neighborhood affluence and racial composition had statistically significant unique associations, whereas no school factors evidenced statistically significantly relationships with risk of obesity after controlling for other factors. Results of the study were interpreted in terms of contributions to the social determinants literature, as well as recommendations for the improvement of future large-scale surveys.
Thesis:
Dissertation (Ph.D.)--University of South Florida, 2008.
Bibliography:
Includes bibliographical references.
System Details:
Mode of access: World Wide Web.
System Details:
System requirements: World Wide Web browser and PDF reader.
Statement of Responsibility:
by Bethany A. Bell-Ellison.
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains 216 pages.
General Note:
Includes vita.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 002007619
oclc - 405643566
usfldc doi - E14-SFE0002420
usfldc handle - e14.2420
System ID:
SFS0026738:00001


This item is only available as the following downloads:


Full Text
xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 002007619
003 fts
005 20090622133350.0
006 m||||e|||d||||||||
007 cr mnu|||uuuuu
008 090622s2008 flu s 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0002420
035
(OCoLC)405643566
040
FHM
c FHM
049
FHMM
090
LB3051 (Online)
1 100
Bell-Ellison, Bethany A.
0 245
Schools as moderators of neighborhood influences on adolescent academic achievement and risk of obesity :
b a cross-classified multilevel investigation
h [electronic resource] /
by Bethany A. Bell-Ellison.
260
[Tampa, Fla] :
University of South Florida,
2008.
500
Title from PDF of title page.
Document formatted into pages; contains 216 pages.
Includes vita.
502
Dissertation (Ph.D.)--University of South Florida, 2008.
504
Includes bibliographical references.
516
Text (Electronic dissertation) in PDF format.
520
ABSTRACT: Grounded in Bronfenbrenner's (1979) Ecological Systems Theory and through the application of cross-classified random effects models, the goal of this study was to examine simultaneously neighborhood and school influences on adolescent academic achievement and risk of obesity, as well as the moderating effects of schools on these outcomes. By examining concurrently neighborhood and school influences on achievement and risk of obesity, this study aimed to fill gaps in the social determinants literature. For example, it is unclear if where an adolescent lives or where she/he attends school has a stronger influence on academic achievement. We also do not know if schools can moderate neighborhood influences on adolescent achievement, nor do we know much about the relationships among schools, neighborhoods, and adolescent risk for obesity.Using data from the National Longitudinal Study of Adolescent Health and the Adolescent Health and Academic Achievement study, four research questions were investigated: (1) To what extent are neighborhood influences on U.S. middle and high school students' academic achievement moderated by school environments? (2) What are the relative influences of neighborhood and school environments on U.S. middle and high school students' academic achievement? (3) To what extent are neighborhood influences on U.S. middle and high school students' risk of obesity moderated by school environments? (4) What are the relative influences of neighborhood and school environments on U.S. middle and high school students' risk of obesity? Findings did not suggest a moderating relationship between neighborhood and school factors examined in this study.In terms of relative relationships with academic achievement, three neighborhood factors (affluence, racial composition, and urbanicity) and two school characteristics (student body racial composition and school socioeconomic status) appeared to have the strongest relationships with adolescent achievement after controlling for individual and other neighborhood and school characteristics. For adolescent risk of obesity, neighborhood affluence and racial composition had statistically significant unique associations, whereas no school factors evidenced statistically significantly relationships with risk of obesity after controlling for other factors. Results of the study were interpreted in terms of contributions to the social determinants literature, as well as recommendations for the improvement of future large-scale surveys.
538
Mode of access: World Wide Web.
System requirements: World Wide Web browser and PDF reader.
590
Advisor: John M. Ferron, Ph.D.
653
Social influences
School effects
Adolescent development
Hierarchical linear modeling
Ecological systems theory
690
Dissertations, Academic
z USF
x Measurement and Evaluation
Doctoral.
773
t USF Electronic Theses and Dissertations.
4 856
u http://digital.lib.usf.edu/?e14.2420



PAGE 1

Schools as Moderators of Neighborhood In fluences on Adolescent Academic Achievement and Risk of Obesity: A Cro ss-Classified Multilevel Investigation by Bethany A. Bell-Ellison A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy Department of Educational Measurement and Research College of Education University of South Florida Major Professor: John M. Ferron, Ph.D. Robert F. Dedrick, Ph.D. Melinda S. Forthofer, Ph.D. Jeffrey D. Kromrey, Ph.D. Anthony J. Onwuegbuzie, Ph.D. Date of Approval: March 7, 2008 Keywords: social influences, sc hool effects, adolescent develo pment, hierarchical linear modeling, ecological systems theory Copyright 2008, Bethany A. Bell-Ellison

PAGE 2

Acknowledgements I would like to acknowledge the memb ers of my dissertation committee whose expertise, encouragement, and support we re always appreciated. Each of you has influenced my professional development in uni que ways and my gratitude to all of you is ineffable. Thank you for not only welcomi ng me to the program and providing an environment that allowed me to discover my pa ssion for research and statistics, but more importantly, thank you for embracing (and encour aging) my enthusiasm and for nurturing my inquisitiveness. I consider myself fort unate to have had each of you guide me throughout my doctoral journey. Additionally, I would also like to acknowledge “all of th e giants” on whose shoulders I stood on while designing and c onducting this research. This dissertation would not have been possible without the wo rk of other scholars who share my passion for advancing the fields of hierarchical linear modeling and social determinants of adolescent development. Finally, I would like to thank my family and friends for respecting my choice to complete my Ph.D. and providing encour agement along the way. Your unwavering support and understanding throughout my docto ral journey have been invaluable. Nothing great was ever achieved without enthusiasm. Ralph Waldo Emerson

PAGE 3

iii Table of Contents List of Tables................................................................................................................. .. vi List of Figures................................................................................................................ viii Abstract....................................................................................................................... .........x Chapter One: Introduction...................................................................................................1 Statement of the Problem.........................................................................................1 Rationale for the Study............................................................................................3 Purpose of the Study................................................................................................6 Research Questions..................................................................................................7 Overview of the Study Design................................................................................ 7 Data Sources............................................................................................................9 Significance of the Study.........................................................................................9 Delimitations..........................................................................................................10 Limitations ...........................................................................................................11 Definition of Terms................................................................................................15 Organization of Remaining Chapters.....................................................................19 Chapter Two: Literature Review.......................................................................................20 Introduction............................................................................................................20 Theoretical Framework..........................................................................................22 Neighborhood Influences on Adolescent Academic Achievement.......................23 Neighborhood SES.....................................................................................24 Neighborhood male joblessness.................................................................27 Neighborhood social disorganization........................................................27 Perceived neighborhood quality................................................................28 Other neighborhood measures...................................................................29 Neighborhood Influences on Adolescent Risk of Obesity.....................................30 Neighborhood SES.....................................................................................30 Built environment......................................................................................31 Other neighborhood measures...................................................................32 School Influences on Adolescent Academic Achievement...................................34 School sociodemographic characteristics..................................................35 School resources and sector.......................................................................36 Teacher characteristics...............................................................................38 Perceived social clim ate and school quality..............................................39 Organizational climate...............................................................................41

PAGE 4

iv School Influences on Adolescent Risk of Obesity................................................42 Summary................................................................................................................47 Chapter Three: Method......................................................................................................50 Purpose of the Study .............................................................................................50 Research Questions................................................................................................51 Study Design .........................................................................................................51 Overview of the Add Health Study........................................................................52 Study design..............................................................................................52 In-School sampling frame.........................................................................53 In-School Questionnaire............................................................................55 School Administrator Questionnaire.........................................................56 In-Home sampling.....................................................................................57 In-Home Interview....................................................................................57 Parent Questionnaire.................................................................................58 Contextual data..........................................................................................59 Sample weights.........................................................................................59 Overview of AHAA Study.....................................................................................60 Study Sample.........................................................................................................62 Measures................................................................................................................62 Criterion variables......................................................................................65 Predictor variables.....................................................................................68 Data Analysis.........................................................................................................74 Data management.......................................................................................74 Univariate and bivariate analyses..............................................................79 Multivariate analyses.................................................................................79 Model interpretation...................................................................................99 Chapter Four: Results......................................................................................................104 Study Sample.......................................................................................................104 Univariate Analyses.............................................................................................107 Bivariate Analyses...............................................................................................109 Multivariate Analyses..........................................................................................116 Research Question 1................................................................................119 Research Question 2.................................................................................120 Research Question 3.................................................................................125 Research Question 4.................................................................................126 Summary of Findings...........................................................................................131 Chapter Five: Discussion ................................................................................................133 Summary of Findings...........................................................................................134 Neighborhoods, schools, and academic achievement .............................134 Neighborhoods, schools, and risk of obesity...........................................136 Limitations of the Study.......................................................................................138 Implications for the Field.....................................................................................144

PAGE 5

v Directions for Future Research............................................................................145 Conclusions..........................................................................................................147 References..................................................................................................................... ...149 Appendix A: Summary Tables of Prev ious Neighborhood and School Research...........165 Appendix B: BMI Box-and-Whisker Plots......................................................................190 Appendix C: Analysis of Missing Data ..........................................................................193 Appendix D: Investigation of Model Assumptions.........................................................200 About the Author...................................................................................................End Page

PAGE 6

vi List of Tables Table 1 List of Operationalized Variables and Data Source...................................64 Table 2 Intercorrelation of Variables Comprising the Family SES Composite Variable ( n = 10,860).............................................................................. 71 Table 3 Intercorrelation of Variable s Comprising the Neighborhood Affluence Composite Variable ( n = 10,860).......................................................... 73 Table 4 Intercorrelation of Variab les Comprising the Neighborhood Poverty Composite Variable ( n = 10,860)........................................................... 73 Table 5 Intercorrelation of Variables Comprising the School SES Composite Variable ( n = 10,860).............................................................................. 74 Table 6 Intercorrelation of Variable s Comprising the Weight Education Composite Variable ( n = 10,860)........................................................... 76 Table 7 Summary of the Model Structur e for each Cross-Classified Random Effects Model ......................................................................................... 81 Table 8 Unweighted Indi vidual, Neighborhood, and Sc hool Characteristics for Original Sample and Study Sample................................................ 105 Table 9 Descriptive Statistics of Individual, Neighbor hood, and School Characteristics ( n = 10,860).................................................................. 107 Table 10 Unweighted Bivariate Corre lation Matrix for all Criterion and Predictor Variables ( n = 10,860)........................................................... 110 Table 11 Model PseudoR2 Comparisons for Academic Achievement CCREMs............................................................................................... 120 Table 12 Summary Table for Academic Achievement CCREMs ( n = 10,860)........................................................................................... 122 Table 13 Model PseudoR2 Comparisons for Risk of Obesity CCREMs............ 126 Table 14 Summary Table for Risk of Obesity CCREMs ( n = 10,860)................ 128

PAGE 7

vii Table 15 Correlation Coefficient Comparis ons for Different Adolescent Risk of Obesity Measures ( n = 10,860)............................................................. 141 Table A-1 Summary of Neighborhood In fluences on Adolescent Academic Achievement Research Studies ............................................................ 166 Table A-2 Summary of Nei ghborhood Influences on Adolescent Risk of Obesity Research Studies .................................................................................. 174 Table A-3 Summary of School Influences on Adolescent Academic Achievement Research Studies .................................................................................. 178 Table A-4 Summary of School Influen ces on Adolescent Risk of Obesity Research Studies .................................................................................. 187 Table C-1 Frequency of Missing Variables across Observations in the Original Sample ( n = 11.841).............................................................................. 194 Table C-2 Frequency of Missing Variables across Observations after Deleting Cases Missing Household Income Data ( n = 9.919)............................ 197 Table D-1 Tolerance Values for Each Variable Included in Academic Achievement CCREMs......................................................................... 201 Table D-2 Tolerance Values for Each Va riable Included in Risk of Obesity CCREMs................................................................................................ 208

PAGE 8

viii List of Figures Figure 1. Example schematic of cross-clas sified data with adolescents nested within schools and neighborhoods...............................................................8 Figure 2. General overview of Add Health Wave I data sources..............................53 Figure 3. School SES*ne ighborhood affluence......................................................111 Figure 4. School SES*neighborhood poverty.........................................................112 Figure 5. Teacher education*neighborhood affluence............................................113 Figure 6. Teacher education*neighborhood poverty...............................................114 Figure 7. Weight promo tion*neighborhood affluence............................................115 Figure 8. Weight prom otion*neighborhood poverty...............................................116 Figure B-1. Age-and-gender-adjusted BM I box-and-whisker plots for girls..............191 Figure B-2. Age-and-gender-adjusted BM I box-and-whisker plots for boys.............192 Figure C-1. Stem-and-leaf display of correlations between missingness on variables using th e original sample..........................................................195 Figure C-2. Stem-and-leaf display of correlations betw een missingness and observed values using the original sample...............................................196 Figure C-3. Stem-and-leaf display of correlations between missingness on variables after deleting cases missing household income data ...............198 Figure C-4. Stem-and-leaf display of correlations betw een missingness and observed values after deleting ca ses missing household income data.....199 Figure D-1. Box-and-whisker plot for Le vel-1 residuals (academic achievement)....202 Figure D-2. Box-and-whisker plot for neighborhood Level-2 residuals (academic achievement)............................................................................................203

PAGE 9

ix Figure D-3. Box-and-whisker plot for school Level-2 residuals (academic achievement)............................................................................................204 Figure D-4. Level-1 residuals*predicted academic achievement................................205 Figure D-5. Level-2 neighborhood residua ls*predicted academic achievement........206 Figure D-6. Level-2 school residua ls*predicted academic achievement....................207 Figure D-7. Box-and-whisker plot for Level-1 residuals (risk of obesity)..................209 Figure D-8. Box-and-whisker plot for neighborhood Level-2 residuals (risk of obesity)...................................................................................... ..210 Figure D-9. Box-and-whisker plot for school Level-2 residuals (risk of obesity)........................................................................................211 Figure D-10. Level-1 residuals *predicted risk of obesity.............................................212 Figure D-11. Level-2 neighborhood residua ls*predicted risk of obesity......................213 Figure D-12. Level-2 school resi duals*predicted risk of obesity..................................214 Figure D-13. Academic achievement neighborhood Level-2 residuals* neighborhood size....................................................................................215 Figure D-14. Risk of obesity neighborhood Level-2 residuals* neighborhood size....................................................................................216

PAGE 10

x Schools as Moderators of Neighborhood In fluences on Adolescent Academic Achievement and Risk of Obesity: A Cro ss-Classified Multilevel Investigation Bethany A. Bell-Ellison ABSTRACT Grounded in Bronfenbrenner’s (1979) Ec ological Systems Theory and through the application of cross-classified random ef fects models, the goal of this study was to examine simultaneously neighborhood and school influences on adolescent academic achievement and risk of obesity, as well as the moderating effects of schools on these outcomes. By examining concurrently neighborhood and school influences on achievement and risk of obesity, this study aime d to fill gaps in the social determinants literature. For example, it is unclear if where an adolescent lives or where she/he attends school has a stronger influence on academi c achievement. We also do not know if schools can moderate neighborhood influences on adolescent achievement, nor do we know much about the relationships among sc hools, neighborhoods, and adolescent risk for obesity. Using data from the National L ongitudinal Study of A dolescent Health and the Adolescent Health and Academic Achiev ement study, four research questions were investigated: (1) To what extent are neighborhood infl uences on U.S. middle and high school students’ academic achievement moderated by school environments?

PAGE 11

xi (2) What are the relative influences of neighborhood and school environments on U.S. middle and high school students’ academic achievement? (3) To what extent are neighborhood influences on U.S. middl e and high school st udents’ risk of obesity moderated by school environments? (4) What are the relative influences of neighborhood and school environments on U.S. middle and high school st udents’ risk of obesity? Findings did not suggest a moderating relationship between neighborhood and school factors examined in this study. In te rms of relative relationships with academic achievement, three neighborhood factors (afflue nce, racial composition, and urbanicity) and two school characteristics (stude nt body racial composition and school socioeconomic status) appeared to have th e strongest relationships with adolescent achievement after controlling for indi vidual and other neighborhood and school characteristics. For adolescent risk of obesity, neighborhood affluence and racial composition had statistically significant uni que associations, wher eas no school factors evidenced statistically significantly relationships with risk of obesity after controlling for other factors. Results of the study were interpre ted in terms of contri butions to the social determinants literature, as well as recommenda tions for the improvement of future largescale surveys.

PAGE 12

1 Chapter One Introduction Statement of the Problem Academic achievement has been an outcome of interest to educational researchers since the beginning of educati on in the United States. To date, students’ achievement has been studied from several perspectives. In th e past, researchers tended to focus more on individual and family char acteristics (e.g., Marsh & Yeung, 1997; Muijs, 1997; Wentzel, 1998; White, 1982) whereas, recently, an incr easing amount of research has focused more on possible social determinants re lated to academic achievement, including neighborhood characteristics a nd school environments (e.g., Baker, Robinson, Danner, & Neukrug, 2001; Boardman & Saint Onge, 2005; Bowen & Bowen, 1999; Crosnoe & Muller, 2004; Darling-Hammond, 1999; Ever son & Millsap, 2004). However, even though there has been an increase in the num ber of studies that have investigated academic achievement from a social determin ants perspective, it is by no means a new concept. For example, Equality of Educational Opportunity (Coleman et al., 1966) was the first comprehensive, nation-wide investig ation into school influences on academic achievement (Dyer, 1972). Similarly, in his re sponse to Coleman et al.’s (1966) findings and through a reexamination of the data, Armo r (1972) attempted to look past the school environment and examined neighborhood influences on academic achievement. Albeit

PAGE 13

2 Armor’s neighborhood measure was crude and base d solely on aggregated characteristics of students’ families, it was still an early attempt to understand how a child’s social environment relates to academic achieve ment. Likewise, Bronfenbrenner’s (1979) Ecological Systems Theory emphasizes the fact that youth do not live in isolation. Instead, they develop in a variety of cont exts, each of which interacts with their development (Bronfenbrenner, 1979). However, despite previous research fi ndings and suggestions that schools might be powerful moderators of neighborhood eff ects on adolescent development (Leventhal & Brooks-Gunn, 2000), few researchers ha ve examined neighborhood and school influences simultaneously. For example, in their review of 42 neighborhood influence articles on child and adolescent developmen tal outcomes published using both local and national data, Leventhal and Brooks-Gunn (2000) found only two articles that examined neighborhoods and schools simultaneously. Mo reover, in my own review of social context articles published usi ng data from three nationally representative adolescent studies, I found 16 studies involving the ex amination of neighborhood influences on adolescent education and health outcomes, 12 studies wherein school environments were examined, and 4 studies involving the ex amination of the two environments simultaneously. Yet, none of the studies, from either of the reviews, which included both neighborhood and school characteristics, empl oyed the appropriate analytic techniques necessary to understand the simultaneous influe nces of these two social environments, nor did they examine the inte raction, or moderating relati onship, between these social

PAGE 14

3 environments. One exception, not included in either review, is Raudenbush and Bryk’s (2002) discussion of neighborhood and school co ntributions to educational attainment among adolescents in Scotland. However, they also did not investig ate whether schools were moderators of neighborhood influences on achievement. In addition to previous researchers’ lack of investigating multiple environments in relation to adolescent de velopment, they have also tended to limit their investigations to single areas of development a nd well-being. For example, w ithin educational research, dependent variables are often related to cognitive development (e.g., IQ, grade point average, standardized test performance) whereas criteri on variables in public health research are typically relate d to aspects of physical deve lopment (e.g., weight status, drinking and smoking, sexual initiat ion). However, an adolescent’s development is often perceived to include four se parate, yet related areas of well-being: spiritual, mental (intellectual), emotional, and physical (Seawar d, 1999). Thus, consistent with the need to examine simultaneously neighborhood and school influences, it is also necessary for social and behavioral scientists to lo ok beyond single areas of development and investigate multiple realms of adolescent well-being. Rationale for the Study Bronfenbrenner's (1979) Ecological Sy stems Theory posits that human development is influenced by the interrelati ons among settings in which a person actively participates (e.g., family, school, neighborhoods, religious institutions); thus, to study human development effectively, we need to look beyond a single environment and analyze the interactions among multiple e nvironments. When neighborhoods and schools

PAGE 15

4 are conceptualized as representing interrelated social environments, as advocated by Bronfenbrenner (1979), they are no longer simp ly places where an adolescent resides or simple institutions for educating our youth. Inst ead, they are viewed as intricate social structures that impact a child’s overall well -being, including intell ectual, emotional, and physical development, through complex social processes. Distinguishing between people and places is artificial—as not ed by McIntyre and Ellaway (2 003), “people create places and places create people” (p. 26). In a quest to understand fact ors associated with adoles cent educational outcomes, researchers have focused on indi vidual and family characteristics, as well as on social and environmental influences. Over the past few decades, an increasing number of researchers have investigat ed possible environmental fa ctors related to adolescent academic achievement, including neighborhood ch aracteristics and school environments. Examples of significant neighborhood and school characteristics re lated to academic achievement include: neighborhood afflue nce, perceived neighborhood quality, aggregated school poverty, teacher quality, a nd school social climate (Bowen & Bowen, 1999; Crosnoe & Muller, 2004; DarlingHammond, 1999; Everson & Millsap, 2004; Halpern-Felsher et al., 1997). In addition, in their example of cross-classified random effects models (CCREMs), Raudenbus h and Bryk (2002) found neighborhood deprivation to be significantly related to attainment, while statistically controlling for individual and school characteristics. However, the simultaneous investigati on of neighborhood and school influences on adolescent achievement is rare and the examination of schools as moderators of

PAGE 16

5 neighborhood influences appears to be non-ex istent. In addition, am ong studies in which neighborhoods and schools have been examin ed separately, most did not take into account the nested structure of the data. Consequently, results from these studies do not delineate how much variation in the educat ional outcome of interest is related to individual characteristics and how much is related to diffe rences in the neighborhoods in which they live or the schools youth attend. Interestingly, whereas neighborhoods a nd schools have been investigated separately for their influences on educational outcomes, as well as other health behaviors (e.g., smoking and drinking), considerably less research has been conducted on neighborhood and school influences on adolesce nt risk of obesity. Furthermore, although schools and school policies have been suggested as representing impor tant channels to help prevent child and adoles cent obesity (Carter, 2002), the limited social determinants research that has been conducted in this area is relatively new and has primarily focused on neighborhood, not school, influences on adol escent obesity. To date, based on the handful of studies that have involved an examination of neighborhood characteristics related to adolescen t risk of obesity, initial fi ndings suggest that neighborhood socioeconomic status (SES), recreational facili ties, and collective effi cacy are related to adolescent obesity (Cohen, Finch, Bower, & Sastry, 2006; Gordon-Larsen, Nelson, Page, & Popkin, 2006; Nelson, Gordon-Larsen, Song, & Popkin, 2006). These initial findings and suggestions support further investigation of neighborhood and school influences on adolesce nt risk of obesity. Moreover, because of the growing epidemic of adolescent obesity as well as research findings that suggest

PAGE 17

6 being at risk of obesity not only affects a teen ager’s future health as an adult, but also negatively impacts adolescent academic achieve ment during the middle and high school years (Crosnoe & Muller, 2004), investiga tion of the simultaneous and moderating neighborhood and school influences on adol escent risk of obesity is crucial. Purpose of the Study Grounded in Bronfenbrenner’s (1979) Ec ological Systems Theory and through the application of advanced multilevel modeling techniques (Raudenbush & Bryk, 2002), the primary goal of this study was to exam ine simultaneously neighborhood and school influences on academic achievement and adolescent risk of obesity and to examine the moderating effects of schools on these out comes. By examining concurrently neighborhood and school influences on academic achievement and adolescent risk of obesity, this study aimed to fill an important ga p in the social determinants literature. For example, it is unclear if where an adolescen t lives or where she/he attends school has a stronger influence on academic achieveme nt. We also do not know if schools can moderate neighborhood influences on adoles cent academic achievement, nor do we know much about the relationships among schools, neighborhoods, and adolescent risk for obesity. Similarly, by investig ating outcomes related to bot h mental and physical wellbeing, this study helps expand the traditional single-domain approach often undertaken in social and behavioral science research.

PAGE 18

7 Research Questions The following four research questions were investigated in the current study: Research Question 1. To what extent are neighb orhood influences on U.S. middle and high school students’ academic achievem ent moderated by school environments? Research Question 2. What are the relative infl uences of neighborhood and school environments on U.S. middle and high sc hool students’ academic achievement? Research Question 3. To what extent are neighb orhood influences on U.S. middle and high school students’ risk of obes ity moderated by school environments? Research Question 4. What are the relative infl uences of neighborhood and school environments on U.S. middle and high school students’ risk of obesity? Overview of Study Design This study employed a nonexperimental, retr ospective, correl ational research design. Secondary data analyses of the natio nally representative National Longitudinal Study of Adolescent Health (Add Health; National Longitu dinal Study of Adolescent Health [Add Health], 2005c) and Adolescen t Health and Academic Achievement (AHAA; Adolescent Health and Academic Ac hievement Study [AHAA], n.d.) restricteduse data were conducted. The study design was al so cross-sectional in nature because the data represented one point in time. Although multilevel modeling techniques are being used with increasing frequency by educational and other soci al science researchers, use of CCREMs (Raudenbush & Bryk, 2002) is still rare in edu cational research. The lack of CCREMs in education is particularly tr oubling given the cross-classifi ed nature of many education

PAGE 19

8 data structures. For example, Level-1 units (students) are often cross-classified by two Level-2 factors (schools and neighborhoods ) such that students from Neighborhood A might attend a school that students from Neighborhood B and Neighborhood C also attend, and students from the same neighbor hood might attend different schools (Figure 1). When cross-classification of data is ignor ed, models are misspecified, causing them to lack the level of control n ecessary to detect important and possible confounding effects, which, in turn, can lead to spurious conclusions. Figure 1 Example schematic of cross-classified data with adolecents nested within schools and neighborhoods. For this study, the cross-classified multilevel analyses allowed the examination of the influence of multiple contexts on academic achievement and risk of obesity, while statistically controlling for one another. That is, because neighborhood and school environments were analyzed simultaneously, results represent each environment’s unique influence on achievement and risk of obesit y. Further, use of inte ractions within the 1 2 3 4 5 6 7 8 9 10 11 12 A B C D i ii iii iv School Adolescent Neighborhood

PAGE 20

9 CCREMs allowed the investigation of the school environment as a moderator of neighborhood influences on each of the outcomes. Data Sources Data for the study were drawn from Wave I of Add Health (2005c) and AHAA (n.d.)—nationally representative studies with foci on the relations hip between social environments and adolescent education and h ealth outcomes. Within these studies, data were obtained from numerous sources includi ng questionnaires, in terviews, and existing contextual databases (e.g., U.S. Census). Cu rrently, Add Health is the largest, most comprehensive study of adolescents ever conduc ted, with data at the individual, family, school, and neighborhood levels collected in three waves—1994 (Wave 1), 1996 (Wave 2), and 2001-2002 (Wave 3). AHAA data expand Add Health data by providing detailed measures of Add Health part icipants’ educational experien ces, including information on the educational contexts of A dd Health schools. All data used for the current study came from the restricted-use version of the data sources. More information about the studies and the sampling procedures employed is provided in Chapter Three. Significance of the Study By examining simultaneously neighborhood and school influences on multiple adolescent outcomes, this study contribut es to our understanding of the dynamic relationship between neighborhoods and sc hools and their relative influences on adolescent academic achievement and risk of obesity. Before this study, neighborhood and school environments had not been studied together; therefore, previous research findings needed to be interpreted with cau tion (i.e., when studying neighborhood effects,

PAGE 21

10 it was unclear if neighborhood factors were responsible or if school factors were operating as well, and vice versa). However, given the advanced multilevel modeling techniques employed in the current study, findi ngs from this study are likely to be less biased than previous findings Nonetheless, given the correl ational design of the current study, results from the current study still cannot be used to guide policies or programs related to adolescent development. Instead, the most significant contribution of the current study is its addition to the social determinants literature. This study helps to advance our knowledge of social determinants of adolescent development and provides new findings for future researchers to build upon in the creation of experimental quasi-experimental, and qualitative studies focused on the complex relationships between social environments and adolescent wellbeing. Likewise, by investigating academic ach ievement and risk of obesity, this study helps expand the single-domain focus often followed by social and behavioral science researchers. Delimitations The following delimitations were imposed on this study: 1. The study was limited to adolescents who pa rticipated in both the Wave I InSchool Questionnaire and Wave I In-Home Interview, were in 7th through 12th grade at regular middle and high sch ools during the 1994-1995 academic year, and had responses to all vari ables included in the study. 2. The operationalization of academic achieve ment was restricted to adolescent’s Add Health Picture Vocabulary Test (AHPVT) scores.

PAGE 22

11 3. The operationalization of risk of obesity was constrained to self-report measures of height and weight. 4. The operationalization of neighborhood wa s restricted to neighborhoods defined at the census tract level. 5. The operationalization of school was limite d to regular public and private junior high, middle, and high schools (i.e., not magnet or alternative schools). 6. The operationalization of school was cons trained to the school building level. Limitations Although this study contributes to the social determinants literature and enhances our understanding of neighborhoods and schools and their relationships with adolescent academic achievement and risk of obesity, it is not without limitations. For example, this study utilized a non-experimental design, thus the most that could be concluded about the findings was whether the data contradicted or did not contradict the models used to answer the research questions. This limitati on is strong enough that some would not use the term ‘influence’ in the ti tle of a study such as this. Ho wever, acceptable use of the word ‘influence’ is not as clear and we ll-defined as many perceive it to be. The degree to which causal inferences can be drawn from any study lies along a continuum (e.g., Frazier, Tix, & Barron, 2004) and the cut-points delineating such inferences are not the same across researcher s or across disciplines. For example, in the social and behavioral sciences, studies that utilize a true experime ntal design are often deemed worthy of making causal inference statements whereas non-experimental and quasi-experimental studies ar e not (e.g., Games, 1990). However, even among studies

PAGE 23

12 that appear to fulfill the three commonly refere nced criteria for inferring causality in the social and behavioral sciences (i.e., relati onship exists between X and Y, X precedes Y, and ruling out of alternative explanations; Shadish, Cook, & Campbell, 2002), true causal statements are still questionable. First, to make sound causal statements each person in a study needs to be exposed to all of the conditions (i.e., each person needs to be in the control group and treatment group; Holland, 1986; Sobel, 1995), whic h is virtually impossible in the social and behavioral sciences. For example, it is not possible to place a pers on in the treatment group first and then undo any knowledge or ch ange that occurred as result of the treatment or intervention and then place him or her in the control group. Similarly, issues such as history and maturation prohibit rese archers’ ability to expose a person to the control condition first and th en to the treatment group. Unless a person is in both conditions at the same time, he or she is ne ver exactly the same entity, thus researchers are not able to fulfill the requirement of each person in a study being exposed to both conditions. To address the impossibility of exposi ng people to both control and treatment groups, social and behavioral scientist often conduct their research under the st able-unittreatment-value assumption (SUTVA), an a priori assumption that the value of Y for unit u when exposed to treatment t will be the same no matter what mechanism is used to assign treatment t to unit u and no matter what treatments the other units receive (Rubin, 1986, p. 961).

PAGE 24

13 Operating under SUTVA, social and behavioral scientists also appl y various statistical solutions that allow them to estimate th e average causal effect of X over a population (Holland, 1986). Consequently, even the results from well-designed experiments only represent the “average causal effect” and not causal effects at the individual level. Second, even when possible “average causa l effects” are discovered, social and behavioral scientists rarely address the mechanisms behind such relationships (i.e., the nature of the causal effect is usually ignored). In doing so, we are left with an incomplete understanding of the relationship between X and Y. Third, all alterna tive explanations are rarely able to be ruled out. Most researchers assume that random assignment creates equal groups, but we can never be 100% cer tain that even rando mly assigned groups are equal on all possible extraneous variables (i.e., there is always the possibility of committing a Type 1 error). In addition to true experiments, repli cation and extensions of non-experimental studies are other common methods for gather ing evidence to support causal inferences in the social and behavioral sciences. Through this process, researchers aim to gather data, of varying quality, to rule out possible altern ative explanations and to accumulate data that are consistent with causal effects. It is wi thin this part of the research process that the current study fits. Although findings from a single correlationa l study cannot provide evidence of causation, they can and should be used to help inform hypotheses for experimental studies (Games, 1990). This study was developed by “standing on the shoulders of giants who have gone before” and it is hoped that the findings from this study will help inform hypotheses to be ex amined in future experimental research.

