Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

An examination of neighbourhood built and social environment influences on child physical activity patterns VanLoon, Joshua 2011

You don't seem to have a PDF reader installed, try download the pdf

Item Metadata

Download

Media
24-ubc_2011_fall_vanloon_joshua.pdf [ 16.82MB ]
Metadata
JSON: 24-1.0103330.json
JSON-LD: 24-1.0103330-ld.json
RDF/XML (Pretty): 24-1.0103330-rdf.xml
RDF/JSON: 24-1.0103330-rdf.json
Turtle: 24-1.0103330-turtle.txt
N-Triples: 24-1.0103330-rdf-ntriples.txt
Original Record: 24-1.0103330-source.json
Full Text
24-1.0103330-fulltext.txt
Citation
24-1.0103330.ris

Full Text

AN EXAMINATION OF NEIGHBOURHOOD BUILT AND SOCIAL ENVIRONMENT INFLUENCES ON CHILD PHYSICAL ACTIVITY PATTERNS  by JOSHUA VANLOON B.Sc., Queen’s University, 1998 M.Sc., University of Toronto, 2001  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY  in  THE FACULTY OF GRADUATE STUDIES (Planning)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)  May 2011  © Joshua vanLoon, 2011  ABSTRACT Trends in increasing prevalence of childhood overweight and obesity, coupled with evidence that children engage in inadequate physical activity, have prompted considerable interest in the study of correlates of child physical activity. The purpose of this thesis is to investigate potential influences of neighbourhood built and social environment characteristics on child physical activity using data corresponding to two samples of children aged 8-11, attending schools in Vancouver and the surrounding Lower Mainland region of British Columbia. This research benefits from the use of several complementary data sources, including: survey data on parent and child perceptions and travel behaviour; and, objective measures of physical activity and neighbourhood environment characteristics. A variety of analytical methods were used, including Generalized Estimating Equations to account for the clustering of students within schools resulting from a two stage sampling design. Objective measures of built environment characteristics were found to be significantly associated with average daily moderate to vigorous physical activity (MVPA) when controlling for age, gender, ethnicity and household income. These include measures of distance to school, commercial density and intersection density. Measures based on larger spatial units were found to have the strongest associations with MVPA. When modeled as a composite built environment index these characteristics accounted for 5.0% of the variance in MVPA. This index was also significantly negatively associated with sedentary activity. In gender stratified models, different correlates were found for boys and girls, and overall built environment influences on MVPA were found to be stronger for boys than for girls. When controlling for the objective built environment index, neither parent nor child perceptions of safety contributed to explaining MVPA.  ii  The generalizability of specific results is limited by the sampling strategy used and also because of unique characteristics of the Lower Mainland region. Nonetheless, many results are consistent with findings of previous studies, providing further support for policies that promote compact, mixed use developments with high street connectivity to support physical activity. Findings also point towards several important research implications, including the need to further study how best to define neighbourhoods for the purpose of assessing environmental characteristics.  iii  PREFACE While I was responsible for the development of built and social environment measures used in this thesis, several other data sources used were collected by and/or developed in conjunction with other researchers. Physical activity data described in S. 3.2.1 were collected by the Action Schools! British Columbia (AS!BC) research team led by Dr. Heather McKay (University of British Columbia) and Dr. Patti-Jean Naylor (University of Victoria). Following collection and initial screening of the data, I worked as a member of a team cleaning accelerometer data to produce the physical activity measures used in this thesis. As part of this team, I was responsible for processing the data for four of the schools involved in the study. During this process, I worked closely with several other students and AS!BC staff, and in particular Lindsay Nettlefold (a PhD Candidate at UBC), who coordinated the process and produced the final integrated data set. Elements of the accelerometer data cleaning described below have also been outlined in Nettlefold et al. (2010). Travel behaviour and neighbourhood perception data described in S. 4.2.1 were collected as a component of the AS!BC project led by Dr. Lawrence Frank. These data were obtained using surveys developed through a collaborative effort by Jennifer Niece, a Masters Student at the School of Community and Regional Planning (SCARP), together with Dr. Frank, and Dr. James Sallis at San Diego University. I used raw data collected from these surveys to derive a variety of measures for analyses as described in Sections 4.2.2 and 4.2.3. Ethics approval for this investigation was granted by the Behavioural Research Ethics Board at UBC, certificate number B05-0505. Finally, portions of the literature review (S. 2) were written as part of a review article co-authored with my supervisor, Dr. Larry Frank, and recently accepted for publication in the Journal of Planning Literature. I was the primary author for this article.  iv  TABLE OF CONTENTS Abstract ........................................................................................................................................ ii Preface ......................................................................................................................................... iv Table of Contents......................................................................................................................... v List of Tables ............................................................................................................................... ix List of Figures ............................................................................................................................. xi List of Abbreviations ................................................................................................................. xii Acknowledgements ................................................................................................................... xiv Dedication................................................................................................................................... xv 1. Introduction ............................................................................................................................. 1 1.1 Overview.............................................................................................................................. 1 1.2 Research Questions Investigated ......................................................................................... 3 1.2.1 Research Questions Investigated Using Sample 1 ....................................................... 3 1.2.2 Research Questions Investigated Using Sample 1B..................................................... 5 1.2.3 Classification of Research Questions ........................................................................... 6 1.3 Background and Context ..................................................................................................... 6 1.4 Study Significance ............................................................................................................... 9 2. Literature Review .................................................................................................................. 13 2.1 Overview............................................................................................................................ 13 2.2 Child Physical Activity Patterns ........................................................................................ 14 2.2.1 Characterizing Physical Activity ................................................................................ 14 2.2.2 Compensation Effects................................................................................................. 20 2.2.3 Child Physical Activity Guidelines ............................................................................ 21 2.2.4 Current Physical Activity Levels and Trends............................................................. 22 2.2.5 Health Implications of Physical Inactivity ................................................................. 25 2.2.6 Health Risks Associated with Physical Activity in Neighbourhood Environments... 27 2.3 Environmental Correlates of Physical Activity ................................................................. 31 2.3.1 Theories of Physical Activity Behaviour.................................................................... 31 2.3.2 Ecological Models of Health Behaviour .................................................................... 33 2.3.4 Individual and Household Correlates of Physical Activity ........................................ 38 2.3.5 Neighbourhood Environment Correlates of Physical Activity................................... 55 3. Sample 1: Objective Built Environment and Objective Physical Activity Data.............. 74 3.1 Overview............................................................................................................................ 74 v  3.2 Data Sources and Measure Development .......................................................................... 75 3.2.1 Physical Activity Data ................................................................................................ 75 3.2.2 Control Data – Individual Characteristics .................................................................. 84 3.2.3 Built and Social Environment Data ............................................................................ 85 3.3 Sample Description and Regional Context...................................................................... 102 3.3.1 Student Characteristics ............................................................................................. 102 3.3.2 School and School Neighbourhood Characteristics ................................................. 103 3.4 Descriptive Analysis ........................................................................................................ 108 3.4.1 Built and Social Environment Characteristics.......................................................... 108 3.4.2 Physical Activity Behaviour ..................................................................................... 120 3.5 Special Topic - Methods for Dealing with Clustering..................................................... 125 3.5.1 Method Alternatives Considered .............................................................................. 125 3.5.2 Preliminary Analysis and Final Method Selection ................................................... 127 3.6 Research Question 1 (Primary Research Question)......................................................... 131 3.6.1 Research Question and Hypotheses Investigated ..................................................... 131 3.6.2 Analytical Framework and Methods ........................................................................ 132 3.6.3 Results and Discussion ............................................................................................. 134 3.7 Research Question 2 ........................................................................................................ 145 3.7.1 Research Question and Hypothesis Investigated...................................................... 145 3.7.2 Analytical Framework and Methods ........................................................................ 146 3.7.3 Results and Discussion ............................................................................................. 147 3.8 Research Question 3 ........................................................................................................ 152 3.8.1 Research Question and Hypothesis Investigated...................................................... 152 3.8.2 Analytical Framework and Methods ........................................................................ 152 3.8.3 Results and Discussion ............................................................................................. 153 3.9 Research Question 4 ........................................................................................................ 164 3.9.1 Research Question and Hypotheses Investigated ..................................................... 164 3.9.2 Analytical Framework and Methods ........................................................................ 165 3.9.3 Results and Discussion ............................................................................................. 167 4. Sample 1B: Survey Data Combined with Objective Built Environment and Physical Activity Data............................................................................................................................. 169 4.1 Overview.......................................................................................................................... 169 4.2 Survey Data Collection, Variable Selection and Data Cleaning ..................................... 170 4.2.1 Survey Design and Sampling Process ..……………………….………………………..170 vi  4.2.2 Variable Selection …..……………………….…………………………………………..171 4.2.3 Data Cleaning and Coding…..……………….…………………………………………..171 4.3 Sample Description and Regional Context...................................................................... 176 4.4 Descriptive Analysis ........................................................................................................ 178 4.4.1 Objective Built and Social Environment Characteristics ......................................... 178 4.4.2 Survey Data .............................................................................................................. 181 4.4.3 Physical Activity Data ……………………………………………………………...........186 4.5 Research Question 5 ........................................................................................................ 188 4.5.1 Research Question and Hypothesis Investigated...................................................... 188 4.5.2 Analytical Framework and Methods ........................................................................ 188 4.5.3 Results and Discussion ............................................................................................. 189 4.6 Research Question 6 ........................................................................................................ 191 4.6.1 Research Question and Hypotheses Investigated ..................................................... 191 4.6.2 Analytical Framework and Methods ........................................................................ 192 4.6.3 Results and Discussion ............................................................................................. 196 4.7 Research Question 7 ........................................................................................................ 203 4.7.1 Research Question and Hypotheses Investigated ..................................................... 203 4.7.2 Analytical Framework and Methods ........................................................................ 204 4.7.3 Results and Discussion ............................................................................................. 206 5. Discussion and Conclusion.................................................................................................. 209 5.1 Overview.......................................................................................................................... 209 5.2 Summary and Synthesis of Results.................................................................................. 209 5.2.1 Major Findings ......................................................................................................... 209 5.2.2 Integration and Interpretation of Findings: Possible Causal Mechanisms of Neighbourhood Built Environment Influences on Child Physical Activity Behaviour .... 215 5.2.3 Integration and Interpretation of Findings: Illustrative Scenarios............................ 219 5.3 Limitations and Strengths of Research ............................................................................ 221 5.3.1 General Methodological Limitations........................................................................ 221 5.3.2 Limitations on Generalizability ................................................................................ 223 5.3.3 Strengths ................................................................................................................... 226 5.4 Implications for Future Research..................................................................................... 227 5.5 Implications for Planning Policy and Practice................................................................. 229 5.5.1 Overview .................................................................................................................. 229 5.5.2 School Siting and Active Transportation to School ................................................. 230 5.5.3 Supports for Physical Access ................................................................................... 232 vii  5.5.4 The Environment or Perceptions of the Environment - Which should be Targeted?232 5.5.5 Options for Focusing Policies................................................................................... 233 5.5.6 Policy Recommendation Examples .......................................................................... 236 5.5.7 Potential External Challenges to Policies Targeting Neighbourhood Environment Characteristics to Influence Child Physical Activity Patterns........................................... 239 5.5.8 Supporting Rationale for Policies Targeting Neighbourhood Environment Characteristics to Influence Child Physical Activity Patterns........................................... 241 5.6 Conclusion ....................................................................................................................... 242 References................................................................................................................................. 244 Appendices ............................................................................................................................... 272 Appendix 1. Physical Activity Measurement Methodologies .............................................. 272 Appendix 2. Accelerometer Data .......................................................................................... 274 Appendix 3. Built and Social Environment Data .................................................................. 276 Appendix 4. Vancouver Walkability Surface and Walkability Index ................................... 280 Appendix 5. Matched Parent and Child Surveys................................................................... 282  viii  LIST OF TABLES Table 1: Available Data for Samples 1 and 1B ............................................................................. 2 Table 2: Examples of Correspondence Between Measures Used in Analysis and Planning Instruments ..................................................................................................................... 9 Table 3: Objective Measures Contrasted with their Alternatives................................................ 11 Table 4: Commonly Used Thresholds for Sedentary Activity and Light, Moderate and Vigorous Physical Activity .......................................................................................................... 16 Table 5: Dependent Variables Commonly Used in Studies of Environmental Correlates of Physical Activity .......................................................................................................... 19 Table 6: Theoretical Rationale for Selection of Objective Built and Social Environment Measures....................................................................................................................... 90 Table 7: Definition of Neighbourhood Environment Measures .................................................. 97 Table 8: Descriptive Statistics for Sample 1 Control Variables. ............................................... 102 Table 9: Characteristics of Schools, Their Neighbourhoods and Students with Valid Accelerometry Data.................................................................................................... 104 Table 10: Descriptive Statistics for Built and Social Environment Measures for Students in Sample 1 (n=366). ...................................................................................................... 110 Table 11: Proximity to Three Types of Recreation Sites: School Sites, Parks and Non-park Recreation Sites. ......................................................................................................... 112 Table 12: Descriptive Statistics for Physical Activity Variables, by Level of Intensity, for Students in Sample 1 (n=366). ................................................................................... 121 Table 13: Descriptive Statistics for MVPA Outside of School and Total MVPA Stratified by Gender and Ethnicity, for Students in Sample 1 (n=366). ......................................... 123 Table 14: Control Only Model (Model 1), Overall MVPA as Dependent Variable ................. 134 Table 15: Summary of Model Results for Single Covariate Models, Overall MVPA as Dependent Variable .................................................................................................... 136 Table 16: Proximity of Sample 1 Students to Parks .................................................................. 138 Table 17: Variation in Selected Built Environment Measures by Buffer Size.......................... 141 Table 18: Final Model Predicting Overall MVPA (Model 2) ................................................... 144 Table 19: Control Only Models. Model 3a for Average Daily MVPA Outside of School on School Days, and Model 3b for Average Daily MVPA on Weekends ...................... 147 Table 20: Summary of Model Results for Single Covariate Models Predicting Average Daily MVPA Outside of School on School Days ................................................................ 149 Table 21: Summary of Model Results for Single Covariate Models Predicting Average Daily MVPA on Weekends .................................................................................................. 150 Table 22: Model 4 Predicting Overall MVPA with Interaction Terms for Age and Gender .... 154 Table 23: Predicted Changes in Overall MVPA Associated with Changes in Built Environment Index Values, Based on Model 4............................................................................... 156 Table 24: Summary of Model Results for Single Covariate Gender Stratified Models Predicting Overall MVPA for Male Children (n=174)................................................................ 157 Table 25: Summary of Model Results for Single Covariate Gender Stratified Models Predicting Overall MVPA for Female Children (n=192) ............................................................ 158 Table 26: Final Gender Stratified Female Model (Model 5) Predicting Overall MVPA (n = 192). .................................................................................................................................... 162 Table 27: Final Gender Stratified Male Model (Model 6) Predicting Overall MVPA (n = 174). .................................................................................................................................... 163  ix  Table 28: Association of Built Environment Index with Percentage of Time Spent in Sedentary Activity and Light, Moderate and Vigorous Physical Activity .................................. 168 Table 29: Predicted Changes in Percentage of Time Spent in Sedentary Activity, MPA and VPA Associated with Changes in Built Environment Index Values................................... 168 Table 30: Survey Variables Selected for Sample 1B Models ................................................... 173 Table 31: Survey Variable Coding ............................................................................................ 176 Table 32: Descriptive Statistics for Sample 1B Control Variables. .......................................... 177 Table 33: Descriptive Statistics for Household Income. ........................................................... 177 Table 34: Descriptive Statistics for Built and Social Environment Measures for Students in Sample 1B. ................................................................................................................. 179 Table 35: Model 2, Predicting Overall MVPA, Tested Using Both Sample 1 (n=366) and Sample 1B (n=255)..................................................................................................... 180 Table 36: Car Ownership for Households in Sample 1B .......................................................... 181 Table 37: Child Commute To and From School by Mode ........................................................ 182 Table 38: Descriptive Statistics for Parent Perception Variables .............................................. 185 Table 39: Descriptive Statistics for Child Perception Variables ............................................... 185 Table 40: Descriptive Statistics for Physical Activity Variables, Samples One and Two. ....... 187 Table 41: Model 11 Predicting Overall MVPA Using Sample 1B Data................................... 190 Table 42: Fit Measures for Model 12 Measurement Model ...................................................... 197 Table 43: Fit Measures for Model 12, Predicting Overall MVPA. SEM Model Incorporating Measures of Parent and Child Perceptions. ................................................................ 200 Table 44: Fit Measures for Model 13b ...................................................................................... 202 Table 45: Mediation Model Predicting MVPA Outside of School on School Days Results (Model 14a) ................................................................................................................ 207 Table 46: Mediation Model Predicting Overall Average Daily MVPA Results (Model 14b).. 207 Table 47: Summary of Results Based on Sample 1................................................................... 210 Table 48: Summary of Results Based on Sample 1B. ............................................................... 211 Table 49: Proportion of Variance in Dependent Variables Explained by Control and Built Environment Variables ............................................................................................... 214 Table 50: Policy Recommendations .......................................................................................... 236  x  LIST OF FIGURES Figure 1: Study Area...................................................................................................................... 2 Figure 2: Classification of Research Questions............................................................................. 7 Figure 3: Generic Ecological Model Illustrating Multiple Possible Environments Influencing Individual Behaviours .................................................................................................. 34 Figure 4: Ecological Classification of Correlates of Physical Activity....................................... 36 Figure 5: Three Alternative Approaches to Buffer Definition. ................................................... 56 Figure 6: Neighbourhood Physical Environment Correlates of Physical Activity...................... 61 Figure 7: Location of Nine Study Schools Within the Lower Mainland..................................... 77 Figure 8: Path from Initial Participant Recruitment to Final Sample One .................................. 79 Figure 9: Physical Activity Data Cleaning Process..................................................................... 80 Figure 10: Examples of Measures Incorporated in Analyses, Categorized in an Ecological Framework.................................................................................................................... 92 Figure 11: Street Network Data Preprocessing for Calculating Intersection Density. ................ 95 Figure 12: Median Household Income by School Neighbourhood. .......................................... 105 Figure 13: Walkability of School Neighbourhoods Compared to Walkability for Selected Lower Mainland Municipalities............................................................................................. 106 Figure 14: School Locations in Relation to Walkability. .......................................................... 107 Figure 15: School Zone 30 kilometer/hour Speed Limit ........................................................... 111 Figure 16: Typical Recreational Amenities on School Sites ..................................................... 112 Figure 17: Typical Commercial Developments in Study Neighbourhoods............................... 114 Figure 18: Mixed Density Housing in Selected Study Neighbourhoods................................... 116 Figure 19: Low Density Housing in Selected Study Neighbourhoods...................................... 117 Figure 20: Comparison of Street Connectivity.......................................................................... 119 Figure 21: Average Daily MVPA by School............................................................................. 124 Figure 22: Conceptual Model for Research Question 1, Hypotheses 1.1-1.3 ........................... 133 Figure 23: Conceptual Model for Research Question 2, Hypotheses 2.1, 2.2........................... 146 Figure 24: Conceptual Model for Research Question 3, Hypothesis 3.1 .................................. 153 Figure 25: Conceptual Model for Research Question 4, Hypotheses 4.1-4.5. .......................... 166 Figure 26: Path from Initial Survey Distribution to Final Sample Two .................................... 171 Figure 27: Parameter Estimates and 95% Confidence Intervals for Built Environment Index, when Predicting Overall MVPA, Using Sample One and Sample Two. ................... 180 Figure 28: Active and Non-active Commuters by Distance Between Home and School ......... 183 Figure 29: Conceptual Model for Research Question 5, Hypothesis 5.1 .................................. 189 Figure 30: Conceptual Model for Research Question 6. ........................................................... 193 Figure 31: Measurement Model Corresponding to Structural Model Illustrated in Figure 30.. 196 Figure 32: Significant Paths and Standardized Path Coefficients for Model 12 Final Structural Model.......................................................................................................................... 198 Figure 33: Significant Paths and Standardized Path Coefficients for (a) Model Predicting Overall MVPA When Excluding the Built Environment Index, and (b) Reduced Model Retaining only Significant Predictors of MVPA........................................................ 201 Figure 34: Mediation Models 14a Predicting MVPA Outside of School on School Days, and 14b Predicting Overall Average Daily MVPA.................................................................. 205  xi  LIST OF ABBREVIATIONS AS!BC  Action Schools! British Columbia  BC  British Columbia  BE  Built Environment  CANPLAY  Canadian Physical Activity Levels among Youth study  CCHS  Canadian Community Health Survey  CDC  Centers for Disease Control and Prevention (United States)  CFLRI  Canadian Fitness and Lifestyle Research Institute  CFI  Comparative Fit Index  CV  Coefficient of Variation  DA  Dissemination Area (Canadian Census)  DOT  Department(s) of Transportation (United States)  FAR  Floor Area Ratio  GEE  Generalized Estimating Equations  GIS  Geographic Information Systems  GPS  Global Positioning Systems  HBSC  Health Behaviour in School-Aged Children survey  ICC  Intraclass Correlation Coefficient  km  kilometers  km/h  kilometers per hour  LPA  Light Physical Activity  NFI  Normed Fit Index  NLSCY  National Longitudinal Survey of Children and Youth  NPHS  National Population Health Survey  MET  Metabolic Equivalent  MLM  Multilevel Models  MPA  Moderate Physical Activity  MVPA  Moderate to Vigorous Physical Activity  OLS  Ordinary Least Squares regression  PCCF  Postal Code Conversion File  PHAC  Public Health Agency of Canada  RMSEA  Root Mean Square Error of Approximation xii  SCARP  School of Community and Regional Planning, University of British Columbia  SEM  Structural Equation Modeling  SES  Socioeconomic Status  SRTS  Safe Routes to School  TPB  Theory of Planned Behaviour  VIF  Variance Inflation Factor  VPA  Vigorous Physical Activity  xiii  ACKNOWLEDGEMENTS I would first like to thank my supervisor Dr. Lawrence Frank, for both constantly encouraging and challenging me to push myself, but also for providing key resources that made this research possible. Dr. Frank’s dedication to his research is inspirational and my academic experience was greatly enriched through the opportunities he presented to engage with other students, practicing professionals, and researchers. I would also like to thank my other committee members, Dr. Penelope Gurstein and Dr. Michael Hayes for consistently providing thoughtful and balanced feedback. I am particularly grateful to Drs. Hayes and Gurstein for playing a pivotal role in helping me refine and clarify research questions early on, and for helping me step back from quantitative analyses to consider broader contextual considerations surrounding my research in its later stages. I would also like to acknowledge the Action Schools! BC research team lead by Dr. Heather McKay at the University of British Columbia and Dr. Patti-Jean Naylor at the University of Victoria, and including Lindsay Nettlefold, Dona Tomlin, Ashlee McGuire, Karen Strange and Melonie Burrows. I am very thankful to Dr. McKay for her leadership on the AS!BC project and for her insight on approaches to analyzing physical activity data. I am also particularly indebted to Lindsay Nettlefold who patiently and thoroughly answered countless questions I had regarding processing of physical activity data and with whom I engaged in many lengthy discussions about physical activity measurement methodology and analysis. Thank you also to Dr. Brian Klinkenberg who provided an excellent introduction to GIS through several courses at UBC Geography, enabling me to develop skills critical for my thesis research. Finally, I would also like to express my gratitude to the Canadian Institutes of Health Research, and Heart and Stroke Foundation of Canada for their funding support for this research.  xiv  DEDICATION  DEDICATION I dedicate this thesis to Catherina and to all of my family, for their love and support.  