PAGE 25

14 However, in order for this study to adequate ly contribute to the social determinants literature and future research, it was important that the language used in the this study is consistent with the language currently used in the social determinants literature [i.e., use of the word influence because this is the term commonly used in the literature (e.g., Beale Spencer, Cole, Jones, & Phillips Sw anson, 1997; Boyle, Georgiades, Racine, & Mustard, 2007; Chase-Lansdale, Gordon, Brooks-Gunn, & Klebanov, 1997; Cohen et al., 2006; Dornbusch, Ritter, & Steinberg, 1991; Eamon, 2005; French, Story, & Jeffery, 2001; Janssen, Boyce, Simpson, & Pickett, 2006; Wickrama, Wickrama, & Bryant, 2006)]. If it is not consistent, ot her researchers in the field wi ll be less likely to read and build upon the findings. However, with this said, it is also important to note that use of the word influence in the title of this study was not intended to show causal relationships. As previously stated, the most that could be concluded about the findings from this study was whether the data contradict ed or did not contra dict the models used to answer the research questions. Other study limitations include several threat s to external and internal validity. Specifically, ecological validity, specificity of variables, tempor al validity, and crud factor (Onwuegbuzie, 2003) are four threats to external va lidity of the current study. Ecological validity is a threat because statistical software packages cannot include sampling weights with CCREMs, thus findi ngs from the current study have limited generalizability and cannot be generalized to the national population. Similarly, because the variables included in the current study were collected at a specific location, under specific circumstances and are used under a specific operational definition

PAGE 26

15 (Onwuegbuzie, 2003), specificity of variables is also a threat to external validity. Temporal validity is a threat because the da ta were from 1990 and 1994, thus, it is likely that neighborhood and school char acteristics are different toda y. Crud factor is a threat because the large sample size increases the likelihood of rejecting a null hypothesis even if the relationship between variables is trivia l, thus leading to the potential interpretation of statistical artifacts and not meaningful associations between variables (Onwuegbuzie, 2003). Instrumentation and model misspecification ar e two threats to internal validity in the current study. Instrumentation refers to the limitations that (a) individual-level variables included from the Add Health data were self-reported, (b) neighborhoods were defined administratively (i.e ., at the census tract leve l) and not by respondents’ definitions of their neighborhoods, and (c) schoo ls were defined at the building level and not at a more specific unit such as classr ooms or curricular track. Model misspecification refers to the limitations that variable selec tion was limited to variables available from the data sources and that the multilevel analysis only included two of the many social environments that adolescents navigate on a daily basis. Definition of Terms Academic achievement For the current study, adoles cents’ Add Health Picture Vocabulary Test (Add Health, 2004 c) standardized scores were used to operationalize academic achievement. Add Health Picture Vo cabulary Test (AHPVT). The AHPVT was a computerized, abridged version of the Peabody Picture Vocabulary Test—Revised, Form L; a

PAGE 27

16 commonly used screening test of verbal ability (Dunn & Dunn, 1981). In this test, the interviewer reads each of the 87 words aloud and the adolescent selected one answer from four black-and-white i llustrations that best fit its meaning (Add Health, 2004c). Body mass index (BMI). Body mass index is a number calculated from a person’s weight and height [wei ght (lbs)/height (in)2*703]. BMI is considered a reliable indicator of body fatness for most people and is used to screen for weight categories (i.e., underweight, normal, overweight, and obese; Ce nters for Disease Control and Prevention [CDC], 2007). Census tract. A census tract is an administrativ ely defined statistical subdivision of U.S. counties that typi cally contain between 1,500 a nd 8,000 residents (U.S. Census Bureau, 2000). Cross-classified random e ffects models (CCREMs). Cross-classified random effects models refer to an advanced multilevel modeling technique used when hierarchical data are not purely nested; lower-level units (e.g., students) share memberships in a unit of one factor (e.g., a neighborhood) and can belong to different units of a second factor (e.g., diffe rent schools; Raudenbush & Bryk, 2002). Federal poverty level (FPL). Based on the Office of Management and Budget's Statistical Policy Directive 14, FPL is a set of money income thresholds that vary by family size and composition to determine who is living in poverty (U.S. Census Bureau, 2007). Hierarchical linear modeling (HLM). Also commonly referred to as multilevel modeling, HLM is an analytic te chnique that is useful to ex amine data that are nested

PAGE 28

17 within one another, such as individuals w ithin neighborhoods or students within schools. HLM controls for the non-indepe ndence of observations that occu rs due to this nesting as individuals who belong to a group (i.e., nei ghborhood) are likely to be similar to one another resulting in correlated data. Furthermor e, HLM allows for the examination of the variability within and between individuals and groups as well as thei r interactions (DiezRoux, 2003; Hox, 2002; Subramanian, Jones, & Duncan, 2003). Influence. According to The Merriam-Webster Dictionary (Mish et al., 2004), influence is defined as “the power or capacity of causing an effect in indirect or intangible ways” (p. 372). Intraclass correlation coefficient (ICC). The intraclass correlation coefficient represents “the proportion of variance in a de pendent variable that is between groups (i.e., Level-2 units)” (Rauden bush & Bryk, 2002, p. 36). Methodological variables For this study, methodological variables refer to variables required to analyze complex sa mple data correctly—sample weights, neighborhood identification number, a nd school identification number. Moderator. A moderator is a type of variable that affects the relationship between an independent and dependent variable; comm only referred to as an ‘interaction effect’ (Barron & Kenny, 1986; Frazi er et al., 2004). Neighborhood. A neighborhood refers to a geographi cal area where people reside, usually having distinguishing characteristic s (Mish et al., 2004). In this study, these geographical areas corresponded to 1990 census tracts.

PAGE 29

18 Neighborhood affluence. Neighborhood affluence is a measure often used to characterize the quality of a neighborhood; commonly operationalized as a composite measure of neighborhood-level income, percentage of people in a neighborhood with professional positions, and the percentage of neighborhood residents with a college education (Leventhal & Brooks-Gunn, 2003) For this study, the standardized neighborhood affluence composite variable was created from th ree variables: the proportion of families with income equal to or greater than $50,000, proportion of employed persons aged 16 and over in manage rial and professional occupations, and the proportion of residents age 25 and older with at least a college degree. Neighborhood poverty. Neighborhood poverty is a measure often used to characterize the quality of a neighborhood; commonly operationalized as a composite measure of the percentage of people in a neighborhood who are poor, the percentage of female-headed households in a neighborhood, the percentage of neighborhood residents who receive public assistance, and percen tage of residents who are unemployed (Leventhal & Brooks-Gunn, 2003). For this study, the standardized neighborhood poverty composite measure was created from three variables: the proportion of families living below the poverty line, proportion of female-headed households, and the proportion of unemployed adult residents. Risk of obesity. For this study, risk of obes ity was operationalized through standardized age-and-gender-adjusted BMI sc ores, calculated using the National Center for Health Statistics weight by age by gender tables (CDC, 2000a, 2000b).

PAGE 30

19 School. According to The Merriam-Webster Dictionary (Mish et al., 2004), a school is “an institution for teaching and learning” (p. 646). For the current study, school was limited to traditional (i.e., no magnet or alternative schools) U. S. public middle and high schools that taught Grades 7 -1 2 during the 1994-1995 academic year. Socioeconomic status (SES). Socioeconomic status is a prestige-based measure referring to a person’s position within a hierar chical social structur e typically linked to occupation, education level, and income (Kri eger, 2001). For this study, the standardized individual SES composite measure was create d from three variables: parental education, parental occupation, and family income. Organization of Remaining Chapters The remaining chapters present pertinen t information to the study. Chapter Two offers an overview of Bronfenbrenner’s ( 1979) Ecological Systems Theory followed by a review of the literature regarding neighborhood and school influences on adolescent academic achievement and risk of obesity. Chapter Three provides a discussion of the research method, including a description of th e data sources, study sample, measures, and data analysis. Chapter Four describes the resu lts yielded from the da ta analyses. Finally, Chapter Five offers a discussion of the results of the research, includ ing limitations of the study, implications for the field, and directions for future research.

PAGE 31

20 Chapter Two Literature Review Introduction This chapter provides an overview of the theoretical framework, Ecological Systems Theory (Bronfenbrenner, 1979), that guided the study, followed by a synthesis of research that has addressed neigh borhood and school influences on adolescent academic achievement and risk of obesity. The chapter concludes with a summary of significant neighborhood and school at tributes that have been id entified in the literature and a discussion on how the current study builds upon the existing knowledge base. A brief discussion on the methodological advanc es of the current study in relation to previous social determinants research also is provided at the end of this chapter. When possible, information presented in this chapter is limited to studies that focused on neighborhood and school influences on adolescent academic achievement and risk of obesity. This decision was made base d on the different developmental trajectories of adolescents versus younger children. Fo r example, compared to younger children, adolescents spend more time away from home interacting with people in the physical and social spaces and places out side their homes (Boardman & Saint Onge, 2005; HalpernFelsher et al., 1997). Not only does this time spent outside the home provide more opportunities for exposure to nonfamilial infl uences including positive and negative adult

PAGE 32

21 role models (Halpern-Felsher et al., 1997), but adolescents tend to identify with and view themselves in terms of their daily activiti es, often drawing cues from their surrounding contexts (Boardman & Saint Onge, 2005). Because adolescence is a time of identity formation (e.g., Erikson, 1963), it is likely that adolescents link their identities to the “normative” environment of their neighborhoods (Connell & Halpern-Felsher, 1997). For example, a key psychological change that occurs during adolescence is the need to “make meaning” of personal experiences, and most adolescents accomplish this through interactions with adults and peers outside the family (Connell & HalpernFelsher, 1997). Through these interactions and observations of others’ behaviors, adoles cents form beliefs about themselves, their abilities, acceptable behaviors, and thei r futures (Connell & Halpern-Felsher, 1997). However, this process is not the same for all youth. For example, the nature and availability of role models and the physical conditions of neighborhoods and schools of youth living in impoverished areas are likely different than for youth living in more affluent areas, thus the “normative” envir onments that serve as reference points for adolescent identify formation also va ry (Connell & Halpern-Felsher, 1997). The research reviewed in this chapter also has been restricted to U.S.-based studies. Given the large amount of variati on from country to country in terms of population heterogeneity and economic, social, and political contexts, findings from countries outside the U.S. are not generali zable to the population of interest for the current study. Therefore, in an effort to pr esent concisely the most relevant research

PAGE 33

22 related to neighborhood and school influences on academic achievement and risk of obesity among U.S. adolescents, I chose to li mit this chapter to U.S.-based studies. Theoretical Framework Bronfenbrenner’s (1979) Ecological Syst ems Theory emphasizes the idea that youth do not live in isolation. Instead, they de velop in a variety of contexts, each of which interacts with their development. Acco rding to Bronfenbrenner (1979), individuals exist among four interrelated systems—the mi crosystem, the mesosystem, the exosystem, and the macrosystem. The microsystem, which consists of the proximal environments in which an individual is active (e.g., family, school, peer group, and neighborhood), has the most immediate and earliest influence on a pe rson, whereas the mesosystem, which is a system of microsystems, or connections am ong the different environments in which a person is active, has the second strong est influence on individual development (Bronfenbrenner, 1979). The next two levels the exosystem and the macrosystem, are farther removed and have more indir ect influences on human development (Bronfenbrenner, 1979). The exosystem contains settings in which an individual is not an active participant, but can still be affected by events that occur at this level (e.g., a parent’s place of employment) whereas the macrosystem represents the larger cultural context in which a child lives (e.g., cultural norms, policies, politics; Bronfenbrenner, 1979). According to Bronfenbrenner (1979), to study human development effectively, we need to look beyond a single environm ent and look at the interactions among individuals and multiple environments. In th e past, although the majority of researchers

PAGE 34

23 who have applied an ecological systems framework have focu sed their investigations at the mesosystem level, for the most part, they have primarily addressed the nature of a single environmental interaction (e.g., fa mily influence on development or school influence on development). Some have focused on the influences of multiple environments at the same time (i.e., simulta neous neighborhood and school influences on development), but few appear to focus on the interrelations of two different microsystems within the mesosystem (e.g., the interaction between fam ily and school contexts in relation to development). These less-investig ated interactions be tween two different microsystems were the focus of this study; in stead of examining the influence of a single environment on adolescent academic achievement and risk of obesity, the current study examined the nature of the interc onnectedness between two microsystems-neighborhood and school influences on adolescent academic achievement and risk of obesity as well as the interaction e ffect of these two microsystems. Neighborhood Influences on Adolescent Academic Achievement The investigation of neighborhood in fluences on adolescent academic achievement is not new. In fact, Br ooks-Gunn, Duncan, and Aber (1997a, 1997b) published their two-volume collection on ne ighborhood poverty and child development a decade ago, in which they proposed six important neighborhood characteristics potentially related to child and adolescen t outcomes: income, human capital, ethnic integration, social capital, social disorgan ization, and safety, w ith neighborhood income being the most important neighborhood characte ristic related to educational outcomes. Of these six important neighborhood characterist ics proposed by Brooks-Gunn et al. (1997a,

PAGE 35

24 1997b), neighborhood effects on adolescent academ ic achievement research has most often focused on income (i.e., neighborhood SES) human capital (i.e., male joblessness), and social disorganization. Other neighborhoodlevel variables that researchers have examined include neighborhood racial and ethnic diversity and perceived neighborhood quality, cohesion, and resources. The followi ng sections contain an overview of how these neighborhood-level variables relate to various measures of adolescent academic achievement. More details about each of the studies summarized in this section are provided in Appendix A, Table A1 (e.g., type of statistical anal ysis conducted, list of all variables included in the models). Neighborhood SES. Across studies, neighborhood affluence, and not neighborhood poverty, appears to be the most c onsistent characteristic associated with adolescent academic achievement (Boyl e et al., 2007; Brooks -Gunn et al., 1997a; Leventhal & Brooks-Gunn, 2000) Common indicators used to operationalize highSES/affluent neighborhoods in clude neighborhood-level inco me, percentage of people with professional positions, and percentage of residents with a college education (Leventhal & Brooks-Gunn, 2003). Low-ne ighborhood SES/poverty is typically operationalized through the percentage of poor residents, percentage of female-headed households, percentage of residents who receiv e public assistance, a nd the percentage of unemployed residents (Leventhal & Brooks-Gunn, 2003). In Atlanta, Halpern-Felsher et al. (1997) found high-neighborhood SES to be positively associated with Iowa Test of Basi c Skills scores among African American girls aged 11 to 16. Similarly, using two different samples (12 to 15 year olds and 15 to 20

PAGE 36

25 year olds) from an urban, upstate New York sc hool district, Halpern-Fe lsher et al. (1997) found that White boys’ educational risk, in cluding achievement, was reduced with a higher concentration of middl e-class neighbors. Dornbusch et al. (1991) also found a positive association between neighborhood afflue nce and adjusted self-reported grades in a study of San Francisco high school students. Conversely, using data from a sample of youth aged 10 to 16 in New York City, Ba ltimore, and Washington, D.C., HalpernFelsher et al. (1997) found a negative rela tionship between standardized reading and mathematics test scores and neighborhood poverty among White girls. Within the Gautreaux (Rosenbaum, 1995) and Moving to Opportunity (MTO; Kling & Liebman, 2004; Levent hal, Fauth, & Brooks-Gunn, 2005) programs, researchers also have focused on the relationship be tween neighborhood socioeconomic status and adolescent academic achievement. Inte restingly, unlike the findings from nonexperimental studies, results from th ese quasi-experimental (Gautreaux) and experimental (MTO) programs do not reveal statistically significant improvements in adolescent academic achievement based on neighborhood affluence (Kling & Liebman, 2004; Leventhal et al., 2005; Rosenbaum, 1995) More specificall y, in the Gautreaux program, Rosenbaum (1995) found no differences in grade point average (GPA) between high school youth who moved to the suburbs an d those who stayed within Chicago city limits. Similarly, using MTO data from all fi ve participating cities (Baltimore, Boston, Chicago, Los Angles, and New York City), Kling and Liebman (2004) reported no differences in high school Woodcock-Johnson reading and mathematics test scores between adolescents, aged 15-20, who move d to low-poverty neighborhoods and their

PAGE 37

26 peers who remained in impoverished urban housing projects. Conversely, Leventhal et al.’s (2005) 5-year followup study of New York City MTO youth suggests that control group youth, aged 14-19, who remained in trad itional housing projects had statistically significantly higher GPAs than did their similarly aged p eers who moved to low-poverty neighborhoods and those who were allowed to move out of the projects and reside in unrestricted Section 8 housing. When thinking about the conflicting findi ngs between non-experimental studies and quasi-experimental and experimental studi es, several factors should be considered. Foremost, is the issue of model and variable specification—not only were the statistical models used in the studies different, but the research was conducted during different periods. Similarly, in terms of the variab les examined in each study, not only was academic achievement operationalized different ly across the studies, but when GPA was used as the criterion variable, it is importa nt to remember that this measure is often considered unstable as it can vary from school to school. Furthermore, within the Gautreaux and MTO programs, the opera tionalization of neighborhood was weak. Poverty was the only variable examined to determine where participants could move—no other social contexts of th e neighborhoods were considere d. In addition, by moving Black families to White suburbs, theoretically this could have diminished adolescents’ social support, which, in turn, could impact thei r well-being, including achievement. Lastly, given the aforementioned differences and weaknesses in the various neighborhood SES and academic achievement studies, more res earch, in particular, more theory-based research, is needed.

PAGE 38

27 Neighborhood male joblessness. Albeit used less often than neighborhood SES, researchers also have used male joblessne ss as a measure of neighborhood quality in the investigation of neighborhood influences on adolescent academic achievement. For example, among 11to 16-year-old African Am erican boys in Atlanta, male joblessness was negatively associated with Iowa Test of Basic Skills scores (Halpern-Felsher et al., 1997). Male joblessness also was negatively a ssociated with educa tional risk, including achievement, among 12to 15year-old African American boys and White females in an urban, upstate New York school district (H alpern-Felsher et al ., 1997). The negative relationship between male joblessness and New York students’ educational risk also was observed among White 15to 20-year-old females in the same upstate, urban school district; however, the relationship for Afri can American boys was not statistically significant among the older sample of stude nts (Halpern-Felsher et al., 1997). Neighborhood social disorganization Originally developed to explain crime, Social Disorganization Theory (i.e., lo w-neighborhood SES, ethnic heterogeneity, and high residential mobility; Shaw & McKay, 1942) also has been used in the investigation of community influences on adolescent acad emic achievement. First, among eighth-grade students in Virginia public schools, commun ity social disorganiz ation was shown to explain a statistically signifi cant amount of variance in Stanford 9 performance (Baker et al., 2001). Second, using a nationally represen tative sample of middle and high school youth and focusing on process variables linked to Social Disorganization Theory (i.e., lack of neighborhood support, perceptions of pr o-social behaviors, and perceptions of

PAGE 39

28 neighborhood crime and violence), Bowen, Bo wen, and Ware (2002) reported a direct negative effect of neighborhood social diso rganization and self -reported grades. Perceived neighborhood quality. A variety of perceived neighborhood quality measures also have been shown to be a ssociated with adolescent achievement. For example, Eamon (2005) found a positive relations hip between mothers’ ratings of overall neighborhood quality and Peabody Individua l Achievement Test (PIAT) reading comprehension scores for young Latino adoles cents aged 10 to 14. However, the same relationship was not observed for PIAT math ematics scores. Similarly, urban, African American adolescent girls aged 11 to 14 years in a southeastern city who perceived their neighborhoods as being non-cohesive reported lower grades than did their peers who reported high levels of neighborhood cohe sion (Plybon, Edwards, Butler, Belgrave, & Allison, 2003). Using a national probability sample of middle and high school students from the National School Success Profile (SSP) data Bowen and Bowen (1999) also found a statistically significant rela tionship between adolescents’ perceptions of neighborhood quality and school grades. More specifica lly, among middle and high school students, both perceived neighborhood peer culture a nd adolescents’ persona l experience with neighborhood crime and violence were negativel y related to self-repo rted school grades (Bowen & Bowen, 1999). The associati ons between perceived neighborhood deterioration and resourcefulness and GPA also have been examined (Williams, Davis, Miller Cribbs, Saunders, & Williams, 2002). Among urban, African American ninth graders living in a large metropolitan ar ea in the Midwest, perceived neighborhood

PAGE 40

29 deterioration was inversely correlated with youth’s official GPA; however, the relationship between GPA and perceive d neighborhood resourcefulness was not statistically significant (Williams et al., 2002). Other neighborhood measures Neighborhood ethnic and ra cial diversity and socioeconomic resource inequality also have been examined in relation to adolescent academic achievement. For example, using data from the High School Effectiveness Study, Blau, Lamb, Stearns, and Pellerin (2001) investigated the relationship between cosmopolitan communities, characterized by lo w levels of socioeconomic resource inequality and high levels of ethnic and raci al diversity, and two-y ear gain scores in social studies. Neighborhood socioeconomic resource inequality was negatively associated with gains in social studies achievement; neighborhood diversity was not statistically significan tly related to social studies achievement (Blau et al., 2001). Lastly, in an effort to understand better the impact of re sidential context on various elements of adolescent well-being (e .g., risk behaviors, educational outcomes, physical and mental health, a nd social integration), Boar dman and Saint Onge (2005) used Add Health data to calculate adjusted intra-class correlation coefficients (ICC) for 34 adolescent outcomes. Two achievement outcomes included in the study were selfreported GPA and performance on the Add H ealth Picture Vocabulary Test (AHPVT). Based on ICC values, of all 34 outcomes, ne ighborhoods appeared to have the strongest impact on AHPVT performance (ICC = .25); the ICC for self-reported GPA was .10 (Boardman & Saint Onge, 2005).

PAGE 41

30 Neighborhood Influences on Adoles cent Risk of Obesity Unlike neighborhood influences on adol escent academic achievement, the investigation of neighborhood influences on adol escent risk of obesity is a more recent area of inquiry. Not only is there a paucity of publ ished articles in this area, but all of the articles that have examined neighborhood influences on adolescent obesity were published between 2004 and 2007. Even though there is scant published research, to date, common neighborhood factors that have been exam ined in relation to adolescent risk of obesity include neighborhood SES, the built envi ronment, availability of food outlets, and urban sprawl. The following sections cont ain an overview of how these neighborhoodlevel variables relate to adolescent risk of obe sity. More details about each of the studies summarized in this section are provided in Appendix A, Table A-2 (e.g., type of statistical analysis conduct ed, list of all variables included in the models). Neighborhood SES. When studying neighborhood SES and its relationship with adolescent weight status, researchers have used traditional indicators of SES (e.g., education, income, and occupation information) as well as new indi cators (e.g., clustered characteristics of neighborhoods). For example, by applying cluster analysis to measures of neighborhood environments asso ciated with the home street addresses for Wave I Add Health participants, Nelson et al. (2006) id entified six robust nei ghborhood patterns: rural working class, exurban, new suburban devel opment, older suburban development, mixedrace/ethnicity urban, and low-SES inner city. In relation to adolescen t weight, adolescents living in rural working class, exurban, and mixed-race urban neighborhoods were 30% to

PAGE 42

31 40% more likely to be overweight than were their peers livin g in newer suburban developments (Nelson et al., 2006). Next, in terms of traditional indicators of neighborhood SES, Chen and Paterson (2006) reported neighborhood education and neighborhood employment as predictors of St. Louis high school students’ BMI, beyond the effects of family education and family occupation status. However, neighborhood in come and neighborhood assets were not statistically significant predicto rs beyond the effects of family income and family assets (Chen & Paterson, 2006). Similarly, Kling a nd Liebman (2004) did not report any statistically significant differences in adoles cent obesity status be tween MTO adolescents whose families moved to low-poverty neighborhoods and their peers who remained in impoverished urban housing projects. Also interested in the relationship between neighborhood SES and adolescent weight status, Wickrama et al. (2006) used Add Health data to investig ate if the impact of community poverty on adolescent obesity was moderated by adolescent race/ethnicity. Interestingly, community poverty had less of an impact on obesity status among racial and ethnic minorities (Asian, Hispanic, and African American) compared to White adolescents (Wickrama et al., 2006). In other words, being a racial or ethnic minority appeared to buffer the effect of co mmunity poverty on adolescent obesity. Built environment. In addition to examining neighborhood sociodemographic influences on adolescent weight, two recent studies investigated the relationship between neighborhood recreational facilities and adoles cent risk of being overweight or obese. For example, based on a sample of 11 to 15 years olds in San Diego County, Norman, Nutter,

PAGE 43

32 Ryan, Sallis, Calfas, and Patrick (2006) repor ted no statistically significant relationship between the number of recrea tion facilities located within a one-mile radius of an adolescent’s residence and BMI. On the ot her hand, using nationally representative Add Health data, Gordon-Larsen et al. (2006) f ound that an adolescent’s relative odds of being overweight decreased as the number of recr eational facilities per census-block group increased. For example, compared to living in a census block-group with no recreational facilities, residing in a cens us block-group with at least one recreational facility was associated with a 5% decrease in the relativ e odds of being overwei ght (Gordon-Larsen et al., 2006). Furthermore, adolescents living in a census-block with seven recreational facilities were 32% less likely to be overweight compared to their peers residing in census block-groups with no such facil ities (Gordon-Larsen et al., 2006). Other neighborhood measures. Residential context, urban sprawl, availability of food outlets, and collective efficacy also have been examined as neighborhood correlates of adolescent risk of obesity. For exam ple, in addition to adolescent academic achievement, Boardman and Saint Onge (2005) also examined the relationship between residential context and adolescent risk of being overweight. However, unlike the relatively important relationship between neighborhoods and a dolescent verbal achievement (ICC = .25), area of residence appe ared to have a much smaller association with being overweight (ICC = .05; Boardman & Saint Onge, 2005). In terms of urban sprawl’s relationship with adolescent risk of obesity, findings are mixed. For example, based on crosssectional analysis of the 1997 National Longitudinal Survey of Youth (NLSY) data, ur ban sprawl appeared to be correlated with

PAGE 44

33 being overweight/risk of being overweight among U.S. adolescents (Ewing, Brownson, & Berrigan, 2006). However, when examined longitudinally, five years later, the relationship between urban sprawl and we ight status was no longer statistically significant (Ewing et al., 2006). Regarding availability of food outlets a nd adolescent risk of obesity, availability chain supermarkets and convenience stores ha ve both been found to have statistically significant associations with a dolescent BMI. More specific ally, using MTF data, Powell, Auld, Chaloupka, O’Malley, and Johnston ( 2007) found a statisti cally significant negative association between neighborhood av ailability of chain supermarkets and adolescent BMI and a statistically significant positive relationship between the number of neighborhood convenience stores and adoles cent BMI. Furthermore, the negative association between supermarket availability and adolescent BMI was larger for AfricanAmerican youth compared to White or Hispanic youth (Powell et al., 2007). Lastly, neighborhood collective efficacy (i.e ., a measure of social cohesion and informal social control; Cohen et al., 2006) also has been suggeste d as a statistically significant predictor of adol escent weight. Adolescents ag ed 12 to 17 residing in Los Angeles County neighborhoods with high levels of collective efficacy were predicted to have BMI values one unit below their peers who lived in neighborhoods with low levels of collective efficacy (Cohen et al., 2006). In terms of being overwei ght, adolescents who lived in neighborhoods with low efficacy we re 52% more likely to be overweight compared to their peers who lived in neighborhoods with av erage levels of collective efficacy (Cohen et al., 2006).

PAGE 45

34 School Influences on Adolescen t Academic Achievement Just as the investigation of nei ghborhood influences on adolescent academic achievement is not new, nor is the inves tigation of school infl uences on adolescent academic achievement. For example, a lthough criticized for its many methodological limitations, the well-known Equality of Educational Opportunity report (also commonly referred to as the Coleman Report; Coleman et al., 1966) was the first comprehensive, nationwide investigation into school influen ces on academic achievement (Dyer, 1972). However, based on the results of their ex amination of student body, school, and teacher influences on verbal achievement, Coleman et al. (1966) concluded: That schools bring little in fluence to bear on a child’s achievement that is independent of his [her] b ackground and general social c ontext; and that this very lack of an independent e ffect means that the inequalities imposed on children by their home, neighborhood, and peer envir onment are carried along to become the inequalities with which they confront adult life at the end of school. (p. 325) Despite the less-than-promising results presented in the Coleman Report (Coleman et al., 1966), social and behavioral scientists continued inve stigating the relationship between school-level charac teristics and academic achievement. More specifically, school characteristics commonl y examined in relation to adolescent academic achievement include school sociodem ographic characteristics, school resources and sector, teacher characteristics, per ceived social climate and school quality, and organizational climate. The following sections contain an overview of how these schoollevel variables relate to various measures of adolescent academic achievement. More

PAGE 46

35 details about each of the studies summarized in this section are provided in Appendix A, Table A-3 (e.g., type of statisti cal analysis conducted, list of all variables included in the models). School s ociodemographic characteristics. In recent years, several researchers have published findings that appear to contra dict Coleman et al.’ s (1966) findings that schools had little influence on academic ach ievement beyond what youth brought with them to school. For example, among U.S. hi gh school students who graduated from high school in 1995 and had taken the Scholastic Ap titude Test (SAT) dur ing their junior or senior year of high school, school size, sc hool poverty, and school racial and ethnic composition were meaningful predictors of self-reported high school GPA (Everson & Millsap, 2004). Both school size and sc hool racial and ethnic composition were negatively correlated with high school GP A, whereas, surprisingly, school poverty exhibited a positive association with high school GPA (Everson & Millsap, 2004). Similar findings were also found among Bl ack and White public school 10th-grade students in Louisiana (Caldas & Bankston, III, 1997). School-level racial minority composition was negatively associated with standardized test performance whereas poverty and social class status of adolescents’ schoolmates wa s positively associated with 10th-grade achievement. Data from the base year of the Na tional Education Long itudinal Study of 1988 (NELS:88) also suggest that th e percentage of minority student s in a school is inversely related to middle school students’ read ing achievement (Lee & Croninger, 1994). However, school locale, school SES, school sector, grade grouping, and grade size were

PAGE 47

36 not statistically sign ificant school-level predictors of reading achievement among U.S. middle school students (Lee & Croninger, 1994). Crosnoe (2004) also found a statistically significant, yet surprising, relationship betw een school sociodemographics and adolescent academic achievement. Am ong middle and high school students included in Wave I and II Add Health data, school-lev el parental educati on revealed a negative association with self -reported grades in school (Crosnoe, 2004). Next, in their investigation of cosm opolitan environments and academic achievement, Blau et al. (2001) also ex amined the relationship between schools’ sociodemographic environments and two-year gains in social studi es achievement among high school students who participated in th e High School Effectiveness Study. However, results from their study did not suggest that a school’s sociodemographic environment was an important predictor of gains in soci al studies achievement (Blau et al., 2001). Lastly, in addition to their study of comm unity social disorganization and academic achievement of eighth-grade students in Virginia Baker et al. (2001) al so investigated the relationship between school so cial disorganization and Stan ford 9 scores among the same set of students. Results revealed an invers e association between sc hool-level organization and eighth-grade students’ Stanford 9 performance (Baker et al., 2001). School resources and sector. In their meta-analysis of the effect of school resources on student achievement, Greenwal d, Hedges, and Laine ( 1996) concluded that school resources, such as per-pupil expenditure (PPE), teacher salary, teacher/pupil ratio, and school size, appeared to be important fact ors related to students’ standardized test achievement. More specifically, based on findi ngs from 14 studies, th e half-standardized

PAGE 48

37 regression coefficient for PPE’s relationshi p with achievement was .0003, with the units measured as dollars. Based on five studies, th e half-standardized regression coefficient for teacher salary’s association with achie vement was .0263, with units as thousands of dollars. Also, using data from 21 and 15 studi es, respectively, the standardized regression coefficients for teacher/pupil ratio an d school size were .0295 and .0299 with > 0 indicating greater achievement in smaller cla sses and smaller schools (Greenwald et al., 1996). To understand better the magnitude of thes e effect sizes, Greenwald et al. (1996) also presented the information in terms of the effect of $500 per student on achievement. In this circumstance, the effect size for PPE increased to 0.15, teacher salary increased to 0.16, and teacher/pupil ratio incr eased to 0.04 (Greenwald et al., 1996). However, when interpreting these results, it is important to note that it is not possible to tell if the studies included in the meta-analysis focused on child and/or adolescent achievement; therefore, these findings cannot be interpre ted solely in terms of adol escent academic achievement. Attending religious schools also has been suggested as a positive correlate of Black and Hispanic adolescent academic achievement. For example, in their metaanalysis of studies that examined the imp act of school sector on Black and Hispanic adolescent academic achievement, Jeynes (2002) found that middl e school students who attended religious schools perf ormed, on average, 0.25 standa rd deviations higher, for both GPA and achievement tests, than did their peers who did not attend religious schools. The same level of improvement (Hedges's g = 0.26) also was observed among high school students’ GPA and achie vement tests (Jeynes, 2002).