xv  1. INTRODUCTION 1.1 Overview The primary purpose of this thesis is to explore relationships between physical activity patterns of children and characteristics of the built and social environments they are situated within. This thesis draws extensively on ecological models of health behaviour (S. 2.3.2), according to which multiple environments external to individuals shape behaviour, ranging from home and school environments to national institutional environments (Sallis and Owen 2002). The focus of this thesis lies between these extremes, at the scale of community or neighbourhood environment influences (S. 2.3.5.1). Examples of neighbourhood environment characteristics studied in this thesis include: residential and commercial density, land use mix, distance to school, and median household income. Analyses were conducted using two samples of children aged eight to eleven attending nine schools in Vancouver and the surrounding Lower Mainland region of British Columbia. For the purposes of this thesis, the Lower Mainland is defined as the area consisting of the jurisdictions of Metro Vancouver and the adjacent municipality of Mission (Figure 1). Sample 1 consists of 366 students for whom objective data (defined in S. 1.4) on physical activity patterns were collected. These data were used to create a variety of measures of physical activity to use as outcome variables in the analyses, including average daily minutes of moderate to vigorous physical activity (MVPA) as the primary measure. Predictor variables were derived using objective built and social environment data integrated within a Geographic Information Systems (GIS) framework, while data on child age, gender and ethnicity were used to create control variables.  1  Figure 1: Study Area. Metro Vancouver and the Municipality of Mission illustrated as the area bounded by . Created using data from DMTI Spatial Inc. (2006).  Sample 1B consists of a subsample of 255 students from sample 1, for whom additional survey data were available. These data were used to create additional predictor variables relating to parent and child perceptions of their neighbourhood environment, travel behaviour, as well as measures of car ownership and household income. Table 1 summarizes key elements of the two samples. Table 1: Available Data for Samples 1 and 1B Available data 1. objectively measured physical activity 2. objectively measured built and social environment 3. child and parent perceptions of neighbourhood environment 4. travel behaviour 5. other covariates of physical activity – control data a) age b) gender c) ethnicity d) household income e) car ownership  sample 1 (n=366) X X  X X X  sample 1B (n=255) X X X X X X X X X 2  1.2 Research Questions Investigated The following summarizes the research questions investigated. These questions reflect available data, and thus are divided into two groups, corresponding to each of the two samples described above.  1.2.1 Research Questions Investigated Using Sample 1 Research Question 1 (Primary Research Question): Are objectively measured average daily minutes of MVPA significantly associated with objective measures of characteristics of built and social environments when controlling for child age, gender and ethnicity? Rationale: This question directly addresses the primary purpose of this thesis: to explore relationships between physical activity patterns and children’s built and social environments. Average daily minutes of MVPA was chosen as the dependent variable because of its widespread use as an indicator of physical activity patterns and policy relevance. This measure is commonly used to specify physical activity recommendations in national policy guidelines (CDC 2008, PHAC 2007a, PHAC 2007b). Age, gender and ethnicity were included as control variables given their well documented relationships with physical activity (Sallis and Owen 1999).  Research Question 2: Are objective measures of average daily MVPA outside of school significantly associated with objective measures of characteristics of built and social environments when controlling for child age, gender and ethnicity? Rationale: While overall average daily minutes of MVPA (as used in Research Question 1) is an important policy relevant measure of physical activity, it also includes a component of MVPA which takes place during school. By focusing solely on MVPA outside of school, research question two is posed to explore direct influences of built and 3  social environments on physical activity patterns without possible confounding effects of MVPA during school.  Research Question 3: Are relationships between objectively measured average daily minutes of MVPA, and objective measures of characteristics of built and social environments moderated by child age and gender? Rationale: Age and gender are well documented as covariates of physical activity for children (Sallis and Owen 1999). Parental restrictions on their children’s independent mobility may vary by age and gender (Hillman 1993, Carver et al. 2008a), which in turn may result in interaction effects between age, gender and neighbourhood environment influences on MVPA.  Research Question 4: Are objective measures of characteristics of built and social environments significantly associated with the percentage of time spent in: sedentary activity, light activity, moderate activity and vigorous activity, when controlling for child age, gender and ethnicity? Rationale: This question further expands on Research Question 1 in part by addressing environmental influences on different intensities of physical activity: sedentary activity and light activity. A secondary purpose of this question is to distinguish between environmental influences on moderate activity from those influences on vigorous activity because while commonly aggregated together, it is possible that moderate and vigorous activity are differentially influenced by built environment characteristics.  4  1.2.2 Research Questions Investigated Using Sample 1B Research Question 5: How are previously observed relationships between built and/or social environment characteristics and average daily MVPA (Research Question 1) modified when car ownership and household income are controlled for, in addition to age, gender and ethnicity. Rationale: While documented as important covariates of child physical activity (Frank et al. 2007a, Kerr et al. 2007), car ownership and household income data were not available for all students in sample 1. The purpose of this question was to re-examine the results associated with Research Question 1 in light of additional available data.  Research Question 6: When controlling for relevant socio-demographic covariates, are both objective measures of environmental characteristics, and measures of parent and child perceptions of environmental characteristics significantly associated with average daily MVPA? Rationale: While environmental characteristics may directly influence physical activity behaviour, it is possible that perceptions also play an important role. This research question was introduced to explicitly explore relationships between objective measures of environmental characteristics, perceptions, and average daily MVPA.  Research Question 7: Does a child’s mode of transportation to school mediate associations between built environment characteristics and: a) MVPA outside of school on school days, and b) average daily MVPA? Rationale: Built environment characteristics have been postulated to influence physical activity patterns through multiple types of behaviours, including walking or biking to school (Cooper 2005). That is, built environment characteristics may influence mode  5  choice for a child’s trip to school, which may in turn influence broader physical activity patterns. This research question is introduced to explore such possibilities.  1.2.3 Classification of Research Questions Figure 2 presents a classification of the research questions listed above. As illustrated in this figure, the Primary Research Question (Research Question 1), addresses relationships between built and social environment characteristics and overall average daily MVPA. Several of the other questions in turn build on this question, by incorporating additional variables (e.g. Research Question 3 incorporates age and gender as moderators). In contrast to these questions, Research Questions 2 and 4 address different relationships, focusing on measures of MVPA outside of school and percentage of time spent in sedentary, light, moderate and vigorous activity as dependent variables. Finally, Research Question 7 employs both a measure of overall average daily MVPA and measures of MVPA outside of school as dependent variables.  1.3 Background and Context This thesis is situated within a broader literature on neighbourhood environment correlates of physical activity. Barnett et al. (2006) note that intrapersonal models of behaviour have only been able to explain approximately 20 to 40% of the variance in health behaviours such as physical activity. Because of the limited explanatory power of such models, researchers have increasingly turned towards conceptual models incorporating a broader range of factors, such as ecological models accounting for both intrapersonal characteristics as well as environmental characteristics. Built and social environment correlates of physical activity for adults have been extensively studied, with findings documented in numerous literature reviews (Badland and Schofield 2005, Frank and Engelke 2001, Saelens et al. 2003, TRB 2005), interdisciplinary synthesis articles (Handy et al. 2002, King et al. 2002, Sallis et al. 6  Figure 2: Classification of Research Questions. Shaded boxes indicate questions based on sample 1B.  Main Relationships Studied built and social environment measures of MVPA outside of school,  built and social environment overall average daily MVPA, controlling for age, gender and ethnicity  controlling for age, gender and ethnicity  address main relationships build on research questions addressing main relationships, with additional variables noted  built and social environment percentage of time in sedentary, light, moderate and vigorous activity controlling for age, gender and ethnicity  Research Question 2  Research Question 1  Research Question 4  Primary Research Question Research Question 3 additional moderators = age, gender  Research Question 5 additional controls = car ownership and household income  Research Question 6 additional explanatory variables = measures of parent and child perceptions  Research Question 7 additional mediator = mode of transportation to school  7  2004) and books (Frank et al. 2003, Frumkin et al. 2004). In contrast, although correlates of physical activity for children in general have received a great deal of attention, built environment correlates specifically were initially studied less extensively. Further, relationships between built environment characteristics and child physical activity cannot be assumed to be the same as with adults. Children spend a large proportion of their time at school, are not able to drive, are generally subject to restrictions placed on them by their parents and other adults, and are likely interested in different out-of-home destinations than adults (Davison and Lawson 2006, Timperio et al. 2006, Frank et al. 2007a). Trends in increasing prevalence of childhood overweight and obesity have more recently prompted considerable interest in the physical activity patterns of children (Shields 2006), and have resulted in a proliferation of research on neighbourhood environment correlates of child physical activity. In Canada, recent national level data on child physical activity patterns indicate that 88% of children do not engage in adequate physical activity, based on Canada’s physical activity guidelines for children and youth (CFLRI 2009). Such statistics are of consequence because there are numerous health costs associated with inadequate physical activity, including increased risk of overweight or obesity (LeBlanc 2003, Yeung and Hills 2007). In 1978/9, 12% of 2-17 year olds in Canada were overweight, and 3% were obese (Shields 2006). By 2004, the numbers had climbed to 18% overweight and 8% obese (ibid). In addition to being negatively associated with the prevalence of overweight and obesity, regular physical activity is associated with other health benefits. These include: lower risk of heart disease and type 2 diabetes, improved maintenance of muscle mass, and reduced prevalence of anxiety (Armstrong and Welsman 1997, Byrne and Hills 2007, Sallis and Owen 1999).  8  1.4 Study Significance As a result of current interest in the study of neighbourhood environment correlates of physical activity in planning and public health, this thesis addresses questions relevant to researchers in both fields. Given the emphasis on built environment correlates of physical activity in particular, this research is also relevant to practicing planners. With these audiences in mind, the significance of this thesis may be described in terms of its relevance to future research and planning practice. This study will contribute to a rapidly expanding field of research on neighbourhood environment correlates of physical activity behaviour which is cumulatively beginning to shape land use planning practice. Many of the objective measures incorporated in this analysis correspond to metrics used in a variety of planning instruments, as indicated in Table 2. Evidence from this and related studies can thus be used to inform such instruments. Table 2: Examples of Correspondence Between Measures Used in Analysis and Planning Instruments Planning Instrument regional plans and growth strategies  Relevant measure(s) used in the present analysis residential and commercial density  zoning bylaws  land use mix, residential and commercial density  subdivision design guidelines and official community plans for new neighbourhoods  street connectivity, cul de sac density  school siting and closure policies  distance to school  While contributing to the broader literature shaping such policies, this study is also important on its own because it is one of relatively few studies conducted in the context of a major Canadian metropolitan area.  9  The proposed research will also help to fill an important gap in the existing literature, through the use of objective measures for both environmental characteristics and physical activity, supplemented by a variety of survey measures. Although objective measures of both environmental characteristics and physical activity have been employed in some studies (e.g. Jago et al. 2006, Kligerman et al. 2007), these measures remain underutilized (Davison and Lawson 2006, Pont et al. 2009). In the context of research on correlates of physical activity, the term ‘objective measures’ generally refers to quantitative data collected directly by researchers using standardized methodologies with replicable results. Such measures can be contrasted with those based on information provided by research subjects which may be subject to considerable uncertainty and bias arising from respondent perceptions. Examples of objective measures are contrasted with their alternatives in Table 3. Objective measures have some notable advantages relative to the alternatives. Objective measures of built environment characteristics may, for example, be more easily interpretable and translatable into policy recommendations than measures of perceptions, because they are typically already expressed in terms of metrics used by planners. Surveys and travel diaries are generally considered to be less reliable than objective measures for the assessment of certain critical dimensions of physical activity in children (Armstrong and Welsman 1997, Trost 2007), and considerable discrepancies have been documented between objective and survey measures of physical activity (Active Healthy Kids Canada 2007). Survey measures of physical activity specifically are prone to reporting bias, with children or their families often overstating their physical activity levels (Reilly et al. 2008). Despite the advantages of objective measures, it is worth noting that survey measures can be used to gauge distinct, complementary characteristics that may not be measurable using objective means. For example, children’s or parent’s perceptions of the environment may motivate behaviour as much or more than the physical characteristics being perceived (Davison and Lawson 2006). This thesis benefited from the use 10  of data on measures of perceptions of built and social environments in the analysis, in addition to objective measures. Table 3: Objective Measures Contrasted with their Alternatives. Objective Measures  Alternative Measures  Neighbourhood Built / Social Environment  Objective built or social environment measures are predominantly derived from municipal or third-party spatially referenced databases on land uses, street networks, and sociodemographics, using GIS. Examples include: measures of residential density and street connectivity, calculated using a 1 kilometer radius around a research subject’s residence (e.g. Liu et al. 2007).  The main alternative to objective built or social environment measures are measures of perceptions of the environment, generally assessed using survey instruments (Hoehner et al. 2003). Surveys may for example be used to gauge a respondent’s perceptions of traffic levels, or of safety in their neighbourhood (e.g. Mullan 2003).  Physical Activity  A wide variety of objective physical activity measurement methodologies are available, including (Sallis and Owen 1999): • accelerometers, which were used to collect data used in the present study (S. 3.2.1.1); • pedometers, to count daily steps; • heart rate monitors, to gauge variation in heart rates over the course of the day; and, • observational assessment of physical activity by trained researchers.  Alternatives to objective physical activity measures include: • data collected using surveys or activity diaries, completed either directly by children (self-reported measures) or by adults knowledgeable of their behaviour (Trost 2007).  To take full advantage of the objective built environment data available, this thesis also explicitly addressed an important methodological consideration: how best to define neighbourhood environments for the purposes of assessing environmental characteristics. This was achieved by creating measures based on multiple geographic scales, and comparing results across these scales, as recommended in a recent review (Brownson et al. 2009). Finally, in 11  addition to incorporating widely used measures of environmental characteristics (and thus enabling some degree of comparability to past research), this study also incorporated some novel built and social environment measures, including for example, a measure of the proportion of streets in a child’s neighbourhood with speed limits of 30 kilometers per hour or less.  12  2. LITERATURE REVIEW 2.1 Overview The following review is structured around two sections: an overview of child physical activity patterns and associated health implications, and a review of the literature on neighbourhood environment correlates of these activity patterns. The first section begins with a description of how physical activity is characterized, emphasizing its characterization in studies of neighbourhood environment correlates. This is followed by a brief discussion of possible compensation effects relevant to the study of specific types of physical activity. Next, physical activity guidelines are discussed to indicate points of reference relevant to the quantitative measurement of physical activity. Data on physical activity engagement and trends are then highlighted in relation to these guidelines. Following this, health implications associated with physical inactivity are considered. Finally, health risks associated with physical activity in the neighbourhood environment are noted, with an emphasis on spatial patterns of these risks within urban areas. The second section begins by briefly introducing theories of physical activity behaviour, as they relate to specific classes of correlates of physical activity. This discussion then focuses on ecological models of health behaviour which underlie many studies of neighbourhood environment correlates of physical activity, and which form the theoretical basis for this thesis. As part of this discussion, a typology of correlates of physical activity is presented. Following from this typology, covariates of physical activity often used as control variables in studies of neighbourhood environment correlates are highlighted (e.g. age, gender and ethnicity), in addition to other relevant individual and household characteristics. Specific neighbourhood environment correlates are then introduced with a discussion of alternative approaches to the definition of neighbourhood environments. Finally, empirical studies of a broad range of built 13  and social environment correlates of physical activity are reviewed. The chapter concludes by considering some of the limitations of studies to-date. The studies reviewed in this chapter primarily examine subjects ranging in age from young childhood to adolescence, thus surrounding the age range of children participating in the present study (8-11 years). In a few instances, studies examine a broader age range, extending to young adulthood (for example, Frank et al. 2007a).  2.2 Child Physical Activity Patterns 2.2.1 Characterizing Physical Activity The technical definition of physical activity is very broad, referring generally to any form of muscular movement that results in energy expenditure (Sallis and Owen, 1999). Thus, it is evident that a first step in any research on correlates of physical activity is to identify specific types of physical activity of interest. There are many different approaches to characterizing physical activity; this section highlights some of the major dimensions of physical activity and classification schemes relevant to children. Physical activity is commonly described in terms of duration, frequency, intensity and type or mode (Katmarzyk and Tremblay 2007). The three dimensions of frequency, duration, and intensity together represent the major quantitative dimensions of physical activity, and can therefore be used to create measures of overall volume of physical activity which can subsequently be linked to particular health benefits. For example, guidelines for children in the United States specify a minimum of one hour per day of physical activity, most of which should be comprised of moderate or vigorous intensity aerobic activities (CDC 2008). In this example, the duration specified is one hour, at a daily frequency, with a recommended intensity of moderate or vigorous.  14  In general, duration refers to the length of time that children engage in physical activity, often expressed in terms of specific episodes or bouts of physical activity. Children spend most of their time in low intensity activities interspersed with sporadic bouts of high intensity physical activity (Tomson et al. 2007). Longer bouts of physical activity may, however, be associated with additional health benefits above and beyond those associated with more sporadic physical activity (Mark and Janssen 2009). Alternately, duration may simply refer to the total amount of time spent in physical activity over a specified time frame, such as the one hour per day recommendation in the US guidelines noted above. Frequency refers to the number of bouts of physical activity engaged in over a given time frame, usually a day or a week (Marshall and Welk 2008). Specific activities may occur frequently in a child’s regular routine, and targeting these types of activities may present opportunities for increasing overall physical activity levels. For instance, walking to and from school may be an important source of physical activity for children because it can become part of their regular routine. More generally, children may benefit more from interventions supporting unstructured activities like active free play which can occur in short bouts spontaneously throughout the day, rather than structured activities which may require more preplanning. Intensity describes the level of exertion at which a physical activity is performed, or the “magnitude of the physiologic response to physical activity” (Marshall and Welk 2008, p.8). Intensity is substantially more difficult to gauge than either duration or frequency, and debates continue regarding the definition and measurement of different levels of intensity (Guinhouya et al. 2006). Intensity of physical activity is usually expressed in terms of three specific thresholds: light, moderate or vigorous (CDC 2007). These may in turn be contrasted with lower intensity sedentary activities. Specific activities may be performed at different levels of intensity. Walking, for instance, may commonly be performed at either light or moderate 15  intensity (Andersen et al. 2006, Jago et al. 2006). Some health benefits may only be associated with a certain minimum level of intensity. The majority of health benefits are generally assumed to be associated with physical activity of at least moderate intensity (Guinhouya et al. 2006). Because of this, measures of moderate-to-vigorous physical activity (MVPA) are commonly cited dependent variables (e.g. de Vries et al. 2007, Kligerman et al. 2007). Intensity thresholds may be quantitatively defined in terms of metabolic equivalents (METs), which are multiples of resting metabolism (Trost 2007). Commonly used thresholds for the different intensities of physical activity, expressed in terms of METs, are highlighted in Table 4, together with examples of activities corresponding to the different intensities.  Table 4: Commonly Used Thresholds for Sedentary Activity and Light, Moderate and Vigorous Physical Activity (Guinhouya et al. 2006, PHAC 2007a) Intensity  METs  Examples of corresponding activities  sedentary  <1.5  watching TV, accessing the internet  light  1.5-3  strolling / slow walking**  moderate  3-6*  brisk walking**, biking, swimming, skating, outdoor play  vigorous  >6  jogging, running, hockey, soccer  * Although the threshold of 3.0 METs is commonly used (e.g. de Vries et al. 2007, Roemmich et al. 2007, Trost et al. 2002), there is some controversy as to the appropriateness of this (Pate et al. 2006). Specifically, Pate et al. (2006) note that although there is consensus that this threshold is appropriate for adults, resting energy expenditure for children is higher, so a value of 3 METs for children represents a lower relative activity for children. Differences in metabolic and biomechanical efficiency due to growth and maturation further complicate matters (ibid). Finally, Pate and colleagues criticize the use of abrupt thresholds as they represent an artificial categorization of continuous data. ** A fine line may be drawn between slow walking, which may be considered a light intensity physical activity (Jago et al. 2006) and brisk walking, which is generally considered to be MVPA (Andersen et al. 2006). Andersen et al. (2006) specifically equate walking at about 4 km/h with the low end of MVPA. Type or mode of physical activity are terms used to distinguish between qualitatively different types of physical activity. They may be used to refer to a distinction between activities 16  such as walking, cycling, running or climbing (as in Trost 2007) or between different physical activities based on the types of physical fitness they target: endurance, flexibility or strength (as in Janssen 2007). They may also be used to refer to the main physiologic systems employed: aerobic or anaerobic (as in Marshall and Welk 2008). Studies of environmental correlates of physical activity often reference two specific types of physical activity in particular: independent mobility or activities, and active transportation. Independent activities are of interest in part because of the wide ranging benefits for children beyond benefits of supervised activity, including: experiential learning and the discovery of personal limits (Malone 2006, Ungar 2007), the acquisition of environmental knowledge (Risotto and Tonucci 2002), and greater opportunities for social interaction (Boardman and Saint Onge 2005). In addition, the more mobile children are, the more likely they are to interact with their environment and acquire environmental knowledge. This in turn creates an increased motivation to explore and be more mobile (Kytta 2004). Finally, the supervision and chauffeuring of children may also reduce the quality of life for caregivers, by adding trips or limiting work schedule or job opportunities (McMillan 2005). Both independent and supervised activities occur throughout the environment, but the extent of independent mobility granted to children may be particularly sensitive to neighbourhood built and social environment characteristics as perceived by parents. While site specific design might encourage limited independent mobility (e.g. design of a local playground to incorporate adequate lighting and clear sight lines), truly independent mobility may require that the built environment is designed for safe accessibility by children in a much more encompassing sense, incorporating elements such as neighbourhood traffic calming (Freeman 2006, Tranter 2006). Parent perceptions of neighbourhood safety and the resultant restrictions placed on children are discussed further below (S. 2.3.4.2.1).  17  Active transportation refers to physical activity for the purposes of transportation, primarily walking and cycling (Sallis et al. 2004). Active transportation may be an independent activity, or supervised (e.g. walking school buses). The predominant form of active transportation for children of interest to researchers to-date is travel to school (e.g. Evenson et al. 2006, Timperio et al. 2006). This form of active transportation provides an important opportunity for increasing daily physical activity levels because going to school is a regular activity, and active commuting to school can thus become part of a child’s regular routine. Active transportation is of particular interest to planners because it takes place in large part on transportation facilities (e.g. roads, sidewalks) and is dependent on the design and connectivity of these facilities. Active transportation is not only a form of physical activity in and of itself, but may also grant children access to more physical activity opportunities at parks or recreational facilities. It may also hold numerous other benefits for children if it is undertaken as an independent activity. Finally, active transportation is also associated with health benefits that extend beyond the individual child, to the broader neighbourhood as a whole, especially when it substitutes for transportation by private vehicles. These benefits include reduced noise, air pollution and traffic hazards associated with reduced traffic (Gleeson and Sipe 2006, Wilson et al. 2007). In contrast to active transportation, certain types of structured activities may be restricted to very specific settings such as ice rinks and basketball courts. Thus, participation in such activities may primarily be influenced by the degree of access to these specialized settings. Access to structured activities like organized sports is not only a function of physical proximity, but also socioeconomic status because these activities often have costs associated with access. In contrast, activities such as active transportation may be hypothesized to be associated with more wide ranging built environment characteristics because they occur throughout the built environment. In this vein, Frank et al. (2005) suggest that walking is expected to be more 18  sensitive to community design than vigorous activities such as running or team sports. Jago et al. (2006) speculate that youth engage in a larger proportion of structured physical activity in specific settings than adults, and thus may be less sensitive to environmental characteristics than adults. They further speculate that activities occurring in the broader built environment which are common to adults (such as walking and jogging) may play a less dominant role for children. Given the wide ranging options for characterizing physical activity, studies of environmental correlates of physical activity specify a variety of different measures of physical activity as dependent variables. While the variety of measures used often limits direct comparability across studies, broad groupings of similar variables can be identified, as illustrated in Table 5.  Table 5: Dependent Variables Commonly Used in Studies of Environmental Correlates of Physical Activity Type of Dependent Variable overall volume of physical activity  Examples •  Average daily minutes of light physical activity per day or minutes of MVPA per day, as in Hume et al. (2005), Jago et al. (2005)  walking or active transportation in general  •  Walking above a threshold distance or more than a minimum number of trips per specified time period, as in Frank et al. (2007a) Walking frequency per week, as in Alton et al. (2007)  active transportation to school  •  independent activities or mobility  •  •  •  •  Number of times active commuting per week, as in Kerr et al. (2006), Schlossberg et al. (2006) Mode of commute to school, as in Cooper et al. (2005). Child allowed to cross roads or play outside without adults, as in O’Brien et al. (2000) Percentage of trips made alone, as in Clifton (2003).  An additional factor limiting direct comparability of absolute outcomes across studies relates to differences in measurement methodology. Options include both indirect measures (e.g. travel diaries completed by children or their parents), and direct or objective measures  19  derived using equipment such as pedometers. Measurement methodologies are discussed further in Appendix 1. 2.2.2 Compensation Effects Studies finding associations between specific types of physical activity and neighbourhood environment correlates may need to be interpreted with caution because an increase in one type of physical activity does not necessarily mean that the overall volume of physical activity also increases. One type of physical activity which has been particularly well studied in this regard is active commuting to school. That is, researchers have examined whether children who commute to school on foot or by bike also engage in more physical activity in general. In this vein, Cooper et al. (2005) found that both boys and girls who walked to school were significantly more physically active in general than those who traveled by car. Similarly, Cooper found that boys (but not girls) who cycled to school were more physically active than those traveling by car. Cooper noted that the journey to school itself contributed relatively little to overall physical activity levels, possibly suggesting that the act of active commuting might somehow engender in children a motivation to engage in further physical activity. An alternate hypothesis is that naturally active children choose to travel by active means. Based on an evaluation of weekend physical activity data, Cooper concluded that observed differences in overall activity levels could not be explained by the natural differences in activity behaviour of children, but rather was related to active commuting. Faulkner et al. (2009), based on a broader review of studies examining associations between active transportation to school and physical activity outcomes, also concluded that children who actively commute to school tend to be more physically active overall than those driven or bused to school. Nonetheless, such findings do not necessarily extend to other activities. Notably, active commuting to school may be a special 20  case because it is a regular, routine source of physical activity. Other incidental recreational activities (like unstructured play in a park) may in contrast substitute for each other, with children compensating for an increase in one type of physical activity by limiting other types of activities. Hargströmer et al. (2009) lend support to this possibility, finding that an organized weekly exercise intervention occurring over 13 weeks resulted in decreased total daily physical activity among obese adolescents participating in the trial, compared to a control group. 