PAGE 49

38 Teacher characteristics. In their meta-analysis, Gr eenwald et al. (1996) also found teacher ability, teacher education, and te acher experience to be important variables related to student achievement. For exampl e, results from six studies produced a standardized regression coefficient of .0724 for teacher ability. However, based on 15 and 12 studies, respectively, the effects of teacher experience ( = .0482) and teacher education ( = .0003) were less than the effect of teacher ability (Greenwald et al., 1996). In terms of the effect of $500 per st udent on achievement, the effect sizes for teacher experience and education become 0.18 and 0.22, respectively (Greenwald et al., 1996). Next, to examine the relationship betw een teacher qualifications and student achievement at a national level, Darling-Hammo nd (1999) used teacher qualification data from the Schools and Staffing Survey and ei ghth-grade achievement data from the 1996 National Assessment of Educa tional Progress (NAEP). Findings from her study revealed both positive and inverse correlations betw een mathematics achievement and teacher qualifications. For example, the percentage of teachers out-of-field a nd the percentage of newly hired uncertified teachers were invers ely correlated with eighth-grade mathematics achievement, whereas the percentage of well-qua lified teachers was positively correlated with mathematics achievement (Darling-Hammond, 1999). However, when data were aggregated and examined at the state-level, the only statistically significant teacher quality predictor of eighth-grade mathema tics achievement was the percentage of wellqualified teachers (D arling-Hammond, 1999).

PAGE 50

39 Teacher practices and teacher empowerment also have been investigated as possible correlates of adolescent achievement. For example, in her examination of teacher practices and year-end grades among suburban sixth graders, Wentzel (2002) found an inverse relationship between negativ e feedback and achievement and a positive relationship between high expectations and sixth-grade achievement. Teacher practices that were not statistically significant pred ictors of sixth-grade achievement included fairness, teacher motivation, and rule se tting. In terms of teacher empowerment, Sweetland and Hoy (2000) reported school-l evel teacher empowerment to be a statistically significant predictor of standa rdized reading and mathematics achievement among eighth graders in New Je rsey public middle schools. Perceived social climate and school quality. In a study focused on the relationship between risk of obesity, self -reported grades in school, and school social climate, Crosnoe and Muller (2004) reported some inte resting findings. First, using Wave I and II Add Health data, Crosnoe and Muller (2004) found no statistically significant relationships between school climate vari ables and middle and high school students’ academic achievement. However, they did repo rt several cross-level interactions between individual risk of obesity and three school climate va riables (rate of athl etic participation, mean student romantic behavior, and mean BMI; Crosnoe & Muller, 2004). That is, the relationship between school climate variab les and adolescent academic achievement varied based on adolescent ri sk of obesity status. For example, adolescents who were at risk of obesity had lower levels of achievement when they attended schools with higher levels of mean student romantic

PAGE 51

40 activity (Crosnoe & Muller, 2004). Conversely, adolescents who were at risk of obesity performed better academically in schools w ith higher average BMI values (Crosnoe & Muller, 2004). However, it is important to no te that this relationship was reported as statistically significant at the .10 level. Adol escents who were at risk of obesity also performed better in schools w ith greater levels of athletic participation (Crosnoe & Muller, 2004). Surprised by this last fi nding, Crosnoe and Muller (2004) undertook further analyses and found that adolescents who were at risk of obesity became more academically involved when they attended sc hools with increased rates of athletic participation. Various measures of school quality also have been suggested as being predictors of adolescent academic achievement. For example, among Latino adolescents, age 10 through 14, perceived school quality has been found to have a positive relationship with reading and mathematics achievement (Eamon, 2005). Also, in addition to examining perceived neighborhood peer cu lture and adolescents’ pe rsonal experience with neighborhood crime and violence, Bowen and Bowen (1999) also explored the relationship between perceived school danger and se lf-reported grades using data from a national probability sample of middle and hi gh school students. Both composite measures of school danger (perceived crime and viol ence, and personal threats) had inverse associations with achieveme nt (Bowen & Bowen, 1999). Factors such as school and student-teacher bonding also have been examined in relation to adolescen t academic achievement. Among Af rican American adolescents, aged 11 to 14, in a large Mi dwestern city, adolescents who reported feeling bonded to

PAGE 52

41 their school were also more likely to re port higher school grades (Zand & Thomson, 2005). In terms of student-teacher bonding, using Add Health data, Crosnoe (2004) reported that the relationship between stude nt-teacher bonding and self-reported grades depended on how close an adolescent felt to hi s or her parents. Adolescents who were not close to their parents benefited less from a ttending schools with high levels of studentteacher bonding compared to their peers who fe lt close to their parents (Crosnoe, 2004). On the other hand, perceived teacher support wa s not shown to be related to self-reported GPA among urban, African American eighth graders (Sanders, 1998). Organizational climate. In addition to examining th e relationship between a school’s social climate and adolescent academ ic achievement, researchers also have investigated how schools’ or ganizational climate (from the teacher or principal’s perspective) relates to adol escent academic achievement. For example, in New Jersey middle schools, two of the six dimensions of organizational climate were associated with youth performance on all three ar eas of New Jersey’s Eighth Grade Early Warning Test (Hoy & Hannum, 1997). More spec ifically, teacher affiliation and institutional integrity were both positively associated with eighth-grade mathematics, reading, and writing achievement. Academic emphasis also was f ound to have a positive association with eighth-grade achievement; however, it was only related to mathematics and reading achievement (Hoy & Hannum, 1997). Henderson, Buehler, Stein, Dalton, Robinson, and Anfara, Jr. (2005) also found a positive co rrelation between academic emphasis and eighth-grade standardized test scores in a sample of 10 Tennessee middle schools.

PAGE 53

42 School social and academic organization also have been suggested as being significant correlates of adolescent academ ic achievement. Using NELS:88 data, Lee, Smith, and Croninger (1997) reported that high school students who attended schools with higher levels of social organization, more mathematics and science course offerings, and higher levels of authentic instructional practices in mathematics and science had larger gains in science and mathematics ach ievement than did their peers who attended schools with low levels of social organizat ion, fewer mathematics and science course offerings, and lower levels of authentic instructional practices. Analysis using NELS:88 data also suggested that teacher cooperat ion and the number of books used in eighthgrade English classes were pos itive correlates of eighth-gr ade reading achievement (Lee & Croninger, 1994). However, when school ac ademic organization within the NELS:88 data was conceptualized in terms of authoritativeness, school environment was not a statistically significant predictor of eighth-gr ade standardized mathematics test scores (Gill, Ashton, & Algina, 2004). School Influences on Adolescen t Risk of Obesity Whereas there has not been much res earch conducted on neighborhood influences on adolescent risk of obesity there has been even less research focused on school influences on adolescent risk of obesity. Fu rthermore, unlike research that has examined school influences on adolescent academic achie vement, the school influence and risk of obesity research has focused less on the soci al and demographic aspects of the school environment and more on the effectiveness of school-based interven tions. In fact, there appears to be only one published study that has investigated th e relationship between

PAGE 54

43 various school characteristics a nd adolescent risk of obesity (O’Malley, Johnston, Delva, Bachman, & Schulenberg, 2007). Below is an ov erview of the limited literature on school influences on adolescent risk of obesity. De tails about each of the studies summarized in this section are provided in Appendix A, Table A-4 (e.g., type of statistical analysis conducted, list of all variable s included in the models). Regarding school social and demographi c attributes and adolescent risk of obesity, using MTF data, O’Malley et al. ( 2007) reported a statistically significant positive association between school SES and adolescent BMI. However, other school variables included in the analys is (school type, school size, and student body racial/ethnic composition) exhibited statistically non-signi ficant relationships with adolescent BMI (O’Malley et al., 2007). O’Malle y et al. (2007) also found that most of the variation in adolescent BMI was within, not between schools (ICC = .03). Next, in terms of adolescent risk of obe sity and school-based interventions, all three school-based interventions that have focused on adolescent obesity prevention targeted different elements within school e nvironments. For example, in an effort to reduce obesity among the gene ral population of Boston ar ea middle school students, Planet Health worked with teachers to deve lop sessions that could be easily incorporated into existing curricula (Gor tmaker et al., 1999). More sp ecifically, the intervention curricula aimed to decrease the amount of time youth spent watching television, increase the amount of time youth spent engaging in moderate and vigorous physical activity, decrease consumption of high-fat foods, and increase daily fruit and vegetable consumption (Gortmaker et al., 1999). Anot her key component of the Planet Health

PAGE 55

44 curriculum was that the intervention materials were incorporated into multiple academic subject areas (i.e., language arts, math, scienc e, social studies) as well as PE classes (Gortmaker et al., 1999). The New Moves intervention also was an education-focused program; however, unlike Planet Health, New Moves provided physical activity and nutrition education through girls-only alternative physical education classes in three Twin City area high schools (Neumark-Sztainer, Stor y, Hannan, Stat, & Rex, 20 03). New Moves also differed from Planet Health in terms of its targ et population. Instead of focusing on obesity prevention among the general student populati on, New Moves was developed specifically for high school girls who were overweight or at risk of being overweight (NeumarkSztainer et al., 2003). The speci fic aims of the New Moves in tervention were to increase physical activity and improve eating behavior s as well as help girls avoid unhealthy dieting behaviors and feel bett er about themselves in a th in-oriented society (NeumarkSztainer et al., 2003). The third school-based adolescent ob esity prevention trial, Middle-School Physical Activity and Nutrition study (M-SP AN; Sallis et al., 2003), was different from both Planet Health and New M oves in that it did not contai n any classroom education. Instead, it included broad policy and social ma rketing interventions aimed at increasing middle school students’ physical activity both in physical education classes and throughout the day, as well as marketing and providing low-fat foods at all food sources within the schools (Sallis et al., 2003). As a secondary outcome of interest, M-SPAN also aimed to reduce BMI among students in the intervention schools (Sallis et al., 2003).

PAGE 56

45 Example components from the policy and so cial marketing interventions included providing funds for new PE equipment and addi ng signs to promote low-fat food options (Sallis et al., 2003). Each of the three school-based interven tions also reported di fferent levels of program effectiveness. At the conclusion of the 2-year interven tion, Planet Health researchers reported a statistically signifi cant decrease in obesity for girls in the intervention schools compared to girls in th e control schools; howev er, the decrease in obesity prevalence among boys in the inte rvention schools was not statistically significantly different than the post-intervention obesity prevalence among boys in the control schools (Gortmaker et al., 1999). Results from th e New Moves post-intervention (16 weeks from baseline) and 8-month fo llow-up evaluations did not reveal any statistically significant differe nces in BMI between girls in the intervention schools and girls in the control schools (Neumark-Sztainer et al., 2003). Moreover, as with Planet Health, M-SPAN’s effectiveness in reduc ing BMI appeared to vary by gender. Specifically, this program appeared to be more effective for boys than it was for girls. At the end of the 2-year intervention, boys in intervention schools had greater BMI reductions compared to boys in the control schools, but ther e was no effect on girls’ BMI (Sallis et al., 2003). The last study with published findings re lated to schools and adolescent risk of obesity is from the Trial of Activity in Adolescent Girls (TAAG; Scott et al., 2007). However, unlike Planet Health, New Move s, and M-SPAN, TAAG was not a randomized trial designed to test the effectiveness of a specific school-based in tervention. Instead, it

PAGE 57

46 was a coordinated school-and community-based project affiliated with six U.S. universities (Universit ies of Arizona, Maryland, Minne sota, and South Carolina; San Diego State; and Tulane University) with a primary goal of reducing the normal decline in physical activity in middle school girls (Scott et al., 2007 ). As part of assessing the “healthiness” of participants’ neighb orhoods, TAAG researchers examined the relationship between weekend acce ssibility of school recreational facilities and obesity and found a statistically signi ficant association between th e number of locked schools within a half-mile of a sixt h-grade girl’s home and BMI; each additional locked school was associated with a predicted 3% increase in BMI (Scott et al., 2007). Lastly, although there is curre ntly limited evidence of the role schools play in the prevention of adolescent obesity, several papers have been published that postulate arenas within the school environment that likely influence adolescen t risk of obesity (Carter, 2002; Dietz & Gortmaker, 2001; Story, Ka phingst, & French, 2006). In addition to increasing physical activity opportunities and improving the healthfulness of food both served and sold in schools, schools should also provide health education and other programs aimed to increase both student and parent knowledge and attitudes toward nutrition and weight control (C arter, 2002; Dietz & Gortmake r, 2001; Story et al., 2006). Story et al. (2006) also discuss the important role that school health services can play in addressing adolescent risk of obesity. Endor sed by the Institute of Medicine, BMI reporting through health report car ds also has been suggested as a way schools can help prevent adolescent obesit y (Story et al., 2006).

PAGE 58

47Summary Neighborhood SES has been commonly us ed in the investigation of both neighborhood influences on adolescent ach ievement and neighborhood influences on adolescent risk of obesity. However, nei ghborhood SES is often measured differently across these two outcomes. For example, when investigating the relationship between neighborhood SES and adolescent achievement studies have often included composite measures of neighborhood affluence and/or neighborhood poverty, whereas the majority of research focused on neighborhoods and adolescent risk of obesity has relied on individual indicators of ne ighborhood SES (i.e., neighborhood education or neighborhood employment). By using composite m easures of neighborhood affluence and neighborhood poverty, the cu rrent study provides a ne w perspective into the neighborhood and adolescent risk of obesity literature. In addition to neighborhood affluence a nd poverty, male joblessness, social disorganization, and perceived neigh borhood quality are other commonly documented neighborhood correlates of adolescent academic achievement. However, to date, these same neighborhood characteristics have not b een included in the investigation of neighborhood influences on adolescent risk of obesity. Besides neighborhood SES, availability of recreational f acilities is the only other nei ghborhood-level vari able that has been examined in relation to adolescent risk of obesity. Unlike the neighborhood and academic achievement research, school and academic achievement research has tended to use single variables more often than composite variables when measuring SES (e.g., school-level pove rty or school-level

PAGE 59

48 parental education). Thus, use of a composite measure of school SES in the current study makes an important contribution to the school and academic achievement literature. In addition to SES, other common school-level variables that have been examined in relation to adolescent academic achievement include school-level racial composition, teacher quality, perceived social climate, and school resources. In terms of school characteristics and adolescent risk of obesity the current study adds to the paucity of literature in this area by including a composite measure of weight promotion education as a potential predictor of adolescent risk of obesity. Lastly, 68% of the neighborhood and school influence research reviewed in this chapter did not use hierarchical linear m odeling techniques even though the data were hierarchical in nature. Thus, fi ndings from studies that utiliz ed nested data but that did not account for the nesting of the data in thei r analytic techniques n eed to be interpreted with caution. Also, even though some studies included variables from multiple social environments (e.g., neighborhood and school vari ables or family and school variables), the lack of appropriate HLM techniques in these studies prevents us from understanding each environment’s unique influence on achievement. Furthermore, except for Crosnoe (2004), none of the research that included measures of two social environments investigated interactions be tween the environments. By utilizing advanced multilevel modeling techniques (i.e., CCREMs), the curr ent study makes an important contribution to both the academic achievement and risk of obesity literature not only by providing information on each environment’s unique influence on both outcomes, but also by

PAGE 60

49 offering insight into the interconnecte dness between neighborhoods and schools and adolescent academic achievement and risk of obesity.

PAGE 61

50 Chapter Three Method Purpose of the Study Grounded in Bronfenbrenner’s (1979) Ec ological Systems Theory and through the application of advanced multilevel modeling techniques (Raudenbush & Bryk, 2002), the primary goal of this study was to exam ine simultaneously neighborhood and school influences on academic achievement and adolescent risk of obesity and to examine the moderating effects of schools on these out comes. By examining concurrently neighborhood and school influences on academic achievement and adolescent risk of obesity, this study aimed to fill an important gap in the social determinants literature. For example, it is unclear if where an adolescen t lives or where she/he attends school has a stronger influence on academic achieveme nt. We also do not know if schools can moderate neighborhood influences on adoles cent academic achievement, nor do we know much about the relationships among schools, neighborhoods, and adolescent risk for obesity. Similarly, by investig ating outcomes related to bot h mental and physical wellbeing, this study helps expand the traditional single-domain approach often undertaken in social and behavioral science research.

PAGE 62

51Research Questions The following four research que stions were investigated: Research Question 1. To what extent are neighb orhood influences on U.S. middle and high school students’ academic achievement moderated by school environments? Research Question 2. What are the relative infl uences of neighborhood and school environments on U.S. middle and high sc hool students’ academic achievement? Research Question 3. To what extent are neighb orhood influences on U.S. middle and high school students’ risk of obes ity moderated by school environments? Research Question 4. What are the relative infl uences of neighborhood and school environments on U.S. middle and high school students’ risk of obesity? Study Design This study employed a nonexperimental, retr ospective, correl ational research design. Secondary data analyses of nationa lly representative Add Health (2005c) and AHAA (n.d.) restricted-use data were conducted. The study design also was crosssectional in nature because the data represented one point in time. Although multilevel modeling techniques ar e used with increasing frequency by educational and other social science researchers, use of CCREMs (Raudenbush & Bryk, 2002) is still rare in educa tional research. The lack of CCREMs in education is particularly troubling given the cross-classified nature of ma ny education data structures. For example, Level-1 units (students) are of ten cross-classified by two Level-2 factors (schools and neighborhoods) such that st udents from Neighborhood A might attend a school that students from Neighborhood B a nd Neighborhood C also attend, and students

PAGE 63

52 from the same neighborhood might attend differe nt schools. When cro ss-classification of data is ignored, models are misspecified, cau sing them to lack the level of control necessary to detect importan t and possible confounding effects, which, in turn, can lead to spurious conclusions. For this study, the cross-classified multilevel analyses allowed the examination of the influence of multiple contexts on academic achievement and risk of obesity, while statistically controlling for one another. That is, because neighborhood and school environments were analyzed simultaneously, results represent each environment’s unique influence on achievement and risk of obesity Further, use of interactions within the CCREMs allowed the investigation of the school environment as a moderator of neighborhood influences on each of the outco mes. All procedures for the study were approved through the University of South Florida’s Institutional Review Board. Overview of the Add Health Study Study design. Add Health is a nationally repres entative longitudinal study that seeks to advance the understanding of the rela tionships between indi viduals and different social contexts (family, friends and p eers, schools, and neighborhoods) and U.S. adolescents’ development. To date, three waves of data have been collected—Wave 1 (1994-1995), Wave II (1995-1996), and Wave III (2001-2002). Wave IV is scheduled to occur in 2007-2008. Data were collected thr ough a complex sampling design that utilized a cluster sample, at the school level, with unequal probabili ty of selectio n (Chantala & Tabor, 1999). Schools were selected to repres ent all high schools a nd middle schools in the U.S., thus the students attending the sc hools constitute a nationally representative

PAGE 64

53 sample of adolescents in Grades 7 to 12 (Tourangeau & Shin, 1999) Because this study only included data from Wave I (1994-1995), the following information only pertains to the sampling and data collection for Wave I. Similar information for subsequent waves can be found on the Add Health website (Add Health, 2004b). Before presenting details about the sampling and data co llection for Wave I, an overview of the different Add Health data sources is shown in Figure 2. Figure 2 General overview of Add Hea lth Wave I data sources. In-School sampling frame. A total of 132 schools (80 hi gh schools and 52 feeder schools) were included in the Add Health st udy. The initial 80 high schools approached about participating in the st udy were selected from the co mprehensive Quality Education Data, Inc. (QED) database (Tourangeau & Shin, 1999). In creating the sampling frame, U.S. Middle and High Schools (1994-1995) In-School Questionnaire School Administrator Questionnaire In-Home Interview Parent Questionnaire Contextual Data

PAGE 65

54 all schools that included an 11th grade and enro lled more than 30 students were classified as high schools. Similarly, if the grade sp an of a school was not clear, the school was included in the original sampling frame. Th rough this process, a sampling frame of 26,666 public and private high schools in the QED database was generated (Tourangeau & Shin, 1999). Before sampling, the schools in the sampling frame were sorted by size, school type, census region, level of urbanicity and percentage of Wh ite students to help ensure that the sample of schools selected were representative along the specified dimensions (Tourangeau & Shin, 1999). Schools then were systematically selected from the sorted lists with selection probabi lities proportional to th e school’s enrollment (Tourangeau & Shin, 1999). This process, often referred to as implicit stratification, helped ensure that the sample of school s was representative along the previously mentioned stratificati on variables (Tourang eau & Shin, 1999). Only 52 of the original 80 sampled hi gh schools were eligible and agreed to participate in the study. The remaining 28 sc hools were replaced by similar high schools. Replacement schools were identified by first so rting the sampling frame by school size, school type, urbanicity, percentage of White students, grade span, percentage of Black students, census region, and census divi sion (Tourangeau & Shin, 1999). Within each category, schools were sorted in a rando m order and the replacement school was the school that followed the originally sample school. If the first replacement school was not eligible or did not want to participate, this process was continued until an eligible and cooperative replacement school was found (Tourangeau & Shin, 1999).

PAGE 66

55 To identify the feeder schools, high school administrators were asked to provide a list of all junior high and middle schools expect ed to send at least five students to the high school’s entering class. High school administ rators also were asked to indicate what percentage of the entering class was exp ected to come from each feeder school (Tourangeau & Shin, 1999). From these lists, researchers attempted to select a single feeder school for each high school; however, three different situations prevented the inclusion of one feeder school for every hi gh school (Tourangeau & Shin, 1999). First, four of the high schools had no eligible f eeder schools because students entered their school from a vast number of junior high and middle schools. Second, 20 of the high schools included in the sample had grade span s that included sevent h and eighth grade, thus they served as their own feeder schools. Third, 4 of the 56 feeder schools that were asked to participate in the study declined; ther efore, the final Add H ealth sample included 80 high schools and 52 feeder schools (Toura ngeau & Shin, 1999). The probability of selection for each feeder sc hool was proportional to the estimated percentage of the entering class that came from the feeder school (Tourangeau & Shin, 1999). In-School Questionnaire. No sampling of students within the schools occurred for administration of the In-School Questionnaire. Instead, administrators at the sample schools were asked to have all students in the eligible grades (7th through 12th) complete the In-School Questionnaire (Tourangeau & Sh in, 1999). All but four of the participating schools allowed their students to complete the In-School Questionnaire (Tourangeau & Shin, 1999). However, the schools that did no t allow the In-School Questionnaire were

PAGE 67

56 retained in the sample because they did allow students to be sampled for the in-home data collection (Tourangeau & Shin, 1999). The In-School Questionnaire was self-administered during 45to 60-minute class periods to 90,118 students between Septem ber 1994 and April 1995 (Add Health, 2004c). Schools notified parents in advance of the date the questionnaire was going to be administered so they could decide if their child was to participat e or not (Add Health, 2004c). Also, there was no make-up day for students who were absent the day the questionnaire was administered at their schools The following nine topics were included on the In-School Questionnaire: social and de mographic information, parental education and occupation, household structure, risk beha viors, expectations for the future, selfesteem, health status, friendships, and extr acurricular activities (Add Health, 2004c). In order to identify students for subsequent da ta collection points, each school provided a student roster and Add Health staff assigned identification numbers to each student. Also, to help gather data on students’ peers, st udents were provided c opies of their school roster to identify their friends as they comp leted the questionnaire (Add Health, 2004c). School Administrator Questionnaire. In addition to the In-School Questionnaire given to the students, administrators at the 132 sample schools also were asked to complete a self-administered School Admini strator Questionnaire (Chantala & Tabor, 1999). Areas covered on the questionnaire incl uded issues dealing with school policy and procedures, teacher characteristics, health-ser vice provision or referral, and student body characteristics (Add Health, 2004c). A total of 164 School Administrator Questionnaires were collected between September 1994 and April 1995 (Add Health, 2005b).

PAGE 68

57In-Home sampling. In addition to obtaining inform ation from students through the In-School Questionnaire, 20,745 adolescents al so participated in In-Home Interviews (Add Health, 2005b). Students were eligible for the In-Home Interview sample if they completed the In-School Questionnaire and/or we re listed on a school roster. To generate a nationally representative sample of adolesce nts in Grades 7 through 12, students in each school were first stratified by grade and sex (Add Health, 2004c). Next, approximately 17 students from each stratum were randomly c hosen for each of the 80 pairs of schools. This selection process yielde d a core In-Home Interview sample of 12,105 adolescents (Add Health, 2004c). The remaining 8,640 adoles cents included in the In-Home sample were from the special oversamples. Oversampling was conducted for different et hnicities, students with disabilities, and genetic siblings who lived in th e same household (Add Health, 2004c). To investigate social networks, oversampling, or saturation, also was conducted in 16 schools. All students enrolled in 14 small school s (enrollment less than 300) and 2 large schools (total combined enrollment exceeding 3,300) also were included in the In-Home Interview sample (Add Health, 2004c). In-Home Interview. Wave 1 In-Home Interviews were conducted between April 1995 and December 1995. The In-Home Interviews varied in length from one to two hours, depending on the adolescent’s age a nd experiences (Add Health, 2004c). For example, additional questions were asked of adolescents who indicated multiple behaviors (e.g., if a respondent indicated that he or she had used drugs and had sexual intercourse, he/she was also asked if he or she used drugs while engaging in sexual

PAGE 69

58 intercourse; Add Health, 2004c). All interview data were r ecorded on laptop computers. Interviewers read less sensitive questions aloud and recorded each adolescent’s responses. For more sensitive questions, adolescen ts listened to prerecorded questions via headphones and entered their re sponses into the computer themselves (Add Health, 2004c). This process of data collection helped maintain data security and helped minimize interviewer and parental influence. The content of the In-Home Interviews covered a variety of topics including health status, healthcare utilization, nutriti on, peer networks, deci sion-making processes, family composition and relationships, edu cational aspirations and expectations, employment experiences, romantic relationshi ps, sexual experiences, substance use, and criminal activities (Add Health, 2004c). Re spondents also were administered the Add Health Picture Vocabulary Test (AHPVT) at the beginning of the In-Home Interview sessions. This test was a computerized, shortened version of the Peabody Picture Vocabulary Test-Revised (Add Health, 2004c). Parent Questionnaire. In addition to gathering in formation from adolescents during the Wave I In-Home In terview sessions, Add Health researchers also collected information from a parent of each adolescen t respondent. When possible, the preferred parent was the adolescent’s resident mother (Add Health, 2004c). Information obtained through the interviewer-assisted questionnair e included inheritable health conditions; marriages and other marriage-like relations hips; perceived neighbor hood characteristics; civic, volunteer, and school activity involve ment; health-affecting behaviors; education and employment; household income and economic assistance; and parental

PAGE 70

59 communication, interaction, and monitori ng (Add Health, 2004c ). A total of 17,700 Parent Questionnaires were complete d between April 1995 and December 1995 (Add Health, 2005b). Contextual data. Data about the neighborhoods wh ere adolescents lived were based on state, county, tract, and block group levels derive d from the Wave I addresses and were gathered from a variety of existing sources including but not limited to the U.S. Census, the Centers for Disease Control and Prevention, the National Center for Health Statistics, and the Federal Bureau of I nvestigation (Add Health, 2004a). Variables available in the Add Hea lth Contextual data include geographic and household characteristics, labor force participation a nd unemployment, crime, social programs and policies, income and poverty, social integrati on, and availability of health services (Add Health, 2004a). Sample weights. Add Health data contain multiple sampling weights to be used with different categories of analyses—a nalyses fitting population-average models, analyses fitting multilevel models that include adolescents and schools as the two levels of analysis, and analyses fitting populationaverage models for special subpopulations (binge drinkers, romantic partners of Add H ealth participants, and educational analyses involving high school transcript data; Chanta la, 2006). Although sampling weights could not be used in the cross-classified random effects models conducte d in this study, they were included in some preliminary univari ate analyses. This section provides an overview of the creation of the sampling wei ght used in this study—the Wave I sampling weight for fitting population-average models. Information on sampling weights for other

PAGE 71

60 waves and analytic procedures can be f ound on the Add Health website (Add Health, 2004d). Adolescents in 1995 who were enro lled in Grades 7-12 during 1994-1995 represent the population of inte rest for the sampling weight used in this study—Wave I sampling weight for use with single-level an alytic procedures (i .e., population-average models; Chantala, 2006). To calculate this sampling weight, Add Health researchers weighted Wave I In-Home samples using a f our-step process. The first step included calculating a preliminary school weight (W1) to compensate for probability selection differences among schools (Tour angeau & Shin, 1999). Next, W1 was adjusted for feeder school ineligibility and nonresponse. The th ird step accounted for student selection probabilities across schools and across grades and sexes within school s in the creation of an initial student-level wei ght (Tourangeau & Shin, 1999). The final weight calculated during the fourth step of the weighting pro cess was derived to compensate for student nonresponse to the Wave I In-Home Questionnair e. Thus, the sampling weight used in this study had been adjusted for both school-lev el and student-level selection probability and non-response (Tourangeau & Shin, 1999). Overview of AHAA Study AHAA is an educational supplement to Add Health. Whereas Add Health provides a great deal of data on a variety of social contexts, it has limited academicrelated information (Muller et al., 2007b). By collecting official high school transcripts from all Wave III respondents who signed a Transcript Release Form (TRF) and by compiling contextual information about the schools adolescents attended, AHAA

PAGE 72

61 provides the rich education-related data th at Add Health is mi ssing. Although AHAA was developed to supplement Add Health data, the data were selected separately from Add Health and were designed to create an e ducational data set that can be used in conjunction with Add Health or independently (Muller et al., 2007b). When used with the Add Health data, researchers are able to cap ture a more holistic view of the adolescent social, educational, and health-re lated behaviors and outcomes. AHAA’s study design is comparable to the 1987, 1990, 1994, 1998, and 2000 National Assessment of Educa tional Progress (NAEP) High School Transcript Studies; AHAA data collection and processing were modi fied from those used in NAEP transcript studies (Muller et al., 2007b) AHAA data were collected from a variety of sources including official student tran scripts, course catalogs, text book lists and course syllabi, School Information Forms, and several sec ondary data sources including two National Center for Education Statistics databases-Common Core of Data (CCD) and Private School Survey (PSS; Muller et al., 2007b). Although the AHAA data contain detailed information about the educational trajectories of Add Health re spondents, this study did not utilize the individual-level AHAA data. Instead, this study used the AHAA school context data obtained from the CCD. The CCD data included in the AHAA data were obtained from the 1990-1991, 1993-1994, 1994-1995, and 1999-2000 surveys. Example variables from the education contextual data include school-wide Title I el igibility, proportion of free lunch students, district size, school size, a nd racial composition indicators (e.g., proportion of White students, proportion of Black st udents; Muller et al., 2007a).

PAGE 73

62Study Sample Data for this study were drawn from the combined Wave I Add Health and AHAA studies. Starting with th e original sample of youth who completed both the InSchool Questionnaire and In-Home Interview (n = 15,356), the sampling frame for this study was limited to adolescents who attended regular public middle or high schools (i.e., not magnet or alternative schools) duri ng the 1994-1995 school year and who had complete data for all methodological variables (n = 11,841). Although limiting the sampling frame to regular public middle and hi gh schools reduces external validity, doing so allowed for more parsimonious models to be examined (i.e., eliminated the need to statistically control for school type). Thus, given the complex ity of the CCREMs used in the study, a reduction in external validity wa s deemed acceptable in exchange for models that were more parsimonious. This restrict ion removed 2,459 adolescents nested in 24 schools and 803 neighborhoods from the analyses More details on the study sample are provided in the Data management portion of the Data Analysis section. Measures Two criterion variables, adolescent acad emic achievement and risk of obesity, were examined in the study. Individual control variables (Level-1) consisted of adolescent biological sex, age, race/ethnicity, family SES, and athletic participation. Neighborhood-level variables (Level-2) consisted of neighborhood affluence, neighborhood poverty, neighborhood racial co mposition, and urbanicity. School-level variables (Level-2) consisted of school-level SES, stude nt body racial composition, teacher education, weight management educatio n, and school-level athletic participation.