2.2.3 Child Physical Activity Guidelines Sallis and Owen (1999) observe that recommendations for physical activity in children and adolescents have historically received less attention than those for adults. When first developed by the American College of Sports Medicine in 1988, youth physical activity guidelines were thus based on adult guidelines, and recommended 20-30 minutes of vigorous exercise per day (Andersen et al. 2006). Guidelines for children were not revised to be substantially different from those for adults until 1988, when the UK Health Education Authority commissioned a series of reviews to update the guidelines. The updated guidelines specified a minimum of one hour per day of at least moderate physical activity, and recommend that those children who did little activity should participate in at least 30 minutes of MVPA per day (ibid). The adopted guidelines also recommended that some activities should be performed specifically to enhance muscular strength, flexibility and bone health at least two days a week (Sallis and Owen 1999). Finally, more recent reviews have reconfirmed the validity of the Health Education Authority one hour per day recommendation (Andersen et al. 2006, Byrne and Hills 2007). International guidelines are now generally consistent with the Health Education Authority recommendations, including for example the US and Australian guidelines, both of which specify one hour per day of MVPA (Byrne and Hills 2007, CDC 2008). Canada’s 21  physical activity guidelines for children and youth were recently revised and are now also generally consistent with the Health Education Authority guidelines, specifying a minimum accumulation of 60 minutes of MVPA daily for children aged 5-11 (CSEP 2011a). The guidelines further recommend vigorous-intensity activities and activities that strengthen muscle and bone at least three days per week. In addition to physical activity guidelines, guidelines have been adopted for sedentary behaviour, recommending a maximum screen time of two hours per day, and more generally limiting sedentary transport, sitting time and time spent indoors (CSEP 2011b). 2.2.4 Current Physical Activity Levels and Trends Despite the high levels of interest in trends in child physical activity, it is widely acknowledged that Canadian data on physical activity patterns of children is lacking (Active Healthy Kids Canada 2007, Tremblay 2007), and especially, reliable data on trends over time (Katmarzyk and Tremblay 2007). Nonetheless, existing data, with few exceptions (such as Eisenmann et al. 2004), provides strong indications that Canadian children are engaging in inadequate levels of physical activity (Active Healthy Kids Canada 2010, CFLRI 2005), and that this may be associated with rising rates of overweight and obesity (Shields 2006, Tremblay and Willms 2003). Major sources of physical activity data on Canadian children and youth include: the Canadian Fitness and Lifestyle Research Institute (CFLRI)’s Physical Activity Monitor (PAM), the 2005 Survey of Canadian Schools, and the 2005-2009 Canadian Physical Activity Levels Among Youth (CANPLAY) study; Statistics Canada’s National Longitudinal Survey of Children and Youth (NLSCY), the National Population Health Survey (NPHS) and more recent Canadian Community Health Survey (CCHS); and the World Health Organization’s Cross  22  National Collaborative Health Behaviour in School-Aged Children (HBSC) surveys (Active Healthy Kids Canada 2007, CFLRI 2009, PHAC 2007b, Tremblay 2007). In 2007, Active Healthy Kids Canada published a comprehensive assessment of Canadian children’s physical activity patterns, drawing on the CANPLAY data together with many of the other data sources noted above (HBSC, PAM, CCHS and NLSCY) and other regional surveys (Active Healthy Kids Canada 2007). This report included assessments of physical activity, inactivity (e.g. measures of television and computer use or ‘screen time’) and sports participation. Based on these assessments, Active Healthy Kids Canada concluded by giving Canada a grade of ‘F’, representing a decline from previous years. However, this decline was based primarily on the CANPLAY data, which was given heavier weighting because it was objectively collected (using a pedometer), and therefore considered to be more reliable than other self-reported data. The HBSC data was one of the data sources which was discounted because it was self-reported. In contrast to the CANPLAY data, the HBSC indicated a 9.2% increase in physical activity levels of children and youth from 2002 to 2006. Active Healthy Kids Canada thus acknowledges that although their rating was lower than in previous years, this change does not necessarily reflect real changes over time in physical activity patterns, but may rather reflect improved measurement methodology (i.e. the use of pedometers). More recent CANPLAY data indicates that a slightly higher proportion of children are engaging in adequate levels of MVPA than in previous years (CFLRI 2009). However, absolute numbers of children obtaining adequate MVPA are still very low, at approximately 12% in 2007-2009, based on guidelines established by Canada’s Physical Activity Guide (ibid). Results across studies also relatively consistently indicate differences by gender and age, whereby boys engage in more physical activity than girls, and younger children engage in more physical activity than older children (CFLRI 2009, Iannotti et al. 2009, Shields 2006).  23  Data on active transportation as a specific form of physical activity is also limited, but is consistent with low and potentially decreasing levels of physical activity in children and youth. A survey conducted almost a decade ago by Environics Canada found that only 37% of 5- to 13year-olds and 33% of 14- to 18-year-olds walked to school at least half the time (Go for Green, Environics 1998). Further, only 2% and 4% of students in these respective age groups cycled (ibid). More recent survey data from CFLRI (2005) suggests that only 26% of children and youth aged 5-17 use only active modes of transportation to commute to and from school. However, these data are not directly comparable to the Environics data, as 13% of children were reported to use a combination of both active and inactive modes of transportation (CFLRI 2005). Another recent study, focusing on the Greater Toronto Area specifically, similarly found low levels of active commuting, at approximately 42.5% for 11-13 year olds and 30.7% for 1415 year olds in 2006 (morning walking mode share; Buliung et al. 2009). Notably, this study also examined trends over time, finding substantial decreases for both age groups between 1986 and 2006 (ibid). Trends towards higher levels of sedentary leisure activities have been well documented, likely contributing towards decreases in overall energy expenditure and time for physical activity (Eisenmann 2006, King 2007, Molnar et al. 2004). Notably, children now have access to a wider range of reinforcing sedentary activities at home than in the past, including not only TV, but also the internet (enabling the use of social networking web-sites such as FaceBook) as well as video games. In Canada, data on sedentary activity in children and youth indicates that (Spanier et al. 2006): 20-30% of 11-15 year olds watch more than 4 hours of TV per day, large numbers of children are beginning to watch TV at an earlier age and in greater amounts, and  24  more Canadian children report playing more than 4 hours of video games per week than children in other countries. Increases in sedentary activities may be important not only because they take away from time available for physical activity, but also because eating often occurs in conjunction with such activities, and the food eaten is often energy-dense snack foods (King 2007). Such unhealthy behaviour may be further reinforced by TV advertisements for energy dense snack foods and drinks (Deforche et al. 2007). 2.2.5 Health Implications of Physical Inactivity There are substantial health consequences associated with inadequate physical activity in children. Notably, inadequate physical activity may be directly linked to trends in increasing childhood overweight and obesity (WHO 2006), because weight status is a function of both energy intake (diet) and energy expenditure (via resting metabolic processes and physical activity). That is, overweight or obesity result from energy intake exceeding energy expenditure over prolonged periods of time (LeBlanc 2003, Yeung and Hills 2007). In a recent review, Jiménez-Pavón et al. (2009) found consistent evidence of negative associations between objectively measured physical activity and adiposity of children. Overweight and obesity in childhood are in turn associated with a range of health problems including: increased prevalence of type 2 diabetes; sleep apnea and asthma; hypertension and adverse lipid profile; social stigma and decreased self esteem; back pains and other orthopaedic problems; and non-alcoholic fatty liver disease and other gastrointestinal problems (Denney-Wilson and Baur 2007, Yeung and Hills 2007). In Canada, in 1978/9, 12% of 2-17 year olds were overweight, and 3% were obese (Shields 2006). By 2004, the numbers had climbed to 18% overweight and 8% obese (ibid). Thus more than one quarter of Canada’s children and youth were either overweight or obese in  25  2004. In youth aged 12 to 17 in particular, the prevalence of obesity tripled from 1978/9 to 2004 (ibid). Interventions to reduce the prevalence of obesity or overweight may increase energy expenditure through physical activity and/or reduce energy intake through dietary modifications. In contrast to interventions based solely on dietary modifications, the incorporation of increased physical activity can also promote the maintenance of muscle mass as well as other health benefits described below. Further, in addition to directly affecting energy balance, physical activity may also contribute to energy expenditure by increasing the resting metabolic rate, leading to favorable energy utilization, although specific mechanisms are unclear (Goran et al. 1999, King 2007). Broader benefits of physical activity among children include: promotion and maintenance of muscle mass and improved aerobic fitness; reduced risk of osteoporosis in later life by enhancing peak bone mineral density; and improvements in self-esteem as well as reductions in anxiety, stress and depression (Armstrong and Welsman 1997, Byrne and Hills 2007, Sallis and Owen 1999, Tomson et al. 2007). Independent mobility has also been linked to specific health benefits as noted above, and one recent study found that walking to school might be protective against cardiovascular disease by decreasing cardiovascular reactivity and perceived stress during the school day (Lambiase et al. 2010). In addition to providing short term health benefits, some evidence exists of tracking of physical activity behaviour, weight status, and the health consequences of childhood weight status, from childhood to adulthood (Kristensen et al. 2008, Chin A Paw et al. 2007, DenneyWilson and Baur 2007, Gordon-Larsen et al. 2004). For example, adults who were overweight as children or adolescents have an increased risk of various morbidities such as cardiovascular diseases, an effect which may remain even after adjustment for adult weight (Chin A Paw et al. 2007, Denney-Wilson and Baur 2007). Gordon-Larsen et al. (2004) found stability of sedentary 26  behavior between adolescence and young adulthood in a longitudinal study of national level US data, concluding that while preventive efforts are needed at all life cycle stages, “they are critically needed before adolescence” (p. 282). The need to promote physical activity starting at an early age has also been argued by many others (including Armstrong and Welsman 1997, Byrne and Hills 2007, Sayre and Gallager 2001). 2.2.6 Health Risks Associated with Physical Activity in Neighbourhood Environments 2.2.6.1 Exposure to Pollution Exposure to air pollution has been associated with wide ranging health consequences for children, including: an increased general rate of mortality, slowed lung growth, reduced lung function, increased sickness rates, aggravation or causation of asthma and increased prevalence of respiratory symptoms more broadly (Bates 1995, Brunekreef and Holgate 2002). Common outdoor air pollutants having documented associations with child health outcomes include Nitrogen dioxide (NO2), particulate matter (PM10, PM2.5 ; where 10 and 2.5 refer to the maximum particle diameter in microns) and Ozone (O3) (Kunzli et al. 2003). Air pollutants can originate from a variety of point sources such as factories, dispersed area sources such as lawn mowers, or mobile sources including traffic (Frumkin et al. 2004). In urban areas in developed countries, air pollution associated with traffic density is a particularly important problem, with children facing potentially higher exposures than adults as a result of outdoor physical activity (Bates 1995). Spatial variations in pollutant levels within urban areas can be substantial and may result in highly variable levels of exposure. In one recent study involving children in the San Francisco Bay Area, Ostro and Kim (2009) found that the highest risks of asthma were for those exposed to very high levels of traffic density or living within 75 meters of a freeway or highway. They conclude that proximity to traffic is associated with respiratory health 27  consequences for children, even in a region with good air quality. Numerous other studies also indicate that exposure to traffic pollutants poses serious health consequences for children (Potera 2008). A recent study set in Vancouver, British Columbia mapped spatial variations in air pollutant concentrations in relation to walkability (Marshall et al. 2009). The walkability measure used in this study incorporated measures of residential density, commercial density, land use mix and street connectivity and is discussed further below (including in Appendix 4). This study found that neighbourhoods with high walkability tend to have high levels of Nitric Oxide (NO) but low concentrations of Ozone. In contrast, low walkability suburban areas were characterized by high levels of Ozone. Relatively few high walkability neighbourhoods with low levels of air pollution were identified. These neighbourhoods tended to be high income neighbourhoods located near, but outside of the city center. Air pollution exposure is a function of both concentration of pollutants and exposure time (Setton et al. 2008). Thus, children’s exposure to pollutants may be strongly impacted by the location of specific sites or facilities which children spend large amounts of time in. Schools located near major roads may therefore result in high exposures to pollutants (Potera 2008). In contrast, exposure to air pollution during transportation to and from school may be limited due to relatively short exposure times. Travel mode choice may nonetheless influence air pollution exposure, with rates of exposure to certain pollutants of up to 10 times higher in cars than outdoors (Kunzli et al. 2003). Several studies have similarly found elevated levels of diesel exhaust inside school buses (Solomon et al. 2001, Wargo et al. 2002). Conversely, walking to school may increase children’s exposure relative to being driven if walking routes in relation to traffic patterns around schools are not taken into account (Kunzli et al. 2003). Children are also exposed to noise pollution in urban settings. As with air pollution, urban noise levels may vary considerably spatially, and such variation is due in part to traffic levels. In one study, Seto et al. (2007) found that urban noise increased by 6.7 decibels with a 28  10 fold increase in street traffic. Bus and heavy trucks were found to be particularly important contributors to noise. Exposure to noise pollution has several health consequences for children, including impairments of specific cognitive functions (e.g. difficulties concentrating and poorer reading ability), decreased motivation, increased blood pressure, increased endocrine levels and noise annoyance (Stansfeld and Matheson 2003).  2.2.6.2 Child Pedestrian and Cyclist Injuries due to Traffic Child pedestrian and cyclist injury trends in Canada are comprehensively outlined in Safe Kids Canada (2007). Among the findings of this report, child pedestrians constitute the second highest category of unintentional injury deaths for children aged 0-14. This is second only after child passengers of motor vehicles. In absolute numbers, approximately 56 child pedestrians under age 14 die annually in Canada, and approximately 780 are hospitalized with serious injuries. Child cyclists comprise 5% of total unintentional injury deaths for this age group. According to the report, approximately 70% of deaths and 50% of serious injuries occur in the absence of traffic controls, for example when children are attempting mid-block crossings or walking out from between parked cars. While remaining a major source of injuries and deaths, pedestrian accidents were also characterized by the highest decline in combined death and hospitalization rates from 1994 to 2003 (ibid). Although traffic related deaths have been falling in Canada and in other developed countries, this may be in large part attributed to the increasing withdrawal of unsupervised children from the urban environment as greater restrictions are placed on their independent travel (UNICEF 2001, Sonkin et al. 2006). It is for this reason that the US Transportation Research Board has recommended that efforts to encourage more walking and cycling among children be implemented with care, and specifically that mitigation measures are simultaneously implemented to reduce risk of injury (TRB 2005).  29  Young children are susceptible to traffic related injuries for a number of reasons, including, for instance, slower movement which translates into longer times to cross streets and effectively increases their risk exposure (Wier et al. 2009). In addition, the cognitive ability of children to safely cross streets may not be sufficiently developed until beyond middle childhood (Barton 2006). Both age and gender are risk factors for pedestrian injury rates, with rates of injury higher for boys and peaking between the ages of 5-11 (ibid). A social gradient has also been consistently observed, whereby higher rates of pedestrian injury are noted in lower income children, although the reasons for such patterns are poorly understood (Laflamme and Diderichsen, 2000, Lascala et al. 2004). Wide ranging factors influence the spatial distribution of injuries, but important mediating factors include traffic volume and speed, and also volume of pedestrians and cyclists (Jacobsen 2003). In a recent review, Ewing and Dumbaugh (2009) conclude that the traffic environments of dense urban areas are safer than those of the suburbs, an observation which they attribute to lower per capita travel distances, lower speeds and more restrictive design treatments (e.g. narrower lanes) in urban areas. The influence of traffic volume and speed has been documented in a number of studies of injuries around schools, two of which are now considered as examples. First, in a study of children and youth in Orange County, Florida, Abdel-Aty et al. (2007) found that high crash involvement was associated with high and middle school children. They observe that this is because middle and high schools are frequently located adjacent to major roads with high posted speed limits. In contrast, in a recent study of pedestrian injury ‘hot spots’ in Vancouver, Schuurman et al. (2009) found that schools were not a type of land use associated with hot spots, a finding which they attribute to road safety engineering in the proximity of schools, including speed humps and reduced speed limits.  30  2.3 Environmental Correlates of Physical Activity 2.3.1 Theories of Physical Activity Behaviour As outlined by Spence and Lee (2003), there is no single agreed upon universal theory or model of physical activity behaviour. What is increasingly agreed upon is that there are a wide variety of correlates of physical activity, spanning individual, physical environmental and social environmental characteristics (Barnett et al. 2006, Sallis and Owen 1999, TRB 2005). The influence of such wide ranging correlates on physical activity behaviour reflects the complex nature of physical activity, as described above. Theories of physical activity behaviour may be broadly described as specialized, emphasizing a specific correlate or group of related correlates, or integrative, emphasizing a wider range of correlates. Examples of the former include many theories developed in the health sciences emphasizing psychological characteristics, including the theory of planned behaviour, the transtheoretical model and the health belief model: Theory of Planned Behaviour (TPB). The TPB posits that the intent to perform a behaviour is the immediate correlate of a given behaviour (Craig et al. 1996). The TPB in turn highlights the importance of three cognitive factors in shaping intent: attitudes towards the behaviour, subjective norms and perceived behavioural control (ibid). Transtheoretical Model. The Transtheoretical model specifies discrete ‘stages of change’ through which an individual progresses as they make behavioural changes such as increasing the amount of physical activity they engage in. The stages are: precontemplation, contemplation, preparation for action, action, and maintenance (Sallis and Owen 1999). Individuals can be identified as falling in one of the stages, and specific processes of change necessary to successfully transition to the next stage can be identified. Specific correlates of physical activity may thus be associated with specific stages of change. For instance, the primary difference between precontemplators and contemplators is that contemplators intend 31  to change. Thus, as with the TPB, intent is identified as an important correlate of physical activity (ibid). Health Belief Model. The health belief model is based on the premise that the likelihood of engaging in health related behaviour depends on how an individual perceives the benefits of that behaviour, the barriers to engaging in that behaviour, and the perceived threat posed by not engaging in the behaviour (Armstrong and Welsman 1997). Correlates of physical activity specified by this theory thus include measures of perceived benefits and barriers (Sallis and Owen 1999).  Such theories have many important applications, including, for example, the application of the transtheoretical model to the study of physical activity interventions. However, a limitation of such theories is that individual psychological characteristics comprise only one potential branch of the entire set of correlates of physical activity. Acknowledging this limitation, researchers are increasingly drawing on models which emphasize a wider variety of correlates, and in particular, environmental correlates (e.g. de Brujin et al. 2006, Wechsler et al., 2000). A seminal theory built on this premise is the Social Cognitive Theory (SCT) developed by Bandura (1986). This theory has been widely applied to studies of youth physical activity patterns (e.g. Craig et al. 1996, Pate et al. 2000). In contrast to the theories noted above, SCT recognizes a variety of factors influencing behavior, including intrapersonal, social and physical environmental influences (Sallis and Owen 1999, Wechsler et al. 2000). According to the theory, interrelationships between environment, individual and behaviour are seen as reciprocal rather than unidirectional, a concept described as “triadic reciprocality” (Merrell 2008, Salmon et al. 2008). Examples of such reciprocal relationships include (Pate et al. 2000, Wechsler et al. 2000): 32  the physical or social environment may either facilitate or constrain individual behaviour, but individuals can also change their environment to facilitate behaviour change both personal beliefs and environmental factors may influence behaviour, which can in turn result in changes in beliefs SCT thus highlights the importance of constructs corresponding to each of the individual, environment and behaviour, but emphasizes as central the role of self-efficacy, an individual’s confidence in their ability to engage in specific activities (Sallis and Owen 1999). Many recent physical activity studies rely on a set of integrative models referred to as ecological models, which draw extensively on Social Cognitive Theory (Lee and Moudon 2004, TRB 2005). These models explicitly recognize that physical activity is influenced by a range of factors corresponding to multiple levels of environmental influence (Carver et al. 2005, GilesCorti 2006) and have been argued to be more realistic than more since they are based on broad conceptions of influences on behaviour (Sallis and Owen 2002). Because this thesis is based on an ecological modeling framework, ecological models are now considered in greater depth. 2.3.2 Ecological Models of Health Behaviour Ecological models may be defined by a number of common features. First, these models are premised upon the nesting of individuals within multiple levels of environments or sociospatial contexts (Sallis and Owen 2002; Figure 3). These environments may correspond to specific physical settings, or they may describe social environments such as workplaces or jurisdictions responsible for policies that may influence behaviour. For children, the home and the school are often specified as relevant environments because of the proportion of time children spend in these settings. Of greater interest to planners, neighbourhood environments may also play an important role in shaping physical activity behaviours of children. 33  Figure 3: Generic Ecological Model Illustrating Multiple Possible Environments Influencing Individual Behaviours  broader society neighbourhood family individual  Most empirical studies of urban form influences on physical activity behaviour focus on the neighbourhood surrounding a child’s home (e.g. Copperman and Bhat 2007, Roemmich et al. 2007). However, it is logical that multiple neighbourhoods where children spend their time might be of interest. The need to consider multiple neighbourhood environments is highlighted by an example of an ecological model developed by Lee and Moudon (2004), the ‘Behavioral Model of Environment’ or BME. According to this model, characteristics of the origin, route, and destination of a trip are highlighted as potentially influencing physical activity, in addition to characteristics of the areas (neighbourhood environments) surrounding the origin and destination (ibid). Applied to child physical activity patterns, the BME might, for instance, specify the following as important environments: the child’s home, the child’s school, the route between the home and the school, and the neighbourhoods surrounding both the home and the school. A second characteristic of ecological models of physical activity behaviour, following from the first, is that specific characteristics of environments may be identified which influence the behaviour being investigated (Sallis and Owen 2002). An ecological model might therefore specify that in addition to individual characteristics (e.g. age and gender), household characteristics (e.g. household income and car ownership) and neighbourhood characteristics 34  (e.g. level of access to parks) all exert independent influences on physical activity patterns. In some cases, similar measures may be used to characterize multiple environments. For instance, measures of socioeconomic status assessed at both the household and neighbourhood levels may be independently associated with physical activity patterns. Because they acknowledge wide ranging correlates, a multitude of ecological models of physical activity behaviour have been developed, each with their own unique model specifications. Ecological models have thus been criticized for their lack of clear and consistent definition, and as a result, some authors have attempted to outline frameworks for comprehensive models of physical activity (Krizek et al. 2004, Spence and Lee 2003). The emphasis of ecological models on environmental correlates reflects the belief that environments sets limits on the behaviour that occurs within them (Wechsler et al. 2000). Further, in contrast to lower level interventions, modifying an environment can have implications for many individuals (Spence and Lee 2003). Thus knowledge of which environmental factors influence behaviour may allow for more effective interventions, highlighting the potentially important role of actors such as planners in influencing behaviour. However, the implications of an environmental intervention may not always be clear-cut, as they may negatively influence some individuals while benefiting others. Because ecological models are premised upon multiple levels of influence, they also lend insight into more opportunities for interventions than simpler, single level models (ibid). Interventions may, for example, occur via government initiatives at the federal, provincial, or municipal level. With regards to physical activity promotion specifically, these might include diverse strategies ranging from media messaging to local traffic calming initiatives to school based interventions. Figure 4 situates neighbourhood environment correlates of physical activity within a broader ecological classification of correlates. Each of the areas highlighted in this illustration represent environments that might be characterized by a number of correlates. For example, 35  correlates corresponding to the individual child include age, gender, weight status and psychological characteristics (Kahn et al. 2008, Van der Horst et al. 2007). Physical correlates corresponding to the school environment include availability of recreational features (Barnett et al. 2006, Nichol et al. 2009), school and play area size (Cradock et al. 2007) and classroom design (Lanningham-Foster 2008). Social correlates related to the school include peer and staff support (Barnett et al. 2006, Hohepa et al. 2007). With regards to the household environment, physical correlates include yard size (Spurrier et al. 2008), level of access to sedentary activities such as TVs (Singh et al. 2008), and availability of home exercise equipment, although findings in this regard are mixed (Davison and Lawson 2006). Social correlates associated with the home environment largely center around parents, who can influence physical activity patterns of their children through engagement and modeling of physical activity behaviours (Bauer et al. 2008, Hohepa et al. 2007, Ornelas et al. 2007). Given the breadth of possible correlates of physical activity, an overview of literature on all possible correlates of physical activity is beyond the scope of this review. This review instead focuses on two categories of correlates of physical activity relevant to the present study: 1) Individual and household correlates of physical activity which are commonly incorporated in analyses of neighbourhood environment correlates. These include variables recognized as important covariates of physical activity in general and therefore used as control variables (e.g. age, gender), as well as characteristics more specific to the study of neighbourhood environment correlates (e.g. household car ownership).  2) Neighbourhood environment correlates, including both physical and social components. Studies employing objective measures of these characteristics, and measures of perceptions are discussed.  36  Figure 4: Ecological Classification of Correlates of Physical Activity correlates of physical activity individual  demographic (e.g. age, gender) psychological (e.g. self efficacy) health status / genetic (e.g. disability, weight status)  perceptions of environmental characteristics proximate environments  components of environments social  physical  school environment  school physical education, peer influences  playground facilities, gym equipment  household environment  household income, parental attitudes towards physical activity  car ownership, yard size  neighbourhood environment  neighbourhood socioeconomic status (SES), neighbourhood social capital  topography, residential density, street connectivity  broader regional / jurisdictional environments  social marketing campaigns, school district policies  climate, regional recreational facilities  In discussing these correlates, the following sections draw extensively on primary sources as well as review articles and books (including Davison and Lawson 2006, Sallis et al. 2000, Carver et al. 2008a, Ferreira et al. 2006, Panter et al. 2008).  37  2.3.4 Individual and Household Correlates of Physical Activity An understanding of individual and household characteristics is critical for the analysis of neighbourhood environment correlates of physical activity because these characteristics may act as confounding variables or otherwise interact with neighbourhood environment correlates in more complex ways. For example, as discussed below, the age of children may moderate parental restrictions on their independent mobility, which in turn may mediate the relationships between built environment characteristics and child physical activity patterns. At a minimum, it is common practice to control for some of these characteristics when investigating relationships between the neighbourhood environment and physical activity, although the choice and definition of control variables is inconsistent from study to study. 2.3.4.1 Individual Characteristics 2.3.4.1.1 Age and Gender A wide variety of individual characteristics have been shown to be associated with child physical activity. Of primary relevance to studies of neighbourhood environment influences are child age and gender (Sallis and Owen 1999). As noted by Armstrong and Welsman (1997), international data on age and gender-related trends in physical activity are generally highly consistent, although absolute levels of physical activity cannot be accurately compared across studies. In general, older children engage in less physical activity than younger children (ibid). The decline in physical activity is most dramatic during the transition from childhood to adolescence (Carver et al. 2005, Godin et al. 2005), but may begin even earlier. Sallis and Owen (1999) for instance document decreases in physical activity starting as early as 6 years, based on a review of several international studies. Canadian data is consistent with a sharp and continuous decline in physical activity with age, starting in early childhood and continuing beyond adolescence (Active Healthy Kids Canada 2007, CFLRI 2009). In a review of several  38  studies, Armstrong and Welsman (1997) found that the rate of decline in physical activity was approximately 2.5 times greater for girls than boys. The decline in physical activity also begins earlier with girls (LeBlanc 2003). With regards to activities within the neighbourhood environment, older youth may also be less likely to engage in active transportation. For example, Frank et al. (2007a) found that youth of driving age were less likely to make walking trips. Children of different ages may also respond differently to specific neighbourhood environment characteristics. Frank et al. (2007a) found that living near a recreation or open space was consistently positively associated with walking across age groups from 5-20, but other built environment characteristics such as residential density were only related to walking for specific age groups within this range. This may be because destinations of interest or use by children change with age. In addition, because children and youth are allowed to independently access a wider variety of destinations as they age (Kytta 2004, Korpela et al. 2002), their physical activity patterns may become increasingly sensitive to neighbourhood environment conditions (Malone 2006). Sallis and Owen (1999) suggest an alternative hypothesis, speculating that environmental influences on physical activity behaviour might decline in importance relative to social and psychological factors, during the transition from childhood to adolescence. In addition to older children, girls represent another population of children particularly vulnerable to physical inactivity. Armstrong and Welsman (1997) in a review of international studies found that males between the ages of 6 and 17 were between 15-25% more physically active than females of the same age. A similar review of five studies by Sallis and Owen (1999) found that on average, boys spent 23% more time in physical activity than did girls. CANPLAY pedometer data for Canadian children and youth aged 5-19 also indicates that girls take fewer steps per day than boys (CFLRI 2005, 2009). Based on 2007-2009 data, 16% of boys meet the threshold number of steps corresponding to Canadian Physical Activity Guidelines (PHAC 39  2007a, 2007b), compared to 8% of girls (CFLRI 2009). As with age, gender may also moderate neighbourhood environment influences on physical activity, in part because destinations of interest to girls may be different than those for boys. For example, favorite place research in environmental psychology suggests that between ages 5 and 12, boys tend to favour outdoor places, while girls tend to favour indoor places (Korpela et al. 2002). Commercial sites may also be of more interest to girls. Carver et al. (2005), for instance, found that girls were significantly more likely to report existence of take-away shops and convenience stores near home, and Sener and Bhat (2007) found that male children were less likely to participate in shopping than females. In their literature review on built environment-physical activity relationships for children, Davison and Lawson (2006) observe that associations between environmental characteristics and physical activity patterns were more commonly noted for girls than boys. The reason for this may be that youth at high risk for physical inactivity (including girls) may rely more on external supports due to a lack of intrinsic motivators and potentially fewer opportunities to be active (Davison and Schmalz 2006). Finally, age and gender also moderate parental perceptions and restrictions, with younger children and girls generally subject to greater restrictions, as discussed below. 2.3.4.1.2 Child Perceptions Many studies have found that children’s perceptions of urban form characteristics are associated with physical activity behaviour. For example, Timperio et al. (2004) found that children who perceive that there are no parks in their neighbourhood are less likely to walk or bicycle in their neighbourhood. Similarly, Hume et al. (2005) conducted a cognitive mapping exercise in which children were asked to draw pictures of their home and neighbourhood. They found that those girls who drew more opportunities for physical activity in their neighbourhood engaged in more objectively measured low intensity physical activity. Holt et al. (2008) 40  similarly examined mental maps completed by children aged 6-12 depicting sites for recreation, finding that children in a high walkability neighbourhood depicted more active and less nonactive transportation than children in a lower walkability neighbourhood. With regards to perceptions of traffic safety, Carver et al. (2005) found that girls who perceived local roads to be safe spent more time walking for exercise on weekends and for transportation on weekdays. Liu et al. (2007) similarly found a positive association between a child’s perception of safe drivers and pleasant walks in the neighbourhood with self reported physical activity. While not considering physical activity outcomes specifically, Mullan (2003) further highlights the importance of traffic safety in shaping activity patterns. Specifically, Mullan found that adolescents in Wales who reported living with busy traffic and car parking were less likely to have positive perceptions of their local area, effectively viewing it as less safe and pleasant for play. While other studies have found no association between perceptions of safety and overall levels of physical activity, this may be in part due to the fact that children compensate for reduced physical activity in their neighbourhood with activities in other locations (Davison and Lawson 2006). Children’s perceptions of their neighbourhood conditions and opportunities for play are likely also shaped by characteristics of the social environment. In addition to examining perceptions of the traffic environment, Carver et al. (2005), examined associations between measures of physical activity and adolescent’s perceptions of their neighbourhood social environment. Carver and colleagues concluded that important explanatory factors for walking and cycling within their neighbourhoods included adolescent’s reports of: having friends nearby, having young people of similar age nearby to socialize with, knowing neighbors, and waving or talking to neighbors. Findings that adolescent’s perception of crime threat in their neighbourhoods is negatively related to active transportation to recreational sites (Grow et al. 2008) also highlight the importance of perceptions of the neighbourhood social environment. 41  Children’s perceptions of characteristics of the neighbourhood built and social environments are not solely shaped by the characteristics themselves, but also by the matching of environmental features with individual characteristics such as their physical abilities and social needs (Kytta 2004). Thus, individual children with different interests and physical abilities may perceive their environments in diverse ways. Perceptions of opportunities for physical activity in the neighbourhood environment may also be subject to characteristics of the household environment, including level of access to sedentary activities in the house, which may moderate the exposure of children to their environment. For example, access to TVs in a house with few restrictions may effectively ‘pull’ children inside, limiting their exposure to characteristics of the neighbourhood environment (Epstein et al. 2006, Roemmich et al. 2007, Wong et al. 2010), although findings on the displacement of physical activity by TV viewing are mixed (Biddle et al. 2004, Smith et al. 2008).  2.3.4.1.3 Ethnicity and Immigrant Status Both ethnicity and immigrant status may be important factors underlying physical activity differences in children. Sayre and Gallagher (2001), for instance, speculate that cultural factors associated with ethnicity may be associated with preference for specific types of activities. Ethnicity is commonly incorporated in research on neighbourhood environment correlates as a control variable, but its influences and the reasons for these influences are not well understood. Sallis and Owen (1999) in a review of literature on correlates of child physical activity note that racial and ethnic differences in physical activity are unclear, with some US studies finding ethnic minorities to be less active than European Americans, but other studies finding no differences. Further, the ethnicity classifications used in international studies may not be relevant in the multi-ethnic context of major Canadian metropolitan areas. For example, in one recent US study (Kligerman et al. 2007), children were classified as either Mexican42  American or white whereas major ethnic groups in the Lower Mainland may be described as European/North American, East/Southeast Asian and South Asian (S. 3.3.1 below). Tremblay et al. (2006) lend some insight into the importance of ethnicity and immigrant status in the Canadian context, in a study of leisure time physical activity behaviour of adults (based on pooled data from 2000/2001 and 2003 CCHS survey). Notably, Tremblay et al. found that immigrants are less likely to be physically active in leisure time than non-immigrants, and that the proportion of those who are active increases with time since immigration. Tremblay and colleagues also found evidence of ethnic variations, noting, for instance, that the largest differences in MVPA between immigrants and non-immigrants was found for Latin Americans. While this study focused on adults, it may have some bearing on child physical activity behaviour because parents are important role models for their children, but more research on the influence of immigrant status and ethnicity needs to be conducted with regards to children (Active Healthy Kids Canada 2007). 2.3.4.2 Household Environment Characteristics This section considers characteristics of the household environment in terms of their influence on child physical activity patterns. First, parental influences are considered as parents may both play important roles in constraining child physical activity patterns (for example, through restrictions on independent mobility due to safety concerns), and in promoting them (for example, through behavioural modeling, or direct encouragement). Second, the broader household environment is considered insofar as it may act as an important setting for physical activity, or conversely, as a setting for sedentary activities which compete for physical activity. The parental influences discussed in the following sections relate primarily to short term considerations such as decisions on travel mode choice and restrictions on children’s independent mobility. However, certain longer-term decisions on the part of parents are also 43  likely to frame or constrain opportunities for their children’s physical activity patterns. In particular, household location decisions serve to fix the environment within which their children grow-up, thus potentially influencing their physical activity patterns over a long time-frame. Historically, the suburbanization of families with children has been a widespread phenomenon (Karsten 2003). Such location decisions may, for instance, be based on perceptions of safety, proximity to what are perceived to be good schools and parks, proximity to other children, and housing costs (Baum 2004, McGahan 1995). 2.3.4.2.1 Parental Perceptions Parents may directly influence physical activity patterns of their children by preferring certain venues for play and accompanying their children to these venues, but perhaps more importantly, parents may play an important role in shaping neighbourhood environmentphysical activity relationships because they act as ‘gate keepers’. In this role, parents are concerned with wide ranging characteristics of the neighbourhood built and social environment, centering around safety (Carver et al. 2008a, Prezza et al. 2005). With regards to characteristics of urban form specifically, fear of injuries or fatalities from traffic are major concerns that parents have about active transportation to school or leaving children unattended outdoors (Ahlport et al. 2008, Gielen et al. 2004). Such fears are in turn influenced by urban form and traffic conditions. Timperio et al. (2004), for instance, found that parental perceptions of the need to cross several roads to reach play areas (for older girls), and lack of traffic lights or crossings (for older boys) were negatively associated with children regularly walking or cycling to local destinations. Carver et al. (2005) similarly found that parental reports of heavy traffic was associated with lower rates of cycling and walking. In contrast to this finding, however, Timperio et al. (2004) found that parental perceptions of heavy traffic in the neighbourhood were positively associated with children’s active transportation. Carver et al. (2005) suggest 44  that this anomalous finding might imply that the parents of children who walk or cycle frequently are more aware of local traffic issues. Parental concerns about traffic may also help to explain anomalous findings regarding the relationship between street connectivity and physical activity. Specifically, findings that connectivity is negatively or not associated with physical activity measures (Braza et al. 2004, Copperman and Bhat 2007, Timperio et al. 2006) may arise because parents perceive cul-de-sacs as safer than through streets (Veitch et al. 2006) and may therefore consider neighbourhoods with disconnected street networks to be safer for their children than highly connected networks. Perceptions of traffic safety have also in some instances been associated with neighbourhood socioeconomic status. Timperio et al. (2004), for instance, found that children in families of higher socioeconomic status (SES) walked and cycled more frequently compared to their lower SES counterparts because their neighbourhoods were perceived as safer due to decreased traffic exposure, more sidewalks and the absence of visual obstacles. Parental perceptions of safety of the social environment may influence relationships between urban form and child physical activity patterns. These perceptions contribute to parental decisions on restrictions of the independent mobility of their children. Major parental concerns relating to social safety include fears of ‘stranger danger’ and bullying (Freeman 2006, Carver et al. 2008a). McDonald (2007b) and McDonald et al. (2010), for instance, found that a measure of social control and cohesion was an important explanatory variable for walking to school, suggesting that parents require a neighbourhood with high levels of social trust to allow their children to walk to school. Another important characteristic that may influence parent’s perceptions about the safety of their children is the perception of the presence of other children (Salmon et al. 2007, Timperio et al. 2006). The presence of other children might make parents feel more confident about neighbourhood safety because it suggests a level of confidence in other parents regarding neighbourhood safety (Tranter 2006). Alternately, the presence of other 45  children might be seen as important as it implies more ‘eyes on the street’ (both those of the children, and of other parents watching after them). Finally, as with parental perceptions of the traffic environment, parental concerns may also be associated with neighbourhood socioeconomic conditions. For instance, Gielen et al. (2004) found that parents in low income neighbourhoods reported the highest rates of unpleasant walking environments, citing concerns about crime, drug dealers, violence, and garbage. Parental perceptions of neighbourhood conditions, whether related to traffic or social safety, may also be moderated by a number of other factors. For example, the amount of time parents spend in their neighbourhood may influence their perceptions of the neighbourhood environment. Lam (2001) found that parents with full-time jobs felt their local environment was safer than those with part time jobs. Lam hypothesized that this might be a result of those parents spending more time at home being more likely to observe firsthand the dangers that their children are exposed to. Gebel et al. (2009) found that simply having children changes environmental perceptions of adults. They noted that adults with children were significantly more likely than other adults to misperceive a high walkable environment as low walkable, speculating that due to time constraints, such adults would be less familiar with their environment. In general, it is likely that perceptions of risk and fears are modified by a range of sociopsychological factors including prior experiences and preconceptions (Loukaitou-Sideris, 2006). In contrast to studies examining individual components of parental concerns, Kerr et al. (2006) developed a composite parental concerns scale incorporating measures of fears of stranger danger, gangs, bullying, traffic, availability of sidewalks, among others. They found that parental concerns as gauged by this measure were associated with active commuting such that parents with few concerns were significantly more likely to have children who actively commute to school.  46  Subsequent to forming perceptions about the environment, parents may choose to influence how their children engage with the environment through either defensive behaviour (e.g. accompanying their children during play or teaching their children how to cross streets safely), or avoidance behaviour (e.g. restrictions; Carver et al. 2008a). Parental restrictions on the independent mobility of their children are often conceptualized in terms of ‘mobility licenses’, which include licenses to: cross roads, use buses, go to school on their own, go to other places on their own, cycle on the road, and go out after dark (Hillman 1993). Alternately, parents may place restrictions on children’s territorial range, limiting the distance from home that children are permitted to travel when playing and socializing (Kytta 2004, Korpela et al. 2002). Parents may for instance advise their children not to travel beyond a particular playground, thus specifying a limited range (Korpela et al. 2002). This type of restriction has formed the basis of studies on children’s independent mobility, which have attempted to explicitly measure children’s home range (e.g. Cornell et al. 2001), and is consistent with the idea of buffering around a child’s home when calculating built environment characteristics (S. 2.3.5.1). Specific restrictions that parents place on their children are highly dependent on the age and gender of their children, with greater restrictions on independent mobility generally imposed on younger children (Hillman et al. 1990, Veitch et al. 2008) and girls (Korpela et al. 2002, Kytta 2004, Carver et al. 2008a, O’Brien et al. 2000). As a result, younger children and girls may be less independently mobile than their peers (Carver et al. 2005, Kytta 2002, Korpela et al. 2002). The influence of parental restrictions on child physical activity patterns was studied by Evenson et al. (2006), who found that girls engaged in higher levels of physical activity when they had parents who allowed them to walk on their own, use public transit on their own, or walk/bike to transit from their home. Roberts et al. (1997) found that age was positively associated with the number of streets children cross. Frank et al. (2007a), in a study 47  of 5-20 year olds, similarly found that youth aged 12 to 20 were more likely to walk 0.5 mile or more per day than the youngest children, and that although youth of driving age were less likely to make a walk trip, when they did, it was of a longer distance than younger age groups. According to Tranter (2006), parents start to give their children more independent mobility licenses between the ages of 8-12. While such findings intuitively confirm the importance of parental influence on physical activity patterns, even this relationship may not be direct, because children may perceive and respond to their parental restrictions in a variety of ways. Although a great deal of emphasis in the literature is placed on parents as gatekeepers (Davison and Lawson 2006, Carver et al. 2008a), children themselves extend the spatial extent of their activities independently, and through the influence of peers (Cornell et al. 2001). Further, parents often do not have a good sense of what their children are actually doing, including how far or where they are going independently (ibid). In many instances, children do not accept parental restrictions and may even prefer places outside of the control of their parents (Korpela et al. 2002). For example, in a study of favorite places, Korpela et al. (2002) did not find significant associations between mobility licenses given by the parents and the distance of children’s favorite places from home. Korpela and colleagues also found that many parents did not know their children’s favourite places. One reason for non-compliant reactions to parental restrictions may be that children perceive their environment in very different ways than their parents. Carver et al. (2005), for example, found that adolescents had more favourable views of their neighbourhoods than did parents in terms of safety while walking or cycling. Carver et al. speculate that such views reflect adolescent tendencies to seek increased autonomy. While parental perceptions are potentially very important to consider in the study of neighbourhood environment influences, a limitation of many studies employing measures of parental perceptions of the built environment is that they often do not simultaneously assess objectively measured characteristics. This points to a conceptual limitation of the sole use of 48  perceived measures of the environment, in that it may be difficult to disentangle the influences of perceptions of environmental characteristics from the influences of the environmental characteristics themselves. 2.3.4.2.2 Parental Behaviour and Attitudes towards Physical Activity While parents may restrict the mobility of their children in response to perceived dangers, they may also play many positive roles in influencing physical activity patterns. First, parents may offer a number of monetary or other material supports to their children to engage in structured physical activity (Deforche et al. 2007). For example, while the chauffeuring of children to a particular recreational facility may be seen as curtailing their opportunities for active transportation and independent mobility, it may also be viewed as an important support for their overall levels of physical activity, by giving them access to places where they can be active (Davison and Schmalz 2006). Second, parents may offer direct verbal encouragement to their children (Dowda et al. 2007, McMillan 2005). Canada’s family guide for child physical activity outlines numerous techniques parents can employ in this regard (PHAC 2007a). Third, parents may take a number of more direct approaches in engaging children in physical activity, including, for example, organizing family activities, involving children in chores and limiting television watching (Deforche et al. 2007). A fourth supporting role that parents may play is the role modeling of healthy attitudes towards physical activity (Brug et al. 2006). Numerous studies have confirmed that parents who are physically active positively influence their children’s physical activity patterns (McMillan 2005, Godin et al. 2005, Armstrong and Welsman 1997). For example, Cleland et al. (2005) found that parental exercise was positively associated with children’s extracurricular sports participation. In speculating on the mechanism for this association, Cleland et al. highlight the importance of role modeling, but also suggest that it may be explained in part by a genetic 49  predisposition to physical activity. Armstrong and Welsman (1997) also note that parent’s inactivity may exert more influence on behavioural modeling than physical activity. In the literature on built environment – physical activity relationships for adults, a major topic of debate is the extent to which attitudes towards physical activity factor into household location decisions. According to one argument, adults may ‘self-select’ into neighbourhoods based in part on such attitudes (Frank et al. 2007b). This may result in adults who favour driving over active-transportation being more likely to locate in car-oriented suburban developments. Thus, the physical activity behaviour of adults may reflect their attitudinal predispositions rather than the nature of the built environment in which they reside. While a direct analogy for this situation does not exist for children, because it is parents who generally choose where to live, there could be an indirect self-selection effect, whereby: 1) parental attitudes towards physical activity influence their residential selection decisions, and 2) parental attitudes towards physical activity also influence their children’s physical activity behaviour. Thus children who grow up in an auto-oriented environment with parents who prefer an auto-oriented lifestyle may be influenced either by their parental preferences, the environmental characteristics they are exposed to, or both parental preferences and the environmental characteristics (TRB 2005). It may be difficult to disentangle the influences of each because of the potential self-selection effect outlined above, unless both the environment and parental attitudes towards physical activity (and/or measures of parent physical activity patterns) are addressed explicitly through research design. The joint effects of living in an auto-oriented environment together with parents who prefer an auto-oriented lifestyle may condition children into a ‘car culture’ whereby transport by modes such as walking and biking is seen as less attractive than driving (Tomson et al. 2007), 50  and may have a lasting influence on children’s values and behaviour as they grow up. Yeung et al. (2007) for instance suggest that children who are accustomed to being driven short distances are not likely to appreciate the benefits of walking as adults. More generally, Freeman (2006) suggests that children can become disconnected from the urban environment, as a result of modern suburban design which can result in children’s lives becoming “a fragmented mosaic of places – school, childcare, clubs, shops and playgrounds lacking the linkages provided by walking along streets, past familiar people and places” (p. 73). 2.3.4.2.3 Parental Car Ownership and Travel Behaviour Mode choice decisions of parents and their travel behaviour more broadly are also likely to influence child physical activity patterns. For example, mode choice decisions of parents directly influence the extent to which children are chauffeured or travel on their own, thus engaging in active transportation. A complex suite of factors in turn influence parental travel habits and preferences. Once the sunk costs of car ownership are incurred, driving may be seen as a cheap and highly convenient mode of transportation for families with children. Johansson (2006), for example, found that the number of cars in a household was one of the most important variables correlating positively with chauffeuring of children. Similarly, Wen et al. (2008) found that number of cars in the household was associated with children being driven to school. Traveling by transit, bike or foot with young children may be seen as tiresome for parents (Dieleman et al. 2002), while driving may be seen as relatively convenient. Perhaps indicative of such mode choice preferences, a number of studies have found associations between car-ownership and physical activity related variables or found that car ownership moderates the relationship between urban form and physical activity (Frank et al. 2007a, Kerr et al. 2007, Timperio et al. 2004). Frank et al. (2007a), for instance, found that children in households with few or no cars were more likely to walk than in other households. 51  In this study, having no car in the household was most strongly associated with the walking variables examined, but having one car rather than three was also significant. Kerr et al. (2007) found that household car ownership strongly moderates the influence of urban form on youth physical activity patterns. In general, Kerr and colleagues found weaker associations between built environment characteristics and walking for children in households with fewer cars, suggesting that youth in such households were much more likely to walk for transportation regardless of the environmental conditions. A variety of other factors likely also influence parental mode choice decisions, including household size. In households with children, the coordination of a variety of activities becomes necessary, reflecting complex intra-family dynamics and trip planning, which in turn may lead to complex trip chaining patterns (Lee and McNally 2006, Srinivasan and Ferreira 2002). Thus parents with more children may be more likely to chauffeur their children because of the convenience of joint trip planning. Srinivasan and Ferreira (2002) provide evidence of substantial trip chaining in families with children, finding that less than 6% of tours in two worker households with children were without non-work activity chaining. Finally, an extensive body of research has found that urban form also influences mode choice of adults (Frumkin et al. 2004), including parents. In general, a highly dispersed urban form with little mixing of uses is associated with higher rates of car-use and lower rates of active transportation (ibid). With regards to the driving of children in particular, trends towards increased structured activities rather than unstructured play (Prezza 2007) may result in more adults chauffeuring their children as such activities are often located at specialized facilities which may be seen as being too far from home for children to transport themselves. In a study of children’s mode choice for getting to soccer games, for example, Tal and Handy (2008) found that driving was the predominant mode of transport, with 76.8% of players driven versus 18.4% biking and 4.8%  52  walking. Major factors underlying this modal split included distance to the game and logistical barriers such as the need to carry equipment and snacks (ibid). 2.3.4.2.4 The Broader Household Environment The household environment may be of particular importance for children simply because they spend a large proportion of their time in this environment. Its importance likely varies considerably from child to child, with some children preferring it as a setting for physical activity more than others. Carver et al. (2005), for example, suggest that less active teenage girls may prefer to exercise at home than in the community. As with the broader built environment, the household environment may be characterized by the level of access to recreational facilities that it affords. These may be either internal or external to a house. Thus, although the lack of external yards common in some new suburban environments (when large houses and their garages fill most of the lot they are located on) has been harshly criticized on the basis that it may reduce play opportunities for children (Gleeson et al. 2006), parents may compensate by setting up play areas within the house. However, only two of the six studies reviewed by Davison and Lawson (2006) on this topic found positive associations between the availability of home exercise equipment and child physical activity, while the other four found no association. While access to private recreational resources may act as an enabler for child physical activity, access to sedentary activities may serve as an inhibitor. Roemmich et al. (2007) highlight this possibility because access to reinforcing sedentary behaviours such as watching television within the home increases the time that youth spend in sedentary activities. Notably, children now have access to a wider range of reinforcing sedentary activities at home than in the past, including not only TV, but also the internet (with a wide array of social networking and other web-sites of interest to youth) as well as video games. Findings summarized in Sallis et al. (2000) are consistent with these changes, indicating that participation 53  in sedentary activities after school and on weekends is significantly negatively associated with adolescent physical activity. Another potentially enabling or constraining dimension of the household environment is household income. This may, for example. influence physical activity patterns indirectly through car ownership, but may also have independent effects by either limiting or enabling access to commercial or fee-charging public recreational resources (Armstrong and Welsman 1997). As a result of such potential influences, household income is regarded as an important control variable for child physical activity research, and is almost universally included in empirical analyses of child physical activity patterns. However, household income does not consistently correlate either positively or negatively with physical activity variables, possibly because the above noted influences serve to counteract each other. For instance, on the one hand, data from the 2005/2006 HBSC and 2005 CANPLAY surveys indicate that children in families of higher SES have higher levels of physical activity compared to their lower SES peers (Active Healthy Kids Canada 2007). Evidence that such differences may be attributable to differences in access to commercial recreational facilities is provided by data from the 2001/2002 NLSCY survey which indicates that approximately 55% of low-income children participate regularly in organized sports, compared with 65% of middle-income children, and 79% of high income children, whereas the observed gaps are smaller for unorganized sports (ibid). In contrast, negative associations between household income and child physical activity related variables have been found by Frank et al. (2007a), Kerr et al. (2007) and Liu et al. (2007). Two possible explanations have been put forward for these associations. Frank et al. (2007a) and Kerr et al. (2007) speculate that such associations may result from the association between income and car ownership, with lower income households having lower car ownership, and thus having children who engage in more active transportation. Liu et al. (2007) put 54  forward an alternate explanation that lower SES families have decreased access to sedentary forms of entertainment such as TV and video games. 2.3.5 Neighbourhood Environment Correlates of Physical Activity 2.3.5.1 Defining Neighbourhoods for the Objective Assessment of Environmental Influences on Physical Activity A first step in the objective assessment of environmental influences on physical activity behaviour is to identify a specific neighbourhood or neighbourhoods thought to be of relevance in influencing physical activity. Approaches to neighbourhood identification may in the future involve the use of GPS (Global Positioning Systems) to track individuals as they move around in the course of their daily activities (Cummins et al. 2007). However, in the absence of such detailed data, most studies use a single neighbourhood corresponding to an area around a child’s home. As noted above, another relevant neighbourhood for children might be the area surrounding their school. Once point locations such as the home or school are identified, specific areas around them may be delimited and used as the spatial unit of analysis for assessing built or social environment characteristics. A number of approaches have been adopted in this regard. For analytical purposes, neighbourhood definition may involve use of predefined spatial units such as census tracts or municipally defined neighbourhoods. However, use of such predefined neighbourhoods is problematic because the delimitation of specific spatial units used for the aggregation of data is likely to strongly influence the outcomes of the analysis (the Modifiable Areal Unit Problem; Green and Flowerdew 1996), and the boundaries of such neighbourhoods generally have no clear behavioural basis. The alternate, more common approach is to define a ‘buffer’ centered around a child’s home (or school), of a size and shape that is informed by theory or empirical precedent.  55  Figure 5 illustrates three approaches to defining the shape of a buffer. The first approach is to use a straight line or circular buffers (as in Liu et al. 2007, Nelson et al. 2006; Figure 5(a)). This approach is premised upon individuals having equal access to spaces surrounding their home within a fixed radius. This might be appropriate if free movement in any direction were possible, for example if widespread availability of informal pathways and parks permitted such movement. In an urban context, use of a network buffer (as in Frank et al. 2007a; Figure 5(b)) may be more realistic. This type of buffer is defined as a polygon whose vertices are points on the road network at a fixed distance from the subject’s home, reflecting the fact that movement is primarily restricted to the road network. Network buffers are generally preferred to circular buffers, and are now widely used, because they more accurately define the environment accessible to an individual. A third buffer type, perhaps even better reflecting realistic levels of access is illustrated as Figure 5(c). This approach represents a refinement of network buffers, based on the premise that individuals may only access land uses directly abutting streets (Oliver et al. 2007). The hypothesized mechanism(s) of influence of neighbourhood environment characteristics may be used to inform buffer size selection. For example, one approach might be premised upon parents acting as gatekeepers who effectively limit their children’s access to the neighbourhood environment. Assuming that this is the case, the assessment of neighbourhood characteristics might best be gauged by relatively large areas which parents may perceive when considering the safety of their neighbourhood for their children. Larger areas might also be more relevant if the mechanism of influence is via parental travel habits and mode choice decisions. That is, insofar as parents mode choices influence the physical activity patterns of their children, larger areas which influence the travel behaviour of parents may be most relevant. In contrast, smaller buffers may be most appropriate if it is hypothesized that built environment  56  Figure 5: Three Alternative Approaches to Buffer Definition. All buffers illustrated are of 800 meters radius, created using the same road network.  (a) circular  (b) network  (c) street based characteristics influence physical activity behaviour of children directly, assuming that children have relatively limited independent mobility and may thus be most influenced by areas relatively close to home. There is currently no consensus on the appropriate size of buffers (Kligerman et al. 2007), and although radii of approximately 800 meters to 1 kilometer around children’s homes are commonly used to define the relevant spatial boundaries of a child’s environment (e.g. Frank et al. 2007a, Kerr et al. 2007, Roemmich et al. 2007), some studies have used radii of up to 3 km (Nelson et al. 2006). To address this issue more explicitly, a small number of studies have examined multiple radii to determine which best explains differences in physical activity. 57  Kligerman et al. (2007) for instance initially used three buffer sizes in their analysis of the association of neighbourhood design characteristics with physical and body mass index in adolescents: 0.25 miles, 0.5 miles and 1 mile (approximately 400, 800 and 1600 meters). Based on the results of bivariate models, the authors decided to include only two environmental variables using the 0.5 mile buffer in their overall multivariate model. Liu et al. (2007) generated buffers of 200, 400, 600, 800 and 1000 meters to calculate various environmental characteristics for association with self-reported physical levels, and found that different sized buffers best explained built environment-physical activity relationships for different built environment variables. For example, a measure of street connectivity was only found to be significantly associated with differences in physical activity when it was calculated using a 1000 meter buffer size (ibid). As an alternative to buffer based neighbourhood environment measures, some measures may be based on a route between a specific origin and destination. For example, the route between a child’s home and school may be used to assess specific measures. These may be as simple as measures of distance between origin and destination. Alternately, characteristics of the route, such as number of major intersections crossed, may also be assessed. When deriving such measures, the specific route needs to be specified. The simplest approach is to approximate a route based on the shortest distance along the street network (analogous to the definition of network buffers described above).  2.3.5.2 Neighbourhood Social Environment Correlates As recognized by ecological models of physical activity behaviour, neighbourhood social environment characteristics may have important, independent influences on health and behaviour above and beyond individual and household level characteristics (Molnar et al. 2004). In some cases, it may be difficult to separate influences of broader social environments from 58  parental perceptions as parents are often used to assess social environments. Some studies have, however, used measures of social environments independent of parental perceptions, examples of which are now highlighted. In one study, Carver et al. (2005) examined associations between physical activity and various measures of adolescents’ perceptions of their neighbourhood environment. Carver and colleagues found that important explanatory factors for adolescent physical activity included adolescents’ reports of: having friends nearby, having young people of similar age nearby to socialize with, knowing neighbours, and waving or talking to neighbours. Carver et al. thus concluded that levels of social interaction may be an important predictor for adolescent walking and cycling within their neighbourhoods. Research to date on the patterning of child physical activity on the basis of measures of neighbourhood SES has had mixed results (Molnar et al. 2004), but recent analyses appear to lend support to the existence of some association between neighbourhood SES and physical activity. Carson et al. (2010), for instance found girls living in low SES neighbourhoods to engage in significantly more weekly screen time and TV/movie minutes than girls living in higher SES neighbourhoods. Oliver and Hayes (2005), similarly found that the prevalence of child and youth overweight in Canada is negatively and significantly related to neighbourhood SES. Oliver and Hayes also found that participation rates in organized sports exhibited a social gradient (lower prevalence of ‘never or almost never’ participation were noted in higher SES neighbourhoods), a trend which was not observed in rates for unorganized physical activity. Although they did not assess specific mechanisms to explain their findings, Oliver and Hayes speculate that the patterns identified reflect social and physical environments in lower SES neighbourhoods that are less conducive to maintaining healthy bodyweights, including lack of local facilities, lack of resources to access such facilities, and a lack of awareness of such facilities. Research on the distribution of recreational facilities by neighbourhood SES is, however, mixed (Dowda et al. 2007, Timperio et al. 2007). 59  Molnar et al. (2004) also found that in addition to family-level socioeconomic factors, neighbourhood level socioeconomic factors were positively associated with physical activity, accounting for approximately 2% of the variation in individual level physical activity of the children and adolescents studied. The authors found that part of the variation in individual level physical activity was explained by neighbourhood social disorder, gauged by examining videotapes of activity such as adults fighting or arguing in a hostile way, drinking of alcohol in public, and people selling drugs. In order to highlight the importance of social disorder, Molnar and colleagues note that decreasing their social disorder measure by moving from a neighbourhood at the midpoint of the top quartile to the midpoint of the bottom quartile was associated with an increase in 29 minutes per week of physical activity. In contrast to such findings, other studies have found no associations between measures of neighbourhood disorder, children’s perception of stranger danger and physical activity patterns of children, leading Davison and Lawson (2006) to speculate that only high levels of social disorder (e.g. people selling drugs) such as those highlighted by Molnar et al. exert an influence child physical activity. 2.3.5.3 Neighbourhood Physical Environment Correlates As illustrated in Figure 6, physical environment characteristics of neighbourhoods can be divided into elements of the natural environment and elements of the built environment. Built environment correlates in turn can be broadly classified into characteristics relating to access, and those relating to design. Access related characteristics include density, land use mix and connectivity of the street network (Frank et al., 2006, Saelens et al. 2003). High density and a fine grained mix of land uses generally improve access by increasing the physical proximity of origins and destinations, while high street connectivity allows for more direct travel between origins and destinations (Frank et al. 2003). In the case of children, access to specific uses such 60  as schools and parks may be particularly important (Frank et al. 2007a). In contrast, design related characteristics do not deal with the broad distribution of land uses and connections between them, but rather address such smaller scale elements of the built environment as intersection design, presence of traffic calming devices, availability of sidewalks, and provision of streetlights and benches (Pikora et al. 2002) . Figure 6: Neighbourhood Physical Environment Correlates of Physical Activity neighbourhood physical environment  built environment  natural environment (e.g. weather, topography)  access  design (e.g. presence of traffic calming, play space design)  connectivity  proximity  (e.g. street connectivity, continuity of sidewalks)  density  land use mix  (e.g. residential density)  (e.g. mix of a variety of uses)  access to specific uses  Monetary costs of access (to  (e.g. proximity to parks, school)  specific recreational facilities)  2.3.5.3.1 Natural Environment Characteristics Two major elements of the natural environment have been investigated in the literature as correlates of child physical activity, but only to a limited extent: weather and topography. In one study addressing topography, an objectively assessed steep incline on the way to school was negatively associated with walking or cycling to school among older children (Timperio et al. 2006). With regards to weather, findings of research to-date are inconclusive or at least of limited generalizability. In their review of environmental correlates of physical activity, Davison and Lawson (2006), for instance, found significant associations between weather and children’s physical activity in only two out of the five studies on the subject that they reviewed. 61  In two reviews focusing solely on influences of season and weather, Tucker and Gilliland (2007) and Shephard and Aoyagi (2009) both observe a decline in activity amongst children or adolescents in winter, across the majority of studies. Carson and Spence (2010) similarly find in their review that as seasons become more extreme in temperature that children and adolescents are less physically active. While such differences are possibly due to shorter days and adverse weather conditions, an alternate explanation for such findings may be that children’s activity patterns are very different due to the structure of the school year, with children typically having holidays during the summer (Tucker and Gilliland 2007). Further, many of the studies to-date are limited because they are based on restricted time ranges, often missing several months of the year. Finally, findings likely are not generalizable to regions characterized by markedly different climatic conditions. The study area for this thesis, Vancouver and the surrounding Lower Mainland, is characterized by relatively unique climatic conditions due to its proximity to the ocean and mountains which have a moderating effect on temperature and result in relatively large amounts of precipitation (Taylor and Langois 2000). No studies have been identified which examine the influence of weather or season on physical activity patterns in this region specifically. 2.3.5.3.2 Access Related Characteristics of Urban Form Access related characteristics of urban form are those which are hypothesized to influence physical activity patterns by increasing the physical proximity of origins and destinations. This may translate into: (1) active transportation: for example, in higher density environments, it may be easier for children to walk or bike to a friend’s house than in lower density environments; and/or (2) physical activity at specific sites: for example, parks located close to where children live may be more likely to be used by children for play. 62  Accessible recreation sites may translate into increased physical activity through both mechanisms (Sallis et al. 2004). In addition to shaping child physical activity patterns directly, the degree of access may also influence children indirectly by influencing their parents. For example, parents living in neighbourhood environments with high level of accessibility may drive less in general, and more specifically, may chauffeur their children less. While such mechanisms may translate into positive associations between density and physical activity, a possible counteracting mechanism is that higher densities may be associated with increased parental fears of stranger danger.  DENSITY AND LAND USE MIX In addition to the above noted mechanisms, both density and land use mix might also encourage physical activity by increasing the number of eyes on the street, thus engendering a feeling of greater safety (Jacobs 1961). With some exceptions (e.g. Ewing et al. 2004), positive associations have generally been noted between objective measures of density and measures of physical activity, including measures of walking (Frank et al. 2007a, Kerr et al. 2007), active transportation to and from school (Braza et al. 2004, McDonald 2007b), and weekly moderate to vigorous physical activity (de Vries et al. 2007). Similarly, objective measures of land use mix have also been positively associated with measures of walking (Frank et al. 2007a, Kerr et al. 2007) and active commuting to school (Larsen et al. 2009, McDonald 2007b, McMillan 2007). While aggregate results are commonly reported, such findings do not necessarily apply to all age or demographic groups. In their study of walking behaviour of Atlanta youth, for example, Frank et al. (2007a) stratified results by age group, and found that while all urban form variables studied were related to walking in 12-15 year olds, only access to recreational space was associated with walking across all age groups. Density, for instance, was associated with walking for youth aged 9-20, but not those aged 5-8, and land use mix was associated with 63  walking for youth aged 12-20, but not 5-12. The importance of a variety of destinations beyond recreational space for older children likely reflects their increasing independence with age (ibid). Research in this area often employs land use categories similar to those used for adults, such as residential, commercial and office (e.g. Frank et al. 2007a, Kerr et al. 2007). Others have used alternate definitions, attempting to customize their measures more specifically to children, by incorporating measures of a variety of park and recreational related land uses (e.g. Liu et al. 2007, Roemmich et al. 2007). Liu et al. (2007), for instance, incorporate recreational areas as a use category, while excluding office uses. In contrast to the other studies noted, Liu et al. (2007) did not find land use mix to be a significant predictor of self reported physical activity. While their choice of use categories is arguably intuitive given that recreational areas would be more likely destinations for children than offices, the omission of offices as a use ignores the possibility that such uses increase eyes on the street during the day, thus potentially improving the quality of a neighbourhood environment as a setting for physical activity for children. Alternately, the findings of Liu et al. may indicate that land use mix primarily influences physical activity patterns of children indirectly, by influencing travel behaviour of their parents.  CONNECTIVITY Studies of associations between objective measures of street connectivity have produced more ambiguous results than those relating to density and land use mix, with some studies indicating positive associations with measures of physical activity (Bungum et al. 2009, Frank et al. 2007a, Kerr et al. 2007, Roemmich et al. 2007) and others indicating that street connectivity is negatively or not correlated with physical activity related variables (Braza et al. 2004, Copperman and Bhat 2007, Timperio et al. 2006, Roemmich et al. 2006, Larsen et al. 2009). 64  Such mixed findings may be explained in part because while increased street connectivity may increase access to destinations, smaller block sizes may also create more conflicts between motorists and children walking or bicycling, making less connected street networks more desirable for recreational cycling (Copperman and Bhat 2007) or because children take advantage of cul-de-sac streets, sheltered from traffic (Timperio et al. 2006). Further complicating the assessment of connectivity as a relevant urban form characteristic, streets themselves may not be the most important components of the transportation network for children engaging in physical activity. Potentially more important components of the transportation network include sidewalks, bike paths and informal paths (e.g. cutting across properties). Some studies considering linkages between children’s physical activity and objective measures of the presence and condition of sidewalks indicate positive associations (Boarnet et al. 2005a, Ewing et al. 2004, Jago et al. 2006), while others indicate null associations (McMillan 2007). In contrast, studies examining the role of bike lanes have not found associations with physical activity variables (Ewing et al. 2004, Jago et al. 2006).  OTHER MEASURES OF ACCESS In contrast to measures of density, land use mix and street connectivity, a variety of objective measures of access to specific land uses have been tested for associations with physical activity related variables. These include measures of access to: commercial destinations, schools, parks and recreational areas. While commercial destinations may not be sites for physical activity, they may act as destinations for children either to shop or more generally for social purposes. Frank et al. (2007a), for instance, found a positive association between the presence of commercial land uses within a 1 km network buffer around residences and walking in children and youth. Perhaps more than commercial destinations, close proximity to school may be important for child physical activity patterns because active transportation to 65  and from school may constitute an regular, habitual form of physical activity. Objectively measured distance to school or travel time has been negatively associated with rates of active commuting to school in a number of studies (Ewing et al. 2004, Schlossberg et al. 2006, Timperio et al. 2006, Larsen et al. 2009, Napier et al. 2010), and with MVPA in at least one study (Cohen 2006). Given their importance as both sites for play and destinations for active transportation, measures of access to parks and other recreational areas have also been extensively studied as correlates of child physical activity patterns. As with other neighbourhood environment characteristics, there are some exceptions to findings indicating support for the hypothesis that access to recreational resources is associated with child physical activity (e.g. Kligerman et al. 2007). However, numerous recent studies have found positive associations between objective measures of access to recreational facilities and measures of physical activity for children and adolescents (Boone-Heinonen et al. 2010, Frank et al. 2007a, Kerr et al. 2007, de Vries et al. 2007, Dowda et al. 2007, Gordon Larsen et al. 2006, Powell et al. 2007, Tucker et al. 2009, Epstein et al. 2006, Dowda et al. 2009). Frank et al. (2007a), for instance, found that access to recreation or open spaces was the only urban form variable related to walking for all age groups examined (ranging from age 5-20). Characteristics of such land uses may also moderate their influences on physical activity behaviour. For example, characteristics of recreational facilities likely to influence physical activity patterns include monetary costs of use, and type of facility. In this vein, Dowda et al. (2009) found that multipurpose, but not ‘individual’ or ‘team’ recreational facilities were significantly associated with vigorous physical activity of high school girls. Dowda et al. speculate that this is because the multipurpose facilities house the types of activities in which girls participate. Similarly, monetary costs of access may also need to be considered in conjunction with measures of physical access, because such costs also influence the overall 66  accessibility of the resource in question (de Vries et al. 2007, Dowda et al. 2007, Kligerman et al. 2007). Design of recreation and transportation facilities more broadly may also influence child physical activity patterns. 2.3.5.3.3 Design Related Characteristics of Urban Form As discussed above, measures of access are used to assess large scale elements of urban form relating to the distribution of land uses and connections between them. In contrast, design characteristics correspond to small scale elements of urban form such as the availability of seating, the presence of street trees, safe street crossings and adequate lighting. Relatively few design elements have been studied extensively with regards to their impact on child physical activity patterns, and findings to-date are generally preliminary.  STREET AND INTERSECTION DESIGN One category of design elements which has received moderate attention in research todate is that of street infrastructure design. For instance, the presence of a single busy street may act as a barrier for child movement on foot or bicycle, reducing opportunities for child physical activity. Timperio et al. (2006) lend support to this hypothesis, finding that an objectively assessed busy road barrier on route to school was a negative correlate of active commuting to school. Given the potential barrier effect of major roads, some studies have also considered possible associations between presence of controlled crossings and child physical activity. Boarnet et al. (2005a) found that children who passed by completed infrastructure improvements including sidewalk improvements, crossing improvements and improved traffic control were more likely to show increases in active transportation to school than children who did not pass by projects. Relating to street design more broadly, De Vries et al. (2007) found a positive association between MVPA and the frequency of parallel parking in a neighbourhood. 67  The authors speculate that this may result from a number of factors, including children using empty parking spaces during the daytime, and an increased perception of safety because of the barrier that parked cars create from traffic. Another possibility is that parallel parking covaries with other environmental characteristics such as slow speed zones and less heavy traffic (ibid). De Vries et al. also found that MVPA of children was negatively associated with neighbourhood traffic levels which may in turn be associated with street design. Finally, Carver et al. (2008b) found variable influences of street design on children, depending on their age, gender and type of physical activity. Relationships noted include positive associations between speed humps and adolescent boys’ MVPA during evenings, and between presence of two to three traffic/pedestrian lights in the neighbourhood and walking/cycling trips of adolescent girls.  PARK AND PLAYGROUND DESIGN Park and playground design may also play an important role in physical activity patterns of children. Ridgers et al. (2007) found that a park design intervention which involved retrofitting a park with color coded areas specifying different types of play zones and physical structure upgrades including soccer goal posts, basketball hoops and fencing resulted in increases in school recess time MVPA. Similarly, Colabianchi et al. (2009) found that renovated playgrounds were used by more adults and children than unrenovated playgrounds, and that the proportion of children who were vigorously active was greater at renovated playgrounds. Numerous other recent studies provide further evidence of associations between playground design (as rated by trained observers or school administrators) and physical activity including Fernandes and Sturm (2010), Haug et al. (2010), Loukaitou-Sideris and Sideris (2010), McKenzie et al. (2010), and Willenberg et al. (2010). Research in environmental psychology also lends support to the hypothesis that the design of park and other play spaces influences their use (Kytta 2002, Kytta 2004) and may therefore shape broader physical activity patterns. Play 68  spaces may, for instance, be made more conducive to play by incorporating elements that can be easily used or manipulated by children, such as climbable features, shelters and moldable features such as sand (Kytta 2004). Diversity of unique play opportunities may also be important to provide variety for individual children as well as for different children of varying ages and interests (Limstrand and Rehrer 2008, Walsh 2006). Design of play areas should also take into account parental needs and preferences, since parents will often be attending to their children at such areas. Given the potential importance of parental perceptions of safety regarding their children, play spaces may need to be designed so that they are perceived to be safe. De Vries et al. (2007) suggest that perceptions of safety may correspond to characteristics such as: visibility from residences, adequate lighting and maintenance. In other respects, parental perceptions of safety may run counter to the preferences of children for exciting, stimulating playgrounds, suggesting the need for a balanced approach. Staempfli (2009) presents an example of such an approach, the development of ‘adventure playgrounds’ that provide opportunities for unstructured and self-directed play within an environment made safe through adult supervision. Other design features of importance to parents may include availability of toilets, drinking water, water attractions, lighting and shade (Sallis and Owen 1999, Tucker et al. 2007). Notably, in a survey of parents attending parks with their children, Tucker et al. (2007) found that fewer than half of parents surveyed frequently used the most accessible parks, suggesting that, in many cases, design may be more important than access. 2.3.5.3.4 Multiple Built Environment Correlates Simultaneously In addition to empirical research investigating associations between individual neighbourhood characteristics and physical activity patterns, some studies have examined influences of multiple characteristics simultaneously. These studies may generally be divided 69  into two categories. The first category consist of studies comparing physical activity patterns of children based on broad urban form morphological categorizations such as ‘urban’ versus ‘suburban’ comparisons. One example of this type of study is Nelson et al. (2006), which relied on cluster analysis to identify neighborhoods characterized by socio-demographic and built environment characteristics. In their final typology, neighborhoods were given labels such as ‘rural working class’, ‘exurban’ and ‘mixed race urban’. Nelson et al. found that adolescents living in older suburban areas were more likely to be physically active than those living in newer suburbs, and also that adolescents living in lower socioeconomic status inner city areas were more likely to be active than residents of mixed-race urban areas. Findings from this and similar studies are complex and variable, suggesting the need for further research in this area (Sandercock et al. 2010, Norman et al. 2010). A second category of studies simultaneously investigating multiple characteristics consists of those analyses employing compound environmental indices, such as walkability indices (e.g. Kligerman et al. 2007, Roemmich et al. 2007). The main rationale for use of this kind of index is that neighbourhood environment variables may exhibit strong spatial multicollinearity, which makes analysis of independent effects of the variables difficult. For example, measures of urban form such as density, land use mix and street connectivity are often strongly correlated, and therefore may be combined into a smaller number of factors or components (Frank et al. 2005, Jago et al. 2006). Kligerman et al. (2007), for example, combined measures of net residential density, retail density, intersection density and land use mix, all calculated using a 0.5 mile buffer around subjects homes, into a single index, which was found to be associated with total average MVPA of adolescents studied. Findings of this and similar studies employing compound environmental indices are generally consistent with the findings of studies of individual built environment correlates of physical activity.  70  2.3.5.4 Limitations of Research on Neighbourhood Environment Correlates of Child Physical Activity Research on neighbourhood environment correlates of child physical activity is rapidly expanding, resulting in continuous improvements in understanding. Given the diversity of studies, it is difficult to identify any single universal limitation in study design. Nonetheless, some common limitations may be noted. One such limitation is the continued reliance on crosssectional research designs. As noted by Davison and Lawson (2006), the vast majority of studies in this area are based on a cross-sectional research design. In their review, 31 of the 35 studies examined were cross-sectional. Similarly, in their review of broader correlates of physical activity for children, Sallis et al. (2000) examined 108 studies, 86 of which were crosssectional. Cross-sectional studies examine physical activity behaviour of children in different contexts at a single point of time, and can only be used to identify associations between variables (Frank et al. 2007b). The overwhelming reliance on cross-sectional studies is thus widely viewed as a limitation because these studies cannot be used to make causal inferences (Godin et al. 2005). Longitudinal studies may be more appropriate for gaining insight into causal effects. One possibility is highlighted by Boarnet et al. (2005a), who investigate changes in physical activity behaviour occurring within a short time frame before and after built environment infrastructure modifications designed to improve the safety of routes to school for children. A second limitation is that empirical analysis is often emphasized with little corresponding development of or linkage to underlying theoretical frameworks. The value of ecological models of physical activity behaviour in general is increasingly recognized, because of their emphasis on wide ranging correlates spanning individual and multiple environmental contexts (Sallis and Owen 1999), but there is no single agreed upon model or theory, ecological or otherwise, describing physical activity behaviour (Spence and Lee 2003). Thus, beyond 71  recognizing the need to account for multiple and wide ranging factors, there is little agreement on which specific measures should be incorporated within ecological models (Owen et al. 2004, TRB 2005). One particular point of debate centers on the relative importance of environmental and individual factors. While ecological models have been developed in large part because of the recognition that individual factors alone cannot adequately explain physical activity behaviour (Barnett et al. 2006), they have also been criticized for their failure to adequately address certain individual factors (Spence and Lee 2003). Finally, theories of physical activity behaviour have been criticized for failing to specify the mechanisms underlying relationships between neighbourhood environment characteristics and specific physical activity behaviours (Brug et al. 2006). More sophisticated models may be required to investigate such mechanisms. Such models could, for instance, be designed to explore multiple possible mediating and moderating effects, and could be tested using statistical techniques such as structural equation models, that better represent complex causal pathways than conventional single equation models. A third limitation concerns the generalizability and comparability across studies due to the different social and built environment contexts studied and inconsistencies in methods and measures used. The degree to which studies can be directly compared is limited by a number of factors which may partly explain inconsistencies in results noted. One important factor is that each individual study occurs within a unique social and built environment context which may have important implications for the outcome of the study (Frank et al. 2007a). For example, relatively few studies of neighbourhood correlates of child physical activity have occurred within Canada, and the ethnicity classifications used in international studies may not be relevant in the multi-ethnic context of major Canadian metropolitan areas. Further, some studies rely on samples acknowledged to exhibit limited social variability relative to the broader populations from which they are drawn (e.g. Davison and Schmalz 2006). More generally, different areas 72  may be fundamentally non-comparable because of a constellation of social and environmental differences. Urban environments are characterized by wide variation along multiple dimensions, and simple analyses considering a limited number of characteristics may thus have little generalizability beyond the specific spatial context in which they occur. A second important factor potentially limiting the generalizability of results is the inconsistency in measures used for both physical activity, and correlates of physical activity. As discussed above, for example, some studies rely on objective measures of neighbourhood environment characteristics, while others rely on measures of perceptions of the environment. Beyond the broad distinction between objective and perceived measures, there are many other more specific differences between measures used, including widely varying buffer sizes for the characterization of the built environment and different land use categories incorporated in measures of land use mix. Despite such limitations, common findings across wide ranging contexts and measures may be interpreted as lending support to the existence of common mechanisms of influence on physical activity behaviours.  