PAGE 74

63 Below is a description of the Add Health and AAHA items used to measure the criterion and predictor variables as well as a descrip tion on how each SES composite variable (i.e., family SES, neighborhood affluence, nei ghborhood poverty, and school-level SES) was calculated. In addition, Table 1 provid es a summary of how each variable was operationalized and the data source for each variable. Family and school SES composite variab les were created following the same standardization process used by Duncan and Aber (1997). First, the mean and standard deviation for each variable included in the composite variable was calculated using data from observations included in the sampling frame for this study. Second, because the variables included in these two measures were not originally measured on the same scale, z-scores were created for each adolescent fo r each variable included in the composite ii i i x x z s Third, the z-scores for each variable in cluded in the composite were averaged into a final composite score; for example, 1233 zzz SESComposite Lastly, although this same general process was followed for the family and school SES composite variab les, the unit of anal ysis included in the creation of each composite varied. For family SES, individual adoles cents were the unit of analysis and for school SES, sc hools were the unit of analysis. Next, because all of the variables in cluded in the neighborhood SES composite variables were originally measured on the sa me scale, these variables were standardized using a slightly different pro cess than was used with family and school SES. Instead of

PAGE 75

64 standardizing each variable before crea ting the composite measure, neighborhood affluence and neighborhood poverty were standardized after the individual variables were averaged into a neighborhood composite score. More specifically, af ter calculating the overall mean level of affluence a nd poverty across neighborhoods, neighborhood affluence and neighborhood poverty z-scores were created for each adolescent using the following formula ii i i x x z s Table 1 List of Operationalized Variables and Data Source Variable Operational Definition Data Source1 Criterion Variables Academic achievement Standardized scores on the Add Health Picture Vocabulary Test (AHPVT). IH Risk of obesity Age-and-gender-adjusted BMI z -scores IH Level-1 control variables Biological sex Girl (0), boy (1) IH Age Age in years, gr and-mean centered IH Race/Ethnicity Non-Hispanic White or non-Hispanic Asian (0), nonHispanic Black, non-Hispanic Other, or Hispanic (1) IH Family SES A composite variable calculated as the mean of standardized ( z -score) measures of family income, parental educational level, and parental occupational prestige PI, IH Athletic participation Number of sports-related activities adolescents reported participating in IS Level-2 neighborhood variables Neighborhood affluence A composite vari able calculated as a standardized ( z score) measure computed fr om the average proportion of neighborhood income, occupational prestige, and educational levels CD

PAGE 76

65Table 1 List of Operationalized Variables and Data Source Variable Operational Definition Data Source1 Neighborhood poverty A composite vari able calculated as a standardized ( z score) measure computed fr om the average proportion of neighborhood poverty, single-parent households, and unemployment CD Neighborhood racial composition Proportion of White residents CD Urbanicity Proportion of residents who live inside an urbanized area CD Level-2 school variables School-level SES A composite variable calculated as the mean of standardized ( z -score) measures of school-level poverty, parental education, and parental occupational prestige PI, IH, AHAA Student body racial composition Proportion of White, non-Hispanic students AHAA Teacher education Proportion of t eachers with a Master’s degree or higher SA Weight management education Average proportion of students who reported being taught about four weight-related health topics--foods to eat, exercise, obesity, and being underweight IH School-level athletic participation Proportion of students involved in at least one sportsrelated activity IS Notes: 1AHAA = Adolescent Health and Academic Achievement, IS = Add Health In-School Questionnaire, SA = Add Health School Administ rator Questionnaire, IH = Add Health In-Home Interview, PI = Add Health Pare nt Questionnaire, and CD = Add H ealth Contextual Database. Criterion variables. Adolescent academic achievement and adolescent risk of obesity were the two criterion variables examined in this study. Academic achievement. In this study, standardiz ed Add Health Picture Vocabulary Test (AHPVT) scores were used as a measure of adolescent academic

PAGE 77

66 achievement. AHPVT is a modified versi on of the Peabody Picture Vocabulary TestRevised (PPVT-R), Form L. One-half of the original PPVT-R items were used in the AHPVT; odd-numbered items from 1 to 87 and even-numbered items from 90 to 175. Scores were standardized by age, with each age group having a mean of 100 and a standard deviation of 15. Score reliability and validity information on the AHPVT is not available (Joyce Tabor, personal communica tion, August 16, 2007). However, score reliability and validity information for the PPVT-R, Form L was obtained and is presented below. Using a sampling plan based on populati on data from the 1970 U.S. Census and stratified by age, gender, geographic regi on, parental occupation, ethnicity, and community size and type, the PPVT-R was stan dardized in 1979 using a sample of 4,200 children and youth aged 2 1/2 years to 18 years and 828 persons aged 19 years to 40 years (Dunn & Dunn, 1981). Based on PPV T-R, Form L tests consisting of approximately 35 items, split-half reliability coefficients, by relevant age for this study (i.e., 11 to 20), ranged from a low of .77 fo r 11-year-olds to a high of .88 for 18-yearolds, with an average of .84 (Dunn & Dunn, 1981). However, the Spearman-Brown adjustment for AHPVT suggests a higher averag e reliability of .91. Immediate retest standard score reliability coefficients, by age, were slightly weaker, with a low of .71 for 17-year-olds, a high of .89 for 11-year-olds, and an average of .82 (Dunn & Dunn, 1981). Delayed retest standard score reliability coe fficients, by age, also were lower than the split-half reliability coefficients, with a low of .56 for 18-year-olds, high of .90 for 11year-olds, and an average of .77 (Dunn & Dunn, 1981).

PAGE 78

67 In terms of content validity, the PPVT-R wa s designed to be representative of the content universe for hearing vocabulary—Webster’s New Collegiate Dictionary (Merriam, 1953, as cited in Dunn & Dunn, 1981). The only restriction in selecting a word from the dictionary was that its meaning had to be able to be de picted by a picture (Dunn & Dunn, 1981). Regarding construct validity, words were included in the PPVT-R when they fit the curve for hearing vocabulary es tablished by using the Rasch-Wright latent trait model (i.e., items with steep or flat item characteristic curves were not included; Dunn & Dunn, 1981). Concurrent validity evidence was the only criterion-related validity available for the PPVT-R (Dunn & Dunn, 1981). Based on 55 corre lations with other vocabulary tests, the PPVT-R was reported to have relatively high levels of correlation with other vocabulary tests (median correlation = .71; Dunn & Dunn, 1981). However, these data were not based on the PPVT-R directly. In stead, because the PPVT-R had a median correlation of .70 with the original PPVT, researchers applied validity research findings from the PPVT to the PPVT-R (Dunn & D unn, 1981). No construct validity evidence, such as that related to convergent vali dity, was reported in the PPVT-R manual. Risk of obesity. In this study, age-and-gender-adjusted BMI z-scores were used as a measure of risk of obesity. Although risk of obesity is often operationalized as having an age-and-gender-adjusted BMI 85th percentile (CDC, 2007), CCREMs cannot be used with a dichotomous criterion variable; therefore, a continuous measure of risk of obesity was created—standard ized age-and-gender-adjusted BMI. The age-and-genderadjusted BMI z-scores were created through a three-step process.

PAGE 79

68 First, adolescent BMI was calculated using the standard BMI formula [weight (lbs)/height (in)2*703]. Second, age-and-gender-adjuste d percentiles (5, 10, 25, 50, 75, 85, 90, and 95) were calculated using the CDC (2000a, 2000b) age-and-gender BMI tables. Third, using the age-a nd-gender-adjusted percentiles li near interpolation was used to calculate more precise BMI percentiles. Thes e percentiles were then standardized (i.e., expected normal scores) to create age-and-gender-adjusted BMI z-scores. Figures B-1 and B-2 in Appendix B contai n box-and-whisker plots for the initial age-and-genderadjusted BMI values. The height and weight data used to cr eate BMI values were ascertained through two In-Home Interview items—What is your height in feet and inches? and What is your weight? Although self-reported height and weight were used to calculate BMI, the correlation between interviewer-measured weight and self-reported weight in the Add Health data was .95 (Goodman, Hinden, & Khandelwal, 2000). Predictor variables. Three categories of predictor va riables were included in the current study: individual-level control va riables, neighborhood-level variables, and school-level variables. Biological sex, age, race/ethnicity, family SES, and athletic participation comprised the individual c ontrol variables in the CCREMs. Neighborhood affluence, poverty, racial composition, a nd urbanicity comprised the neighborhood-level variables in the CCREMs. School-l evel variables consisted of school-level SES, student body racial composition, teacher education, weight management education, and schoollevel athletic participation.

PAGE 80

69Biological sex. Boys were coded one and girls were coded zero; values were obtained from the interview item, Interviewer, please confirm that R’s sex is (male) female. Ask if necessary. Age. Adolescent age was measured by subtracting the adolescent’s date of birth from the Wave I In-Home interview date. In order to assign age-and-gender-adjusted BMI percentiles using the CDC (2007) BMI tables, age was computed and entered into the models as integers (i.e., full years) ranging from 11 to 20. Race/ethnicity. A dichotomous race/ethnicity vari able (0 = non-Hispanic White and non-Hispanic Asian, 1 = non-Hispanic Bl ack, non-Hispanic Other, and Hispanic) was created from two interview items: Are you of Hispanic or Latino origin? (Response options were yes or no) and What is your race? (Response options were White, Black or African American, American I ndian or Native American, Asia n or Pacific Islander, or Other). Family SES. Using the previously mentioned SES composite variable formula, this composite measure was created from three commonly used measures of family socioeconomic status: parental education, parental occupation, and family income. Parental education was ascertain ed during the Parent Interview—How far did you go in school? [Response options were never went to school (0); 8th grade or less (1); more than 8th grade, but did not graduate from high sc hool (2); went to a business, trade, or vocational school instead of high school (3); completed GED (4); high school graduate (5); went to a business, trade, or vocational school after high school (6); went to college,

PAGE 81

70 but did not graduate (7); gr aduated from a college or university (8); and professional training beyond a 4-year colle ge or university (9)]. The parent who participated in the interview also wa s asked about his or her spouse’s/partner’s education—How far did your current (s pouse/partner) go in school? (Response options same as above). When educa tion data were availa ble for two parents, an average parental education z-score was used in the family SES composite. For example, 2momedudadeduzz parentaleducation Household income data also were ob tained through the Parent Interview—About how much total income, before ta xes did your family receive in 1994? The original variable was continuous in $1,000 increments; however, for use in the composite score, income data were converted to ratios of income to 1995 federal poverty level (FPL) and coded 1 to 8: <100% (1), 100%-149% (2 ), 150%-199% (3), 200%-249% (4), 250%299% (5), 300%-349% (6), 350%-399% (7), and 400% (8). Parent occupation data were obtained from the adolescent In-Home Interviews—What kind of work does she do? (for mom) and What kind of work does he do? (for dad). Original response options: professional 1, such as doctor, lawyer, sc ientist; professional 2, such as teacher, librarian, nurse; manager, such as executive, director; technical, such as computer specialist, radiologist; office worker, such as bookkeeper, office clerk, secretary; sales worker, such as insurance agent, store clerk; re staurant worker or personal service, such as waitress, housek eeper; craftsperson, such as toolmaker, woodworker; construction worker, such as ca rpenter, crane operator ; mechanic, such as plumber, machinist; factory work er or laborer, such as assembler, janitor; transportation,

PAGE 82

71 such as bus driver, taxi driver; military or se curity, such as police officer, soldier, fire fighter; farm or fishery worker; other; and none. Occupation data were recl assified following the 1990 U.S. Census Bureau’s occupation classifications included in th e Add Health Contextual data: operators, fabricators, and laborers (1); production, craft or repair (2 ); farming, forestry or fishing (3); service occupations (4); military or secu rity (5); technical, sales or administrative support (6); and managerial or professional (7). As with parental education, when occupation data were available for two pa rents, an average parental occupation z-score was used in the family SES composite, such as the following 2momoccdadocczz parentaloccupation The intercorrelation of the three variables included in the family SES variable and Cronbach’s alpha are provided in Table 2. Table 2 Intercorrelation of Variables Comprising the Family SES Composite Variable (n = 10,860) Parental education Parental occupation Household income Parental education 1.0 Parental occupation .50 1.0 Household income .42 .26 1.0 Note: All variables were z -scores. Cronbach’s =.65 Athletic participation. Adolescent athletic participation was derived from adolescents’ responses to the In-School survey item, Here is a list of clubs, organizations, and teams found at many schools. Darken the ov al next to any of them that you are participating in this yea r, or that you plan to partici pate in later in the school year. Response options consisted of 33 common scho ol activities, 13 of which asked about

PAGE 83

72 different sports (cheerleading/dance team, baseball/softball, basketball, field hockey, football, ice hockey, soccer, sw imming, tennis, track, volleyba ll, wrestling, other sport). To create the athletic particip ation variable for this study, adolescents’ responses to the 13 sports-related response options were first summed and then winsorized such that the derived variable had values ranging from zero to four. The decision regarding how best to winsorize the athletic partic ipation variable was informed by examining the relationship between athletic participation and adoles cent BMI. More specifically, the initial relationship between BMI and athletic pa rticipation was non-lin ear such that BMI decreased as the number of sports-related activities increased until the value four; after four reported sports-activities the relationship between BMI and athletic participation diminished. Therefore, all athletic participa tion values greater than four were collapsed into four such that a value of four on the deri ved variable represents participation in four or more sports-related activities. Neighborhood affluence. Using the previously mentioned neighborhood SES composite variable formula, this composite measure was created from three variables: the proportion of families with income equal to or greater than $50,000, proportion employed persons aged 16 and over in managerial and professional occupations, and the proportion of residents age 25 and older with at leas t a college degree, as reported from the 1990 census in the Add Health contextual data. Th e intercorrelation of these three variables and Cronbach’s alpha for this composite variable are provid ed in Table 3.

PAGE 84

73 Table 3 Intercorrelation of Variables Comprising the Neighborhood Affluence Composite Variable (n = 10,860) Proportion of families with income $50,000 Proportion of managerial & professional occupations Proportion with at least a college degree Proportion of families with income $50,000 1.0 Proportion of managerial & professional occupations .72 1.0 College degree Proportion with at least a college degree .75 .91 1.0 Note: All variables were z -scores. Cronbach’s =.89 Neighborhood poverty. Using the previously mentioned neighborhood SES composite variable formula, this composite measure was created from three variables: the proportion of families living below the pove rty line, proportion of female-headed households, and the proportion of unemployed a dult residents, as reported from the 1990 census in the Add Health contextual data. Th e intercorrelation of these three variables and Cronbach’s alpha for this composite variable are provid ed in Table 4. Table 4 Intercorrelation of Variables Comprising the Neighborhood Poverty Composite Variable (n = 10,860) Proportion of families below the poverty line Proportion of femaleheaded households Proportion of unemployed adults Proportion of families below the poverty line 1.0 Proportion of femaleheaded households .18 1.0 Proportion of unemployed adults .77 .16 1.0 Note: All variables were z -scores. Cronbach’s =.44

PAGE 85

74Neighborhood racial composition. The proportion of White residents in a neighborhood, as reported from the 1990 census in the Add Health c ontextual data, was used to measure neighborhood racial composition. Urbanicity. The proportion of residents who liv e inside an urbanized area, as reported from the 1990 census in the Add Health contextual data, was used to measure urbanicity. School-level SES. Using the previously mentioned SES composite variable formula, this composite measure was created from three variables: aggregated parental education (as previously de fined), aggregated parental occupation (as previously defined), and proportion of st udents not eligible for the fr ee lunch program (as a proxy for income), as reported from the 1994-1995 CCD in the AAHA data. The intercorrelation of these three variables and Cr onbach’s alpha for this composite variable are provided in Table 5. Table 5 Intercorrelation of Variables Comprising the School SES Composite Variable (n = 10,860) School-level parent education School-level parental occupation Proportion of students not eligible for free lunch School-level parent education 1.0 School-level parental occupation .49 1.0 Proportion of students not eligible for free lunch .41 .79 1.0 Note: All variables were measured as z -scores. Cronbach’s =.80

PAGE 86

75Student body racial composition. As reported from the 1994-1995 CCD in the AHAA data, the proportion of White, non-Hispanic students was used to measure student body racial composition. Teacher education. The proportion of teachers at a school with a Master’s degree or higher, as reported by school administrators in response to the School Administrator Questionnaire item, Approximately what percentage of your full-time classroom teachers hold Master’s degrees or higher? (WRITE IN PERCENT). Weight management education. A composite variable created from responses to the In-Home Interview item, Please tell me whether you ha ve learned about each of the following things in a class at school. Response options consisted of 17 health-related topics, 4 of which were related to mainta ining a healthy weight (foods you should and should not eat; the importance of exercise; the problems of being overweight; and the problems of being underweight). To create the weight education variable, first the proportion of students per school who reported learning about each of these four topics was calculated and then the average of the four proportions was derived. The intercorrelation of thes e four variables and Cronbach’s al pha for this composite variable are provided in Table 6.

PAGE 87

76 Table 6 Intercorrelation of Variables Comprising the Weight Education Composite Variable (n = 10,860) Foods you should and should not eat Importance of exercise Problems of being overweight Problems of being underweight Foods you should and should not eat 1.0 Importance of exercise .67 1.0 Problems of being overweight .71 .60 1.0 Problems of being underweight .73 .59 .85 1.0 Cronbach’s =.862 School-level athletic participation. The proportion of student s involved in at least one sports-related activity. Data Analysis Data management. All data used in this study came from the restricted-use data files versus the public-use data files because the public-use data only contain information on 6,504 adolescents and cannot be linked to the contextual neighbor hood data included in this study (Add Health, 2005a ). For security purposes, a ll electronic files associated with and generated from the restricted data (e.g., SAS programs and output) were encrypted and stored on a pa ssword protected external hard drive that was kept in a locked file cabinet when not in use. The researcher was the only person who knew the password to access the encrypted fi les. Similarly, the research er’s laptop, which was used to conduct the data analysis, was password protected and programmed to lock after 10 minutes of inactivity. Only the researcher knew the password to unlock the computer.

PAGE 88

77 Prior to conducting any analysis, several data management tasks were completed. First, to improve data analysis processing time, a smaller data set that contained only methodological variables (e.g., sample wei ghts, respondent identification, strata variables) and substantive variables of interest (e.g., criterion and predictor variables) was created. Second, non-applicab le response options were examined for all variables included in the study to determine if they coul d be recoded into theo retically conceivable responses. For example, not all schools have at hletics, thus, non-appl icable responses to the items used to assess student athletic participat ion could have been conceived of as a response of no. Upon examination of the vari ables, it was determined that none of the variables had non-applicable res ponses that could be recoded in this manner. In fact, the athletic participation items did not contain non-applicable responses. Third, the study sample was restricted to adolescents who participated in the InSchool Questionnaire and In-Home Interview, attended a regular public junior high, middle or high school during the 1994-1995 academic year, and had complete data on all methodological and substantive variables in cluded in the study. Also, because Add Health data contain pairs of siblings, one si bling from the sample of adolescents who met the aforementioned criteria was randomly selected for inclusion in the study sample. Fourth, because employing sample filters can alter the generalizability of findings, missing and refusal data (when applicable) were examined to determine the frequency of missing data across observations and to what extent the missingness and refusals were random (i.e., correlations between missing a nd refusal indicators and all variables included in the analyses were examined). Although researchers typically treat refusal

PAGE 89

78 responses as missing, these responses were analyzed separately because theoretically refusal responses are different than missing (i.e ., a refusal to respond to an item is itself a response and should not be treated as if it were simply missing). However, given the nonsensitive nature of the major ity of variables included in the study, it was not surprising that the only variable with a substantial amount of refusal responses was household income [n =1,060 (11%)]. Therefore, examination of refusal data focused only on the extent to which refusals for household income were random. When systematic missingness and/or refusa ls were observed, statements about conclusions and interpretations of the data ha ve been tempered with appropriate cautions and caveats. For example, because the variab le used to measure household income did not appear to be missing at random, the obtai ned parameter estimate for family SES, as well as the parameter estimates for variable s correlated with household income and/or family SES have been interpreted with additi onal caution as they are likely to be biased. All data management tasks were executed in SAS v9.1.3 (SAS Institute Inc., 2003). Also, although imputation is a common method for dealing with missing data (e.g., Allison, 2002; Rubin, 1996; Schafer & Graham, 2002), it is not always the best missing-data treatment. For example, when data are missing completely at random and the amount of missing data are not extreme researchers have shown that imputation methods do not perform better than listwise deletion as used in the current study (Allison, 2002; Kromrey & Hines, 1994). Furthermore, wh en data are not missing completely at random and less than 30% of da ta are missing, listwise de letion yields less biased regression parameter estimates than do ot her common imputation methods (Kromrey &

PAGE 90

79 Hines, 1994). Thus, even though the data do not appear to be missing at random, less than 30% of data were missing; therefore, limiting the sample to those with complete data on all variables of interest (i.e., listwise deleti on) was an appropriate missing-data treatment. Univariate and bivariate analyses. Descriptive univariate statistics were examined to gain an understanding of the data distribution and bivariate correlational analyses were conducted to gain a better unde rstanding of how the va riables of interest were interrelated. Because sample weights coul d not be used in the multivariate analyses, univariate statistics were examined both we ighted and unweighted and then compared. Doing so helped inform the generalizability of the multivariate findings. All univariate and bivariate analyses were conducted us ing SAS v9.1.3 (SAS Institute Inc., 2003). Multivariate analyses. Research questions were examined using cross-classified random effects hierarchical linear models with individuals nested within schools and neighborhoods. All multivariate data analys es were conducted using PROC MIXED in SAS v9.1.3 (SAS Institute Inc., 2003). However, before conducting any multivariate analyses, data were screened for violati ons of assumptions of ten associated with multilevel models (i.e., multicollinearity, normality, linearity, and homogeneity of variance). Further, the data screening techni ques described below are the same as those recommended by Hox (2002) and Raudenbush and Bryk (2002). First, the data were examined for mul ticollinearity. In addition to the bivariate examination of independent variables vi a zero-order correlation coefficients, multicollinearity was assessed by examining tolerance values from four multiple regression models for each of the criterion va riables. The first multiple regression model

PAGE 91

80 contained the main effects for all Level1 predictor variables, the second multiple regression model contained the main eff ects for all Level-2 neighborhood predictor variables, the third multiple regression model contained the main effects for all Level-2 school predictor variables, and the fourth multiple regression model contained the main effects for all Level-2 neighborhood and school pr edictor variables. All variables from all eight regression models (i.e., four for academic achievement and four for risk of obesity) exhibited acceptable tolerance va lues (Berry, 1993), therefore, all variables were retained and included in the CCREMs. Next, Level-1 and Level-2 residuals from the full academic achievement CCREM (Model 5-AA) were examined for potential violations of normality, linearity, and homogeneity of variance. To examine the nor mality assumption of Level-1 residuals, a box-and-whisker plot of the re siduals was created and the skewness and kurtosis of the residuals were calculated. Normality, linea rity, and heteroscedasticity also were examined by plotting the Level-1 residuals against the predicted values for academic achievement. Because CCREMs contain data for two different Level-2 structures (i.e., neighborhoods and schools), Level-2 resi duals were examined separately for neighborhoods and schools. To examine the normality assumption of neighborhood Level-2 residuals, a box-and-whisker plot of the neighborhood Level-2 residuals was created and the skewness and kurtosis of the residuals were calculated. Normality, linearity, and heteros cedasticity also were examined by plotting the neighborhood Level2 residuals against the predicted values for academic achievement. The same process was

PAGE 92

81 repeated using school Level-2 residuals. Si milarly, the Level-1 and Level-2 residuals from the full risk of obesity CCREM (Model 5-RO) were examined for potential violations of normality, linearity, and homogeneity of variance following the same process as described above for the academic achievement analysis. To allow comparison of models that diffe red in their fixed effects, the crossclassified random effects hierarchical lin ear models were estimated using maximum likelihood estimation. All co ntinuous predictor variables, without a meaningful interpretation of zero, were grand-mean cen tered. Grand-mean centering was used instead of group-mean centering because (a) the fo cus of the study was on Level-2 predictors, while statistically controlling for Level-1 vari ables and (b) the interactions included in the study were between Level-2 predictors (E nders & Tofighi, 2007). To determine the moderating effects of schools on neighborhoods as well as the unique influence of neighborhoods and schools, six CCREMs were examined for each criterion variable. A description of the models examined in this study is presented below. See Table 7 for a general overview of the structure of each CCREM for each criterion variable. Table 7 Summary of the Model Structure for each Cross-Classified Random Effects Model Model Academic Achievement Predictor Variables Risk of Obesity Predictor Variables Model 1: Unconditional model None None Model 2: Level-1 control model Biological sex, age, race, family SES Biological sex, age, race, family SES, athletic participation Model 3: Neighborhood model Affluence, poverty, racial composition, Urbanicity Affluence, poverty, racial composition, Urbanicity

PAGE 93

82Table 7 Summary of the Model Structure for each Cross-Classified Random Effects Model Model Academic Achievement Predictor Variables Risk of Obesity Predictor Variables Model 4: School model School SES, student body racial composition, teacher education School SES, weight management Education, school athletic participation Model 5: Neighborhood and school main effects model Affluence, poverty, racial composition, Urbanicity, school SES, student body racial composition, teacher education Affluence, poverty, racial composition, urbanicity, school SES, weight management education, school athletic participation Model 6: Neighborhood, school, and interaction model Affluence, poverty, racial composition, Urbanicity, school SES, student body racial composition, teacher education, affluence*school SES, poverty*school SES, affluence*teacher education, poverty*teacher education Affluence, poverty, racial composition, urbanicity, school SES, weight management education, school athletic participation, affluence*school SES, poverty*school SES, affluence*weight education, poverty*weight education Following a model-building strategy as discussed by Raudenbush and Bryk (2002), the cross-classified random effects models were examined in order of complexity, starting with the simplest mode l that had no predictors and ending with the most complex model with multiple interaction terms. The fi rst academic achievement model was a fully unconditional model with no predictors (M odel 1-AA). At Level-1, the model was 121212()()() ijjinterceptjjijjYe (1) where 12() ijjYsymbolizes the achievement outcome (AHPVT) for student i in neighborhood 1 j and school 2 j The intercept,12() interceptjj represents the predicted AHPVT score for students from neighborhood 1 j and school2 j The residual, 12()ijje,

PAGE 94

83 represents the deviation of a student’s AHPVT score from the student’s neighborhood and school predicted intercept value and is assumed 2(0,)N At Level-2, the Level-1 intercept,12()interceptjj was modeled as a random effect in the fully unconditional model. 1212()0000 interceptjjinterceptjjbc (2) The overall intercept,intercept represents the grand mean AHPVT score. The neighborhood residual,100jb, represents the neighborh ood effect for neighborhood 1 j (averaged across schoo ls) and is assumed 00(0,)bN The school residual, 200 jc, represents the school effect for school 2 j (averaged across neighborhoods) and is assumed 00(0,)cN Next, a Level-1 control model (Model 2-AA) examined the extent to which academic achievement varied based on i ndividual-level characteristics. 121212121212 1212121212()()_()()()() /()()()()()_ /ijjinterceptjjbiosexjjijjagejjijj raceethjjijjsesjjijjijjYbiosexage raceethsese (3) At Level-1, 12() ijjY still symbolizes the achieveme nt outcome (AHPVT) for student i in neighborhood 1 j and school2 j The intercept,12() interceptjj is now the expected AHPVT score when all predictor variables are set to zero. More specifically, for this model, 12() interceptjj represents the predicted AHPVT scor e for an average age, non-Hispanic Black/non-Hispanic Other/Hispanic fema le with an average family SES. biosex and / raceeth represent the expected difference in AHPVT scores between a student in

PAGE 95

84 neighborhood 1 j and school2 j with a value of 0 for each variable and a student in neighborhood 1 j and school2 j with a value of 1 for each va riable. For example, because males are coded 1, biosex is the expected difference in AHPVT scores between boys and girls in neighborhood 1 j and school2 j while statistically controlling for all other predictors in the model. For age, age represents the expected change in AHPVT score for a student in neighborhood 1 j and school2 j for every one-year change in age while statistically controlling for all other predictors in the model. For family SES, s es represents the expected change in AHP VT score for a student in neighborhood 1 j and school2 j for every one standard deviation cha nge in family SES while statistically controlling for all other pr edictors in the model. At Level-2, the Level-1 intercept,12() interceptjj was modeled as a random effect in the Level-1 control model. 1212 12 12 12 12()0000 _()_ () /()/ () interceptjjinterceptjj biosexjjbiosex agejjage raceethjjraceeth sesjjsesbc (4) The overall intercept,intercept represents the grand mean AHPVT score when all Level-1 predictor variables are set to zero. More specifically, intercept represents the predicted AHPVT score for an average age, non-Hisp anic Black/non-Hispanic Other/Hispanic female with an average family SES. The neighborhood residual,100 jb represents the

PAGE 96

85 neighborhood effect for neighborhood 1 j (averaged across schools). The school residual,200 jc represents the school effect for school 2 j (averaged across neighborhoods). Each x represents the same value as di scussed above in Equation 3 and each x represents the fixed effects for each corresponding Level-1 predictor variable. For example, biosex represents the effect of biological sex that was modeled not to vary across neighborhoods or schools. The Leve l-1 portion of Model 2-AA (Equation 3) served as the Level-1 model for all remaining academic achievement models. Adding to Model 2-AA, the third mode l (Model 3-AA) examined neighborhoodlevel correlates of achievement while statisti cally controlling for indi vidual differences at Level-1 (Equation 3). At Leve l-2, the Level-1 intercept,12() interceptjj was modeled as a random effect and a function of four neighbor hood variables: afflue nce, poverty, racial composition, and urbanicity. 1211 1112 12 12 12 12()__ _0000 _()_ () /()/ ()__ _interceptjjinterceptneighaffljneighpovj neighracejurbanjjj biosexjjbiosex agejjage raceethjjraceeth sesjjseneighafflneighpov neighraceurbanbc s (5) The intercept,intercept now represents the expected adjusted (for Level-1 predictors) AHPVT score when all Level-2 predictor variables are set to zero. More specifically, intercept is the expected adjusted AHP VT score for a student from a neighborhood with average affluence and povert y levels and no urbanicity or White

PAGE 97

86 residents. Each x represents the fixed effect of variab le X that is assumed constant over all neighborhoods (e.g.,_ neighpov represents the effect of neighborhood poverty on AHPVT scores across all neighbo rhoods). The neighborhood residual,100 jb represents the neighborhood effect for neighborhood 1j (averaged across schools) while statistically controlling for all Level-2 pr edictors. The school residual,200 jc represents the school effect for school 2j (averaged across neighborhoods) while statistically controlling for all Level-2 predictors. Each x and x represent the same values as discussed in Equations 3 and 4. Next, also building on Model 2-AA, the f ourth model (Model 4-AA) investigated school-level predictors of achievement while statistically controlling for individual variables (Equation 3). At Leve l-2, the Level-1 intercept,12() interceptjj was modeled as a random effect and a function of three school variables: school SES, student body racial composition, and teacher education. 1222 212 12 12 12 12()__ _0000 _()_ () /()/ ()__ _interceptjjinterceptschsesjsturacej tchedujjj biosexjjbiosex agejjage raceethjjraceeth sesjjsesschsessturace tchedubc (6) The intercept,intercept now represents the expected adjusted (for Level-1 predictors) AHPVT score when all Level-2 predictor variables are set to zero. More specifically, intercept is the expected adjusted AHPVT sc ore for a student who attends an

PAGE 98

87 average SES school with no White, non-Hispanic students and no teachers with graduate degrees. Eachx represents the fixed effect of variable X that is assumed constant over all schools (e.g.,_ s chses represents the effect of school SES on AHPVT scores across all schools). The neighborhood residual,100 jb represents the ne ighborhood effect for neighborhood 1j (averaged across schools) while sta tistically controlling for all Level-2 predictors. The school residual,200 jc represents the school effect for school 2j (averaged across neighborhoods) while statistically cont rolling for all Level-2 predictors. Each x and x represent the same values as discussed above in Equations 3 and 4. Model 5-AA was a combination of Models 3-AA and 4-AA and examined achievement as a function of both neighbor hood and school factors simultaneously, while statistically controlling for i ndividual characteristics (Equation 3). At Level-2, the Level1 intercept,12() interceptjj was modeled as a random eff ect and a function of four neighborhood variables and three school variables: neighborhood affluence, neighborhood poverty, neighborhood racial com position, urbanicity, school SES, student body racial composition, and teacher education. 1211 1122 212 12()__ ___ _0000 _()___ ___ _interceptjjinterceptneighaffljneighpovj neighracejurbanjschsesjsturacej tchedujjj biosexjjbiosneighafflneighpov nei g hraceurbanschsessturace tchedubc 12 12 12() /()/ () ex agejjage raceethjjraceeth sesjjses (7)