73  3. SAMPLE 1: OBJECTIVE BUILT ENVIRONMENT AND OBJECTIVE PHYSICAL ACTIVITY DATA 3.1 Overview This chapter presents methodology and results relating to Sample One as described in S. 1.2 above, consisting of 366 students attending 9 schools for whom objective built environment and physical activity data were collected. First, data sources and measure development are presented, with details on sampling, data cleaning and derivation of final measures for subsequent analysis. Because of the complexity of measure development and the implications of decisions made at this stage for subsequent analyses, this includes a detailed discussion of measure choice and derivation. This is followed by a description of sample characteristics corresponding to both students and the schools they attended, including a summary of selected indicators of physical activity behaviour and built environment characteristics. As a preface to the inferential analysis, alternative methods of analyzing clustered data are reviewed because sample 1 consists of students clustered within schools. Finally, each research question is investigated sequentially. The presentation of each question begins with the statement of hypotheses, followed by an overview of the analytical framework and methods, and concluding with results and discussion. The power to detect a range of effect sizes using this sample was estimated using G*Power (Fraul 2009). Specifically, power was estimated based on the following parameters: n=366, alpha=0.05, 6 predictors. Given these parameters, power was estimated at 0.77 to detect a small effect size (f2=0.02), 0.99 for a medium effect size (f2=0.15), and 1.0 for a large effect size (f2=0.35).All analyses in this thesis are premised upon ecological models of physical activity behaviour. This is evident in the research questions addressed (S. 1.2), which reference multiple levels of influences on behaviour extending from the individual (age, gender, ethnicity) to the household (car ownership, household income), to neighbourhood 74  built and social environments (e.g. residential density, median household income). In addition to informing the research questions, ecological models were drawn upon to inform measure selection, hypotheses, analytical frameworks and interpretation of results. Specific details of the theoretical underpinnings of this research are thus introduced throughout this chapter and the next.  3.2 Data Sources and Measure Development 3.2.1 Physical Activity Data 3.2.1.1 Sampling Process Physical activity data used in this study were collected by the Action Schools! British Columbia (AS!BC) research team between November 2005 and February 2006, as part of a multi-disciplinary, multi-year study of a school based physical activity intervention and child physical activity patterns more broadly (Naylor et al. 2006, Naylor et al. 2008, Reed et al. 2008, Nettlefold et al. 2010). The AS!BC initiative was led by Dr. Heather McKay (University of British Columbia) and Dr. Patti-Jean Naylor (University of Victoria). This initiative was initially tested as a pilot program between 2003 and 2005 (Reed et al. 2008). Physical activity data used in the present study represent baseline measurements following a Province-wide rollout of the program, prior to implementation of the school based intervention. A two stage sampling design was used, with a sample of schools selected first, and students subsequently selected within schools. Initially, 87 schools were randomly selected as a representative sample of schools within four of British Columbia’s five Health Authority Districts (Vancouver Island, Vancouver Coastal, Fraser and Northern Health Authorities were included; Interior Health was not). Of these, 29 schools did not meet a program eligibility criterion requiring that schools were not already engaged in a physical activity or healthy eating program. A further 28 schools declined participation, leaving 30 schools which were randomized for the intervention. Only  75  schools within the Vancouver Coastal and Fraser Health Districts (encompassing the Lower Mainland) were selected for objective physical activity measurement, leaving a possible sample of 12 schools. However, this portion of the study was implemented at a later stage than other portions, at which time three of the 12 schools had been provided with consents that did not include objective physical activity measurement. As a result, nine schools in the Lower Mainland were ultimately selected for objective physical activity measurement. These schools are spread across the region, with two in Vancouver, two in North Vancouver, three in Burnaby, and two in Mission (Figure 7). Within the nine participating schools, all grade 4 and 5 students were invited to participate, resulting in an initial recruitment of 629 children. Participants included the study provided verbal and written assent, and their parents or guardians also provided written informed consent (Nettlefold et al. 2010). The study was approved by the human research ethics board at the University of British Columbia (certificate number B050505).  76  Figure 7: Location of Nine Study Schools Within the Lower Mainland. Created using data from DMTI Spatial Inc. (2006).  77  3.2.1.2 Physical Activity Data Measurement and Measure Development Starting with an initial sample of 629 participants recruited, Figure 8 illustrates steps that were taken to result in a final sample of 366 students with valid physical activity data for subsequent analyses. Physical activity data was objectively measured using accelerometers, which are small, relatively unobtrusive measurement devices similar in appearance to pedometers. While 68 of the original 629 participants recruited were excluded because no measured data was available, and a further 54 students were excluded due to accelerometer malfunction or unavailability of accelerometers, the largest reduction in sample size was a result of cleaning available accelerometer data. In total, 125 students were excluded due to insufficient data, specified as consisting of at least 3 days with 10 hours of wear time per day (outlined below). Raw output data from the accelerometers is highly disaggregate, consisting of thousands of data points for each student. Thus extensive data cleaning was required using a systematic process to ensure that data for each participant were handled consistently. The processes of data collection and cleaning are now detailed, with Figure 9 highlighting major steps in the cleaning process. This description expands on previous summaries produced by the AS!BC team (Nettlefold et al. 2010, McGuire and Nettlefold 2007), with additional background explaining the rationale for major decisions and assumptions. The accelerometer used in this study was an ActiGraph GT1M activity monitor. This is a relatively small accelerometer (3.8x3.7x1.8 centimeters), which measures accelerations ranging in magnitude from approximately 0.05 to 2 g’s (where a g is a unit of gravitational acceleration corresponding to the nominal acceleration of gravity on earth at sea level, approximately 9.8 m/s2). The signal is filtered so that it detects normal human motion and rejects motion from sources such as vehicular movement. The immediate output produced by the accelerometer is analog, but it is digitized at 30 Hz to produce a measure of ‘counts’ at a selected ‘epoch’ or measurement interval. The number of counts varies in proportion 78  Figure 8: Path from Initial Participant Recruitment to Final Sample 1 (derived from Nettlefold et al. 2010)  Total Number of Participants Recruited n=629  Not Measured Absent Moved between consent and measurement Consent process incomplete Total not measured  n=55 n=5 n=8 n=68  Total Number of Participants Consented and Measured n=561  Excluded from Further Analysis Wore pedometer instead of accelerometer Accelerometer malfunctioned Insufficient data for inclusion (<3 days) No weekday measured Total excluded  n=31 n=23 n=125 n=2 n=181  Total Number of Participants with Valid Physical Activity Data n=380  Urban Form Data Unavailable No spatial referencing (address or postal code not provided) Total excluded  n=14 n=14  Total Number of Participants included in Sample One n=366  79  Figure 9: Physical Activity Data Cleaning Process. This highlights major steps in the cleaning process applied to physical activity data for each student. raw physical activity data: counts at 15 second intervals  1. screen for spurious data  2. identify start and end of wear time for each day repeat for all days of data 3. identify and exclude motionless bouts  4. at least 3 days of data with 10 hours of wear time per day?  no  exclude student from further analysis  yes 5. select intensity cut-points and process data using specialized software  final cleaned measures (e.g. average daily minutes of moderate to vigorous physical activity) 80  to the magnitude of acceleration of the accelerometer, and is interpreted as a measure of intensity of physical activity over the measurement interval. In this case, epochs of 15 seconds were selected to ensure that short bouts of activity would be captured. While epoch lengths of one minute are common in accelerometer studies, shorter epochs are increasingly recognized as important for accurately gauging intermittent activities (Reilly et al. 2008, Trost et al. 2005). Empirical evidence to date supports the premise that shorter epoch lengths may be desirable for measurement of child physical activity patterns because many activities such as sports participation and free play may involve intermittent physical activity (McClain et al. 2008). Longer epoch lengths may increase measurement error due to an averaging effect whereby short bouts of intense activity are combined with less intense activity (or inactivity) within a single epoch (McGrath and Hinckson 2009). Research assistants initially distributed accelerometers to participants at each school. The accelerometers were attached to an elastic belt and worn by students at the hip, the most common placement for accelerometers (Ward et al. 2005). Students were asked to wear the accelerometers during all waking hours every day for five days. The only exception to this request is that students were advised to remove them for bathing and swimming, because the devices are not waterproof. Students were also given an information package, and follow-up phone calls were made on the first evening of wear to answer any questions that parents and guardians might have regarding the study (McGuire and Nettlefold 2007). Upon collection of the accelerometers, the following major cleaning steps were implemented. First, the raw data was inspected for spurious data including extreme high values and extended periods with a single specific count value. Next, only students with three or more valid days of measurement were used in the study, including at least one valid school day and one valid weekend day. The minimum of three valid days of data was selected as a compromise between maximizing the reliability of the final measures for estimating habitual physical activity 81  and obtaining a sufficient final sample size for analysis. While guidelines for minimum recommended wear days for children vary, 4-7 days of measurement are commonly cited as a guideline (Trost et al. 2000, Esliger and Tremblay 2007). Despite such guidelines, shorter measurement periods are often used due to sample size or other practical constraints (e.g. Hume et al. 2005, Jago et al. 2006, Mattocks et al. 2008). In the present study, retaining only students with a minimum of four days would translate into an approximate halving of sample size from 366 to 193. Valid days were defined as consisting of days in which students had at least 10 hours of measurement, excluding motionless bouts (characterized by an accelerometer count value of zero) of 30 minutes or more. The minimum wear time was selected to be 10 hours following common practice in the literature (e.g. Stone et al. 2009, Moeller et al. 2008). Guidance on how best to treat motionless bouts is limited, and in the absence of a definitive guideline, motionless bouts of 30 minutes or more were identified as biologically implausible and excluded from the analysis. This duration approximately corresponds to a 95% confidence interval for the longest motionless bout length (27 +/- 2 minutes; Nettlefold et al. 2010). While the start and end of each day were generally easily recognizable through a visual inspection of the data, a decision rule was established to deal with temporal outliers in a consistent manner. Specifically, if non-zero counts occurred within 20 minutes of the last count, they were retained as valid. If counts occurred more than 20 minutes from the last counts, then they were only accepted as valid if there were more than eight epochs with non-zero counts over a four minute period. This approach was chosen to distinguish between possibly invalid counts (e.g. where parents moved the accelerometer after the child removed it) from valid, intermittent counts. More generally, this flexible approach was chosen rather than the more rigid alternative of setting absolute on or off times to acknowledge that individual children have varying sleep patterns (Ward et al. 2005) and arbitrary cut-offs times might therefore distort the results. 82  Once students with three valid days were identified, their data could be summarized using specialized software. While measures of total counts and average counts per minute may be used as outcome variables in quantitative analyses, their absolute values have limited meaningful interpretation because the units of measurements for counts are manufacturer specific (Esliger and Tremblay 2007). Thus it is common practice to summarize the data into physiologically and policy relevant units of measurement. In the present study, Kinesoft version 2.0.94 (Kinesoft Software, New Brunswick, Canada) was used to process the raw data into a variety of outcome measures including total daily MVPA and MVPA ‘windowed’ by time period (including weekend MVPA and MVPA outside of school on school days). In addition to the raw data, the software required specification of a variety of parameters. Key among these are ‘cut points’ used to classify physical activity by intensity. The cut points used in this study were age-specific cut points for sedentary, light, moderate and vigorous physical activity that were originally developed and validated using a group of children and adolescents aged 6-18 years (Trost et al. 2002). While these cut points have been criticized as being relatively low, and thus overestimating time spent in higher intensity activities (Guinhouya et al. 2006, Reilly et al. 2008), recent evidence appears to support their use (Trost et al. 2010). However, the debate over which cut points are most appropriate remains open, and a wide variety of cut points continue to be used by researchers (Trost 2007, Esliger and Tremblay 2007). Additional details on accelerometer data are provided in Appendix 2 and an overview of physical activity measurement alternatives is provided in Appendix 1. The primary measure created for subsequent analyses was daily minutes of MVPA averaged over the entire wear period of three or more days (Research Questions 1, 3, 5, 6, 7). Two related measures were created to assess neighbourhood environment influences on physical activity outside of school specifically. These are both based on 1-2 days of measurement: average minutes of MVPA outside of school on school days, and average minutes of MVPA on 83  weekends (Research Question 2, 7). Finally, physical activity was segmented by level of intensity, to produce measures of daily minutes of sedentary activity and light, moderate and vigorous physical activity averaged over the entire wear period (Research Question 4). While measures of MVPA outside of school were investigated as dependent variables for certain analyses, average daily MVPA across the measurement period was chosen as the primary measure because it is based on a minimum of three wear days and thus closer to the recommended minimum required to produce reliable measures of habitual physical activity. Focusing solely on MVPA outside of school has the advantage of eliminating possible confounding effects of MVPA during school. However, because these measures are based on only 1-2 days, they are considerably less reliable measures of habitual behaviour, and results of any analyses based on these measures accordingly need to be interpreted with caution. Findings by Fairclough et al. (2007) further suggest the need for cautious interpretation of the measure of MVPA outside of school on school days, specifically. In particular, Fairclough et al. note that the number of days of measurement to achieve adequate reliability when gauging physical activity during specific segments of the day may be substantially greater than the number of days required for whole day measures. 3.2.2 Control Data – Individual Characteristics Data on four control variables were collected in conjunction with accelerometer data: gender, age, ethnicity and weather. Gender was coded as a simple binary variable (male / female). Birth dates of participating students were collected, which enabled creation of an continuous age variable, rather than a categorical integer age. Ethnicity was coded based on the birthplace of parents and grandparents, using the following categories for ethnic origins from the Canadian Census: North American/European; North American Aboriginal; Caribbean; Latin, Central and South American; African; Arab; South Asian; East and Southeast Asian; Oceanian; 84  West Asian; and mixed. For example, if both parents or all four grandparents were born in North America or Europe, they would be considered North American/European. For analytical purposes, this variable was collapsed based on major ethnic groups in the study area (see S. 3.3.1) into the following categories: North American/European, East/Southeast Asian, South Asian, Mixed and other. Data on weather was coded based on scheduled accelerometer wear days as 0 for no precipitation or 1 for non-zero precipitation. In preliminary exploratory data analysis, this variable was found to be highly non-significantly associated with physical activity outcome variables. These finding are considered further below (S. 5.3.2), but weather was excluded from subsequent analyses. One important correlate of physical activity was notably not collected for all students in sample 1: household income. These data were only collected for the subsample consisting of students with additional survey data (sample 1B described in S. 1.1). As a result, attempts were made to impute missing household income values using regression based imputation with available data such as estimates of home price and census dissemination area median income. However, none of the imputation models attempted produced adequate fit, and thus household income was not included as a control variable for sample 1. Household income was however included in later models based on sample 1B. 3.2.3 Built and Social Environment Data 3.2.3.1 Data Sources and Integration With the exception of 14 students for whom home spatial referencing information was not provided by parents (Figure 8), built and social environment data was produced for the entire sample of students with valid accelerometer data. The following data sources were incorporated into a Geographic Information Systems (GIS) framework, using ArcGIS 9 and Network Analyst 9.1 (ESRI 2005) for the creation of built and social environment measures:  85  •  2006 parks and street network data produced by GIS Innovations Inc. (GIS Innovations Inc. 2010),  •  2006 street network data produced by DMTI Spatial Inc. (DMTI Spatial Inc. 2006),  •  2006 census population and income data (Statistics Canada 2007, 2008), and  •  2005 parcel level land use data produced by BC Assessment (BC Assessment 2005).  Additional details on the original data sources are provided in Appendix 3. The process through which built and social environment measures were produced may be broken down into the following steps: 1. identification of activity settings, 2. neighbourhood delineation, 3. measure selection, 4. data cleaning, and 5. measure implementation The following sections correspond to each of the above steps, outlining major methodological considerations. 3.2.3.2 Identification of Activity Settings The first step in the creation of built environment measures is the identification of specific neighbourhood settings in which research subjects are anticipated to engage in physical activity (S. 2.3.5.1). An ideal approach might involve the use of GPS technologies to track individuals and thus identify all possible activity settings, as suggested by Cummins et al. (2007). Such an approach would have the advantage of enabling researchers to account for spatially disperse activity settings which may influence physical activity behaviour, such as parks near the homes of children’s friends. This approach remains to be developed and widely implemented, although a small number of preliminary studies have now employed GPS (Cooper 86  et al. 2010a and 2010b, Jones et al. 2009, Quigg et al. 2010). In the absence of such data identifying activity settings specific to individuals, it is necessary to rely upon a theoretical framework to identify relevant settings. In the present study, the theoretical framework used is the BME (Behavioral Model of the Environment), discussed in S. 2.3.2. Based on this model, three primary activity settings were defined for school aged children: a child’s home, their school, and the route between their home and school. By specifying these settings, environmental data could in turn be linked to individual data. In the case of student’s homes and their schools, street addresses were used for this purpose. These were available for the vast majority of students (n=358) and for all schools. In instances where student home addresses were unavailable, the following approach was used. If postal codes were available, these were used for spatial referencing, in conjunction with Statistic Canada’s Postal Code Conversion Files (PCCFs; UBC 2010). PCCFs are designed to assign spatial coordinates to postal codes. Use of postal codes in this manner introduces some degree of spatial imprecision. In most cases, this is relatively minor because a typical urban postal code corresponds to a block face (ibid). However, the area corresponding to a postal code may vary in size depending on its setting, from a single large building in an urban core to disperse communities in rural settings. One recent study examining the degree of spatial imprecision introduced as a result of using PCCFs in an urban setting (Calgary, Alberta) found that approximately 90% of postal code locations were within 200 meters of the true address location (Bow et al. 2004). In the present study, postal codes in conjunction with PCCFs were used to spatially reference eight students living in urban settings. If postal codes were unavailable, students were excluded from further analysis (n=14; Figure 8). Finally, the route between a child’s residence and school was modeled as the shortest distance between their home and school, along the road network. The validity of this approach rests on a number of assumptions, including for instance that children do not take any off road 87  shortcuts to get to school, and that parents allow their children to take the shortest route, although it may not be the route perceived to be the safest. In the absence of additional information, this approach was selected as a reasonable and convenient option with no other clear alternatives. In addition, both parental perceptions and objective measures of route safety were taken into account through other means, as discussed below.  3.2.3.3 Neighbourhood Delineation For measures involving a child’s residence and school, the second step in measure development was to delineate neighourhoods using buffers around these locations to use as the spatial basis for measure estimation (S. 2.3.5.1). The process of buffer generation requires specification of two parameters: type and radius. Following common practice in current research on built environment correlates of physical activity, network buffers were selected as the type of buffer. As noted above, these approximate the area accessible to an individual within a specified radius, along the road network, from specified spatial coordinates (in this case, the address of a child’s home or school; Figure 5(b)). However, because research to-date presents little insight on the appropriate radius of buffers for studying built environment influences on children, the decision was made to explicitly analyze multiple buffer sizes. This is consistent with the recommendations of a recent review acknowledging the uncertainty regarding appropriate buffer size (Brownson et al. 2009). It was anticipated that smaller buffer sizes might be most appropriate for the present study, because children aged 8 to 11 may have limited independent mobility (S. 2.3.4.1.1), and thus be most influenced by the environments immediately surrounding their home and school. However, since parental influences on child mobility and broader activity patterns have been well documented, larger buffer sizes possibly better corresponding to parental perceptions of neighbourhood characteristics were also employed. Four buffer sizes in total were selected: 200 meters, 400 meters, 800 meters and 88  1600 meters. This range of buffer sizes was selected to encompass and extend beyond typical buffer sizes of 400 meters to 1 kilometer. This range also extends beyond the range of buffer sizes examined by Liu et al. (2007), who note that they may not have used as broad a range as necessary to capture environmental influences on child physical activity patterns. 3.2.3.4 Measure Selection As noted in Figures 4 and 6, a wide variety of possible neighbourhood environment correlates of child physical activity patterns have been identified. The selection of measures for the present analysis was not therefore constrained by existing theory, but was primarily constrained by available data. Characteristics hypothesized to influence physical activity behaviour that could not be measured given available data include neighbourhood social capital, and many aspects of design gauging, for example, the quality of neighbourhood play spaces. The available data, however, did permit the creation of a wide variety of measures, as outlined in Table 6, including some novel measures relating to street characteristics, the social environment and access to non-park recreation sites. Based on the typology presented in Figure 6, many of the measures developed relate to physical access, including measures of proximity (e.g. distance to school), and connectivity (intersection density). Other measures relate to design, and in particular street design, including the proportion of streets with posted speed limits of 30 kilometers per hour or less, and cul de sac density (although this could also be considered a measure of access). Finally, some measures were chosen to characterize the neighbourhood social environment, including population density of children and median household income. An ecological classification of measures chosen for this analysis is illustrated in Figure 10.  89  Table 6: Theoretical Rationale for Selection of Objective Built and Social Environment Measures. Operationalization of measures is described below, in S. 3.2.3.6. Measure  Theoretical Rationale (see also Chapter 2)  Route Between Home and School:  Home/school route characteristics may directly influence physical activity by influencing mode choice to school. They may also indirectly influence physical activity by moderating access to recreational resources on the school site.  1. distance to school  Longer distances to school likely limit opportunities for active transportation to school (Ewing et al. 2004, Schlossberg et al. 2006, Timperio et al. 2006, Larsen et al. 2009)  2. average intersection spacing, 3. total number of four way intersections  More intersections en route to school and in particular, more major intersections may translate into greater perceived danger on the part of parents, thus limiting active transportation to school (Timperio et al. 2006).  Home / School Neighbourhood: 4. net commercial density, 5. net residential density, 6. land use mix  These three characteristics all may influence physical activity through multiple mechanisms (S. 2.3.5.3.2). For example, higher density and land use mix may translate into: • more destinations and thus may lead to greater active transportation, and • more ‘eyes on the street’ and thus greater perceptions of safety  7. intersection density  Higher street connectivity, as gauged by intersection density, also increases access to destinations, possibly promoting active transportation. Conversely, higher street connectivity may decrease parental perceptions of safety because children may have to cross more intersections to get to a given destination.  8. cul-de-sac density  Cul de sacs close to home may present safe areas for children to play in, or may at least increase parental perceptions of safety (Timperio et al. 2006).  9. proportion of streets with posted speed limits of 30 kilometers per hour or less  Traffic speed is well documented as an important factor impacting pedestrian safety (Ewing and Dumbaugh 2009). While posted speed limits do not directly reflect traffic speed, they may be a reasonable proxy and therefore may also influence physical activity behaviour through perceptions of safety. 90  Table 6 (continued): Theoretical Rationale for Selection of Objective Built and Social Environment Measures. Operationalization of measures is described below, in S. 3.2.3.6 Measure  Theoretical Rationale (see also Chapter 2)  10. lane-kilometers of roads per square kilometer  This measure reflects both the number and width of roads in an area, and may be conceived of as a measure of the degree of coverage of a neighbourhood by roads. Higher values of this index may therefore correspond to lower perceptions of safety.  11. number of parks in buffer 12. distance to closest park  As with school sites, parks may influence physical activity both by providing destinations for active transportation, and by acting as sites for physical activity. Two measures of access to parks were developed given the demonstrated importance of such measures in past studies and in theory (S. 2.3.5.3.2). Measure 12 represents a simple measure of access to the single closest park, while Measure 11 gauges access in a broader sense, with more parks translating into greater variety for children. Measure 11 may be particularly important because children or their parents may prefer more distant parks over the closest park, depending on their design and amenities (Tucker et al. 2007).  13. distance to closest possible non park recreation site 14. population density of children 15. proportion of population made up by children  In addition to parks, a variety of other sites may provide opportunities for recreation (Limstrand and Rehrer 2008).  16. median household income  Higher median household income may be associated with a number of unmeasured aspects of access to recreational opportunities. For instance, parks in higher income neighbourhoods may have more amenities or be better maintained (Crawford et al. 2008).  Higher numbers of children in a given neighbourhood may foster social interaction among children and result in greater opportunities for physical activity. The presence of other children might also increase the confidence of parents (Tranter 2006), encouraging them to let their children play outdoors.  91  Figure 10: Examples of Measures Incorporated in Analyses, Categorized in an Ecological Framework components of environments  social  physical residential density, street connectivity  neighbourhood environment  car ownership  household environment  age, gender, ethnicity  individual  proportion of population consisting of children household income  3.2.3.5 Data Cleaning Distinct validation and cleaning steps were developed for each data source. Street network and park data provided by GIS Innovations Inc. (2010) and DMTI Spatial Inc. (2006) were validated by cross referencing with each other, and with secondary sources (Google 2007, MapArt 2006). No unexplainable discrepancies were identified in the study area. This is not surprising given that such data are used by a variety of corporate clients requiring accurate and complete information and undergo rigorous quality control procedures including extensive field validation (GIS Innovations 2006). These data sources were subsequently used as part of the creation of the majority of measures outlined in Table 6 above (measures 1-3 and 7-13). Census data from Statistics Canada were also considered to be of high quality, and the primary cleaning step required for these was to impute certain population characteristics in instances where data were suppressed due to high non-response rates (Luebbe 2009). Specifically, populations of children in a small number of Dissemination Areas (DA) were imputed based on the total DA  92  population (which was not suppressed), and the average age distribution of adjacent DAs prior to the creation of measures 14 and 15 (Table 6). In contrast to the other data sources, an extensive cleaning process was required for BC Assessment data (BC Assessment 2005). As a result, only very brief highlights are discussed here, with additional details outlined in Appendix 3. In general, use of assessment data is complicated by the fact that it may originate from multiple sources. In British Columbia, spatial data is created by individual municipal jurisdictions, while the BC Assessment Authority maintains a central tabular database with attributes of interest (e.g. land use codes, residential floor area; Setton et al. 2005). Simply linking these data was therefore complex because of the different formats used by the jurisdictions involved. In addition, it was necessary to impute floor area data in a small proportion of cases, by developing jurisdiction and use class specific regression formulas to predict floor area based on property improvement values. Following these and other cleaning steps, a final product consisting of parcel level data on land use, land area and building area was produced. This was subsequently used to create measures of density, land use mix and access to non-park recreational sites (measures 4-6 and 13, Table 6 above) with a high degree of spatial precision.  3.2.3.6 Measure Implementation Following the cleaning of data, individual measures were created through an iterative process of development, testing and fine tuning. This section highlights aspects of this process with implications for measure interpretation. Emphasis was placed on creating measures which accurately represent built environment features and are also easily interpretable and policy relevant. In this regard, starting points for measure creation include work by Frank and colleagues (Frank et al. 2005, Frank et al. 2006, Kligerman et al. 2007), as well as a very  93  detailed GIS protocol produced at the University of Minnesota specifically for studies of neighbourhood environment influences on physical activity (Forsyth et al. 2005). Even with valid, complete data, certain data sources required additional preprocessing steps prior to measure preparation, to ensure that certain spatial features were accurately represented for the intended purpose. For example, when calculating the shortest distance between a child’s home and park, the software used (Network Analyst 9.1; ESRI 2005) required both origin and destination to be represented as individual point locations. The simplest implementation of this measure involved representing a park by its centroid (center point), and the measure thus created was the distance between a child’s home and the park centroid. While this simplification was reasonably accurate for small urban parks, substantial spatial error was introduced when applied to large regional parks of highly irregular shapes. The final iteration of this measure therefore involved representing the edges of parks as a series of points, a step which dramatically increased processing time, but resulted in a much more accurate spatial representation of park access. On average the difference between the initial and refined measure was 175 meters, but in some instances, the difference was as great as two kilometers. A second example of preprocessing of data with major implications for measure interpretation concerns the calculation of intersection density. The first iteration of this measure involved an automated procedure to calculate the number of intersections falling within individual network buffers. However, street network data is commonly prepared for a variety of clients interested in accurately modeling driving conditions for applications such as fleet routing and drive time analysis (ESRI 2010). Given this emphasis, certain road features such as median separated road segments and major intersections are represented as complex, highly interconnected networks that do not accurately reflect connectivity from the perspective of cyclists or pedestrians. In effect, such features result in an inflation of intersection density estimates, unless simplified as shown in Figure 11. If ignored, this problem would in particular 94  result in highly inflated estimates of street connectivity in certain suburban settings characterized by intersecting arterial roadways.  