PAGE 99

88 The intercept,intercept now represents the expected adjusted (for Level-1 predictors) AHPVT score when all Level-2 predictor variables are set to zero. More specifically,intercept is the expected adjusted AHPVT score for a student from a neighborhood with average affluence and povert y levels and no urbanicity or White residents and who attends an average SES school with no White, non-Hispanic students and no teachers with graduate degrees. Each x represents the fixed effect of variable X that is assumed constant over all neighborhoods (e.g.,_ neighpov represents the effect of neighborhood poverty on AHPVT scores across all neighborhoods). Each x represents the fixed effect of variable X that is assumed constant over all schools (e.g.,_ s chses represents the effect of school SES on AHPVT scores across all schools). The neighborhood residual,100 jb represents the neighborh ood effect for neighborhood 1j (averaged across schools) while statistically controlling for all Level-2 predictors. The school residual,200 jc represents the school effect for school 2j (averaged across neighborhoods) while statistically control ling for all Level-2 predictors. Each x and x represent the same values as discussed in Equations 3 and 4. Lastly, Model 6-AA extended Model 5-AA a nd examined whether the association between achievement and neighborhoods a nd schools depended on four different moderating effects while statistically controlling for individual differences at Level-1 (Equation 3). At Level-2, the Level-1 intercept,12() interceptjj was modeled as a random effect and a function of four neighborhood va riables, three school variables, and four interactions: neighborhood affluence, neighborhood poverty, neighborhood racial

PAGE 100

89 composition, urbanicity, school SES, student body racial composition, teacher education, neighborhood affluence*school SES, nei ghborhood poverty*school SES, neighborhood affluence*teacher education, and ne ighborhood poverty*teache r education. 1211 1122 21()__ ___ __*___ ___ __*interceptjjinterceptneighaffljneighpovj neighracejurbanjschSESjsturacej tchedujneighafflschsesjneighafflneighpov neighraceurbanschsessturace tcheduneighaffl 2 1212 1212 12 12 12_*__*_ _*_0000 _()_ () /()_ _*__*_ _*_j neighpovschsesjjneighaffltchedujj neighpovtchedujjjj biosexjjbiosex agejjage raceethjjraschses neighpovschsesneighaffltchedu neighpovtchedubc 12/ () ceeth sesjjses The intercept,intercept now represents the expected adjusted (for Level-1 predictors) AHPVT score when all Level-2 predictor variables are set to zero. More specifically,intercept is the expected adjusted AHPVT score for a student from a neighborhood with average affluence and po verty levels and no urbanicity or White residents and who attends an average SES sc hool with no White, non-Hispanic students, and no teachers with graduate degrees. Each x represents the fixed effect of variable X that is assumed constant over all neighborhoods (e.g.,_ neighpov represents the effect of neighborhood poverty on AHPVT scores across all neighborhoods). Each x represents the fixed effect of variable X that is assumed constant over all schools (e.g.,_ s chses represents the effect of school SE S on AHPVT scores ac ross all schools). (8)

PAGE 101

90 The interactions, _*_ neighafflschses represents the moderating effect of school SES on neighborhood affluence (i.e., the relati onship between neighborhood affluence and AHPVT scores may differ depend ing on the level of school SES). _*_ neighpovschses represents the moderating e ffect of school SES on neighborhood poverty (i.e., the relationship between neighborhood poverty a nd AHPVT scores may differ depending on the level of school SES). _*_ neighaffltchedu represents the moderating effect of teacher education on neighborhood affluence (i.e., the relationship between neighborhood affluence and AHPVT scores may differ depe nding on the level of teacher education). _*_ neighpovtchedu represents the moderating effect of teacher education on neighborhood poverty (i.e., the relationship between neighborhood poverty and AHPVT scores may differ depending on the level of teacher education). The neighborhood residual,100 jb represents the neighborhood effect for neighborhood 1j (averaged across schools) while sta tistically controlling for all Level-2 predictors. The school residual,200 jc represents the school effect for school 2j (averaged across neighborhoods) while statistically cont rolling for all Level-2 predictors. Each x and x represent the same values as discussed in Equations 3 and 4. When predicting risk of obesity, th e same model-building procedure was conducted. The first risk of obesity m odel was a fully unconditional model with no predictors (Model 1-RO). At Level-1, the model was 121212()()() ijjinterceptjjijjYe (9)

PAGE 102

91 where 12() ijjY symbolizes the risk of obesity outcome [age-and-gender-adjusted BMI z score] for student i in neighborhood 1j and school 2j The intercept,12() interceptjj represents the predicted BMI z -score for students from neighborhood 1j and school2j The residual, 12() ijje represents the devia tion of a student’s BMI z -score from the student’s neighborhood and school predic ted intercept value and is assumed 2(0,) N At Level-2, the Level-1 intercept,12() interceptjj was modeled as a random effect in the fully unconditional model. 1212()0000 interceptjjinterceptjjbc (10) The overall intercept,intercept represents the grand mean BMI z -score. The neighborhood residual,100 jb represents the neighborhood effect for neighborhood 1j (averaged across schools) and is assumed 00(0,)bN The school residual, 200 jc represents the school effect for school 2j (averaged across neighborhoods) and is assumed 00(0,)cN Next, a Level-1 control model (Model 2-RO) examined the extent to which risk of obesity varied based on indivi dual-level characteristics. 121212121212 12121212121212()()()()_()() ()()/()()()()()_ /ijjinterceptjjathletejjijjbiosexjjijj agejjijjraceethjjijjsesjjijjijjYathletebiosex ageraceethsese (11) At Level-1, 12() ijjY still symbolizes the risk of obesity outcome (BMI z -score) for student i in neighborhood 1j and school2j The intercept,12() interceptjj is now the expected BMI z -score when all predictor variables ar e set to zero. More specifically, for this model, 12() interceptjj represents the predicted BMI z -score for an average age, non-

PAGE 103

92 Hispanic Black/non-Hispanic Other/Hispanic female with an average family SES. biosex and / raceeth represent the expected difference in BMI z -scores between a student in neighborhood 1j and school2j with a value of 0 for each variable and a student in neighborhood 1j and school2j with a value of 1 for each va riable. For example, because males are coded 1, biosex is the expected difference in BMI z -scores between boys and girls in neighborhood 1j and school2j while statistically controlling for all other predictors in the model. For age, age represents the expected change in BMI z -score for a student in neighborhood 1j and school2j for every one-year change in age while statistically controlling for all other predictors in the model. For family SES, s es represents the expected change in BMI z -score for a student in neighborhood 1j and school2j for every one standard deviation chan ge in family SES while statistically controlling for all other predictors in the model. At Level-2, the Level-1 intercept,12() interceptjj was modeled as a random effect in the Level-1 control model. 1212 12 12 12 12 12()0000 () _()_ () /()/ () interceptjjinterceptjj athletejjathlete biosexjjbiosex agejjage raceethjjraceeth sesjjsesbc (12)

PAGE 104

93 The overall intercept,intercept represents the grand mean BMI z -score when all Level-1 predictor variables are set to zero. More specifically, intercept represents the predicted BMI z -score for an average age, non-Hispan ic Black/non-Hispanic Other/Hispanic female with an average family SES. The neighborhood residual,100 jb represents the neighborhood effect for neighborhood 1j (averaged across schools). The school residual,200 jc represents the school effect for school 2j (averaged across neighborhoods). Each x represents the same value as discussed above in Equation 11 and each x represents the fixed effects for each corresponding Level-1 predictor variable. For example, biosex represents the effect of biol ogical sex that was modeled not to vary across neighborhoods or schools. Th e Level-1 portion of Model 2-RO (Equation 11) served as the Level-1 model for a ll remaining risk of obesity models. Adding to Model 2-RO, the third model (Model 3-RO) examined neighborhoodlevel correlates of risk of obesity while stat istically controlling for individual differences at Level-1 (Equation 11). At Level-2, the Level-1 intercept,12() interceptjj was modeled as a random effect and a function of four neighbor hood variables: affluence, poverty, racial composition, and urbanicity.

PAGE 105

941211 1112 12 12 12 12()__ _0000 () _()_ () /()__ _interceptjjinterceptneighaffljneighpovj neighracejurbanjjj athletejjathlete biosexjjbiosex agejjage raceethjjneighafflneighpov neighraceurbanbc 12/ () raceeth sesjjses (13) The intercept,intercept now represents the expected adjusted (for Level1predictors) BMI z -score when all Level-2 predictor variables are set to zero. More specifically, intercept is the expected adjusted BMI z -score for a student from a neighborhood with average affluence and po verty levels and no urbanicity or White residents. Each x represents the fixed effect of variable X that is assumed constant over all neighborhoods (e.g.,_ neighpov represents the effect of neighborhood poverty on BMI z scores across all neighborhoods). The neighborhood residual,100 jb represents the neighborhood effect for neighborhood 1j (averaged across schools) while statistically controlling for all Level-2 predictors. The school residual,200 jc represents the school effect for school 2j (averaged across neighborhoods) while statistically controlling for all Level-2 predictors. Each x and x represent the same values as discussed in Equations 11 and 12. Next, also building on Model 2-RO, the f ourth model (Model 4-RO) investigated school-level predictors of risk of obesity wh ile statistically controlling for individual variables (Equation 11). At Level-2, the Level-1 intercept,12() interceptjj was modeled as a

PAGE 106

95 random effect and a function of three school variables: school SES, weight management education, and school-level athletic participation. 1222 212 12 12 12 12 12()_ 0000 () _()_ () /()/ ()_interceptjjinterceptschsesjweightj athleticsjjj athletejjathlete biosexjjbiosex agejjage raceethjjraceeth sesjjsesschsesweight athleticsbc (14) The intercept,intercept now represents the expected adjusted (for Level-1 predictors) BMI z -score when all Level-2 predicto r variables are set to zero. More specifically, intercept is the expected adjusted BMI z -score for a student who attends an average SES school with no weight manage ment education and no student athletes. Eachx represents the fixed effect of variable X that is assumed constant over all schools (e.g.,_ s chses represents the effect of school SES on BMI z -scores across all schools). The neighborhood residual,100 jb represents the neighborhoo d effect for neighborhood 1j (averaged across schools) while statistically controlling for all Level-2 predictors. The school residual,200 jc represents the school effect for school 2j (averaged across neighborhoods) while statistically controlling for all Level-2 predictors. Each x and x represent the same values as disc ussed above in Equations 11 and 12. Model 5-RO was a combination of Models 3-RO and 4-RO and examined risk of obesity as a function of both neighborh ood and school factors simultaneously, while statistically controlling for individual char acteristics (Equation 11). At Level-2, the

PAGE 107

96 Level-1 intercept,12() interceptjj was modeled as a random effect and a function of four neighborhood variables and three school variables: neighborhood affluence, neighborhood poverty, neighborhood racial composition, urbanicity, school SES, weight management education, and schoollevel athletic participation. 1211 1122 212 12()__ __ 0000 ()__ __interceptjjinterceptneighaffljneighpovj neighracejurbanjschsesjweightj athleticsjjj athletejjathleneighafflneighpov neighraceurbanschsesweight athleticsbc 12 12 12 12_()_ () /()/ () te biosexjjbiosex agejjage raceethjjraceeth sesjjses (15) The intercept,intercept now represents the expected adjusted (for Level-1 predictors) BMI z -score when all Level-2 predictor variables are set to zero. More specifically,intercept is the expected adjusted BMI z -score for a student from a neighborhood with average affluence and po verty levels and no urbanicity or White residents and who attends an average SES sc hool with no weight management education and no student athletes. Each x represents the fixed effect of variable X that is assumed constant over all neighborhoods (e.g.,_ neighpov represents the effect of neighborhood poverty on BMI z -scores across all neighborhoods). Each x represents the fixed effect of variable X that is assumed c onstant over all schools (e.g.,_ s chses represents the effect of school SES on BMI z -scores across all schools). The neighborhood residual,100 jb

PAGE 108

97 represents the neighborh ood effect for neighborhood 1j (averaged across schools) while statistically controlling for all Leve l-2 predictors. The school residual,200 jc represents the school effect for school 2j (averaged across neighborhoods) while statistically controlling for all Level-2 predictors. Each x and x represent the same values as discussed in Equations 11 and 12. Lastly, Model 6-RO expanded Model 5-RO and examined whether the association between risk of obesity and neighborhoods and schools depended on four different moderating effects while statistically controlling for individual differences at Level-1 (Equation 11). At Level-2, the Level-1 intercept,12() interceptjj was modeled as a random effect and a function of four neighborhood va riables, three school variables, and four interactions: neighborhood affluence, neighborhood poverty, neighborhood racial composition, urbanicity, school SES, weight ma nagement education, school-level athletic participation, neighborhood affluence*school SES, neighborhood poverty*school SES, neighborhood affluence*weight manage ment education, and neighborhood poverty*weight management education.

PAGE 109

981211 1122 21()__ __ _*___ __ _*interceptjjinterceptneighaffljneighpovj neighracejurbanjschsesjweightj athleticsjneighafflschsesjneighafflneighpov neighraceurbanschsesweight athleticsneighaffl 2 1212 1212 12 12 12_*__* _*0000 () _()_ ()_ _*__* _*j neighpovschsesjjneighafflweightjj neighpovweightjjjj athletejjathlete biosexjjbiosex agejjageschses neighpovschsesneighafflweight neighpovweightbc 12 12/()/ () raceethjjraceeth sesjjses The intercept,intercept now represents the expected adjusted (for Level-1 predictors) BMI z -score when all Level-2 predictor variables are set to zero. More specifically,intercept is the expected adjusted BMI z -score for a student from a neighborhood with average affluence and po verty levels and no urbanicity or White residents and who attends an average SES sc hool with no weight management education and no student athletes. Each x represents the fixed effect of variable X that is assumed constant over all neighborhoods (e.g.,_ neighpov represents the effect of neighborhood poverty on BMI z -scores across all neighborhoods). Each x represents the fixed effect of variable X that is assumed c onstant over all schools (e.g.,_ s chses represents the effect of school SES on BMI z -scores across all schools). For the interactions, _*_ neighafflschses represents the moderati ng effect of school SES on neighborhood affluence (i.e., the rela tionship between neighborhood affluence and BMI z -scores may differ depending on the level of school SES). _*_ neighpovschses represents (16)

PAGE 110

99 the moderating effect of school SES on neighborhood poverty (i.e., the relationship between neighborhood poverty and BMI z -scores may differ depending on the level of school SES). _* neighafflweight represents the moderating effect of weight management education on neighborhood affluence (i.e., the relationship between neighborhood affluence and BMI z -scores may differ depending on the level of weight management education at an adolescent’s school). _* neighpovweight represents the moderating effect of weight management education on neighborhood poverty (i.e., the relationship between neighborhood poverty and BMI z -scores may differ depending on the level of weight management education at an adolescent’s school). The neighborhood residual,100 jb represents the neighborhood effect for neighborhood 1j (averaged across schools) while sta tistically controlling for all Level-2 predictors. The school residual,200 jc represents the school effect for school 2j (averaged across neighborhoods) while statistically cont rolling for all Level-2 predictors. Each x and x represent the same values as discussed in Equations 11 and 12. Model interpretation. To determine what percentage of adolescent academic achievement variance was among neighborho ods, what percentage was among schools, and what percentage was among adolescents within neighborhoods and schools, three different ICC values were calculated based on the results from the unconditional academic achievement model (Model 1AA). See Equations 17, 18, and 19 for more details on how each ICC was be calculated. Neighborhood ICC = 00 2 0000 b bc (17)

PAGE 111

100 School ICC = 00 2 0000 c bc (18) Neighborhood and School ICC = 0000 2 0000 bc bc (19) Next, to assess the relative strength of asso ciation between sets of independent variables and adolescent academic achievement, model pseudoR2 values were calculated for each academic achievement model. See Equations 20 to 24 for details on how the model pseudoR2 values were calculated for Model 2-AA, Model 3-AA, Model 4-AA, Model 5AA, and Model 6-AA, respectively. 0000100002 00001[[]] []bcModelAAbcModelAA bcModelAA (20) 0000100003 00001[[]] []bcModelAAbcModelAA bcModelAA (21) 0000100004 00001[[]] []bcModelAAbcModelAA bcModelAA (22) 0000100005 00001[[]] []bcModelAAbcModelAA bcModelAA (23)

PAGE 112

101 0000100006 00001[[]] []bcModelAAbcModelAA bcModelAA (24) After calculating the pseudoR2 values, a series of model pseudoR2 comparisons were made. First, to determine if the pro portion of variance accounted for by the set of neighborhood and school interactions was st atistically significantly above and beyond the main effects of neighborhood and school characteris tics, the pseudoR2 for Model 6-AA was compared to the pseudoR2 for Model 5-AA (Equation 25). Second, to determine if the proportion of variance accounted for by neighborhoods and sc hools together was statistically significantly great er than the proportion of variance accounted for by school characteristics alone, the pseudoR2 for Model 5-AA was compared to the pseudoR2 for Model 4-AA (Equation 26). Third, the pseudoR2 for Model 5-AA was compared to Model 3-AA to determine if the proportion of variance accounted for by neighborhoods and schools was statistically significantly greater than the proportion of variance accounted for by neighborhood characteristics alone (Equation 27). 2265(Pseudo-)(seudo-) M odelAAModelAARPR (25) 2254(Pseudo-)(seudo-) M odelAAModelAARPR (26) 2253(Pseudo-)(seudo-) M odelAAModelAARPR (27) Also, although the research questions did not focus on individual characteristics, to gain a more holistic understa nding of the data, the pseudoR2 for Model 5-AA was compared to the pseudoR2 for Model 2-AA to determine if the proportion of variance accounted for by neighborhoods and schools was st atistically signifi cantly greater than

PAGE 113

102 the proportion of variance accounted for by individual characteristics alone (Equation 28). The pseudoR2 for Model 2-AA also was examined to determine how much variability in adolescent academic achievement was accounted for by individual characteristics alone. 2252(Pseudo-)(seudo-) M odelAAModelAARPR (28) To determine if each of the abovementioned differences in pseudoR2 values were statistically significant, like lihood ratio tests were conducted on the difference between the -2 Log Likelihood values for each of the model comparisons. For example, to determine if the difference in the proporti on of variance accounted for between Model 6AA and Model 5-AA was statistically significan t, a likelihood ratio test was conducted on the difference between the -2 Log Li kelihood from Model 6-AA and the -2 Log Likelihood from Model 5-AA, where the degree s of freedom equaled the difference in the number of fixed effect parameters between the models. When the difference in model fit was statistically significant (i.e., the 2 statistic associated with the likelihood ratio test was statistically significant), then it wa s inferred that the difference in pseudoR2 values was statistically significant. Each model comparison was conducted at =.05. Lastly, in an effort to unpack furthe r the magnitude of the relationship among neighborhoods, schools, and adolescent acad emic achievement, the parameter estimates from Model 5-AA were also examined and tested for significance using =.05. Statistically significant parameter estimates from Model 5-AA were also transformed by dividing each obtained estimate by the AHPVT sample standard deviation, thereby, allowing interpretation of these estimates of predicted change in terms of standard

PAGE 114

103 deviation units. The results from the risk of obesity cross-classified random effects hierarchical linear models were examined an d interpreted following the same process, except for the parameter estimate transformation, as described for the academic achievement cr oss-classified random effects hierarchical linear models.

PAGE 115

104 Chapter Four Results Study Sample The sampling frame for this study consisted of adolescents in Grades 7 – 12 who participated in the Add Hea lth Wave I In-School Questionnaire and In-Home Interview; who attended regular, public middle and hi gh schools during the 1994-1995 school year; and who had data for all methodological variables ( n = 11,841). From this sampling frame, the study sample was then restricted to adolescents with complete data on substantive variables of interest related to the study and one randomly sampled sibling from families that had more than one child in the Add Health data. After applying the inclusion criteria, 10,860 adoles cents were included in the study sample. The adolescents in the study sample were dispersed across 99 schools (density = 5 to 1,135) and 1,111 neighborhoods (density = 1 to 189). As s hown in Table 8, there were no substantial characteristic differences of adolescents in the original sampling frame and those included in the study sample.

PAGE 116

105 Table 8 Unweighted Individual, Neighborhood, and School Characteristics for Original Sample and Study Sample Original sample ( n = 11,841) Study sample ( n = 10,860) % ( n ) % ( n ) Biological sex Female Male 51.92 (6147) 48.08 (5692) 51.55 (5598) 48.45 (5262) Race White/Asian Underserved minority 60.29 (7133) 39.71 (4699) 60.25 (6543) 39.75 (4317) M ( SD ) M ( SD ) Age 15.64 (1.70) 15.66 (1.68) Family SES -0.08 (0.76) -0.07 (0.75) Athletic Participation 1.03 (1.18) 1.04 (1.18) Neighborhood affluence -0.09 (0.87) -0.09 (0.86) Neighborhood poverty -0.08 (0.91) -0.08 (0.91) Neighborhood racial composition .76 (.28) .76 (.28) Urbanicity .56 (.48) .56 (.48) School SES -0.04 (0.73) -0.03 (0.73) Teacher education .44 (.27) .44 (.26) Student body racial composition .60 (.36) .60 (.36) Weight education .76 (.08) .76 (.08) School athletic participation .55 (.50) .55 (.50) Add Health Peabody Vocabulary Test 98.92 (14.77) 99.06 (14.62) Age-and-gender-adjusted BMI z -score 0.33 (0.92) 0.37 (0.88) In terms of missing data, the amount of missing data for each adolescent ranged from 0 to 13 variables ( M = 0.55, SD = 0.91).Overall, two-thirds of adolescents had no missing data and another 30% had missing data on one or two of the variables examined (Appendix C, Table C-1). Most of the phi co efficients (i.e., the correlations between missingness on pairs of variables) were within an acceptable range of -.02 to .35;

PAGE 117

106 however, a few slightly str onger correlations were observ ed (Appendix C, Figure C-1). The strongest associations between mi ssingness on two variables were found among Level-1 demographic variables. More speci fically, missingness on household income and parental education was th e strongest correlation ( = .90), followed by missingness on age and each of the five race variables ( = .51) and missingness on age and biological sex ( = .50). Because the missingness on these demographic variables did not appear to be random, caution was used when interpreting the parameter estimates for these variables, as well as the parameter estimates of composite variables that include any of the original variables (i.e., family SES) a nd the parameter estimates of other variables correlated with these demographic variable s. Conversely, no strong correlations were found between missingness and observed values ; correlation coefficients ranged from .15 to .19 (Appendix C, Figure C-2). Next, the data also were examined fo r possible correlations between a refusal response for the household income variable and other variables included in the study. After removing cases that were missing hous ehold income data and converting a refusal response for household income into missing, less than 1% of adolescents had missing data on more than two variables and 71% ha d no missing data (Appendix C, Table C-2). Unlike the strong correlati on between missing household in come and missing parental education, refusing to provide household income did not appear to be systematic (phi coefficients ranged from -.02 to .39; A ppendix C, Figure C-3). Similarly, no strong correlations were found between missingne ss and observed values; correlation coefficients ranged from -.13 to .20 (Appendix C, Figure C-4).

PAGE 118

107 Univariate Analyses To help inform the generalizability of the multivariate findings, both weighted and unweighted descriptive statistics were examined for level-1 variables and schoollevel variables. However, based on the Add Health study design, sample weights could not be used with neighborho od-level variables; therefor e, only unweighted descriptive statistics were calculated for neighborhood variables. As shown in Table 9, although the majority of differences between unweighted and weighted descriptive statistics were relatively small, differences in the race variab le were rather pronounced. Given this large difference and the inability to use sample weights with neighborhood-level variables, unweighted statistics were interpreted for all statistical analyses and findings are not considered generalizable at the national level. Table 9 Descriptive Statistics of Individual, Neighborhood, and School Characteristics (n = 10,860) Unweighted Statistics Weighted Statistics % ( n ) % ( n ) Biological sex Female Male 51.55 (5598) 48.45 (5262) 50.06 (5437) 49.94 (5423) Race White/Asian Underserved minority 60.25 (6543) 39.75 (4317) 74.08 (8045) 25.92 (2815) M ( SD ) M ( SD ) Age 15.66 (1.68) 15.35 (1.76) Family SES -0.07 (0.75) -0.04 (0.74) Athletic participation 1.04 (1.18) 1.10 (1.22) Neighborhood affluence -0.09 (0.86) NA Neighborhood poverty -0.08 (0.91) NA Neighborhood racial composition .76 (.28) NA Urbanicity .56 (.48) NA School SES -0.03 (0.73) -0.03 (0.76) Teacher education .44 (.26) .49 (.25)

PAGE 119

108Table 9 Descriptive Statistics of Individual, Neighborhood, and School Characteristics (n = 10,860) Unweighted Statistics Weighted Statistics Student body racial composition .60 (.36) .72 (.31) Weight education .76 (.08) .76 (.10) School athletic participation .55 (.50) .56 (.50) Add Health Peabody Vocabulary Test 99.06 (14.62) 100.85 (14.01) Age-and-gender-adjusted BMI z -score 0.37 (0.88) 0.37 (0.88) Overall, adolescents included in the st udy sample were primarily non-Hispanic White and non-Hispanic Asian (60%) and lived in slightly below-average SES households ( M = -0.07, SD = 0.75). There were slightly more girls than boys (52% vs. 48%) and the mean age was 15.66 years ( SD = 1.68). Also, on average, adolescents in the study sample reported participating in on e school sport. In terms of the criterion variables, the average achievement for adolescen ts in the study sample was slightly less than the Add Health standardized average of 100 ( M = 99.06, SD = 14.62). Conversely, for risk of obesity, the study sample had slig htly above average age-andgender-adjusted BMI scores ( M = 0.37, SD = 0.88). In terms of the neighborhoods where the study sample resided, on average, adolescents lived in neighborhoods with high proportions of White residents ( M = .76, SD = .28) and moderate levels of urbanicity ( M = .56, SD = .48). In terms of neighborhood socioeconomic status, adolescents in the study sample lived in neighborhoods with slightly below-average levels of affluence and slightly belowaverage levels of poverty ( M = -0.09. and -0.08., respectively). Similarly, adolescents in the study sample attended schools w ith slightly below-average SES ( M = -0.03, SD = 0.52). Regarding other school characteristics, on average, adolescents in the study sample

PAGE 120

109 attended schools with high prop ortions of White students ( M = .60, SD = .36), high levels of weight education ( M = .76, SD = .08), and moderate leve ls of masters educated teachers ( M = .44, SD = .26). Bivariate Analyses Correlation coefficients for the bivariate relationships between all of the variables included in the model ranged from -.002 to .78. Only 18 bivariate associations had absolute values equal to or greater than .30. Furthermore, of these 18 relationships, only 4 were between a criterion variable and a pr edictor variable; the other 14 were between pairs of predictor variables. For example, the academic achievement criterion variable (AHPVT) had four bivariate relationships st ronger than .30 or -.30 (neighborhood racial composition, .31; school racial composition, .3 4; family SES, .36; and race, -.32). No bivariate relationships between standardized age-and-gender-adjusted BMI were stronger than .30 or -.30. All of the bivariate asso ciations between predictor variables and ageand-gender-adjusted BMI z -scores had absolute values less than .10. The two strongest bivariate associations were between neigh borhood racial composition and school racial composition (.75) and between individual athlet ic participation and school-level athletic participation (.78). Table 10 contains the co mplete correlation matrix of criterion and predictor variables.

PAGE 121

110Table 10 Unweighted Bivariate Correlation Matrix for all Criterion and Predictor Variables (n = 10,860 AHPVT BMI Affluence Poverty Neigh racial comp Urbanicity Teacher education School racial comp School SES Age Family SES Biological sex Race Athlete School athletic BMI -.012 Affluence .166 -.092 Poverty -.121 .054 -.436 Neigh racial comp .314 -.066 .096 -.224 Urbanicity -.119 -.006 .232 -.002 -.197 Teacher education .130 -.028 .018 .126 .352 -.020 School racial comp .338 -.042 -.011 -.047 .748 -.384 .409 School SES .250 -.090 .587 -.238 .252 -.005 .138 .374 Age -.066 -.086 .006 -.036 -.068 .079 -.055 -.106 .032 Family SES .356 -.066 .337 .158 .182 -.033 .095 .216 .359 -.098 Biological sex .064 .056 .008 -.022 .018 -.008 -.002 .012 .011 .040 .040 Race -.321 .088 -.10 2 .192 -.506 .2 51 -.130 -.560 -.228 .041 -.25 8 -.022 Athlete .058 .006 .028 -.016 .084 -.087 .008 .150 .054 -.162 .132 .099 -.076 School athletics .066 .003 .026 -.026 .054 -.068 -.022 .112 .035 -.110 .126 .093 -.062 .782 Weight education .192 -.022 .018 -.080 .308 -.224 -.026 .418 .310 .134 .127 .022 -.263 .048 .065

PAGE 122

111 The bivariate relationships between vari ables included in the interaction terms examined in the CCREMs were also examined Overall, there was not much cross-over between the variables included in the interaction terms (e.g., the majority of youth that lived in high-affluent neighborhoods also attended high-SES schools). Plots of each of these relationships are pres ented in Figures 3 to 8. School SES -3 -2 -1 0 1 2 Affluence -2-1012345 Figure 3 School SES*neighbor hood affluence. Sixty-four percent of kids livi ng in low-affluent neighborhoods ( z -score < 0) attended low-SES schools ( z -score < 0). Sixty-seven percent of kids living in highaffluent neighborhoods ( z -score 0) attended high-SES schools ( z -score 0).

PAGE 123

112 School SES -3 -2 -1 0 1 2 Poverty -3-2-1012345 Figure 4 School SES*neighbor hood poverty. Forty-eight percent of kids living in non-poor neighborhoods ( z -score < 0) attended high-SES schools ( z -score 0). Fifty-two percent of kids living in poor neighborhoods ( z -score 0). attended low-SES schools ( z -score 0).

PAGE 124

113 Teacher Education 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Affluence -2-1012345 Figure 5 Teacher education*neighborhood affluence. Fifty-seven percent of kids livi ng in low-affluent neighborhoods ( z -score < 0) attended schools with low levels of teacher education (proportio n of teachers with graduate degree <.50). Forty-two percent of kids in high-affluent neighborhoods ( z -score 0) attended schools with high levels of teacher educati on (proportion of teachers with graduate degree .50).

PAGE 125

114 Teacher Education 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Poverty -3-2-1012345 Figure 6 Teacher education*neighborhood poverty. Thirty-seven percent of kids living in non-poor neighborhoods ( z -score < 0) attended schools with high levels of teach er education (proporti on of teachers with graduate degree .50). Fifty-two percent of ki ds living in poor neighborhoods ( z -score 0) attended schools with low levels of t eacher education (proportion of teachers with graduate degree <.50).

PAGE 126

115 Weight Promotion 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Affluence -2-1012345 Figure 7 Weight promotion*neighborhood affluence. Less than one percent of kids liv ing in low-affluent neighborhoods ( z -score < 0) attended low-weight promoting schools (a verage proportion of students who reported being taught about weight-relate d health topics < .50). Ninety-e ight percent of kids living in high-affluent neighborhoods ( z -score 0) attended high-weight promoting schools (average proportion of students who reported being taught about wei ght-related health topics .50).

PAGE 127

116 Weight Promotion 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Poverty -3-2-1012345 Figure 8 Weight promotion*neighborhood poverty. Ninety-eight percent of ki ds living in non-poor neighborhoods ( z -score < 0) attended high-weight promoting schools (ave rage proportion of students who reported being taught about weight -related health topics .50). Less than one percent of kids living in poor neighborhoods ( z -score 0) attended low-weight promoting schools (average proportion of students who reported being taught about wei ght-related health topics < .50). Multivariate Analyses Research questions were examined using cross-classified random effects hierarchical linear models (CCREMs) with individuals nested within schools and neighborhoods. However, before interpreting any multivariate analyses, data were screened for violations of assumptions associated with multilevel models.

PAGE 128

117 More specifically, data were examined for multicollinearity and Level-1 and Level-2 residuals, from models 5-AA and 5-RO, were screened for potential violations of normality, linearity, and homogeneity of variance. No assumptions appeared to be seriously violated when predicting academic ach ievement or risk of obesity; therefore, it was presumed reasonable to conduct the CCRE Ms for each criterion variable, using the modelbuilding strategy as presented in Ch apter 3. Tables and figures documenting the examination of assumptions are found in Appendix D. For academic achievement, tolerance values for all of the independent variables ranged from .28 to .99 (Appendix D, Table D-1). Thus, with the relatively weak zeroorder correlation coefficients among predic tor variables presented in Table 10 and acceptable tolerance values (Berry, 1993), th ere was no evidence of multicollinearity when predicting adolescent academic achievement. Examination of box-and-whisker plots and skewness and kurtosis values fo r Level-1 residuals and neighborhood and school residuals did not suggest serious violation of the normality assumption (Appendix D, Figures D-1, D-2, and D-3). More specifically, Level-1 re siduals and Level-2 school residuals were relatively normally distribu ted (sk = -0.37, ku = 1.72 and sk = -0.21, ku = 0.06, respectively; Appendix D, Figures D1 and D-3). However, although Level-2 neighborhood residuals were relatively symmetri c (sk = -0.44) they were also leptokurtic (ku = 7.83; Appendix D, Figure D-2). Last ly, an examination of Level-1, school-level, and neighborhood-level residuals plotte d against predicted values for academic achievement revealed no evidence of hetero scedasticity (Appendix D, Figures D-4, D-5, and D-6).