Figure 11: Street Network Data Preprocessing for Calculating Intersection Density. Intersections counted illustrated by circles ( ) (a) intersection with traffic islands – five intersections counted instead of one. Note that in highly complex intersections of major roads, single intersections were represented as upwards of 20 intersections in the original network.  original network  simplified network  (b) median separated road segment – eight intersections counted instead of two  original network  simplified network  The final iteration of the intersection density measure was therefore based on a simplified road network as illustrated above. This simplified network was also used for the measures of total number of four way intersections and average intersection spacing. Notably, while it may be assumed that such steps are commonly taken prior to measure preparation, such basic details are rarely mentioned in the literature, despite the implications if they are ignored. 95  Because of the large number of processing steps and parameter inputs that go into the creation of individual measures, ample opportunities exist for introducing error. An important final step was therefore to examine the output measures at each iteration, both spatially, and aspatially. Examining output measures in their spatial context was particularly useful in identifying spatial outliers that would not be evident simply by looking at the absolute numbers. Aspatial examinations were also employed to identify quantitative outliers, and to compare absolute values of certain measures (e.g. net residential density) to known values from other contexts or sources. Table 7 presents a summary of the final measures as implemented, with detailed definitions.  96  Table 7: Definition of Neighbourhood Environment Measures  Measure  Units  Definition and Assumptions  All of these measures were calculated based on the route with the shortest distance between a child’s home and school along the street network.  Route Between Home and School 1. distance to school  meters  Shortest distance along the road network. Because school sites are quite large in some cases, and have multiple access points, these were identified and spatially referenced so that the distance reflected multiple possible points of entry.  2. average intersection spacing  meters  Calculated as the shortest distance along the road network divided by the total number of three and four way intersections, using the simplified road network described above.  3. total number of four way intersections  dimensionless Total number of four legged intersections, based on the simplified road network (count) described above.  Home / School Neighbourhood 4. net commercial density  All of these measures were calculated using 200, 400, 800 and 1600 meter network buffers as the spatial unit of analysis (described in S. 2.3.5.1 above). dimensionless Calculated as a floor area ratio (FAR): FAR = commercial building floor area for all parcels in buffer commercial site area for all parcels in buffer Where commercial land uses were broadly defined to consist of land uses such as department stores, convenience stores and restaurants. This is a net density measure because the denominator consists of commercial land area, not the entire land area falling within the buffer. As such, this measure can be interpreted as a measure of intensity of commercial land uses on commercial sites. 97  Table 7 (continued): Definition of Neighbourhood Environment Measures Measure  Units  Definition and Assumptions  5. net residential density  dwellings per residential acre  Calculated as: net residential density =  total number of dwellings in buffer total residential site area for all parcels in buffer  Where the total number of dwellings in a buffer was determined using 2006 Census data (Statistics Canada 2007b), and the total residential site area was determined using BC Assessment parcel data. The total number of dwellings was derived via a two step process. First, Census DA data was spatially joined with parcel level assessment data to assign dwellings to residential parcels based on the proportion of overlapping DA area. Second, network buffers were spatially joined with the estimated parcel level dwelling counts to obtain an estimate of total dwellings in a buffer. 6. land use mix  dimensionless, ranging in value from 0 to 1  Land use mix was calculated using methodology developed by Frank and colleagues (Frank et al. 2005, Frank et al. 2006, Kligerman et al. 2007), based on an ‘entropy index’ which varies in value from 0 to 1 with higher values representing a more even mix of land uses:    ∑ pi ln ( pi )  Land Use Mix = − i =1...n ln n Where: i refers to land use i (in this case, single family residential, multi-family residential, entertainment, commercial or office); k is equal to 5, the total number of land uses incorporated in the measure; and pi refers to the proportion of land in use i. Note that because this measure is based on total areas, it does not reflect the ‘grain’ of mixing of land uses. Thus, an urban mixed use development incorporating many small commercial and residential buildings could be characterized by this measure as having the same degree of land use mix as a suburban development with only a few large commercial and residential buildings, if they encompassed the same relative proportions of land uses. 98  Table 7 (continued): Definition of Neighbourhood Environment Measures Measure  Units  Definition and Assumptions  7. intersection density  intersections per square kilometer  Calculated as the number of intersections with three or more legs, per square kilometer of buffer area, excluding water. Based on the simplified street network described above.  8. cul-de-sac density  cul de sacs per square kilometer  Calculated as the number of cul-de-sacs, per square kilometer of buffer area, excluding water. Based on the simplified street network described above.  9. proportion of streets with posted speed limits 30 kilometers per hour or less  dimensionless (proportion, ranging from 0 to 1)  Calculated as the proportion of streets weighted by the length of street segments within a buffer, with posted speed limits of 30 kilometers per hour or less. Thirty kilometers per hour was selected as a reasonable threshold because: a) the distribution of > 30 km/h speed limits in the study area is relatively flat (the vast majority of roads have a speed limit of 50 km/h), and b) pedestrian fatality rates increase exponentially, with speeds of approximately 30 km/h or less resulting in a 5% fatality rate. This increases dramatically to 45% at approximately 50 km/h (Ewing and Dumbaugh 2009).  10. lane-kilometers of roads  lanes·kilometers This measure was calculated using the product of the number of lanes and the per square length of street segment, for all street segments falling inside the buffer. This kilometer measure was calculated using the original street network, not the simplified street network described above, to properly account for all road segments.  11. number of parks in buffer  dimensionless (count)  Calculated as the total number of public parks falling entirely or partly within the buffer. This includes local and regional parks, but excludes land uses included in measure 13 (e.g. cemeteries and golf courses), and other land uses likely to have access restrictions (e.g. protected areas and ecological reserves). Created using GIS Innovations park data rather than assessment data because assessment data spatially represents all land uses, including parks in terms of legal parcels. Thus, a given park may consist of multiple contiguous parcels, distorting the count.  99  Table 7 (continued): Definition of Neighbourhood Environment Measures Measure  Units  Definition and Assumptions  12. distance to closest park  meters  Calculated as the shortest distance to the nearest park, along the road network. As described above, based on park boundaries modeled as a series of points to minimize spatial error. This approach assumes that park boundaries are accessible from the road network at multiple points. Calculated using GIS Innovations Inc. (2010) park data.  13. distance to closest possible non-park recreation site  meters  This was calculated using the shortest distance to specific land uses identified using assessment data, along the road network: campgrounds, bowling alleys, community halls, golf courses, recreational clubs/ski hills, cemeteries, schools and colleges/universities. These land uses were selected either as possible sites for independent recreation (e.g. school sites), or as sites which may offer opportunities for supervised recreation with parents or other adults (e.g. community halls, ski hills). In general, given available data, it was not possible to distinguish between sites based on public accessibility or costs of access, but it was assumed that all of the selected land uses would have at least some degree of public accessibility. For example, some campus sites (schools, colleges/universities) identified by this measure are very likely not completely accessible by children, but were selected because they usually incorporate large open areas or sports fields which may at least offer some opportunities for public recreation.  14. population density of children  children per residential acre  Calculated as net population density =  total number of children aged 5-14 total residential site area for all parcels in buffer  The total number of children aged 5-14 falling in the buffer was calculated using census population data (Statistics Canada 2007) in the same way as the number of dwellings noted in measure 5. The total residential site area was calculated using assessment data. The age range of 5-14 was selected specifically as the smallest range available using census data, encompassing the range of students participating in the study (8-11). 100  Table 7 (continued): Definition of Neighbourhood Environment Measures Measure  Units  Definition and Assumptions  15. proportion of population made up by children  dimensionless (proportion expressed as a percentage ranging from 0100)  Calculated as:  16. median household income  dollars ($)  proportion of children =  total number of children aged 5-14 total population  This measure was calculated using census data as described for measure 14 above. Calculated using Census income data (Statistics Canada 2008). As with measure 5, this measure was calculated by assigning Census DA median household income values to residential parcels falling within each DA. An average median household income value was subsequently calculated by spatially joining network buffers with the parcel data.  101  3.3 Sample Description and Regional Context 3.3.1 Student Characteristics As illustrated in Table 8, the sample is approximately equally comprised of boys and girls. The largest category of ethnicity for participants is European/North American, followed by East/Southeast Asian and then South Asian. The distribution of students’ ages is approximately normal, with a mean age of 9.94.  Table 8: Descriptive Statistics for Sample 1 Control Variables. Data presented for sample 1, n=366; and for students excluded due to insufficient accelerometer data or missing spatial referencing data, n=263. Percent / Value in Sample 1 Gender Male Female Ethnicity European/North American East/Southeast Asian South Asian Mixed Other Age Mean quartile 1 quartile 2 (median) quartile 3  Percent / Value in Excluded  47.6% 52.4%  50.2% 49.8%  44.0%  38.7%  29.3% 11.1% 7.6% 8.0%  29.1% 11.1% 7.7% 13.4%  9.94 9.43 9.94 10.46  9.87 9.46 9.86 10.30  Approximately 70% of Grade 4 and 5 students in participating schools consented to participate, but data on the characteristics of non-consenting students is not available. Characteristics of students who consented, but were subsequently excluded due to insufficient data or missing spatial referencing data (n=263, as illustrated in Figure 8) are very similar to students in the final sample, suggesting that little bias was introduced due to the data cleaning process, except possibly with regards to the ethnic profile of students. Age and gender profiles 102  of the excluded students are very similar as shown in Table 8, with excluded students slightly more likely to be male and younger on average. The ethnicity profiles are also very similar, with nearly identical proportions of East/Southeast Asian, South Asian and mixed ethnicity students. However, the excluded students include a greater diversity of other ethnicities (13.4% compared to 8.0%), and the proportion of European/North American were commensurately lower (38.7% compared to 44.0%). While directly comparable census data do not exist, available data suggests that sample 1 is a reasonable representation of the Lower Mainland, in which 41.2% of residents reported being a member of a visible minority, including 9.8% as South Asian and 26.5% as East/Southeast Asian (Statistics Canada 2008b). However, this ethnically diverse sample is not representative of a broader Canadian population, in which only 16.2% of the population reports being a visible minority (ibid).  3.3.2 School and School Neighbourhood Characteristics Table 9 presents a high level summary of characteristics of schools, their surrounding neighbourhoods and study participants by school, to situate the study sample within its regional context. The sample of schools consists of nine elementary schools (all offering kindergarten to Grade Seven classes) of varying sizes, ranging in enrollment from 229 to 1043. The number of study participants with valid data is similarly varied and the ethnic profile of participants reflects the ethnic diversity of the schools. In addition, the schools are situated in neighbourhoods with widely varying median household incomes and built environment characteristics. Figure 12 illustrates the distribution of school neighbourhood median household incomes in relation to regional statistics on median DA household income. While five of the schools fall between the first and third quartiles for the region, three fall below the first quartile, and one well above the third quartile.  103  Table 9: Characteristics of Schools, Their Neighbourhoods and Students with Valid Accelerometry Data Municipality and School (see Figure 7 for a map of school locations) Vancouver 1  School Characteristics Total % of students Enrollment* whose home language is not English**  School Neighbourhood Characteristics Median Walkability Household Index**** Income***  Participants with Valid Physical Activity and Neighbourhood Environment Data Number of Top Three Ethnicities in School Participants  741  89% $52,000  3.16 45  2  301  72% $38,000  5.82 24  3  303  26% $96,000  -2.14 43  4  315  13% $68,000  -2.00 46  5  229  56% $77,000  -1.03 13  6  412  41% $72,000  -2.06 35  7  1043  67% $47,000  1.40 83  8  402  11% $67,000  -2.21 40  9  280  4% $47,000  -1.91 39  South Asian, East/Southeast Asian, Latin/Central/South American East/Southeast Asian, Latin/Central/South American, Aboriginal  North Vancouver North American/European, East and Southeast Asian, Mixed North American/European, Mixed, Aboriginal  Burnaby South Asian, East and Southeast Asian, North American/European East/Southeast Asian, North American/European, West Asian East and Southeast Asian, North American/European, Mixed  Mission North American/European, South Asian, Mixed North American/European, East and Southeast Asian, Mixed  See following page for notes (*,**,***,**** ) 104  Notes for Table 9: *, ** Numbers for 2005-2006 school year using BC Ministry of Education (2010) data. ***Estimated based on 2006 Census data (Statistics Canada 2008), using an 800m network buffer surrounding school. ****Walkability index based on measures of residential and commercial density, street connectivity and land use mix, as developed by Dr. Larry Frank and applied to the Lower Mainland (Appendix 4). Discussed further in this section.  Figure 12: Median Household Income by School Neighbourhood. Median household income for 9 schools based on an 800 meter network buffer, compared to first quartile, median and third quartile median household incomes for Lower Mainland DAs. $120,000  Median Household Income  $100,000  $80,000  Third Quartile  $60,000  Median First Quartile  $40,000  $20,000  $0 2  7  9  1  8  4  6  5  3  School  The variation in built environment characteristics of school neighbourhoods can be characterized by a ‘walkability index’ based on measures of residential and commercial density, street connectivity and land use mix, as developed by Dr. Larry Frank and colleagues and applied to Metro Vancouver (Appendix 4). Built and social environment characteristics developed for the present study (Table 7) were created solely for the sample of students being studied. In contrast, walkability index values are available for the broader region and thus can be used to situate the environmental characteristics of neighbourhoods surrounding the nine schools within their broader regional context. High walkability index values (e.g. 5.82 for 105  School 2) typically reflect dense, mixed urban environments, with grid street networks while low values (e.g. -2.14 for School 3) typically represent low density, single use suburban environments. As indicated in Figures 13 and 14, the schools are situated in diverse environments. Figure 13: Walkability of School Neighbourhoods Compared to Walkability for Selected Lower Mainland Municipalities. The walkability of school neighbourhoods is indicated by yellow squares/circles. The walkability of selected municipalities (in parentheses below the municipality name) is highlighted for comparative purposes. Morphological characterizations of these municipalities are also indicated for reference.  Figure 13 in particular illustrates that at the one extreme, School 8 is situated in a neighbourhood with walkability similar to new suburban developments. While exurban/rural environments are not represented in this study, it is likely that relationships between urban form and physical activity patterns would be very different in such environments, given the high degrees of car dependency of residents. At the other extreme, School 2 is situated in a neighbourhood with walkability above that of the City of Vancouver’s average. This reflects the fact that the City of Vancouver has varied built environment morphologies ranging from mature suburban developments historically developed on grid street networks along streetcar  106  Figure 14: School Locations in Relation to Walkability. Based on walkability surface data developed by Dr. Larry Frank and colleagues, as described in Appendix 4. Created in part using data from DMTI Spatial Inc. (2006).  107  lines to extremely high density mixed use developments. School 2 is located in a neighbourhood representing a hybrid of such morphologies, with a mix of moderate to high density housing types and main-street commercial development. Given the potential importance of neighbourhood income as a covariate of physical activity, a sampling design explicitly taking into account the simultaneous variation in neighbourhood walkability and income would have been desirable. Using such an approach might involve selection of schools in the following types of neighbourhoods: high walkability, high income; high walkability, low income; low walkability, high income; and low walkability, low income, to adequately capture variation of the broader population of neighbourhoods. As illustrated above, the final sample of schools is situated within a broad range of neighbourhoods as characterized by walkability and median household income. However, none of the schools fall in neighbourhoods characterized by both high income and high walkability. Such neighbourhoods do exist in the Lower Mainland (e.g. Kitsilano). In contrast, a variety of low income environments are well represented, with schools 2, 7 and 9 falling within high, medium, and low walkability environments, respectively. In addition, school 3 falls within a high income, low walkability environment. Detailed descriptions of individual school neighbourhoods are presented further in the next section.  3.4 Descriptive Analysis 3.4.1 Built and Social Environment Characteristics In this section, characteristics of the built environments that study participants live in are further explored. Whereas the emphasis of the previous sections was on broadly situating the study sample within its regional context using selected indicators for school neighbourhoods, the emphasis in this section is on detailing the full range of built environment measures created for the present study (Table 7) at the level of individual students. First, these measures are 108  summarized across all students. Next, variation in these characteristics across schools are discussed to further illustrate contrast in the built and social environments assessed in this study. For ease of comparison, all neighbourhood measures referred to in this section (measures 4-16 in Table 7) are based on a single intermediate network buffer size of 800 meters. 3.4.1.1 Descriptive Statistics for Entire Sample Table 10 summarizes descriptive statistics for built and social environment measures for all students in sample 1. On average, students live approximately 1300 meters from their school along the road network, but there is considerable variation in this measure reflecting both variation within and across schools. The route to school is generally characterized by small blocks (mean intersection spacing of 120 meters), and students correspondingly live in environments with moderately high intersection densities. The mean intersection density of 50 intersections per square kilometer could correspond to an urban grid street network with an average block size of 140 meters, but because this measure does not distinguish between three and four way intersections, it could also correspond to a wide range of street-network variations. Measure three in contrast focuses solely on four way intersections, and hence greater variation between students is evident in this measure. Average cul-de-sac density is considerably lower than intersection density, at approximately eight cul-de-sacs per square kilometer over the entire sample. The mean net commercial density of 0.6 floor area ratio could correspond to single storey strip mall developments with parking in front, or alternately, single storey ‘main street’ commercial developments built tightly to the street, with parking in the rear. Both of these forms of development are widespread in the Lower Mainland. The small absolute value of the standard deviation in this measure might be taken as implying relatively limited variation in urban form, but the difference between a FAR of 0.94 (mean + one standard deviation) and 0.26 109  Table 10: Descriptive Statistics for Built and Social Environment Measures for Students in Sample 1 (n=366). Values reported for home neighbourhood based measures correspond to 800 meter buffer size. Measure (units) Route Between Home and School 1. distance to school (m) 2. average intersection spacing (m) 3. total number of four way intersections Home / School Neighbourhood 4. net commercial density (dimensionless floor area ratio) 5. net residential density (dwellings/residential acre) 6. land use mix (dimensionless) 7. intersection density (intersections/km2) 8. cul-de-sac density (cul-de-sacs/km2) 9. proportion of streets with posted speed limits of 30 kilometers per hour or less (%) 10. lane-kilometers of roads (lane-km/km2) 11. number of parks in buffer 12. distance to closest park (m) 13. distance to closest possible non-park recreation site (m) 14. population density of children aged 5-14 (population/ residential acre) 15. proportion of population made up by children aged 5-14 (%) 16. median household income ($)  Mean (Standard Deviation) 1280(1780) 120(50) 3.7(4.6)  0.60(0.34) 13.33(11.33) 0.33(0.23) 51.52(17.40) 8.25(6.61) 8.24(6.46) 32.28(7.48) 3.00(1.63) 410(860) 485(690) 3.58(1.99) 11.41(2.78) $58,000($14,000)  (mean – one standard deviation) is considerable. Assuming single storey development, the former (0.94) implies nearly complete site coverage, while the latter (0.26) could consist of a mall type development with a building occupying one quarter of the site and the remainder occupied by parking. Similarly, land use mix values correspond to neighbourhoods with a considerable range of non-residential development mixed with residential at one extreme and single use suburban residential developments at the other. The mean proportion of streets with posted speed limits of 30 kilometers per hour or less is moderately high (at 8%), reflecting the fact that some children live in close proximity to their  110  school, and the areas surrounding these schools often have school zone posted speed limits of 30 kilometers per hour (Figure 15). Measure 10 is a relatively novel measure, and in contrast to residential and commercial density measures, it cannot be easily visualized. However, the mean of 32 lane kilometers of roads per square kilometer suggests a high degree of street coverage in student’s neighbourhoods that is consistent with the moderately high intersection density values observed.  Figure 15: School Zone 30 kilometer/hour Speed Limit  The mean for measure 11, the number of parks in a buffer, indicates that students generally have a very high level of access to parks, with three on average within an 800 meter network buffer surrounding their home. The mean distance to the closest park of 410 meters correspondingly falls well within 800 meters. Students have a similar mean level of access to possible non-park recreation sites, with such sites on average 75 meters further than the closest park. Because school sites are equipped with a range of recreational amenities (Figure 16), they may also be considered as recreational sites. In general, students live much closer to parks on average than schools (410 meters versus 1280 meters on average). However the average 111  distance does not adequately capture variation in these proximity measures. Table 11 presents an alternate illustration, indicating that although much further on average, schools are closer than either parks or possible non-park recreation sites for 10.9% of students, and may thus act as important recreation sites for these students in particular.  Figure 16: Typical Recreational Amenities on School Sites  Table 11: Proximity to Three Types of Recreation Sites: School Sites, Parks and Non-park Recreation Sites. Recreation Site  Percent of students for which recreation site is closest school sites 10.9% parks 58.4% non-park recreation sites 30.7% Finally, the social environment measures (14-16) indicate that students on average live in moderate income neighbourhoods, slightly below the regional median DA level income of 112  $61,000 (Statistics Canada 2008). These neighbourhoods also have moderate proportions of children, 11.41% compared to the regional average of 11.36%. Measure 14, the population density of children, is informative because it provides an indication of the absolute number of children of similar age to study participants living nearby. On average, study participants live in areas populated with approximately 4 children aged 5-14 per acre. 3.4.1.2 Variations in Built and Social Environment Characteristics Across Neighbourhoods Vancouver: Schools One and Two Children attending Vancouver schools generally live in dense, mixed use neighbourhoods with high street connectivity, and in close proximity to school and to multiple parks. In addition to having high levels of access to parks, these children live very close to nonpark recreational sites, 270 meters on average for students attending school one, and 170 meters for those attending school two. Despite living close to their schools, many of these students need to traverse several four way intersections en route to school because of the highly connected, largely grid street networks in their neighbourhoods. Students attending school two, for instance, have a mean commute distance of 860 meters, but need to cross an average of 5.9 four way intersections en route to school. This contrasts with students attending school eight in Mission, who have an average commute distance of 1790 meters but only need to cross 2.6 four way intersections on average. The highly connected street networks of these neighbourhoods are also reflected in the low density of cul-de-sacs, 2.1 and 1.8 per square kilometer for schools one and two, respectively. The high street connectivity also translates into a high surface coverage by roads, with 36.8 lane-km/km2 for school one and 46.5 lane-km/km2 for school two. Commercial development is on average a density of 0.6 FAR, but given the proximity of some students to the downtown core, the maximum FAR for these students is 3.8 (Figure 17). Similarly, while the average residential density of their neighbourhoods is moderately high at 23 113  units per acre, the maximum density is 92 units per acre. Students attending school two in particular live in very high density settings, with an average residential density of 39 units per  Figure 17: Typical Commercial Developments in Study Neighbourhoods  (a) typical commercial development with approximate FAR 0.4-0.7 (one storey, parking in front or rear)  (b) typical commercial of FAR >= 1.0 (one storey or more, no on-site parking or underground parking) acre. Figure 18 illustrates typical housing types in these neighbourhoods. Their neighbourhoods are on average lower income than the neighbourhoods of any other students in the sample. On average, the median neighbourhood income is $48,000 for these students. While Vancouver students’ neighbourhoods have very high densities of children (6.4 children per acre compared to 1.8 for children attending Mission schools at the other extreme), this reflects the overall 114  population density. In contrast, the proportion of children in relation to the total population is relatively low compared to other schools, at 11.3%.  North Vancouver: Schools Three and Four Despite their close spatial proximity to Vancouver, children enrolled at North Vancouver schools live in very different environments, characterized by low density, predominantly single use residential development (Figure 19 (a) and (b)). Residential density is approximately 7 units per acre on average, and commercial density is on average FAR 0.3. North Vancouver students live close to even greater numbers of recreational amenities than students in other municipalities, with an average of 4.3 parks within 800 meters of their homes. They also live relatively close to school, on average 930 meters for school three and 960 meters for school four, and need to cross relatively few four way intersections en route to school. In contrast to children attending schools in other municipalities, the median household income in their neighbourhoods is relatively high, at $74,000. Students attending school three in particular both live the closest to parks when compared to other students in the sample (the average distance to the closest park is 180 meters, and students have an average of 4.8 parks within 800 meters) and live in the highest income neighbourhoods ($78,000 on average). While living closer to parks, North Vancouver students live substantially further from non-park recreational sites than children in Vancouver (on average 470 meters and 620 meters for schools three and four, respectively).  115  Figure 18: Mixed Density Housing in Selected Study Neighbourhoods  (a) school two  (b) school six  (c) school seven  116  Figure 19: Low Density Housing in Selected Study Neighbourhoods  (a) school three  (b) school four  (c) school five  117  Burnaby: Schools Five, Six and Seven Students attending Burnaby schools six and seven live in moderately dense, mixed use environments with a variety of housing types and moderate street connectivity (Figure 18 (b) and (c)). These neighbourhoods are also characterized by a relatively high proportion of streets with posted speed limits of 30 km/h or less (11% of streets overall; 15% for students attending school six). Students attending school five live in lower density environments (Figure 19(c)). These students also live in relatively safe traffic environments as characterized by the number of four way intersections they need to cross en route to school. With a mean commute distance to school of 690 meters, students attending school five only need to cross 1.6 four way intersections on average. Children attending school five also live in neighbourhoods with large numbers of cul-de-sacs (11.3 per square kilometer, on average). The street network in the vicinity of school five is contrasted with the highly connected network in the vicinity of school two in Vancouver in Figure 20. Similarly to Vancouver students, Burnaby students live close to multiple parks. In comparison to Vancouver students, their neighbourhoods are characterized by higher median household incomes (on average $54,000). While students attending schools five and six live relatively close to school (880 m and 690 m on average, respectively), students attending school seven live substantially further from the school (1700 m on average). This is likely because French Immersion programs are offered at this school, and thus commute distances for some students are as long as 9.5 kilometers. Students attending school seven also live in highly mixed use environments, higher than those for students attending any other school in the sample on average (average land use mix index of 0.52). Commercial FAR of these neighbourhoods is on average 0.6, the same as for Vancouver students. However, the maximum FAR at 1.4 is relatively low compared to the maximum FAR for Vancouver students (3.8), reflecting the lower intensity of commercial development in the Lower Mainland outside of Vancouver’s commercial core. 118  Figure 20: Comparison of Street Connectivity. Street networks for areas surrounding schools two (right) and five (left), at same scale: each figure is a 1x1 kilometer square. Produced using DMTI Inc. street network data (2006).  