PAGE 129

118 Results from the examination of assumptio ns for predicting risk of obesity were similar to those found for academic achiev ement. Tolerance values for all of the independent variables used to predict adoles cent risk of obesity ranged from .49 to .99 (Appendix D, Table D-2). Thus, with the relatively weak zero -order correlation coefficients among predictor variables pres ented in Table 10 and acceptable tolerance values (Berry, 1993), there was no evidence of multicollinearity when predicting adolescent risk of obesity. Examination of box-and-whisker plots and skewness and kurtosis values for Level-1 and both Level2 residuals from Model 5-RO did not suggest serious violation of the normality assumption (Appendix D, Figures D-7, D-8, and D-9). More specifically, Level-1 residuals and school residuals were relatively normally distributed (sk = -0.32, ku = -0.58 and sk = 0.12, ku = 0.28, respectively; Appendix D, Figures D-7 and D-9), whereas neighborhood resi duals were relatively symmetric (sk = 0.49) but also leptokurtic (ku = 7.69; Appendix D, Figure D8). Lastly, scatter plots of Level-1, school-level, and neighborhood-level residuals plotted against predicted values for risk of obesity revealed no evidence of heteroscedasticity (Appendix D, Figures D-10, D-11, and D-12). Next, by plotting neighborhood residu als*neighborhood size for both academic achievement and risk of obesity, findings sugge st that the high kurtosis values for these residuals are driven by the singletons (i.e ., neighborhoods that contain only one adolescent). As shown in Appendix D, Figures D-13 and D-14, level-2 neighborhood residuals for neighborhoods with only one obs ervation are tightly clustered around zero. This is likely occurring because the residuals for singletons are pulled closer to zero more

PAGE 130

119 than other neighborhoods because the EB adjustment uses sampling error and neighborhoods with only one adolescent have oodles of sampling error in them. Thus, if the singletons contained more adolescents, the standard deviation of the neighborhood residuals would be larger and the ends of tails would not appear as extreme. Tables 11 14 contain summary resu lts from the academic achievement CCREMs and the risk of obesity CCREMs. The intracl ass correlations for academic achievement were relatively small (neighborhood ICC = .049, school ICC = .117, and within neighborhood and school ICC = .166) and the in traclass correlations for risk of obesity were minuscule (neighborhood ICC = .008, school ICC = .014, and within neighborhood and school ICC = .022). Using results from th e model-building process, each of the four research questions are answered below. Research Question 1. To what extent are neighborhood influences on U.S. middle and high school students’ academic achi evement moderated by school environments? Based on the results from the academic achievement CCREMs, the data do not suggest a moderating relationship between these neighborhood and school characteristics in relation to U.S. middle and high school students’ academic achievement. Not only were none of the parameter estimates for the four neighborhood*school interactions statistically significant (Model 6-AA, Table 12), but the change in pseudoR2 values between Model 6-AA and Model 5-AA also was not statistically significant (Table 11). Thus, inclusion of these interaction terms did not account for a greater proportion of variance in academic achievement than individual, neighborhood, and school main effects. Given these results, Model 5-AA was used as the complete academic

PAGE 131

120 achievement CCREM when interpreting the a cademic achievement multivariate findings. See Table 11 for more details about each of the model pseudoR2 comparisons. Table 11 Model Pseudo-R2 Comparisons for Academic Achievement CCREMs Model 6 -AA to Model 5-AA Model 5-AA to Model 4-AA Model 5-AA to Model 3-AA Model 5-AA to Model 2-AA Model 2-AA to Model 1-AA pseudoR2 .003 .028* .073* .272* .585* -2 log likelihood (obtained2 ) 2.4 55.9 36.5 133.5 1108.8 fixed effects (DF) 4 4 3 7 4 2 critical value 9.49 9.49 7.82 14.07 9.49 p <.05 PsuedoR2 Model 6-AA = .862 PsuedoR2 Model 5-AA = .858 PsuedoR2 Model 4-AA= .830 PsuedoR2 Model 3-AA= .785 PsuedoR2 Model 2-AA = .585 PsuedoR2 Model 1-AA = .000 Note: Model 6-AA = Neighborhood, school, and interaction model Model 5-AA = Neighborhood and school main effects model Model 4-AA = School model Model 3AA = Neighborhood model Model 2-AA = Level-1 control model Model 1-AA = Unconditional model Research Question 2. What are the relative influences of neighborhood and school environments on U.S. middle and high school students’ academic achievement? Because the proportion of variance accounted for by neighborhood and school characteristics together was statistically si gnificantly greater than the proportion of variance accounted for by school characteristi cs alone and neighborhood characteristics

PAGE 132

121 alone (Model 5-AA to Model 4-AA and Mode l 5-AA to Model 3-AA in Table 11), the relative influences of neig hborhood and school environments on U.S. middle and high school students’ academic achievement were determined by examining the parameter estimates from Model 5-AA. However, be fore discussing neighborhood and school’s relative influences, it is important to note th at after controlling for all Level-2 predictors, the variability in average achievement acr oss neighborhoods, averaged across schools, and the variability in av erage achievement across schools, averaged across neighborhoods, both remained statistically significant (001.64b and 00 c = 3.30, respectively). Thus, although the proportion of variance accounted for by neighborhood and school characteristics together was statistically significantly greater than the proportion of variance accounted for by each environment alone, the neighborhood and school variables used in this study did not account for all the variability in average adolescent academic achievement among environments. Also, to help the interpretation of th e relationships between neighborhood, school, and individual characteristics and adolescent academic achievement, the obtained parameter estimates from Model 5-AA were divided by the sample standard deviation of AHPVT scores, thereby allowing the observed relationships to be discussed in terms of predicted standard deviation changes in a dolescent academic achievement. Similarly, to ease the interpretation of va riables scaled as proportio ns (e.g., neighborhood racial composition, urbanicity, and student body raci al composition), parameter estimates were multiplied by .10, thus transforming a conceptual unit for these variables to equal 10%. For example, the parameter estimate for ne ighborhood racial composition from Model 5-

PAGE 133

122 AA (4.54) was first multiplied by .10 and th en divided by the study sample standard deviation (14.62) yielding the interpreted value 0.03. Table 12 Summary Table for Academic Achievement CCREMs (n = 10,860) Model 1-AA Model 2-AA Model 3-AA Model 4-AA Model 5-AA Model 6-AA Fixed Effects Intercept 99.57* (0.56) 100.49* (0.42) 93.36* (1.20) 95.60* (0.77) 92.08* (1.23) 91.43* (1.75) Age 0.54* (0.10) -0.45* (0.09) -0.43* (0.08) -0.40* (0.08) -0.40* (0.08) Biological sex 1.46* (0.24) 1.44* (0.24) 1.44* (0.24) 1.44* (0.24) 1.44* (0.24) Race -4.65* (0.32) -3.98* (0.33) -4.22* (0.32) -3.76* (0.33) -3.76* (0.33) Family SES 4.90* (0.18) 4.75* (0.18) 4.84* (0.18) 4.70 (0.18) 4.70* (0.18) Neighborhood affluence 1.13* (0.24) 0.998* (0.26) 0.89 (0.49) Neighborhood poverty 0.20 (0.22) 0.08 (0.21) 0.46 (0.40) Neighborhood racial composition 6.79* (0.92) 4.54* (1.02) 4.31* (1.04) Urbanicity -1.58* (0.54) -1.12* (0.50) -1.08* (0.50) School SES 1.40* (0.34) 0.84* (0.38) 0.20 (1.07) School-level teacher education 0.23 (1.01) 0.15 (1.01) 1.93 (3.02) Student body racial composition 7.16* (0.88) 4.93* (1.08) 5.18* (1.08) Neighborhood affluence* school SES -0.03 (0.28) Neighborhood poverty* school SES 0.25 (0.28) Neighborhood affluence* school-level teacher education 0.22 (0.88)

PAGE 134

123Table 12 Summary Table for Academic Achievement CCREMs (n = 10,860) Model 1-AA Model 2-AA Model 3-AA Model 4-AA Model 5-AA Model 6-AA Neighborhood poverty* school-level teacher education -0.90 (0.80) Error Variance Level-1 175.91* (2.46) 162.40* (2.26) 162.34* (2.25) 162.66* (2.26) 162.40* (2.25) 162.40* (2.25) Intercept (Neighborhood) 10.37* (1.74) 2.86* (0.84) 1.80* (0.70) 2.60* (0.78) 1.64* (0.66) 1.62* (0.66) Intercept (School) 24.64* (4.38) 11.65* (2.24) 5.72* (1.41) 3.33* (0.96) 3.30* (0.96) 3.23* (0.96) Model Fit AIC 87529.9 86429.1 86340. 1 86357.5 86309.6 86315.2 BIC 87521.9 86413.1 86316. 1 86335.5 86279.6 86277.2 Statistically significant--variance estimate and intercept, p <.05. For fixed effects tested in blocks, test for block of fixed effects p <.05 and test for individual fixed effect p <.05. Values based on SAS Proc Mixed. Entries show parameter estimates with standard errors in parentheses. Neighborhood ICC = .049 School ICC = .117 Neighborhood and school ICC = .166 In terms of individual neighborhood char acteristics and adolescent academic achievement, three of the four neighborh ood characteristics (affluence, racial composition, and urbanicity) were statistically significantly associated with adolescent academic achievement after controlling for i ndividual and school characteristics (Model 5-AA, Table 12). The only neighborhood variable not associated with academic achievement was neighborhood poverty. More specifically, for every one standard deviation increase in neighborhood affluence, AHPVT scores were predicted to increase 0.07 standard deviations while controlling for other neighborhood variables and school and individual characteristics. Also, for ev ery 10% increase in White residents in a

PAGE 135

124 neighborhood, AHPVT scores were predicte d to increase 0.03 standard deviations. Conversely, for every 10% increase in resi dents living in urban areas within a neighborhood, AHPVT scores were predicted to decrease 0.008 standard deviations. In terms of school characteristics, both student body racial composition and school SES were statistically significantly associated with adolescent academic achievement, while controlling for individual and neighborhood characteristics (Model 5AA, Table 12). For every 10% increase in White students at a school, AHPVT scores were predicted to increase 0.03 standard devi ations. In addition, for every one standard deviation increase in school SES, AHPVT scores were predicted to increase 0.06 standard deviations. After controlling for individual and neighborhood characteristics, school-level teacher education was not statis tically significantly re lated to adolescent academic achievement. Next, regarding individual-level variables and adolescent academic achievement, the proportion of variance accounted for in academic achievement through the simultaneous inclusion of individual, neighborho od, and school variables was statistically significantly greater than the proporti on of variance accounted for by individual characteristics alone (Model 5-AA to Model 2-AA in Table 11). Also, unlike neighborhoods and schools, all four indivi dual control variables were statistically significant predictors of adolescent a cademic achievement after controlling for neighborhood and school contexts (Model 5-AA, Table 12). More specifically, AHPVT scores among traditionally underserved racial mi nority adolescents were predicted to be 0.26 standard deviations below non-Hispanic White and non-Hispanic Asian adolescents

PAGE 136

125 and girls were predicted to achieve 0.10 st andard deviations below boys. Also, for every year increase in age, adolescents were predic ted to achieve 0.02 standard deviations less. Conversely, in terms of family SES, for ever y one standard deviation increase in SES, AHPVT scores were predicted to increase 0.32 standard deviations. Lastly, when examined alone, approximately 58% of the variability in adolescent academic achievement was accounted for by individual characteristics alone (Model 2-AA to Model 1-AA in Table 11). Research Question 3. To what extent are neighborhood influences on U.S. middle and high school students’ risk of obes ity moderated by school environments? Based on the results from the risk of obesity CCREMs, the data do not suggest a moderating relationship between these ne ighborhood and school characteristics in relation to U.S. middle and high school studen ts’ risk of obesity. As presented in Table 13, the change in pseudoR2 values between Model 6-RO and Model 5-RO was not statistically significant. Thus inclusion of these interact ion terms did not account for a greater proportion of variance in risk of obesity than individual, neighborhood, and school main effects. Given these results, Model 5-RO was used as the complete risk of obesity CCREM when interpreting the risk of obesity multivariate findings. See Table 13 for more details about each of the model pseudoR2 comparisons.

PAGE 137

126Table 13 Model Pseudo-R2 Comparisons for Risk of Obesity CCREMs Model 6-RO to Model 5-RO Model 5-RO to Model 4-RO Model 5-RO to Model 3-RO Model 5-RO to Model 2-RO Model 2-RO to Model 1-RO pseudoR2 .063 .075* .028 .275* .494* -2 log likelihood (obtained2 ) 6.5 24.0 7.1 52.4 210.8 fixed effects (DF) 4 4 3 7 5 2 critical value 9.49 9.49 7.82 14.07 11.07 p <.05 PsuedoR2 Model 6-RO = .833 PsuedoR2 Model 5-RO = .770 PsuedoR2 Model 4-RO= .694 PsuedoR2 Model 3-RO= .742 PsuedoR2 Model 2-RO = .494 PsuedoR2 Model 1-RO =.000 Note: Model 6-RO = Neighborhood, school, and interaction model Model 5-RO = Neighborhood and school main effects model Model 4-RO = School model Model 3-RO = Neighborhood model Model 2-RO = Level-1 control model Model 1-RO = Unconditional model Research Question 4. What are the relative influences of neighborhood and school environments on U.S. middle and high school students’ risk of obesity? Understanding the relative influences of neighborhood and school environments on U.S. middle and high school students’ ri sk of obesity was more challenging than it was for adolescent academic achievement. For example, when the pseudoR2 value from Model 5-RO was compared to the pseudoR2 for Model 4-RO, the proportion of variance accounted for by neighborhood and school char acteristics together was statistically significantly greater than the proportion of va riance accounted for by school characteristics alone (Table 13). However, when the pseudoR2 value from Model 5-RO

PAGE 138

127 was compared to the pseudoR2 for Model 3-RO, the proportion of variance accounted for by neighborhood and school ch aracteristics together was no t statistically significantly greater than the proportion of variance accounted for by neighborhood characteristics alone (Table 13). Thus, these model comp arisons suggest that after controlling for neighborhood and individual characteristics, school characteristics do not uniquely contribute to the proportion of variance accoun ted for in adolescent risk of obesity. Based on the findings from the model comp arisons, the selection of the best risk of obesity model for the interpretation of parameter estimates was less straightforward than model selection for academic achievemen t. However, in terms of the research questions investigated in this study, th e parameter estimates from Model 5-RO (representing the relationships between risk of obesity and school f actors after adjusting for neighborhood factors, and the relationshi ps between risk of obesity and neighborhood factors after adjusting for sc hool factors) best addressed the fourth research question. Furthermore, although the proportion of va riance accounted for in Model 5-RO was not statistically significantly gr eater than was the proportion of variance accounted for in Model 3-RO, at = .05 level, Model 5-RO was a better fitting model in the sample than Model 3-RO (BICModel 5-RO = 27,682.1, BICModel 3-RO = 27,689.2; Table 14).

PAGE 139

128Table 14 S ummary Table for Risk of Obesity CCREMs (n = 10,860) Model 1-RO Model 2-RO Model 3-RO Model 4-RO Model 5-RO Model 6-RO Fixed Effects Intercept 0.38* (0.02) 0.25* (0.02) 0.46* (0.06) -0.008 (0.09) 0.22 (0.12) -0.42 (0.37) Age -0.05* (0.006) -0.05* (0.005) -0.05* (0.005) -0.05* (0.005) -0.05* (0.005) Biological sex 0.11* (0.02) 0.11* (0.02) 0.11* (0.02) 0.11* (0.02) 0.11* (0.02) Race 0.15* (0.02) 0.12* (0.02) 0.14* (0.02) 0.13* (0.02) 0.12* (0.02) Family SES -0.05* (0.01) -0.03* (0.01) -0.04* (0.01) -0.03* (0.01) -0.03* (0.01) Athletic participation -0.004 (0.007) -0.003 (0.007) -0.001 (0.01) -0.001 (0.01) -0.001 (0.01) Neighborhood affluence -0.07* (0.01) -0.06* (0.01) 0.02 (0.11) Neighborhood poverty -0.001 (0.01) -0.001 (0.01) 0.21* (0.10) Neighborhood racial composition -0.07 (0.04) -0.08* (0.04) -0.08 (0.04) Urbanicity -0.01 (0.02) -0.006 (0.02) -0.0006 (0.02) School SES -0.08* (0.02) -0.04 (0.02) 0.02 (0.06) Weight education 0.36* (0.12) 0.29 (0.13) 1.09* (0.48) School athletic participation -0.008 (0.02) -0.008 (0.03) -0.009 (0.02) Neighborhood affluence* school SES -0.01 (0.01) Neighborhood poverty* school SES -0.02 (0.02) Neighborhood affluence* weight education -0.10 (0.14) Neighborhood poverty* weight education -0.27 (0.13)

PAGE 140

129Table 14 S ummary Table for Risk of Obesity CCREMs (n = 10,860) Model 1-RO Model 2-RO Model 3-RO Model 4-RO Model 5-RO Model 6-RO Error Variance Level-1 0.76* (0.01) 0.74* (0.01) 0.74* (0.01) 0.74* (0.01) 0.74* (0.01) 0.74* (0.01) Intercept (Neighborhood) 0.006* (0.002) 0.003 (0.002) 0.002 (0.002) 0.003 (0.002) 0.002 (0.002) 0.002 (0.002) Intercept (School) 0.01* (0.003) 0.006* (0.002) 0.002* (0.002) 0.002 (0.002) 0.002 (0.002) 0.001 (0.001) Model Fit AIC 27953.3 27752.5 27715. 2 27730.1 27714.1 27715.6 BIC 27945.3 27734.5 27689. 2 27706.1 27682.1 27675.6 Statistically significant--variance estimate and intercept, p <.05. For fixed effects tested in blocks, test for block of fixed effects p <.05 and test for individual fixed effect p <.05. Values based on SAS Proc Mixed. Entries show parameter estimates with standard errors in parentheses. Neighborhood ICC = .008 School ICC = .014 Neighborhood and school ICC = .022 After controlling for individual variable s and school factors, neighborhood affluence and racial composition were sta tistically significantly associated with adolescent risk of obesity and neighborhood poverty and urbanicity were not (Model 5RO, Table 14). As with the academic achievement models, to ease the interpretation of variables scaled as proportions, parameter estimates were multiplied by .10, thus transforming a conceptual unit for these vari ables to equal 10 %. More specifically, for every one standard deviation increase in neighborhood affluence, adolescent BMI z scores were predicted to decrease 0.06 st andard deviations. Similarly, for every 10% increase in White residents in a neighborhood, adolescent BMI z -scores were predicted to decrease 0.008 standard deviations. Furthe rmore, after controlling for all individual,

PAGE 141

130 neighborhood, and school predictors, the residual variation between neighborhoods (00 b = 0.002 ) and between schools was close to zero (00 c = 0.002). Thus, it appears that the variables included in Model 5-RO account ed for most of the neighborhood and school variability in adolescent BMI z -scores. In terms of school factors, after controlling for individual and neighborhood characteristics, th e school factors examined do not appear to have a statistically significant relationship to U.S. middle and high school students’ risk of obesity (Model 5-RO, Table 14). Next, regarding individual-level variab les and adolescent risk of obesity, the proportion of variance accounted for in risk of obesity through the simultaneous inclusion of individual, neighborhood, and school variab les was statistically significantly greater than the proportion of variance accounted for by individual characteristics alone (Model 5-RO to Model 2-RO in Table 13). After adju sting for neighborhood and school factors, all individual-level variables were statistica lly significantly associated with adolescent risk of obesity except for adolescent athle tic participation (Model 5-RO, Table 14). More specifically, standardized age-and-gender-a djusted BMI for a traditionally underserved racial minority adolescent was predicted to be 0.13 standard deviations above nonHispanic White and non-Hispanic Asian adolescents, and boys were predicted to have standardized age-and-gender-adjusted BMI valu es 0.11 standard deviations above girls. Also, for every year increase in age, sta ndardized age-and-gender-adjusted BMI was predicted to decrease 0.05 standard deviati ons. A similar relationship was observed for family SES; for every one standard deviat ion increase in SES, standardized age-andgender-adjusted BMI was predicted to decr ease 0.03 standard deviations. Lastly, when

PAGE 142

131 examined alone, approximately 49% of the variability in adolescent risk of obesity was accounted for by individual characteristics alon e (Model 2-RO to Model 1-RO in Table 13). Summary of Findings Adolescents included in the study sample did not appear to be substantially different from adolescents included in th e original sampling frame. However, when sampling weights were used, the difference between the weighted and unweighted race frequencies was rather pronounced. Thus, all statistical analyses were unweighted and findings are not considered generalizable at the national level. In terms of the relationships betw een neighborhood, school, and individual characteristics and adolescent academic achievement and risk of obesity, bivariate relationships among all of the variables incl uded in the study were relatively weak. Similarly, albeit the data suggest several neighborhood and school characteristics were statistically significantly asso ciated with adolescent academic achievement and risk of obesity, the magnitude of the relationships was small. Likewise, the data also do not suggest any moderating relationships be tween the neighborhood and school characteristics examined in this study. Regarding the relative association betw een neighborhood factors and academic achievement, neighborhood affluence, racial composition, and urbanicity appeared to have statistically significant unique relati onships with adolescent achievement after controlling for individual, school, and other neighborhood characteristics. Similarly, two school factors (student body racial compositio n and school SES) evidenced statistically

PAGE 143

132 significant unique relationships with adoles cent achievement after controlling for other factors. Conversely, when examining the re lative associations between neighborhood and school factors, in relation to adolescent risk of obesity, neighborhood affluence and racial composition were the only characteristics that appeared to have statistically significant unique relationships with adolescent risk of obesity after controlling for individual, school, and other neighb orhood characteristics. However, results from this study need to be interpreted with caution. For example, given the systematic missingness of two of the variable s included in the standardized family SES composite variable (household income and parental education), the relationships among family SES and adol escent academic achievement and risk of obesity need to be interpreted with caution. The same caution needs to be used when interpreting the relati onships between neighborhood affluence and school SES and both criterion variables as these two predictor vari ables were correlated with family SES. Lastly, there was little variation in adol escent academic achievement or risk of obesity across neighborhoods and schools; th us, even though Model 5-AA and Model 5RO accounted for 86% and 77% of the variance in academic achievement and risk of obesity, respectively, it is important to remember that these pseudoR2 values represent the proportion of explainable variance, not to tal variance accounted for. For example, the pseudoR2 value for Model 5-AA (.86) does not repres ent the proportion of total variance accounted for in adolescent academic achievement. Instead, Model 5-AA accounts for 86% of explainable variance (35.01) in adolescent academic achievement.

PAGE 144

133 Chapter Five Discussion Using data from the National Longitudi nal Study of Adolescent Health (2005c) and the Adolescent Health and Academic Ac hievement study (n.d.), the purpose of the current study was to examine simultaneously neighborhood and school influences on academic achievement and adolescent risk of obesity and to examine the moderating effects of schools on these outcomes. To help f ill the gap in social de terminants literature related to adolescent academic achievement and risk of obesity, four specific research questions were investigated: Research Question 1. To what extent are neighborhood influences on U.S. middle and high school students’ academic achievement moderated by school environments? Research Question 2. What are the relative infl uences of neighborhood and school environments on U.S. middle and high school students’ academic achievement? Research Question 3. To what extent are neighborhood influences on U.S. middle and high school students’ risk of ob esity moderated by school environments? Research Question 4. What are the relative infl uences of neighborhood and school environments on U.S. middle and high school students’ risk of obesity? The following sections contain a summary of the findings, limitations of the study, implications for the field, directions for future research, and overall conclusions.

PAGE 145

134 Summary of Findings Neighborhoods, schools, and academic achievement. Results from the academic achievement CCREMs do not suggest a moderating relationship between the neighborhood and school environments examined in this study. In terms of each environment’s relative relationship with middle and high school students’ academic achievement, three neighborhood characteris tics (neighborhood affluence, racial composition, urbanicity) and two school characteristics (student body racial composition, school SES) appear to have statistically sign ificant unique relationships with adolescent achievement after controlling for indivi dual and other neighborhood and school characteristics. In relation to the social determinants literature and previous findings related to neighborhoods, schools, and adolescent academic achievement, findings from the current study both complement and contradi ct findings from other published studies. For example, the statistically sign ificant positive relationship between neighborhood affluence and academic achieveme nt and the statistica lly non-significant association between neighborhood poverty and achievement are consistent with other non-experimental research findings (Boyle et al., 20 07; Brooks-Gunn et al., 1997a; Dornbusch et al., 1991; Halpern-Felsher et al., 1997; Leventhal & Brooks-Gunn, 2000). Yet, these associations also contradict fi ndings from previous experimental and quasiexperimental studies that did not reveal statistically significant improvements in adolescent academic achievement based on higher neighborhood socioeconomic levels (Kling & Liebman, 2004; Leventhal et al ., 2005; Rosenbaum, 1995). Similarly, the statistically significant positive asso ciation found between neighborhood racial

PAGE 146

135 composition (i.e., proportion of White residents) and academic achievement in the current study contradicts Blau et al .’s (2001) statistically non-s ignificant findings between neighborhood diversity and social studies achievement. Another contradiction with the literatur e is the magnitude of the neighborhood ICC for academic achievement from the current study. Unlike Boardman and Saint Onge (2005) who reported a relatively large neighbor hood ICC based on Add Health data (.25), the neighborhood ICC for academic achievement in the current study was minuscule (.049). Differences in model specifications a nd the sample used to calculate the ICCs are plausible explanations for the variation in ICC values. For example, not only did Boardman and Saint Onge (2005) use a trad itional two-level hier archical model to generate ICC values whereas the current study used a cross-classified two-level model, but the ICC values they report were not derived from an unconditional model as was undertaken in the current study. Instead, the I CC values were genera ted from models that statistically controlled for a host of level-1 factors such as race, age, gender, family structure, and maternal educati on (Boardman & Saint Onge, 2005). In terms of school sociodemographic characteristics and adolescent academic achievement, findings from the current study ar e more consistent with Coleman et al.’s (1966) findings than with findings from more recent studies (i.e., Caldas & Bankston, III, 1997; Everson & Millsap, 2004; Lee & Croninger, 1994). For example, even though the current study revealed sta tistically significant associat ions between school SES and student body racial composition and adolesce nt academic achievement, the magnitude of these associations was negligible, thus lending support to Coleman et al.’s (1966)

PAGE 147

136 conclusion that after accounting for family background characteristics and general social context, school sociodemographic characterist ics have little rela tionship with academic achievement. The lack of a statistically signi ficant association betw een teacher education and academic achievement in the current study also lends support to Coleman et al.’s (1966) findings and contradicts findings fr om more recent studies (i.e., DarlingHammond, 1999; Greenwald et al., 1996). Neighborhoods, schools, and risk of obesity. Results from the risk of obesity CCREMs do not suggest a moderating relati onship between the neighborhood and school environments examined in this study. In te rms of each environment’s relative relationship with risk of obesity, two neighborhood characteristics (n eighborhood affluence, racial composition) appear to have statistically signi ficant unique relationships with adolescent risk of obesity after controlling for individual, school, and other neighborhood characteristics. After control ling for individual and neighbor hood characteristics, none of the three school factors examined in th is study had statistically significant unique relationships with adolescent risk of obesit y. In relation to the social determinants literature and previous findings related to ne ighborhoods, schools, and adolescent risk of obesity, findings from the current study are not directly comparable to other published studies. More specifically, because most th e neighborhood and school factors examined in the current study are different than those in cluded in other studies a direct comparison of findings cannot be made. Nonetheless, some general, common elements among studies can be discussed.

PAGE 148

137 For example, the statistically sign ificant negative association between neighborhood affluence and adolescent risk of obesity in the current study both supports and contradicts Chen and Paterson’s (2006) findings on neighborhood SES and high school students’ BMI. The statistically si gnificant negative asso ciation found between neighborhood affluence and age-and-gender-adjusted BMI z -scores supports Chen and Paterson’s (2006) results of neighborhood e ducation and employment as statistically significant negative predictors of adolescent BMI; however, it contradicts their findings that neighborhood income and assets were not statistically significant predictors of BMI. The relationship between neighborhood affluen ce and adolescent risk of obesity in the current study also contradicts Kling and Li ebman’s (2004) results of no statistically significant differences in adolescent obes ity status between Moving to Opportunity adolescents whose families moved to low-poverty neighborhoods and those who remained in impoverished urban housing projects. As with academic achievement, the magnitude of the neighborhood ICC for adolescent risk of obesity also is much smaller than the neighborhood ICC for risk of being overweight reported by Boardman an d Saint Onge’s (.008 vs. .05, respectively; 2005). Similarly, as with the academic achievement models, differences in model specifications and the sample used to calculate the ICCs are probable explanations for the observed differences. Differences in how risk of obesity was operationalized also could be related to the different neighborhood ICC values. The lack of a statistically significant a ssociation between urbanicity and age-andgender-adjusted BMI z -scores in the current study also can be viewed as supporting and

PAGE 149

138 contradicting previous resear ch findings. For example, the lack of a statistically significant relationship between urbani city and age-and-gender-adjusted BMI z -scores contradicts Ewing et al.’s ( 2006) cross-sectional findings regarding urban sprawl and adolescent weight status. However, the findings from the current study support their longitudinal findings regarding urban sp rawl and adolescent weight status. As with most of the neighborhood and risk of obesity literature, results related to school characteristics and adolescent risk of obesity both support and contradict previous findings. More specifically, the lack of any statistically significant school characteristic and age-and-gender-adjusted BMI z -scores from the current study contradicts O’Malley et al.’s (2007) findings on school SES an d adolescent BMI. However, the magnitude in school ICC for adolescent risk of obesity in the current study is not considerably smaller than the school ICC for risk of obesity reported by O’Malley et al. (.014 vs. .03, respectively; 2007). Limitations of the Study As with all secondary data analyses this study has several methodological limitations. First and foremost is the issue of variable selection and model misspecification. Not only were limited variable s available related to adolescent risk of obesity, but the quality of so me of the variables that we re available was poor. For example, the variable that was used to assess adolescent participation in physical education classes was only asked of studen ts who completed their In-Home Interview during the active academic year; thus, this variable had more legitimate skips than completed responses. Therefore, even though this information is possibly an important

PAGE 150

139 factor in understanding adolescent risk of obesity, the large amount of missing data precluded its inclusion in the study. A lthough it is unclear why the Add Health researchers only asked the physical educat ion class question to students interviewed during the academic year, the overall lack of variables related to adolescent risk of obesity could be related to th e age of the data, which is another limitation of the study. Wave I Add Health data were collected more than10 years ago (1994-1995). Since that time, not only have neighbor hoods and schools likely changed, but the questionnaire and interview items were likely re lated to the pressing health issues of the early 1990’s, which are not the same as the pressing issues of today. For instance, the current childhood and adolescent obesity epidemic was just beginning to be noticed in the 1990’s. Thus, because obesity was not a public health priority when Add Health was designed and initially implemented, it is no t surprising that the data contain little information that can be used to assess factor s associated with obesit y. If Add Health were conducted today, the focus of the questions would likely be very different (e.g., the recently funded National Children’s Study focu s on understanding social and biological factors associated with obes ity; The National Children’s St udy, 2007). Possible areas of interest that might be examined today incl ude detailed questions related to average caloric intake (e.g., keeping a 2-week food journal), adolescent perceptions about the weight status of their friends, family, and st udents at their schools, and attitudes and beliefs towards weight and body image issues. Furthermore, because cross-classified random effects models can only be used with continuous criterion variables, adolescen t risk of obesity had to be operationalized

PAGE 151

140 differently in this study compared to othe r studies (i.e., age-and-gender-adjusted BMI z scores were used instead of a more traditi onal dichotomous risk/no-risk variable based on age-and-gender-adjusted BMI pe rcentiles). In this manner, although the risk of obesity results from the current study are not directly comparable to findings from studies in which the risk of obesity was operationalized as falling above or below a specific BMI percentile, they are not completely disparat e either. Variables included in the current study had similar bivariate correlations with risk of obesity operationalized as age-andgender-adjusted BMI z -scores ( r1 ) and with risk of obesity operationalized as age-andgender specific BMI 85th percentile ( r2 ; Table 15). After applying the Fisher z transformation, all of the effect sizes for the differences between r1 and r2 were well below Cohen’s (1988) guidelines for a small effect size when comparing correlation coefficients ( q = .10; Table 15). In addition, the correlation between ageand-gender-adjusted BMI z -scores and the dichotomous risk of obesity measure was .74. Therefore, although the difference in how risk of obesity was operationalized in the current study should be noted, the results from the current study need not be considered in complete isol ation from other stud ies that operationalize adolescent risk of obesity as age-and-gender specific BMI 85th percentile.

PAGE 152

141Table 15 Correlation Coefficient Comparisons for Different Adolescent Risk of Obesity Measures (n = 10,860) BMI z -score ( r1 ) BMI 85th percentile ( r2 ) Cohen’s q AHPVT -.012 -.016 0.004 Affluence -.092 -.082 -0.010 Poverty .054 .045 0.009 Neighbor racial comp -.066 -.061 -0.005 Urbanicity -.006 -.009 0.004 Teacher education -.028 -.018 -0.010 School racial comp -.042 -.043 0.001 School SES -.090 -.085 -0.005 Age -.086 -.048 -0.039 Family SES -.066 -.071 0.005 Biological sex .056 .060 -0.003 Race .088 .062 0.026 Athlete .006 -.040 0.047 School athletics .003 -.039 0.042 Weight education -.022 -.038 0.015 Add Health data also only contain two measures of academic achievement—GPA calculated from self-reported grades in Englis h, mathematics, science, and social studies and AHPVT scores, both of which have thei r own limitations. For example, because the lack of standardization in sc hool grades was a serious limitation in using them as a single measure of academic achievement, AHPVT scores were used as a measure of adolescent academic achievement in the current study. However, although this variable is a standardized measure of academic achievement, no reliability or validity studies on this version of the PPVT are available from Add Heal th researchers. Furthermore, it too, is a single measure of achievement at one point in time. An additional limitation of the study is the use of census tracts to operationalize neighborhoods. In doing so, neighborhood m easures included in the study were very

PAGE 153

142 broad and likely did not contain data related to the significant areas of an adolescent’s neighborhood that shape his or her daily experiences. Therefor e, even though the findings from the study help advance our understa nding of neighborhoods’ unique influences on adolescent academic achievement and risk of obesity, findings are still limited to administratively defined neighborhoods. Thus the study does not contribute to our understanding of how smaller, more immediate neighborhood environments might influence adolescent well-being and whether schools moderate these influences. The relatively low correlation among va riables included in the neighborhood poverty composite also is a limitation of the current study. Although the selection of variables used to create the neighborhood poverty composite variable was informed by poverty composites used in previous research (i.e., Duncan & Aber, 1997; Leventhal & Brooks-Gunn, 2003), in this study, these three variables did not appe ar to represent the underlying poverty construct well. More sp ecifically, the propor tion of female-headed households in a neighborhood was not highly correlated with the proportion of families living below the federal poverty level or w ith the proportion of unemployed adults. Thus, even though historically researchers have of ten conceptualized female-headed households as an indicator of poverty, for these data the proportion of female-headed households does not appear to be an accurate component of neighborhood poverty. A further limitation of the study pertai ns to the small neighborhood ICC values and the proportion of singletons (i.e., a neighborhood unit containing only one adolescent) included in the study. The neighborhood ICCs for both academic achievement and risk of obesity were very small (.049 and .008); however, it is unknown

PAGE 154

143 if the variance in adolescent academic achievement and risk of obesity across neighborhoods is truly that small, or if the proportion of singleton neighborhoods (.45) might be diminishing these values. More specifically, with singletons, there is no clustering at the neighborhood level, theref ore, there is no neighborhood variance for these adolescents, which, in turn, co uld be suppressing the neighborhood ICCs. Furthermore, just as we do not know the im pact of having high proportions of singleton neighborhoods, the structure of the Add Health data does not allow for an examination of the degree to which schools are or are not nested in neighborhoods. Theoretically, we would expect some students to attend school in their neighborhoods, whereas other students attend schools not in their neighb orhoods. However, the data do not provide information about which schools are in whic h neighborhoods; thus, it is not possible to determine how many students attended school outside their neighborhoods. The generalizability of findings is anothe r limitation of this study. Not only could sampling weights not be used in the multivaria te analyses, thus prohibiting the results to be generalized to a national level, but, even if sampling weights could have been used, Add Health data do not contain weights at th e neighborhood level. Thus, even though the Add Health schools and sample of adoles cents were selected to be nationally representative, the neighborho ods were not selected to be nationally representative. Therefore, any findings at the neighbor hood level cannot be generalized beyond the sample of adolescents incl uded in the study and their corresponding neighborhoods. The age of the data also requires caution in th e generalizability of findings. For example, the neighborhood and school influences examin ed in the current study do not necessarily

PAGE 155

144 relate to today’s neighborhoods and school s. Instead, they rela te to neighborhood conditions in 1990 and school conditions in 1994-1995. Lastly, even with its many limitations, to date, Add Health data are still the best source for researchers interested in examinin g the relationships between social contexts and adolescent well-being. Although there are many secondary data sources that contain information related to adolescent development and well-being, none include the vast array of individual and contextual data av ailable from Add Hea lth. Thus, albeit not perfect, Add Health’s large sample size and focus on multiple social contexts allows researchers to apply advanced analytic techni ques that other data sources cannot support. Implications for the Field The most notable implication of the curre nt study is its addition to the social determinants literature. By examining simultaneously neighborhood and school environments in relation to adolescent academic achievement and risk of obesity, findings from the current study are likely less biased than are previous findings because the CCREMs used in the current study allo wed for the examination of the unique contributions of each environment. However, even though the current study contributes to our understanding of each environment’s unique relationship with achievement and risk of obesity, given the correlational design of the current study, results from the current study cannot be used to guide policies or programs related to adolescent development. Instead, the strongest implications for the fiel d of social and behavioral science are best discussed in terms of future research.

PAGE 156

145 Directions for Future Research Although the findings from the current study have made an important contribution to the social determinants lit erature, there is still much work to be undertaken in furthering our understanding of neighborhood and school influences on adolescent development and well-being. For example, the criterion variables examined in the current study (academic achievement a nd risk of obesity) were two of many developmental outcomes that might be in fluenced by various neighborhood and school factors. Thus, future research needs to utilize CCREMs to investigate other important social, physical, intellectual, and emotional outcomes. Similarly, just as the criterion variables included in the current study we re two of many possible outcomes to be examined, the neighborhood and school fact ors included in the current study also represent a small proportion of neighborhood and school characteristics that could have been examined. Consequently, as future re search uses CCREMs to investigate different developmental outcomes, it should also inve stigate different neighborhood and school characteristics in relation to these other outcomes. Other neighborhood and school variables that should be examined include those that are more perceptual in nature versus ad ministratively measured variables taken from the census. For example, at the neighborhood leve l, potential variables to be investigated in future research include social capital, social norms regarding health and education, residents’ perceived neighborhood quality/d ilapidation, researchers’ observed neighborhood quality/dilapidation, and an in dex of perceived vs. observed neighborhood quality/dilapidation. At the sch ool level, potential va riables to examine in future studies

PAGE 157

146 include perceived weight status of close fr iends and of students at school, perceived racism, time spent on instruction, school connectedness, and overall academic climate of the school. Future research also is needed to be gin to investigate and understand possible mechanisms behind the relationships am ong neighborhood affluence, adolescent academic achievement, and risk of obes ity. Although the relationship between neighborhood affluence and age-and-gender-adjusted BMI z -scores had not been previously examined, the association between adolescent academic achievement and neighborhood affluence is consistent, albeit weaker, with findings from other social determinants research. Therefore, it seems a ppropriate for future research to further our understanding of these complex social pr ocesses by examining the mechanisms behind these relationships. Qualitative research would be especially useful in this area. For example, future researcher could take a phe nomenological approach to understanding the mechanisms behind neighborhood affluence a nd adolescent wellbeing. In doing so, future researchers would be able to capt ure the meaning of th e lived experience of adolescents in their neighb orhoods (Creswell, 1998). From a methodological perspective, futu re research should focus on several areas. First, given the weak correlations am ong the variables used to operationalize neighborhood poverty, future research should investigate a better composite variable for neighborhood poverty. Second, future research should investigate how much impact using CCREMs had, using Add Health data, compared to the traditional misspecified two-level model with adolescen ts only nested in schools. Given the large proportion of

PAGE 158

147 singletons and low neighborhood ICCs found in the current study, accounting for the theoretical cross-classification of the data might have had little impact on the relationships examined. Third, future large-sc ale studies need to be designed using a better sampling design such that the da ta are nationally representative of both neighborhoods and schools. These better desi gned large-scale studies also need to provide links between neighborhoods and sch ools, thereby allowing researchers to evaluate the extent to which youth are cr oss-classified between neighborhoods and schools. In addition, to allow future resear chers to be able to conduct mixed methods research using secondary data, future large-sc ale studies need to include more than the typical close-ended quantitative items; they n eed to include qualitative, open-ended items that can be used in conjunction with the more traditional quantitative items. Conclusions Bronfenbrenner's (1979) Ecological Systems Theory posits that human development is influenced by the interrelations among settings in which a person actively participates (e.g., family, school, neighborhood s, religious institutions); thus, to study human development effectively, we need to look beyond a single environment and analyze the interactions among multiple en vironments. Although this study did not discover any statistically significant interactions among neighborhood and school characteristics, it was the first to inve stigate school and neighborhood influences simultaneously using national data and cro ss-classified random effects hierarchical models. Thus, findings from the current study are important contributions to the social determinants literature as they are the firs t to present neighborhood associations with

PAGE 159

148 adolescent academic achievement and risk of obesity while statistically controlling for school characteristics and vice versa. Howeve r, given the relatively small magnitude of many of the relationships found in the curre nt study, it is imperative for social and behavioral scientist to continue to explor e the complex relations hips between various social environments and adolescent deve lopment and well-being, while employing proper statistical techniques. Lastly, given the limitations of the curre nt study, the findings do not completely answer the research questions. More specifi cally, the correlational design and model misspecification of the current study prohibit fi ndings from being interpreted as “relative influences.” Instead, the findings should be vi ewed as adding another piece to the social determinants research puzzle. In this fashi on, findings from the current study can be used in conjunction with previous research findin gs to help advance our knowledge of social determinants of adolescent development a nd well-being along the causality continuum. For example, the consistency with findings related to neighborhood affluence underscores the importance of this social c onstruct in the development of achievement and health. Therefore, as more researchers us e findings from the current study to guide new investigations of these complex relationships, policymakers and community leaders will be better informed as they continue to work towards eliminating education inequity and health disparities.

PAGE 160

149 References Adolescent Health and Academ ic Achievement Study. (n.d.). Adolescent Health and Academic Achievement Study Retrieved October 30, 2006, from http://www.prc.utexas.edu/ahaa/index.html Allison, P. D. (2002). Missing data Thousand Oaks, CA: Sage. Armor, D. J. (1972). School and family effects on Black and White achievement: A reexamination of the USOE data. In F. Mosteller & D. P. Moynihan (Eds.), On equality of educational opportunity: Pape rs deriving from the Harvard University faculty seminar on the Coleman Report (pp. 168-229) New York: Random House. Baker, S. R., Robinson, J. E., Danner, M., J., E., & Neukrug, E. S. (2001, April). Community Social Disorganization Theo ry applied to adolescent academic achievement Paper presented at the meeting of the American Educational Research Association, Seattle, WA. (E RIC Document Repro duction Service No. ED453301) Barron, R. M., & Kenny, D. A. (1986). The mo derator-mediator variable distinction in Social Psychological research: Concep tual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 1173–1182.

PAGE 161

150 Beale Spencer, M., Cole, S. P., Jones, S. M., & Phillips Swanson, D. (1997). Neighborhood and family influences on young urban adolescents’ behavior problems: A multisample, multisite analysis. In J. Brooks-Gunn, G. J. Duncan, & J. L. Aber (Eds.), Neighborhood poverty: Vol. I. Context and consequences for children (pp. 200–218). New York: Russell Sage Foundation. Berry, W. D. (1993). Understanding regression assumptions Newbury Park, CA: Sage. Blau, J. R., Lamb, V., L., Stearns, E., & Pe llerin, L. (2001). Cosmopolitan environments and adolescents’ gains in social studies. Sociology of Education, 74, 121–138. Boardman, J. D., & Saint Onge, J. M. (2005). Neighborhoods and adolescent development. Children, Youth and Environments, 15, 138–164. Bowen, N. K., & Bowen, G. L. (1999). Effect s of crime and violence in neighborhoods and schools on the school behavior and performance of adolescents. Journal of Adolescent Research, 14, 319–342. Bowen, N. K., Bowen, G. L., & Ware, W. B. (2002). Neighborhood social disorganization, families, and the educational behavior of adolescents. Journal of Adolescent Research, 17, 468–490. Boyle, M. H., Georgiades, K., Racine, Y., & Mustard, C. (2007). Neighborhood and family influences on educational attainment: Results from the Ontario Child Health Study Follow-Up 2001. Child Development, 78, 168–189. Bronfenbrenner, U. (1979). The ecology of human development Cambridge, MA: Harvard University Press.

PAGE 162

151 Brooks-Gunn, J., Duncan, G. J., & Aber, J. L. (Eds.). (1997a). Neighborhood poverty: Vol. I. Context and consequences for children New York: Russell Sage Foundation. Brooks-Gunn, J., Duncan, G. J., & Aber, J. L. (Eds.). (1997b). Neighborhood poverty: Vol. II. Policy implications in studying neighborhoods New York: Russell Sage Foundation. Caldas, S. J., & Bankston, C., III. (1997). Effect of school population socioeconomic status on individual academic achievement. The Journal of Educational Research, 90, 269–277. Carter, R. C. (2002). The impact of public schools on childhood obesity. JAMA 288, 2180. Centers for Disease Contro l and Prevention. (2000a). 2 to 20 years: Boys body mass index-for-age percentiles Retrieved June 13, 2007, from http://www.cdc.gov/nchs/data/nhanes/gr owthcharts/set1clin ical/cj41c023.pdf Centers for Disease Contro l and Prevention. (2000b). 2 to 20 years: Girls body mass index-for-age percentiles Retrieved June 13, 2007, from http://www.cdc.gov/nchs/data/nhanes/gr owthcharts/set1clin ical/cj41c024.pdf Centers for Disease Contro l and Prevention. (2007). About BMI for children and teens Retrieved August 10, 2007, from http://www.cdc.gov/nccdphp/dnpa/bmi/ch ildrens_BMI/about_childrens_BMI.htm #How%20is%20BMI%20calculated

PAGE 163

152 Chantala, K. (2006). Guidelines for analyzing Add Health data Retrieved June 6, 2007, from http://www.cpc.unc.edu/projec ts/addhealth/files/wt_guidelines.pdf Chantala, K., & Tabor, J. (1999). Strategies to perform a design-based analysis using Add Health data. Retrieved May 20, 2007, from http://www.cpc.unc.edu/projects/addhealth/files/weight1.pdf Chase-Lansdale, P. L., Gordon, R. A., Brooks-Gunn, J., & Klebanov, P. K. (1997). Neighborhood and family influences on the intellectual and behavioral competence of preschool and early school -age children. In J. Brooks-Gunn, G. J. Duncan, & J. L. Aber (Eds.), Neighborhood poverty: Vol. I. Context and consequences for children (pp. 79–118). New York: Russell Sage Foundation. Chen, E., & Paterson, L. Q. (2006). Neighb orhood, family, and subjective socioeconomic status: How do they relate to adolescent health? Health Psychology, 25, 704–714. Cohen, D. A., Finch, B. K., Bower, A., & Sastry, N. (2006). Collective efficacy and obesity: The potential influence of social factors on health. Social Science & Medicine, 62, 769–778. Coleman, J. S., Campbell, E. Q., Hobson, C. J., McPartland, J., Mood, A. M., Weinfeld, F. D., et al. (1966). Equality of educational opportunity Washington, DC: U.S. Government Printing Office.

PAGE 164

153 Connell, J. P., & Halpern-Felsher, B.L. (1997). How neighborhoods affect educational outcomes in middle childhood and adolescence: Conceptual issues and an empirical example. In J. Brooks-Gunn, G. J. Duncan, & J. L. Aber (Eds.), Neighborhood poverty: Vol. 1. Context and consequences for children (pp. 174– 199). New York: Russell Sage Foundation. Creswell, J. (1998). Qualitative inquiry and research design: Choosing among five traditions Thousand Oaks, CA: Sage. Crosnoe, R. (2004). Social capital and the interplay of families and schools. Journal of Marriage and Family, 66, 267–280. Crosnoe, R., & Muller, C. (2004). Body mass index, academic achievement, and school context: Examining the educational experiences of adolescents at risk of obesity. Journal of Health and Social Behavior 45 393–407. Darling-Hammond, L. (1999). Teacher quality and student achievement: A review of state policy evidence. University of Washington, Center for the Study of Teaching and Policy. Retrieved October 30, 2006, from http://depts.washington.edu/ctpmail/Reports.html#TeacherQuality Dietz, W. H., & Gortmaker, S. L. (2001). Pr eventing obesity in children and adolescents. Annual Review of Public Health, 22, 337–353. Diez-Roux, A. V. (2003). The examination of neighborhood effects on health: Conceptual and methodological issues relate d to the presence of multiple levels of organization. In I. Kawachi & L. F. Berkman (Eds.), Neighborhoods and Health (pp. 45–64). New York: Oxford University Press.

PAGE 165

154 Dornbusch, S. M., Ritter, L. P., & Steinber g, L. (1991). Community influences on the relation of family status to adolescent school performance: Differences between African Americans and non-Hispanic whites. American Journal of Education, 38 543–567. Duncan, G. J., & Aber, J. L. (1997). Neighb orhood models and measures. In J. BrooksGunn, G. J. Duncan, & J. L. Aber (Eds.), Neighborhood poverty: Vol. I. Context and consequences for children (pp. 62–78). New York: Russell Sage Foundation. Dunn, L. M., & Dunn, L. M. (1981). Peabody Picture Vocabulary Test-Revised: Manual. Circle Pines, MN: American Guidance Service. Dyer, H. S. (1972). The measurement of educa tional opportunity. In F. Mosteller & D. P. Moynihan (Eds.), On equality of educational opportunity (pp. 513–527). New York: Random House. Eamon, M. K. (2005). Social-demographi c, school, neighborhood, and parenting influences on the academic achievement of Latino young adolescents. Journal of Youth and Adolescence, 34, 163–174. Enders, C., K., & Tofighi, D. (2007). Center ing predictor variables in cross-sectional multi-level models: A new look at an old issue. Psychological Methods, 12, 121– 138. Erikson, E. H. (1963). Childhood and society (2nd ed.). New York: Norton. Everson, H. T., & Millsap, R. E. (2004). Be yond individual differences: Exploring school effects on SAT scores. Educational Psychologist, 39, 157–172.

PAGE 166

155 Ewing, R., Brownson, R., C., Berrigan, D. ( 2006). Relationship between urban sprawl and weight of United States youth. American Journal of Preventive Medicine, 31, 464–474. Frazier, P. A., Tix, A. P., & Barron, K. E. (2 004). Testing moderator and mediator effects in counseling psychology research. Journal of Counseling Psychology, 51, 115– 134. French, S. A., Story, M., & Jeffery, R. W. (2001). Environmental influences on eating and physical activity. Annual Review of Public Health, 22, 309–335. Games, P. A. (1990). Correlation and causation: A logical snafu. Journal of Experimental Education, 58, 239–246. Gill, M. G., Ashton, P., & Algina, J. (2004). Authoritative schools: A test of a model to resolve the school effectiveness debate. Contemporary Educational Psychology, 29 389–409. Goodman, E., Hinden, B. R., & Khandelwal, S. (2000). Accuracy of teen and parental reports of obesity and body mass index. Pediatrics, 106, 52–58. Gordon-Larsen, P., Nelson, M. C., Page, P., & Popkin, B. M. (2006). Inequality in the built environment underlies key health disp arities in physical activity and obesity. Pediatrics, 117, 417–424. Gortmaker, S., L., Peterson, K., Wiecha, J., S obol, A. M., Dixit, S., Fox, M. K., & Laird, N. (1999). Reducing obesity via a school -based interdisciplinary intervention among youth: Planet Health. Archives of Pediatrics & Adolescent Medicine, 153, 409–418.

PAGE 167

156 Greenwald, R., Hedges, L. V., & Laine, R. D. (1996). The effect of school resources on student achievement. Review of Educational Research, 66, 361–396. Halpern-Felsher, B. L., Connell, J. P., Beale Spencer, M., Aber, J. L., Duncan, G. J., Clifford, E., et al. (1997). Neighborhood and family factors predicting educational risk and attainment in African American and White children and adolescents. In J. Brooks-Gunn, G. J. Duncan, & J. L. Aber (Eds.), Neighborhood poverty: Vol. 1. Context and consequences for children (pp. 146–173). New York: Russell Sage Foundation. Henderson, C. L., Buehler, A. E., Stein, W. L., Dalton, J. E., Robinson, T. R., & Anfara, V. A., Jr. (2005). Organizational health and student achievement in Tennessee middle schools. NASSP Bulletin, 89, 54–75. Holland, P. W. (1986). Statis tics and causal inference. Journal of the American Statistical Association, 81, 945–960. Hox, J. J. (2002). Multilevel analysis: Techniques and applications Mahwah, NJ: Lawrence Erlbaum. Hoy, W. K., & Hannum, J. W. (1997). Middl e school climate: An empirical assessment of organizational health and student achievement. Educational Administration Quarterly, 33, 290–311. Janssen, I., Boyce, W. F., Simpson, K., & Pick ett, W. (2006). Influence of individual-and area-level measures on socioeconomic status on obesity, unhealthy eating, and physical inactivity in Canadian adolescents. American Journal of Clinical Nutrition, 83, 139–145.

PAGE 168

157 Jeynes, W. H. (2002). A meta-analysis of th e effects of attending religious schools and religiosity on Black and Hispanic academic achievement. Education and Urban Society, 35, 27–49. Kling, J. R., & Liebman, J. B. (2004, May). Experimental analysis of neighborhood effects on youth Cambridge, MA: National Bureau of Economic Research. Retrieved July 2, 2007, from http://www.nber.org/~kling/mto/483.pdf Krieger, N. (2001). A glossary for social epidemiology. Journal of Epidemiology and Community Health, 55, 693–700. Kromrey, J. D., & Hines, C. V. (1994). Nonr andomly missing data in multiple regression: An empirical comparison of co mmon missing-data treatments. Educational and Psychological Measurement, 54, 573–593. Lee, V. E., & Croninger, R. G. (1994). The re lative importance of home and school in the development of literacy skills for middle-grade students. American Journal of Education, 102, 286–329. Lee, V. E., Smith, J. B., & Croninger, R. G. (1997). How high school organization influences the equitable di stribution of learning in mathematics and science. Sociology of Education, 70, 128–150. Leventhal, T., & Brooks-Gunn, J. (2000). The neighborhoods they live in: The effects of neighborhood residence on child and adolescent outcomes. Psychological Bulletin 126 309–337. Leventhal, T., & Brooks-Gunn, J. (2003). Children and youth in neighborhood contexts. Current Directions in Psychological Science, 12, 27–31.

PAGE 169

158 Leventhal, T., Fauth, R. C., & Brooks-Gunn, J. (2005). Neighborhood poverty and public policy: A 5-year follow-up of children’s educational outcomes in the New York City Moving to Opportunity Demonstration. Developmental Psychology, 41, 933– 952. Marsh, H. W., & Yeung, A. S. (1997). Ca usal effects of academic self-concept on academic achievement: Structural equa tion models of longitudinal data. Journal of Educational Psychology, 89, 41–54. McIntyre, S., & Ellaway, A. (2003). Neighbo rhoods and health: An overview. In I. Kawachi & L. F. Berkman (Eds.), Neighborhoods and health (pp. 20–42). New York: Oxford University Press. Mish et al. (Eds.). (2004). The Merriam-Webster dictionary Springfield, MA: MerriamWebster. Muijs, R. D. (1997). Predictors of academ ic achievement and academic self-concept: A longitudinal perspective. British Journal of Educational Psychology, 67, 263–277. Muller, C., Pearson, J., Riegle-Crumb, C., Ha rris Requejo, J., Frank, K. A., Schiller, K. S., et al. (2007a). National Longitudinal Study of Adolescent Health Wave III education data: Contextual codebook Retrieved June 10, 2007, from http://www.prc.utexas.edu/ahaa/data/EduContext.pdf

PAGE 170

159 Muller, C., Pearson, J., Riegle-Crumb, C., Ha rris Requejo, J., Frank, K. A., Schiller, K. S., et al. (2007b). National Longitudinal Study of Adolescent Health Wave III education data: Design and implementation of the Adolescent Health and Academic Achievement Study Retrieved June 10, 2007, from http://www.prc.utexas.edu/ ahaa/data/edudesign.pdf National Longitudinal Study of Adolescent Health. (2004a). Context explored in Add Health Retrieved May 25, 2007, from http://www.cpc.unc.edu/projects/ addhealth/design/ contexts.html National Longitudinal Study of Adolescent Health. (2004b). Design focus Retrieved May 20, 2007, from http://www.cpc.unc. edu/projects/addhealth/design_focus National Longitudinal Study of Adolescent Health. (2004c). Design focus – Wave I Retrieved May 25, 2007, from http://www.cpc.unc.edu/projects/addhealth/design_focus/wave1 National Longitudinal Study of Adolescent Health. (2004d). User guides Retrieved June 7, 2007, from http://www.cpc.unc.e du/projects/addhealth/pubs/guides National Longitudinal Study of Adolescent Health. (2005a). About public use data Waves I and II Retrieved June 15, 2007, from http://www.cpc.unc.edu/project s/addhealth/data/publicdata National Longitudinal Study of Adolescent Health. (2005b). Design facts at a glance Retrieved May 25, 2007, from http://www.cpc.unc.edu/proje cts/addhealth/designfacts

PAGE 171

160 National Longitudinal Study of Adolescent Health. (2005c). National Longitudinal Study of Adolescent Health Retrieved October 30, 2006, from http://www.cpc.unc.edu/addhealth Nelson, M. C., Gordon-Larsen, P., Song, Y., & Popkin, B. M. (2006). Built and social environments: Associations with adolescent overweight and activity. American Journal of Preventive Medicine, 31, 109–117. Neumark-Sztainer, D., Story, M., Hannan, P. J ., Stat, M., & Rex, J. (2003). New Moves: A school-based obesity prevention program for adolescent girls. Preventive Medicine, 37, 41–51. Norman, G. J., Nutter, S. K., Ryan, S., Sallis, J. F., Calfas, K. J., & Patrick, K. (2006). Community design and access to recreationa l facilities as correlates of adolescent physical activity and body-mass index. Journal of Physical Activity and Health, 3 (Suppl. 1), 118–128. O’Malley, P. M., Johnston, L. D., Delva, J., Ba chman, J. G., & Schulenberg, J. E. (2007). Variation in obesity among American secondary school students by school and school characteristics. American Journal of Preventive Medicine, 33 (4S), S187S194. Onwuegbuzie, A. J. (2003). Expanding the fram ework of internal and external validity in quantitative research. Research in the Schools, 10, 71–89.

PAGE 172

161 Plybon, L. E., Edwards, L., Butler, D., Belgrave, F. Z., & Allison, K. W. (2003). Examining the link between neighborhood cohesion and school outcomes: The role of support coping among Afri can American adolescent girls. Journal of Black Psychology, 29, 393–407. Powell, L. M., Auld, M. C., Chaloupka, F. J ., O’Malley, P. M., & Johnston, L. D. (2007). Associations between access to food stores and adolescent body mass index. American Journal of Preventive Medicine, 33 (4S), S301-S307. Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear mode ls: Applications and data analysis methods (2nd ed.). Thousand Oaks, CA: Sage. Rosenbaum, J. E. (1995). Changing the geography of opportunity by expanding residential choice: Lessons from the Gautreaux Program. Housing Policy Debate, 6, 231–269. Retrieved July 23, 2007, from http://www.fanniemaefoundati on.org/programs/hpd/pdf/hpd_0601_ rosenbaum.pdf Rubin, D. B. (1986). Which ifs have causal answers [Comment]. Journal of the American Statistical Association, 81, 961–962. Rubin, D. B. (1996). Multiple imputation after 18+ years. Journal of the American Statistical Association, 91, 473–489. Sallis, J. F., McKenzie, T. L., Conway, T. L., El der, J. P., Prochaska, J. J., Brown, M., et al. (2003). Environmental interventions for eating and physical activity: A randomized controlled trial in middle schools. American Journal of Preventive Medicine, 24, 209–217.

PAGE 173

162 Sanders, M. G. (1998). The effects of school, family, and community support on the academic achievement of African American adolescents. Urban Education, 33, 385–409. SAS Institute Inc. (2003). SAS/STAT (Version 9.1.3) [Computer software]. Cary, NC: SAS Institute Inc. Schafer, J. L., & Graham, J. W. (2002). Miss ing data: Our view of the state of the art. Psychological Methods, 7, 147–177. Scott, M. M., Cohen, D. A., Evenson, K. R., El der, J., Catellier, D., Ashwood, J. S., et al. (2007). Weekend schoolyard accessibili ty, physical activity, and obesity: The Trial of Activity in A dolescent Girls (TAAG) study. Preventive Medicine, 44, 398–403. Seaward, B. L. (1999). Managing stress: Principles and strategies for health and wellbeing (2nd ed.). Boston: Jones and Bartlett. Shadish, W. R., Cook, T. D., & Campbell, D. T. (2002). Experimental and quasiexperimental designs for gene ralized causal inference Boston: Houghton Mifflin. Shaw, C. R., & McKay, H. D. (1942). Juvenile delinquency and urban areas Chicago: University of Chicago Press. Sobel, M. E. (1995). Causal inference in the social and behavioral sciences. In G. Arminger, C. C. Clogg, & M. E. Sobel (Eds.), Handbook of statistical modeling for the social and behavioral sciences (pp. 1–37). New York: Plenum Press. Story, M., Kaphingst, K. M., & French, S. (2006). The role of schools in obesity prevention. The Future of Children, 16 (1), 109–142.

PAGE 174

163 Subramanian, S. V., Jones, K., & Duncan, C. (2003). Multilevel methods for public health research. In I. Kaw achi & L. F. Berkman (Eds.), Neighborhoods and Health (pp. 65–111). New York: Oxford University Press. Sweetland, S., R., & Hoy, W. K. (2000). School characteristics and educational outcomes: Toward an organizational m odel of student achievement in middle schools. Educational Administration Quarterly, 36, 703–729. The National Children’s Study. (2007). Selection of outcome and exposure measures Retrieved August 17, 2007, from http://www.nationalchildrensstudy.gov/res earch/research_plan/upload/Chapter%2 07%20062007.pdf Tourangeau, R., & Shin, H. C. (1999). National Longitudinal Study of Adolescent Health: Grand sample weight Retrieved May 20, 2007, from http://www.cpc.unc.edu/projects/addhealth/files/weights.pdf U.S. Census Bureau. (2000). Appendix A: Census 2000 geographic terms and concepts Retrieved August 15, 2007, from http://www.census.gov/geo/www/tiger/glossry2.pdf U.S. Census Bureau. (2007). How the Census Bureau measures poverty (official measure) Retrieved August 17, 2007, from http://www.census.gov/hhes/www/poverty/povdef.html Wentzel, K. R. (1998). Social relationships and motivation in middle school: The role of parents, teachers, and peers. Journal of Educational Psychology, 90, 202–209.

PAGE 175

164 Wentzel, K. R. (2002). Are effective teach ers like good parents? Teaching styles and student adjustment in early adolescence. Child Development, 73, 287–301. White, K. R. (1982). The relation between socioeconomic status and academic achievement. Psychological Bulletin, 91, 461–481. Wickrama, K. A., Wickrama, K. A. S., & Br yant, C. M. (2006). Community influences on adolescent obesity: Race/ethnic differences. Journal of Youth and Adolescence 35, 647–657. Williams, T. R., Davis, L. E., Miller Crib bs, J., Saunders, J., & Williams, J.H. (2002). Friends, family, and neighborhood: Understanding academic outcomes of African American youth. Urban Education, 37, 408–431. Zand, D. H., & Thomson, N. R. (2005). Academic achievement among African American adolescents: Direct and indirect effects of demographic, individual, and contextual variables. Journal of Black Psychology, 31, 352–368.

PAGE 176

165 Appendix A: Summary Tables of Previous Neighborhood and School Research

PAGE 177

166 Table A-1 Summary of Neighborhood Influen ces on Adolescent Academic Achievement Research Studies Authors Sample Analytic Technique Academic Achievement Operationalization Neighborhood Operationalization NeighborhoodLevel Variables IndividualLevel Variables Other Variables HalpernFelsher et al. (1997) 11to 16year old African American youth in Atlanta OLS regression National percentile ranking from the Iowa Test of Basic Skills 1980 census tracts Low SES, high SES, male joblessness, family concentration, and ethnic diversity Family income, family structure, and mother’s education, grade in school HalpernFelsher et al. (1997) 12to 15year old White and African American students in an upstate New York urban school district OLS regression Educational risk behavior composite variable that included information on attendance, standardized achievement tests, suspensions, old for grade or recommendation for retention, and two or more core courses were failed in the previous academic year 1980 census tracts Low SES, high SES, male joblessness, family concentration, and ethnic diversity Eligible for reduced price/free lunch

PAGE 178

167Table A-1 Summary of Neighborhood Influen ces on Adolescent Academic Achievement Research Studies Authors Sample Analytic Technique Academic Achievement Operationalization Neighborhood Operationalization NeighborhoodLevel Variables IndividualLevel Variables Other Variables HalpernFelsher et al. (1997) 15to 20year old White and African American students in an upstate New York urban school district OLS regression Educational risk behavior composite variable that included information on attendance, standardized achievement tests, suspensions, old for grade or recommendation for retention, and two or more core courses were failed in the previous academic year 1980 census tracts Low SES, high SES, male joblessness, family concentration, and ethnic diversity Eligible for reduced price/free lunch Dornbusch et al. (1991) High school students in six San Francisco Bay Area schools OLS regression Adjusted selfreported grades in school on a 4point scale U.S. census tracts (year not specified) Community socioeconomic status and community ethnic composition Parental education, family structure, ethnicity, and gender Family process variables: style, parental involvement, decision making, and parental reactions to grades

PAGE 179

168Table A-1 Summary of Neighborhood Influen ces on Adolescent Academic Achievement Research Studies Authors Sample Analytic Technique Academic Achievement Operationalization Neighborhood Operationalization NeighborhoodLevel Variables IndividualLevel Variables Other Variables HalpernFelsher et al. (1997) 10to 16year old White and African American youth in New York City, Baltimore, and Washington, D.C. OLS regression Combined reading and math standardized test scores 1980 census tracts Low SES, high SES, male joblessness, family concentration, and ethnic diversity Family poverty, no father in the home Rosenbaum (1995) High school youth whose families participated in the Gautreaux Program Not stated – was more of an evaluation report High school GPA Not specified – was a comparison between “suburban movers” and “city movers” Suburban movers were families who moved out of the inner city housing projects and into one of 115 suburbs in the six-county area surrounding Chicago City movers were families who moved out of the inner city housing projects and Neighborhood type – urban or suburban Not sure, nothing included in the report

PAGE 180

169Table A-1 Summary of Neighborhood Influen ces on Adolescent Academic Achievement Research Studies Authors Sample Analytic Technique Academic Achievement Operationalization Neighborhood Operationalization NeighborhoodLevel Variables IndividualLevel Variables Other Variables into “revitalized” low-income Black neighborhoods within the city limits Kling & Liebman (2004) Teenage youth (aged 1520) whose families participate d in the MTO program in Baltimore, Boston, Chicago, Los Angles, and New York City OLS regression Woodcock-Johnson reading and mathematics test performance Not clearly stated. Only provided general information on the different treatment and control groups Experimental group could only move to census tracts with a 1990 poverty rate less than 10 % Section 8 group could move to any neighborhood Control group was not allowed to live in Section 8 housing – they remained in the housing projects Poverty level Gender and baseline characteristics (race, gifted classes, special education classes, behavior problems, health problems, school discipline experiences) Leventhal et al. (2005) Youth aged 14-19 whose families OLS regression Self-reported grades in school on a 5point scale Experimental group status – low-poverty group, traditional voucher group, and Fraction poor, fraction rental units, fraction Black, fraction Age, gender, parental characteristics including age,

PAGE 181

170Table A-1 Summary of Neighborhood Influen ces on Adolescent Academic Achievement Research Studies Authors Sample Analytic Technique Academic Achievement Operationalization Neighborhood Operationalization NeighborhoodLevel Variables IndividualLevel Variables Other Variables participated in the New York City MTO program control group Latino, fraction White race, education, employment status, marital status, and number of children in the household Baker et al. (2001) 8th-grade students in the state of Virginia Structural equation modeling Aggregated mean scores on three subtests (reading, language, and mathematics) of the Stanford 9 School district boundaries Economic condition, social organization, and children’s environment Bowen et al. (2002) Nationally representative sample of middle and high school students Structural equation modeling Self-reported grades in school Not defined administratively – youths’ subjective view of their neighborhood Perceived neighborhood support, perceptions of prosocial behaviors of neighborhood peers, and perceptions of neighborhood crime and violence Race/ethnicity and family poverty Supportive parenting and parental educational support Eamon (2005) Latino adolescents aged 10 to 14 whose mothers participated in Hierarchical OLS regression Peabody Individual Achievement Test reading comprehension and mathematics scores Not defined administratively – mothers’ subjective view of their neighborhoods. Overall neighborhood quality Latino origin, gender, age, LEP, maternal characteristics (age when had first child, years of education Youth’s ratings of school environment and parenting processes

PAGE 182

171Table A-1 Summary of Neighborhood Influen ces on Adolescent Academic Achievement Research Studies Authors Sample Analytic Technique Academic Achievement Operationalization Neighborhood Operationalization NeighborhoodLevel Variables IndividualLevel Variables Other Variables the National Longitudinal Survey of Youth completed, percentile score on Armed Forces Qualification Test, LEP, and U.S. born), and family characteristics (average adultto-child ratio and poverty status) (cognitive stimulation, parentyouth conflict, and academic involvement) Plybon et al. (2003) Urban, African American girls aged 11 to 14 living in a southeastern city Hierarchical OLS regression Self-reported grades in school on a 5point scale Not defined administratively adolescents’ subjective view of their neighborhoods Bruckner’s Neighborhood Cohesion Scale Maternal education Bowen & Bowen (1999) National probability sample of middle and high school students from the National Hierarchical OLS regression Composite grade index that included grades and perceptions of grades relative to other students Not defined administratively adolescents’ subjective view of their neighborhoods Negative neighborhood peer culture and neighborhood personal threats Gender, race/ethnicity, school level, free/reduced lunch status, and urbanicity School crime and violence and school personal threats

PAGE 183

172Table A-1 Summary of Neighborhood Influen ces on Adolescent Academic Achievement Research Studies Authors Sample Analytic Technique Academic Achievement Operationalization Neighborhood Operationalization NeighborhoodLevel Variables IndividualLevel Variables Other Variables School Success Profile data Williams et al. (2002) African American 9th-grade students in a large, metropolitan area in the Midwest Hierarchical OLS regression Official 4-point GPA from students’ records Not defined administratively adolescents’ subjective view of their neighborhoods Perceived neighborhood deterioration and perceived neighborhood resources Gender, family structure, religiosity, and exposure to academic success Blau et al. (2001) Public high school students from the High School Effectiveness Study Hierarchical linear modeling Two-year gains in social studies standardized test scores between 10th and 12th grade Zip codes according to 1990 census data Neighborhood diversity and inequality of socioeconomic resources Gender, traditional educational advantage status, SES, previous mathematics and reading performance, family structure, locus of control, educational expectations, and academic motivation School sociodemographic composite variable

PAGE 184

173Table A-1 Summary of Neighborhood Influen ces on Adolescent Academic Achievement Research Studies Authors Sample Analytic Technique Academic Achievement Operationalization Neighborhood Operationalization NeighborhoodLevel Variables IndividualLevel Variables Other Variables Boardman & Saint Onge (2005) Middle and high school youth from the Add Health data Hierarchical linear modeling Self-reported grades and performance on the Add Health Picture Vocabulary Test 1990 census tracts Do not know – not clearly stated in the paper Race/ethnicity, age, gender, mother’s marital status and level of education, and use of public assistance

PAGE 185

174 Table A-2 Summary of Neighborhood Influences on Adolescent Risk of Obesity Research Studies Authors Sample Analytic Technique Risk of Obesity Operationalization Neighborhood Operationalization NeighborhoodLevel Variables IndividualLevel Variables Other Variables Nelson et al. (2006) Adolescents from the Add Health data Cluster analysis and Poisson regression BMI > 95th percentile 3-km buffer around each adolescent’s residential location Income/wealth, race/ethnicity, SES and environment, crime, road type, street connectivity/ walkability, and recreation facilities Race/ethnicity, parental education, and family income Chen & Paterson (2006) Public high school students aged 14 to 19 in the St. Louis, MO area Simultaneous regression BMI (no mention of a specific cut point in the article) Census block groups Education, employment, income, and assets Age, gender, family education, family occupational status, family income, and family assets Kling & Liebman (2004) Teenage youth (aged 15-20) whose families participated in the MTO program in Baltimore, Boston, Chicago, Los Angles, OLS regression BMI > 95th percentile Not clearly stated. Only provided general information on the different treatment and control groups Experimental group could only move to census tracts with a 1990 poverty rate less than 10 % Poverty level Gender and baseline characteristics (race, gifted classes, special education classes, behavior problems, health problems, school

PAGE 186

175Table A-2 Summary of Neighborhood Influences on Adolescent Risk of Obesity Research Studies Authors Sample Analytic Technique Risk of Obesity Operationalization Neighborhood Operationalization NeighborhoodLevel Variables IndividualLevel Variables Other Variables and New York City Section 8 group could move to any neighborhood Control group was not allowed to live in Section 8 housing – they remained in the housing projects discipline experiences) Wickrama et al. (2006) Adolescents from Add Health data Hierarchical linear modeling BMI 95th percentile 1990 census tracts Community poverty Race/ethnicity, gender, and family poverty Norman et al. (2006) Adolescents aged 11 to 15 in San Diego County Pearson Product Moment Correlation BMI-for-age percentile 1-mile radius around adolescent’s home address Number of private recreation facilities, number of schools, number of parks, residential density, intersection density, retail floor area ratio, land use mix factor, walkability index GordonLarsen et al. (2006) Adolescents from Add Health data Relative odds Age and gender adjusted BMI 95th percentile 1990 census block groups Population density

PAGE 187

176Table A-2 Summary of Neighborhood Influences on Adolescent Risk of Obesity Research Studies Authors Sample Analytic Technique Risk of Obesity Operationalization Neighborhood Operationalization NeighborhoodLevel Variables IndividualLevel Variables Other Variables Boardman & Saint Onge (2005) Adolescents from Add Health data Hierarchical linear modeling Age and gender adjusted BMI 85th percentile 1990 census tracts Do not know – not clearly stated in the paper Race/ethnicity, age, gender, mother’s marital status and level of education, and use of public assistance Ewing et al. (2006) Adolescents (12 to 17 years old) from the 1997 National Longitudinal Survey of Youth Hierarchical linear modeling Age and gender adjusted BMI 85th percentile County of residence County sprawl index Age, gender, race/ethnicity, cigarette use, hours worked, household income, and household education level Powell et al. (2007) 8thand 10thgrade students from the 1997 to 2003 MTF data OLS regression BMI School zip-code Per capita income, number of chain supermarkets, number of nonchain supermarkets, number of grocery stores, number of convenience stores, number of full service restaurants, Gender*age, grade, race/ethnicity, fathers’ education, mothers’ education, family composition, urbanicity, students’ weekly income,

PAGE 188

177Table A-2 Summary of Neighborhood Influences on Adolescent Risk of Obesity Research Studies Authors Sample Analytic Technique Risk of Obesity Operationalization Neighborhood Operationalization NeighborhoodLevel Variables IndividualLevel Variables Other Variables number of fast food restaurants, fast food prices, fruit and vegetable prices hours worked by students, maternal employment, year Cohen et al. (2006) Adolescents aged 12 to 17 residing in Los Angles County Hierarchical linear modeling and hierarchical generalized linear modeling BMI-for-age and age and gender adjusted BMI >95th percentile 1990 census tracts in Los Angles County Collective efficacy, neighborhood disadvantage Age, sex, race/ethnicity, nativity, extracurricular activities, hours of TV watched per day, family structure, parental education, family income, employment status, health insurance status, mother’s BMI

PAGE 189

178 Table A-3 Summary of School Influences on Adolescen t Academic Achievement Research Studies Authors Sample Analytic Technique Academic Achievement Operationalization School Operationalization School-Level Variables Individual-Level Variables Other Variables Coleman et al. (1966) U.S. 6th-, 9th-, and 12th-grade students. OLS regression Verbal standardized test scores developed from the ETS Sequential Tests of Educational Progress series Elementary and secondary school buildings Student body characteristics school resource, and teacher characteristics Family structure and size, poverty status, parental education, urbanism, and educational support Everson & Millsap (2004) 1995 U.S. high school graduates Multilevel structural equation modeling Composite achievement measure based on overall high school GPA, class rank, and subject specific GPA High school buildings SES, size, locale, and racial and ethnic composition Gender, race and ethnicity, parental education, household income, and extra curricular activity participation Caldas & Bankston III (1997) Louisiana 10th-grade public school students OLS Regression Louisiana Graduation Exit Examination composite score of mathematics, language arts, and written composition High school buildings Peer family poverty, peer family social status Race, poverty status, social class status, gender, LEP, homework hours, reading hours, TV hours, work hours, and school activity hours

PAGE 190

179Table A-3 Summary of School Influences on Adolescen t Academic Achievement Research Studies Authors Sample Analytic Technique Academic Achievement Operationalization School Operationalization School-Level Variables Individual-Level Variables Other Variables Lee & Croninger (1994) Middle school students included in NELS:88 base year data Hierarchical linear modeling Reading standardized test scores Middle school buildings School composition, environment and organization, and policies and practices Academic background, race and ethnicity, nonnative English speaker, poverty status, parental education, mother’s educational expectations, literacy resources in the home, and family communication about school issues Crosnoe (2004) Middle and high school students from Add Health Wave I and II Hierarchical linear modeling Self-reported grades in school on a 4point scale Middle and high school buildings Student-teacher bonding, parentadolescent relations, and parent educational attainment School-level controls: sector, level, and average academic achievement Gender, age, race and ethnicity, parent education, family structure, parents’ educational expectations, and Wave I academic achievement Parentadolescent emotional distance

PAGE 191

180Table A-3 Summary of School Influences on Adolescen t Academic Achievement Research Studies Authors Sample Analytic Technique Academic Achievement Operationalization School Operationalization School-Level Variables Individual-Level Variables Other Variables Blau et al. (2001) Public high school students from the High School Effectiveness Study Hierarchical linear modeling Two year gains in social studies standardized test scores between 10th and 12th grade High school buildings Sociodemographic composite variable Gender, traditional educational advantage status, SES, previous mathematics and reading performance, family structure, locus of control, educational expectations, and academic motivation Neighborhood diversity and inequality of socioeconomic resources Baker et al. (2001) 8th-grade students in the state of Virginia Structural equation modeling Aggregated mean scores on three subtests (reading, language, and mathematics) of the Stanford 9 Middle school buildings Economic condition, social organization, and children’s environment Greenwald et al. (1996) 60 studies that examined school resources effects on student achieveMetaanalysis – combined significance testing and effect magnitude estimation Standardized achievement tests U.S. school districts or smaller (i.e., schools or classrooms) Per-pupil expenditure, teacher ability, teacher education, teacher experience, teacher salary, Studies included in the review had to control for socioeconomic characteristics in their models

PAGE 192

181Table A-3 Summary of School Influences on Adolescen t Academic Achievement Research Studies Authors Sample Analytic Technique Academic Achievement Operationalization School Operationalization School-Level Variables Individual-Level Variables Other Variables ment teacher/pupil ratio, and school size Jeynes (2002) 15 studies that examined effects of religious schools or religious commitment and academic achievement of Black and/or Hispanic students Metaanalysis – Hedge’s g measure of effect size Overall academic achievement and achievement tests— neither one clearly defined Middle and high school buildings Religious affiliation Race/ethnicity DarlingHammond (1999) 8th-grade U.S. public middle school students included in the 1996 NAEP data OLS regression Mathematics standardized test scores Middle school buildings % well-qualified teachers, % of out-of-field teachers, % of fully certified teachers, % of less than fully certified teachers, % of uncertified new entrants, % of Student poverty

PAGE 193

182Table A-3 Summary of School Influences on Adolescen t Academic Achievement Research Studies Authors Sample Analytic Technique Academic Achievement Operationalization School Operationalization School-Level Variables Individual-Level Variables Other Variables uncertified newly hired teachers, PPE, pupil: teacher ratio, and average class size Wentzel (2002) Suburban 6th graders in a midAtlantic state Hierarchical OLS regression Official end-of-year grades for the subject taught by the teacher students assessed Middle school buildings Teaching practices: fairness, teacher motivation, rule setting, negative feedback, and high expectations Gender and race/ethnicity Sweetland & Hoy (2000) 8th graders in 86 New Jersey public middle schools OLS regression Reading and mathematics standardized test scores from New Jersey’s Eighth Grade Early Warning Test Middle school buildings SES and teacher empowerment None Crosnoe & Muller (2004) Middle and high school students from Add Health Wave I and Hierarchical linear modeling Self-reported grades in school on a 4point scale Middle and high school buildings Rate of athletic participation, mean student romantic activity, mean student peer involvement, Risk of obesity, gender, age, race/ethnicity, family structure, parental education, athletic status,

PAGE 194

183Table A-3 Summary of School Influences on Adolescen t Academic Achievement Research Studies Authors Sample Analytic Technique Academic Achievement Operationalization School Operationalization School-Level Variables Individual-Level Variables Other Variables II and mean BMI School-level controls: SES, racial and ethnic composition, and school level and Wave I achievement Eamon (2005) Latino adolescents aged 10 to 14 whose mothers participated in the National Longitudinal Survey of Youth Hierarchical OLS regression Peabody Individual Achievement Test reading comprehension and mathematics scores School buildings Overall school quality Latino origin, gender, age, LEP, maternal characteristics (age when had first child, years of education completed, percentile score on Armed Forces Qualification Test, LEP, and U.S. born), and family characteristics (average adultto-child ratio and poverty status) Overall neighborhood quality and parenting processes (cognitive stimulation, parentyouth conflict, and academic involvement)

PAGE 195

184Table A-3 Summary of School Influences on Adolescen t Academic Achievement Research Studies Authors Sample Analytic Technique Academic Achievement Operationalization School Operationalization School-Level Variables Individual-Level Variables Other Variables Bowen & Bowen. (1999) National probability sample of middle and high school students from the National School Success Profile data Hierarchical OLS regression Composite grade index that included grades and perceptions of grades relative to other students Middle and high school buildings Perceived school crime and violence and school personal threats Gender, race/ethnicity, school level, free/reduced lunch status, and urbanicity Negative neighborhood peer culture and neighborhood personal threats Zand & Thomson (2005) 11-to-14 year old African American adolescents living in a large Midwestern city Path analysis Self-reported grades in school on a 5point scale School buildings School bonding Global selfworth Sanders (1998) African American 8th-grade students in a Southeastern city OLS Regression Self-reported grades in school on a 4point scale Middle school buildings Teacher support Age, gender, poverty status, household structure, school behavior, academic selfconcept, and Parental support and church involvement

PAGE 196

185Table A-3 Summary of School Influences on Adolescen t Academic Achievement Research Studies Authors Sample Analytic Technique Academic Achievement Operationalization School Operationalization School-Level Variables Individual-Level Variables Other Variables achievement ideology Hoy & Hannum (1997) 8th graders in 86 New Jersey public middle schools OLS Regression New Jersey’s Eighth Grade Early Warning Test reading, writing, and mathematics test scores Middle school buildings SES, academic emphasis, teacher affiliation, collegial leadership, resource support, principal influence, and institutional integrity None Henderson et al. (2005) 10 Tennessee middle schools Pearson Product Moment Correlation Median national percentile scores in reading, language, mathematics, science, and social studies Middle school buildings Academic emphasis, teacher affiliation, collegial leadership, resource support, principal influence, institutional integrity, and overall org. health index score None

PAGE 197

186Table A-3 Summary of School Influences on Adolescen t Academic Achievement Research Studies Authors Sample Analytic Technique Academic Achievement Operationalization School Operationalization School-Level Variables Individual-Level Variables Other Variables Lee et al. (1997) First three waves of NELS:88 data Growth curve analysis Gains in science and mathematics test scores High school buildings Structural practices, social organization, academic organization, and demographics Math and science courses taken in high school, race/ethnicity, gender, SES, 8th-grade ability, and 8thgrade engagement Gill et al. (2004) 8th-grade students include in NELS:88 base year data Hierarchical linear modeling Mathematics standardized test scores Middle school buildings Student perceived school responsiveness, principal perceived demandingness and responsiveness, and mean SES Gender, minority status, SES, and prior grades

PAGE 198

187 Table A-4 Summary of School Influences on Adolescent Risk of Obesity Research Studies Authors Sample Analytic Technique Risk of Obesity Operationalization School Operationalization School-Level Variables Individual-Level Variables Other Variables O’Malley et al. (2007) 1991 to 2004 MTF data Hierarchical linear modeling OLS regression BMI Middle school buildings and high school buildings School type, school size, school SES, racial/ethnic composition Grade, SES, race/ethnicity Region and population density Gortmaker et al. (1999) 6th-and7th grade Boston area students Generalized estimating equation method Age-and-genderadjusted composite indicator based on both BMI and a triceps skinfold measure 85th percentile Middle school classrooms School-based intervention focused on reducing TV viewing, increasing physical activity, decreasing high-fat foods, and increasing fruit and vegetables Age, gender, race/ethnicity, self-reported weight-loss behaviors, and baseline obesity status NeumarkSztainer et al. (2003) High school girls in the Twin Cities area who were Mixedmodel repeatedmeasures with schools as random BMI High school PE classes School-based intervention focused on improving physical activity and Baseline BMI, race/ethnicity, and grade level

PAGE 199

188Table A-4 Summary of School Influences on Adolescent Risk of Obesity Research Studies Authors Sample Analytic Technique Risk of Obesity Operationalization School Operationalization School-Level Variables Individual-Level Variables Other Variables overweight or at-risk of being overweight effects eating behaviors and helping overweight girls feel good about themselves Sallis et al. (2003) Students at 24 San Diego County middle schools Randomized regression models BMI Middle school buildings An environmental and policy focused school-based intervention aimed at increasing the availability of low-fat food choices and physical activity opportunities to promote healthful choices Gender Scott et al. (2007) 6th-grade girls in 6 U.S. cities Hierarchical linear modeling BMI School buildings located within a halfmile radius of participants home School accessibility and amenities and percent of Race and SES Population density, SES index, and median

PAGE 200

189Table A-4 Summary of School Influences on Adolescent Risk of Obesity Research Studies Authors Sample Analytic Technique Risk of Obesity Operationalization School Operationalization School-Level Variables Individual-Level Variables Other Variables addresses in 6 U.S. cities students on free or reduced lunch year construction for each girl’s block group Also, number of parks within study area and presence of one or more schools in each girl’s area

PAGE 201

190 Appendix B: BMI Box-and-Whisker Plots

PAGE 202

191 11121314151617181920 10 20 30 40 50 B M I age Figure B-1 Age-and-gender-adjusted BMI box-and-whisker plots for girls.

PAGE 203

192 11121314151617181920 10 20 30 40 50 60 70 B M I age Figure B-2 Age-and-gender-adjusted BMI box-and-whisker plots for boys.

PAGE 204

193 Appendix C: Analysis of Missing Data

PAGE 205

194 Table C-1 Frequency of Missing Variables Across Observations in the Original Sample (n = 11,841) Number of missing variables Frequency % 13 1 0.01 11 2 0.02 9 3 0.03 7 4 0.03 6 1 0.01 5 11 0.09 4 42 0.35 3 383 3.23 2 1500 12.67 1 2051 17.33 0 7842 66.23

PAGE 206

195 10 0000000000 9 9 0 8 8 7 7 6 6 5 5 011111 4 4 3 5 3 44444 2 88888 2 3 1 58 1 01111122 0 666666678 0 000111111111111111111111222223333344 -0 210000000000000000000 Multiply Stem.Leaf by 10**-1 Figure C-1 Stem-and-leaf display of correlations between missingness on variables using the original sample. Note: The 10 = 1.0 were between each of the five race variables as originally coded in the Add Health data. Given the way these variables were coded (i.e., five dummy coded variables – one for each racial classification) this level of correlation would be expected.

PAGE 207

196 1 9 1 88 1 1 66 1 1 1 3 1 22 1 1 1 00 0 99 0 88 0 7777 0 6666666 0 555555555555555 0 4444444 0 3333333333333333 0 222222222222222222 0 11111111111111111111111111111111111111111 0 000000000000000000000000000000000000000000000000000000000000000 -0 1111111111111111111111111111111111111111111111111111111111111111111 -0 22222222222222222222222222222 -0 33333333333333333 -0 444444444 -0 55555 -0 6666 -0 777 -0 88 -0 99 -1 000000 -1 1 -1 22222 -1 333 -1 4 -1 5 Multiply Stem.Leaf by 10**-1 Figure C-2. Stem-and-leaf display of correlations between missingness and observed values using the original sample.

PAGE 208

197 Table C-2 Frequency of Missing Variables Across Observations after Deleting Cases Missing Household Income Data (n = 9,919) Number of missing variables Frequency % 9 3 0.03 8 1 0.01 7 1 0.01 6 1 0.01 5 6 0.06 4 2 0.02 3 36 0.36 2 427 4.30 1 2371 23.90 0 7071 71.29 Note: For this analysis, adolescents who were missing household income were removed and adolescents whose parent refused to provide household income were marked as missing.

PAGE 209

198 10 0000000000 9 9 8 8 7 7 6 6 5 5 4 4 3 99999 3 2 2 111114 1 57 1 1 0 5555568999999 0 000011111122344444 -0 211111111110000000000000000000000000 Multiply Stem.Leaf by 10**-1 Figure C-3 Stem-and-leaf display of correlations between missingness on variables after deleting cases missing household income data. Note: For this analysis, adolescents who were missing household income were removed and adolescents whose parent refused to provide household income were marked as missing. Also, the 10 = 1.0 were between each of the five race variables as originally coded in the Add Health data. Given the way these variables were coded (i.e., five dummy coded variables – one for each racial classification) this level of correlation would be expected.

PAGE 210

199 2 0 1 1 8 1 1 1 5 1 1 3 1 1 1 0 0 9 0 8 0 77 0 66666666 0 55555555 0 444444 0 33333333333333 0 22222222222222222222 0 11111111111111111111111111111111111 0 00000000000000000000000000000000000000000000000000000000000000000000000000 -0 1111111111111111111111111111111111111111111111111111111111111111111111111 -0 22222222222222222222222222222222 -0 33333333333333333 -0 4444 -0 55555555 -0 6666 -0 7777 -0 888 -0 9999 -1 -1 -1 222 -1 3 Multiply Stem.Leaf by 10**-1 Figure C-4 Stem-and-leaf display of correlations between missingness and observed values after deleting cases missing household income data. Note: For this analysis, adolescents who were missing household income were removed and adolescents whose parent refused to provide household income were marked as missing.

PAGE 211

200 Appendix D: Investigation of Model Assumptions

PAGE 212

201 Table D-1 Tolerance Values for Each Variable Included in Academic Achievement CCREMs Variable Tolerance value Level-1 Model Age .99 Biological sex .99 Race .93 Family SES .92 Neighborhood Level-2 Model Neighborhood affluence .75 Neighborhood poverty .77 Neighborhood racial composition .91 Urbanicity .89 School Level-2 Model School SES .86 Teacher education .83 Student body racial composition .73 Neighborhood & School Level-2 Model Neighborhood affluence .48 Neighborhood poverty .72 Neighborhood racial composition .38 Urbanicity .76 School SES .49 Teacher education .78 Student body racial composition .28

PAGE 213

202 -100 -75 -50 -25 0 25 50Level-1 Residual (sk = -0.37 ku = 1.72) Figure D-1 Box-and-whisker plot for Level-1 residuals (academic achievement).

PAGE 214

203 -2.0000 -1.0000 0 1.0000 2.0000 3.0000Neighborhood Level-2 Residual (sk = -0.44, ku = 7.83) Figure D-2 Box-and-whisker plot for Level-2 neigh borhood residuals (academic achievement).

PAGE 215

204 -6.0000 -4.0000 -2.0000 0 2.0000 4.0000School Level-2 Residual (sk = -0.21, ku = -0.06 ) Figure D-3 Box-and-whisker plot for Level-2 sc hool residuals (academic achievement).

PAGE 216

205 Residual -100 -80 -60 -40 -20 0 20 40 60 Predicted 708090100110120* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * ** * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * ** * * * * * * * * * * * * * ** * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * ** * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * ** * * ** * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * ** * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * ** * * * * * ** * * * ** * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * ** * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * ** * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * ** * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * ** * * * * * * * * * * ** * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * Figure D-4 Level-1 residuals*predicted academic achievement.

PAGE 217

206 Residual -100 -80 -60 -40 -20 0 20 40 60 Predicted 708090100110120* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * ** * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *** * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * ** * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * ** * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** ** * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * ** * * * * * * ** * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * ** * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * ** * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * ** * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * ** * * * * * * *** * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * ** * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * ** * ** * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * Figure D-5 Level-2 neighborhood residuals*pr edicted academic achievement.

PAGE 218

207 Residual -100 -80 -60 -40 -20 0 20 40 60 Predicted 708090100110120* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * ** * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * ** * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * ** * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * ** * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * ** * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * ** * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * ** * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * ** * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * ** ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * ** * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * Figure D-6 Level-2 school residuals*predict ed academic achievement.

PAGE 219

208 Table D-2 Tolerance Values for Each Variable Included in Risk of Obesity CCREMs Variable Tolerance value Level-1 Model Age .96 Biological sex .99 Race .93 Family SES .92 Athletic participation .95 Neighborhood Level-2 Model Neighborhood affluence .75 Neighborhood poverty .77 Neighborhood racial composition .91 Urbanicity .89 School Level-2 Model School SES .90 Weight education .90 School athletic participation .99 Neighborhood & School Level-2 Model Neighborhood affluence .49 Neighborhood poverty .77 Neighborhood racial composition .83 Urbanicity .86 School SES .54 Weight education .79 School athletic participation .99

PAGE 220

209 -3 -2 -1 0 1 2Level-1 Residual (sk = -0.32, ku = -0.58 ) Figure D-7 Box-and-whisker plot for Level-1 residuals (risk of obesity).

PAGE 221

210 -0.06000 -0.04000 -0.02000 0 0.02000 0.04000Neighborhood Level-2 Residual (sk = -0.49, ku = 7.69) Figure D-8 Box-and-whisker plot for Level-2 neighborhood residuals (risk of obesity).

PAGE 222

211 -0.05000 -0.02500 0 0.02500 0.05000 0.07500School Level-2 Residual (sk = 0.12 ku = 0.28) Figure D-9 Box-and-whisker plot for Level-2 school residuals (risk of obesity).

PAGE 223

212 Residual -3 -2 -1 0 1 2 Predicted -0.2-0.10.00.10.20.30.40.50.60.70.80.9* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * ** * * * ** * * * ** * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * Figure D-10 Level-1 residuals*predicted risk of obesity.

PAGE 224

213 Residual -3 -2 -1 0 1 2 Predicted -0.2-0.10.00.10.20.30.40.50.60.70.80.9* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * ** * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * ** * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * ** * * * ** * * * ** * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * ** * * * * * * ** * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * Figure D-11 Level-2 neighborhood residuals*predicted risk of obesity.

PAGE 225

214 Residual -3 -2 -1 0 1 2 Predicted -0.10.00.10.20.30.40.50.60.70.80.91.0* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * ** * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * ** * * * * * * * * * * * * * * * * ** * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * ** ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * ** * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * ** * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * Figure D-12 Level-2 school residuals*predicted risk of obesity.

PAGE 226

215 Residual -2.0000 -1.0000 0 1.0000 2.0000 3.0000 Neighborhood Size 020406080100120140160180200 Figure D-13 Academic achievement neighborhood Level-2 residuals*neighborhood size.

PAGE 227

216 Residual -0.05000 -0.04000 -0.03000 -0.02000 -0.01000 0 0.01000 0.02000 0.03000 0.04000 Neighborhood Size 020406080100120140160180200 Figure D-14 Risk of obesity neighborhood Level-2 residuals*neighborhood size.

PAGE 228

End Page About the Author Bethany A. Bell-Ellison received a B achelor’s Degree in Sociology from Mary Washington College in 1997. After working for AmeriCorps for two years and establishing the Oklahoma Caring Vans progr am, she attended graduate school at the University of Oklahoma Health Sciences Center where she received her MPH in 2002. Following the completion of her MPH, she began her Ph.D. program at the University of South Florida where her program of study fo cused on educational research methods and statistics and community and family health. He r applied research addresses issues related to educational equity, health disparities, a nd social determinants of child and adolescent development and her methodological research focuses on hierarchical linear modeling and complex sample data. Upon graduation, Ms. Bell-Ellison will be employed as an Assistant Professor of Educa tional Assessment, Research Methodology, and Statistics in the College of Education at the University of South Carolina.