Mission: Schools Eight and Nine Mission is located at the outskirts of the Lower Mainland far from the regional center of Vancouver (Figure 1). Students attending Mission schools live in very low residential density neighbourhoods (5 units per acre on average) and live relatively long distances from school (on average 1790 meters for school eight and 2320 meters for school nine). Similarly to school seven, some children attending school nine likely have particularly long commuting distances because of the French Immersion programs offered. However, Mission is not a typical new suburb, but is built around the historic town center of a municipality incorporated in 1892 (Hayes 2005). As a result, Mission neighbourhoods are characterized by moderate intersection density, although the neighbourhoods of students attending school nine are also characterized by a high density of cul-de-sacs (13.7 per square kilometer on average). Students attending Mission schools are also distinguished by the low number of parks within 800 meters of their homes, at 2.0 on average for students attending school eight and 1.6 on average for school nine. 119  In addition, children in these neighbourhoods have to travel greater distances to access non-park recreational sites, 780 meters on average for children attending school eight and 590 meters for those attending school nine. These neighbourhoods are characterized by moderate household incomes, with an average neighbourhood median household income of $55,000. Finally, children make up a relatively large proportion of the population in these neighbourhoods, at 13.4% on average, despite low population densities of children, at 1.8 children per residential acre. 3.4.2 Physical Activity Behaviour Table 12 presents summary statistics for physical activity variables by intensity. As indicated in this table, study participants engaged in sedentary activity for the large majority of their waking hours, on average 69% or 9.0 hours. The amount of time spent in sedentary activity is also relatively stable across students compared to other intensities of physical activity, with a coefficient of variation (CV) of 0.12 (CV =  s _  = standard deviation divided by mean). In  x  contrast, average daily minutes of MVPA vary substantially, with a CV of 0.30. Physical activity levels drop consistently with increasing intensity level, with dramatic differences between sedentary and light physical activity (LPA), and between moderate physical activity (MPA) and vigorous physical activity (VPA). Only 15% of MVPA consists of vigorous physical activity on average. All of the variables noted in Table 12 are normally distributed. Study participants on average engage in 124 minutes of MVPA per day. These high levels of MVPA are inconsistent with large scale studies of physical activity patterns in Canadian children (S 2.2.4), which suggest that children are considerably less active. However, this inconsistency is likely due to methodological differences rather than unique sample  120  characteristics. For example, the CANPLAY studies (CFLRI 2005, 2009) relied upon pedometry, while studies like the HBSC and CCHS used various survey instruments (Active Table 12: Descriptive Statistics for Physical Activity Variables, by Level of Intensity, for Students in Sample 1 (n=366). Variable  Sedentary Activity Light Physical Activity (LPA) Moderate Physical Activity (MPA) Vigorous Physical Activity (VPA) Moderate to Vigorous Physical Activity (MVPA)  mean (standard deviation) – average minutes per day across entire wear period 541(65) 120(21) 105(29) 19(12) 124(38)  Healthy Kids Canada 2007). In a recent evaluation of MVPA using accelerometry, Colley et al. (2011) also found substantially lower levels of MVPA than in the present study, with boys averaging 61 minutes of MVPA per day and girls averaging 47 minutes of MVPA per day. , The high levels of MVPA in the present study are likely at least partly attributed to the cut points chosen, which have been noted as producing relatively high estimates of MVPA (Guinhouya et al. 2006). To illustrate how cut point selection can influence results, average daily MVPA was recalculated for 10% of students randomly selected from Sample 1 (n=37), using cut-points developed by Puyau et al. (2002). This comparison was made because Puyau cut-points are frequently cited as relatively high cut-points for the Actigraph GT1M. In addition, the cutpoints used by Colley et al. (2011) were developed for a different accelerometer and thus cannot be directly applied to the data collected for the present study. As expected, marked differences in average daily MVPA were evident when contrasting the two sets of cut-points. Using the cut-points applied in the present study, average daily MVPA was estimated at 111 minutes per day. In contrast, students were found to obtain an average of 22 minutes per day of MVPA when using the Puyau cut-points. These results are consistent with other similar evaluations, 121  including that of Guinhouya et al. (2006), who found, in a sample of 45 children aged 8-11 that the Puyau cut-points produced an estimate of 28 minutes of MVPA per day in contrast to 141 minutes based on the cut-points used in this study. Further recognizing the uncertainty introduced by cut-point selection, Colley et al. (2011) acknowledge that their low estimates of MVPA in Canadian children may be due to the high cut-point used, that is considerably higher than the comparable adult cut-point and is based on a single study. As noted previously, the selection of cut points and choice of methodology for assessing physical activity more broadly continue to be the subject of substantial debate. Focusing further on MVPA as the primary dependent variable used in subsequent analyses, Table 13 presents some additional descriptive statistics. Study participants engage in an average of 119 minutes of MVPA on weekends, slightly less than the overall daily average. On school days, children engage in an average of 75 minutes of MVPA outside of school. On average, total minutes of MVPA outside of school constitute approximately 65% of total MVPA minutes accumulated. Girls obtain on average 18 minutes less MVPA than boys, a significant difference at p < .001. Similarly North American/European students obtain 15 minutes more MVPA on average than children of other ethnicities, also significant at p < .001. Average daily MVPA is also negatively correlated with age, Pearsons r = -.292 (p <.001). The observed differences in MVPA by gender and age are consistent with previous studies, with many studies indicating declining physical activity of children with age, and studies finding that girls engage in less physical activity than boys (S. 2.3.4.1.1). As discussed in S. 2.3.4.1.3, findings with regards to ethnicity are less consistent and depend on the ethnicity categorization used.  122  Table 13: Descriptive Statistics for MVPA Outside of School and Total MVPA Stratified by Gender and Ethnicity, for Students in Sample 1 (n=366). Variable  outside of school MVPA average daily MVPA outside of school, on school days average daily MVPA on weekends total – average daily MVPA boys girls North American/European all other Ethnicities  Mean(Standard Deviation) minutes per day  75(29)  119(47)  124(38) 134(39) 116(34) 133(39) 118(35)  Finally, it is evident that there are modest differences in MVPA engagement by school, as illustrated in Figure 21. These differences are quantified and their implications are further discussed in the following section.  123  Average Daily Minutes of MVPA  Figure 21: Average Daily MVPA by School. Horizontal bars represent (from bottom to top): minimum, first quartile (lower edge of box), median (inside box), third quartile (upper edge of box), and maximum.  School  124  3.5 Special Topic - Methods for Dealing with Clustering 3.5.1 Method Alternatives Considered Sample 1 consists of 366 students within nine schools. A standard Ordinary Least Squares (OLS) regression ignoring the grouping of students within schools would rely on a model of the form: Y = βo + β1x1 + β2x2 + … βnxn + ε  (1)  Where: Y is the dependent variable βo is the intercept β1… β n are regression coefficients corresponding to predictors x1 … xn, and ε is an error or residual term A critical assumption of this model is that observations on individual students are independent. Students in a given school are, however, likely to share certain characteristics in common that may influence physical activity patterns. Some of these characteristics are observable and incorporated explicitly into the analysis (e.g. the built and social environment characteristics identified in Table 7), but other unmeasured variables will effectively be captured in the residual term ε. In effect, the residual term will be correlated for students within a given school, violating the assumption of independence of observations (Mâsse et al. 2002). Application of such a model to clustered data would produce unbiased estimates of β1… β n, but would produce biased estimates of standard errors, and therefore lead to inflated incidence of Type I error (finding βs to be significant when they are not; Goldstein, 1995). Numerous alternative approaches are available to deal with clustered data. One option would be to aggregate all data to the school level. However, this would dramatically reduce sample size (in this case, from 366 to 9), producing substantially higher standard errors and dramatically limiting power for determining significant effects (Kreft and de Leeuw, 1998). 125  This approach would also effectively discard all within-school variation (Parks and Poston, 2006). In the present study, this would mean ignoring more than 95% of all variance in average daily MVPA, as illustrated in the following section. A second alternative would be to explicitly model schools as a series of dummy variables such that: Y = βo + β1x1 + β2x2 + β3x3 + β4x4 + β5x5 + β6x6 + β7x7 + β8x8 + β9x9 … βnxn + ε  (2)  where: Y is the dependent variable βo is the intercept β1… β8 are regression coefficients corresponding to school dummy variables x1 … x8, β9… βn are regression coefficients corresponding to predictors x9 … xn, and ε is an error or residual term While use of this approach would account for the clustering of students with schools, it would result in two major limitations: 1) because eight dummy variables would be required, any models based on this formulation would not be parsimonious and would limit power of statistical tests (Luke 2004). 2) this model treats the schools as unrelated, ignoring the fact that they may be drawn from a larger population of schools, effectively limiting statistical inference to the sample of nine schools (Diez Roux 2000, Snijders and Berkhof 2008). Given all the limitations of the aforementioned approaches, two additional alternatives were identified and determined to be viable: multilevel models (MLM) and generalized estimating equations (GEE). From these two methods, multilevel models were initially chosen as the best form of statistical model for two reasons. First, they can be used to explicitly model clustering of students within schools, whereas a generalized estimating approach treats clustering as a nuisance term (Diez Roux 2000). Second, multilevel models have been widely used in similar applications and many comparable precedents are available. 126  A multilevel model with students clustered within schools is premised upon individual schools being randomly drawn from a broader population of schools (Snijders 2005, Diez Roux 2002). In this approach, results may be generalized to the broader population of schools, in contrast to modeling schools as dummy variables (equation 2), which limits inference to the specific sample of schools used. In the dummy variable and OLS approaches (equations 1 and 2), all β coefficients are fixed for all subjects, regardless of the school they attend. These coefficients are thus referred to as fixed effects, and models of the type specified by equations 1 and 2 are referred to as fixed effects models. In a multilevel approach, some β coefficients, as specified by the researcher, are allowed to vary across schools, and are thus described as random effects (Hayes 2006). The type of multilevel model used in initial models in the present  study are referred to as random intercept models. This term refers to the fact that only the intercept is modeled as a random effect, and that this intercept thus captures variation in the outcome variable between schools: Y = βoj + β1x1 + β2x2 + … βnxn + ε  (3)  where all of the terms in this model are identical to equation (1), except there is an additional subscript on the intercept term: βoj, indicating that βo may vary between schools. This intercept term can be expressed as: βoj = βo + µj  (4)  where βo is the average intercept for all students, across schools, and µj is a term capturing variation from school (j) to school. 3.5.2 Preliminary Analysis and Final Method Selection A first step in using the multilevel modeling approach was to quantitatively assess the degree of clustering between schools. This was achieved by fitting an unconditional model. An  127  unconditional model is simply a model that does not include any predictors. Combining equations (3) and (4), this can be expressed as: Y = βo + µ j + ε  (5)  Assuming that ε and µj are normally distributed, the total variance of Y can be expressed as: Var(Y) = Var(µj) + Var(ε) = τ2 + σ2  (6)  And the proportion of variance accounted for between schools can be expressed as the intraclass correlation coefficient, or ICC (Mâsse et al. 2002): ICC =  =  τ2 τ2 + σ2  (7)  between schools variance in Y total variance in Y (between + within school variance)  Application of this formula to assess the clustering of average daily MVPA between schools produced an ICC of 0.022, meaning that approximately 2.2% of variance in MVPA can be attributed to the clustering of students in schools. A similar approach was also used to determine the degree of clustering of students in schools when including the control variables of age, gender and ethnicity. This produced an even lower ICC estimate of 1.6% . These results suggest first that the influence of clustering of students within schools on MVPA is very small. Second, the reduction of between school clustering with the introduction of control variables suggests that between school differences are due in part to compositional effects, whereby differences observed between schools are partly explained by differences in the characteristics of students attending each school (Diez Roux 2002). Because of the inclusion of random components, multilevel models cannot be estimated by OLS regression techniques, and thus require specialized algorithms (Fotheringham et al. 2000). There are numerous similar algorithms used to calculate model parameters, all generally 128  based on an iterative procedure starting with OLS estimates of the fixed parameters (Goldstein 1995, Diez Roux 2002). Under certain conditions, the iterative computational algorithms used to produce parameter estimates under MLM may not converge, thus failing to produce valid results (Dedrick et al. 2009). In the present study, as additional explanatory variables were incorporated in the models convergence problems were encountered. This problem can be attributed to two related issues. First, with a small number of grouping units (i.e. schools), power to estimate between-group variability is limited (Diez Roux 2002). In this case, nine schools is below recommended guidelines for precisely estimating between group variability (Maas and Hox 2004). Compounding this problem, between group variability is likely very low to begin with, as indicated by the low ICC estimates produced. Consistent with literature on MLM methodology, a number of diagnostic steps were implemented in attempts to achieve model convergence (Landau and Everitt 2004). These included: adjusting iteration parameters and convergence criteria, checking the scaling of variables, and using software with different computational algorithms (both PASW Statistics 18 and MLWin 2.02 were used). However, none of these attempts were successful, suggesting that the limited between-school variance and small number of schools were likely causing the convergence problems (Landau and Everitt 2004). Instead of disregarding the clustering of students within schools however, the decision was made to use GEE instead of multilevel models. While similarly accounting for clustering of students within schools, GEE models are estimated using different computational algorithms than MLM models and convergence problems were not encountered in the implementation of these models. It is important to note that although convergence problems were not encountered using this approach, GEE is based upon asymptotic normality (Diez Roux 2000), and that nine schools is below the ideal number of clusters. This implies that with nine schools, standard error estimates will be biased downwards, resulting in increased chance of Type I errors. The 129  small number of schools sampled thus presents a fundamental constraint on the analysis, regardless of the analytical method chosen. Given this constraint, GEE was used because it produced more conservative estimates than would be produced by simply ignoring clustering. In contrast to MLM, GEE approaches to analyzing clustered data do not explicitly model between cluster variation. Rather, with GEE, the fixed effects of model coefficients across all groups are estimated, while accounting for clustering (Rice and Jones 1997). To achieve this, GEE algorithms are based on the simultaneous, iterative estimation of an equation to model relationships between the outcome variable and specified covariates and an equation to account for correlation between outcomes based on within cluster similarity of residuals (Hanley et al. 2003). Implementation and interpretation of GEE based models are very similar in many respects to the implementation of OLS models, but some additional decisions need to be made regarding model specification for GEE models. First, the dependent variable is not directly modeled, but rather a link function, g(.) is modeled as a linear function of explanatory variables (Horton and Lipsitz 1999, Kleinbaum and Klein 2010). This approach allows for substantial flexibility because a variety of different types of dependent variables may be modeled. Thus a first step in analysis using GEE is to specify the link function. In the case of continuous normally distributed data, the appropriate link function is an identity link, such that g(a) = a. For all analyses using accelerometry variables as dependents, this was the link function used. Alternately, with a binary dependent, a logit link function may be used, as in the case of analyses with active transportation as a dependent variable, discussed below (S. 4.7). Secondly, a working correlation matrix needs to be specified. This matrix specifies how the responses within clusters are correlated (Kleinbaum and Klein 2010). For the present study, an exchangeable, or compound symmetric correlation matrix was specified. This type of matrix treats all individuals within a given cluster as having the same correlation, and is appropriate 130  where there is no logical ordering of individuals within a cluster (Horton and Lipsitz 1999). Use of this correlation matrix is conceptually equivalent to use of a random intercept model in an MLM context (Equations 3 and 4 above; Koper and Manseau 2009). A third parameter requiring specification in the implementation of GEE models is the choice of estimator for estimation of standard errors. Two options are available: a model based or naïve estimator, or a robust or empirical estimator (Ghisletta and Spini 2004). The latter was chosen because it is  more robust to misspecification of the working correlation matrix (ibid), although application of both options was found to produce virtually identical results in the present study.  3.6 Research Question 1 (Primary Research Question) 3.6.1 Research Question and Hypotheses Investigated This section summarizes results of models designed to investigate Research Question 1: Are objectively measured average daily minutes of MVPA significantly associated with objective measures of characteristics of built and social environments when controlling for child age, gender and ethnicity? To explore this question, three hypotheses are proposed: Hypothesis 1.1: Measures of built environment characteristics will be significantly associated with average daily MVPA in the direction shown: positive: intersection spacing en route to school, net commercial density, net residential density, land use mix, cul de sac density, proportion of streets with posted speed limits of 30 kilometers per hour or less, and number of parks in buffer negative: distance to school, total number of four way intersections en route to school, lane kilometers of roads per square kilometer, distance to closest park, and distance to closest non-park recreational area  131  Hypothesis 1.2: Measures of the social environment characteristics: population density of children, proportion of population made up by children and median household income will all be significantly positively associated with average daily MVPA. Hypothesis 1.3: The spatial unit of measurement of built and social environment characteristics will influence the results of Hypotheses 1.1 and 1.2. Specifically, it is hypothesized that measures based on smaller spatial units will be more strongly associated with average daily MVPA than measures based on larger spatial units. The rationale for hypotheses 1.1 and 1.2 are broadly outlined in Table 6. One measure is notably not identified in these hypotheses, intersection density. That is, because higher intersection density may both increase access to destinations, and decrease parental perceptions of safety, it was unclear whether intersection density would be positively or negatively associated with average daily MVPA. This ambiguity is also evident in the mixed results of previous studies using such measures (S. 2.3.5.3.2). Hypothesis 1.3 was explored by testing models using built and social environment measures based on a range of buffer sizes: 200, 400, 800 and 1600 meters, as outlined in S. 3.2.3.3. Smaller buffer sizes are anticipated to better explain variation in MVPA because children of the ages included in the sample (8-11) are anticipated to have limited independent mobility and therefore be influenced predominantly by environmental characteristics close to their home.  3.6.2 Analytical Framework and Methods Generalized Estimating Equations were used to test models of the type specified in Figure 22, while controlling for clustering of students within schools.  132  Figure 22: Conceptual Model for Research Question 1, Hypotheses 1.1-1.3. BUILT ENVIRONMENT CHARACTERISTICS SOCIAL ENVIRONMENT CHARACTERISTICS  AGE  average daily minutes of Moderate to Vigorous Physical Activity (MVPA)  GENDER ETHNICITY  A process of model building was used whereby the following models were tested sequentially: 1. Control only model with age, gender and ethnicity as the only explanatory variables. 2. Multiple models with a single built or social environment variable in addition to all control variables. One such model was created for each built and social environment variable, using all of the above noted buffer sizes. There are two related reasons for creating such models. First, because of the large number of built and social environment variables being tested, analysis of these models served as a screening step for creating final models. Second, because preliminary analyses revealed many built environment variables to be collinear, these variables could not be entered simultaneously. For brevity, these models will subsequently be referred to as single covariate models.  3. A final model incorporating multiple environment measures in addition to control variables. Because of multicollinearity amongst built environment variables, a compound built environment index was created. An important implication of the limited clustering of student’s average daily MVPA by schools as observed in S. 3.5.2 is that school neighbourhood variables are not significant predictors of MVPA. This is because such variables explain differences between schools, rather 133  than between individuals. And because between school variation is minimal, these variables contribute minimally to explaining the variance of MVPA. This finding was both confirmed in preliminary models and in the models discussed immediately above (#1-3). Because highly non significant associations between all school neighbourhood variables and physical activity outcome variables were found in these models, the results are not illustrated below, for brevity. 3.6.3 Results and Discussion 3.6.3.1 Control Variables Only Model (Model 1) Table 14 presents the results of the control only model. As illustrated in this table, all control variables are significantly associated with average daily MVPA. The results of this model are consistent with the descriptive analysis presented in S. 3.4.2, indicating that when controlling for the clustering of students within schools: •  One year increase in age is associated with an approximately 19 minute decline in average daily MVPA  •  Girls engage in approximately 17 minutes less MPVA per day than boys  •  North American/European children engage in more MVPA than East/Southeast Asian and South Asian children (the difference between North American/European children and those of mixed ethnic origin is not significant).  Table 14: Control Only Model (Model 1), Overall MVPA as Dependent Variable  age gender = female ethnicity – North American/European as reference category East/Southeast Asian South Asian Mixed  Parameter Estimate (B) -19.28*** -16.82***  -25.71*** -8.50** -9.88  Marginal R2 0.240 *significant at p <.05, **significant at p <.01, ***significant at p <.001 134  While the traditional OLS R2 model fit statistic is not available in GEE, it was possible to estimate an analogous marginal R2 for this model, as originally introduced by Zheng (2000), using methods outlined by Tan et al. (2009). Similarly to the OLS equivalent, this measure can be interpreted as the proportion of variance in the response variable explained by the fitted model (Ballinger 2004). Taken together then, the control variables account for 24.0% of variance in average daily MVPA. While these results are generally consistent with the broader literature on correlates of physical activity (S. 2.3.4.1), the results for age may be viewed as somewhat surprising. This is because while strong age effects have been noted in other studies, these effects might be expected to be unobservable over a truncated age range as used in the present study (Sallis et al. 2000). A possible explanation for the significant association observed is that the age range of participating children may overlap with a period of adolescence for many children, during which the decline in physical activity is most dramatic (Carver et al. 2005, Godin et al. 2005).  3.6.3.2 Single Covariate models Table 15 summarizes the results of the single covariate models. While all control variables were entered in each of these models, parameter estimates are not shown for control variables given space constraints. However, in every model tested, all control variables remained significant, with the same sign and similar magnitude as in the control only model described above. Focusing first on the significant results (bolded in the table), all but one of the significant associations were in the direction expected: distance to school (-), total number of four way intersections en route to school (-), distance to non park recreation sites (-), commercial density (+) and residential density (+). Notably, intersection density was positively associated with average daily MVPA. This finding stands in contrast to some previous findings  135  Table 15: Summary of Model Results for Single Covariate Models, Overall MVPA as Dependent Variable Variable Route Between Home and School 1. distance to school+ 2. average intersection spacing++ 3. total number of four way intersections++  Home Neighbourhood 4. net commercial density 5. net residential density+ 6. land use mix 7. intersection density 8. cul-de-sac density++ 9. proportion of streets with posted speed limits of 30 kilometers per hour or less++ 10. lane-kilometers of roads 11. number of parks in buffer 12. distance to closest park+ 13. distance to closest possible non park recreation site+ 14. population density of children aged 5-14++ 15. proportion of population made up by children aged 5-14 16. median household income  Parameter Estimate -5.42*** -0.04 -3.80* Parameter Estimates by Buffer Size 200 meter 400 meter 2.53 -5.99 -0.98 0.43 4.29 -4.65 0.03 0.17 0.16 0.88 0.07 0.19  800 meter 4.74 1.21 -0.63 0.25* -0.48 -0.30  1600 meter 21.59*** 6.67* 18.63 0.52** 2.30 2.59  0.08 0.30 -1.12 -17.53**  0.15 -1.06 -  0.59* 1.99 -  -4.07 -0.52  -2.62 -0.73  -1.73 -0.10  11.14 -1.13  8.13E-5  -2.40E-4  8.89E-5  1.06E-4  0.98* 0.79 -  *significant at p <.05, **significant at p <.01, ***significant at p <.001; all significant results also bolded + logarithmic transformation, ++ square root transformation  136  indicating that street connectivity is negatively or not associated with physical activity measures (Braza et al. 2004, Copperman and Bhat 2007, Timperio et al. 2006). This may be interpreted as indicating that increased access to destinations afforded by greater street connectivity may be a more influential factor than any decreased perceptions of safety due to the higher numbers of intersections. The sole finding contrary to anticipated directions of association was for lanekilometers of roads. It was hypothesized that this measure might be negatively associated with average daily MVPA because higher values reflect a greater extent of paved surfaces in a neighbourhood. However, it is likely that this measure is measuring the same characteristics as the intersection density measure. At a 1600 m buffer size, these measures are highly correlated (r = 0.854). As a result, the lane-kilometers of roads measure was excluded from subsequent analyses. Two home neighbourhood variables were significantly associated with MVPA in the directions expected based on a 1600 meter buffer size, but also had non-significant associations in the opposite direction of what was expected with smaller buffer sizes. Notably, these were both density measures: net commercial density and net residential density. The negative associations in both cases were highly non-significant (p = 0.207 for net commercial density at 400 meters and p=0.665 for net residential density at 200 meters). Nonetheless, these results may reflect parent preferences for moderate neighbourhood densities over large areas, but not immediately proximate to homes. This issue is considered further in the discussion below, regarding correlates of weekend MVPA (S. 3.7.3.2). While many of the measures tested were significantly associated with MVPA, many others were not. These include measures of average intersection spacing, land use mix, cul-desac density, proportion of streets with posted speed limits of 30 kilometers per hour or less, and all of the social environment measures tested. Most notably, however, both measures of access 137  to parks were non-significantly associated with MVPA. This result is particularly surprising because access to parks and other recreational resources have generally been found to be important correlates of MVPA (S. 2.3.5.3.2). The non-significant finding with regards to measure 12 (distance to closest park) could be interpreted as explained by research suggesting that children don’t necessarily use the closest park, but rather travel further to use more desirable parks (Timperio et al. 2008). However this doesn’t explain the non-significant association for measure 11 (number of parks in buffer). One additional explanation for these anomalous findings is that in the study region, the number of parks are sufficient such that children’s level of access to parks is effectively saturated. Evidence of this is presented in Table 16. Table 16: Proximity of Sample 1 Students to Parks Distance to Closest Park (meters) 200 400 800 1600  Percentage of Sample (cumulative) 38.5 68.8 91.6 97.9  As illustrated in this table, almost 40% of children live within 200 meters of a park. This number rises steadily such that more than 90% of children have at least one park within 800 meters of their house. To further assess this explanation, the level of access to parks in the Lower Mainland was quantified in terms of the metric of area per 1,000 population, using DMTI parks data (DMTI Spatial Inc. 2006). This metric was chosen for ease of comparison to the widely used standard of 10 acres per 1,000 population, originally established by the US National Recreation and Parks Association (Williams and Dyke 1997). At 42.9 acres per 1,000 population, the estimate produced for the Lower Mainland greatly exceeds this guideline and is substantially higher than comparable values for several comparable US cities (e.g. 9.1 acres per 1,000 residents in Seattle and 6.7 acres per 1,000 residents in San Francisco; TPL 2010). 138  Another possible reason for the unexpected finding with regards to access to parks measures may be that other recreation sites may play an important role. Notably, both distance to school and distance to the closest possible non-park recreation sites are both significant predictors of average daily MVPA. As illustrated in Table 11, taken together, these sites are closer than parks for almost 42% of children in the study. The significant association of proximity to non-park recreation sites with MVPA may also in part be a result of some of the land uses incorporated in this measure being indoors (bowling alleys, community halls) since physical activity data were collected during the fall/winter. This explanation is also consistent with the lack of significant associations noted with the park measures. School and non-park recreation sites may more generally be important because of specific recreational amenities on site that are unavailable at parks. School sites in particular, might be especially desirable as playgrounds for children because of the variety of age specific amenities present on site that might not be available in more generalized parks (Figure 16). With regards to other measures having non-significant associations with MVPA, a number of possible explanations may apply. In general, given the relatively small sample size used, it is possible that power to detect effects was limited. It is also possible that some of the measures tested simply are not correlates of MVPA. More specific explanations may be applicable to certain measures however. For example, the non-significant association for median household income may be a result of the relatively truncated range for the sample. As noted in S. 3.3.2, high income, high walkability neighbourhoods are underrepresented in the sample used. In other cases, the measures chosen may not adequately gauge the relevant characteristics. This may apply to the measures of population density of children and proportion of population made up by children (Measures 14 and 15). These measures were selected because higher numbers of children in a given neighbourhood may foster social interaction and result in greater opportunities for physical activity. However, other measures may better assess opportunities for 139  social interaction in this regard. While objective measures may be desirable for gauging built environment characteristics, measures based on surveys or observational methods may be more appropriate for gauging the neighbourhood social environment (as in Molnar et al. 2004). Alternately, objective measures may be appropriate, but with a narrower age range than that specified (children aged 5-14) because children may play primarily with other children who are very similar in age. Development of such a measure was not possible given available data because of the broad age categories used in the Census. The age range employed was the smallest age range possible encompassing the age range of participating students. The most unexpected findings based on the above models relate to Hypothesis 1.3. In general, the findings highlighted in Table 15 illustrate the importance of buffer size, suggesting that the implications of alternate neighbourhood definitions need to be further studied. More specifically, it was anticipated that measures based on smaller buffer sizes yield the most significant results, because children aged 8 to 11 would likely have limited independent mobility, and thus be most influenced by the environments immediately surrounding their home. In contrast, the results presented above indicate that for all buffer-based measures, the strongest associations were found for 1600 meter buffer sizes. No measures were significant at either 200 or 400 meter buffer sizes, only two measures were significant at 800 meters, and 4 measures were significant at 1600 meters. One possible explanation investigated was that larger buffer sizes are more sensitive to differences between neighbourhoods by capturing variation not present at the smaller buffer sizes. However, this is contrary to what was found in a subsequent examination of measures at different buffer sizes. Focusing on measures 4, 5 and 7 as significant measures at 1600 meters, coefficients of variation (the standard deviation of a measure divided by its mean; a measure of dispersion with higher values corresponding to greater dispersion) were calculated by buffer size and are illustrated in Table 17. 140  Table 17: Variation in Selected Built Environment Measures by Buffer Size Measure  4. net commercial density 5. net residential density 7. intersection density  Coefficient of Variation by Buffer Size 200 meter  400 meter  800 meter  1600 meter  1.41 1.18 0.57  0.98 1.09 0.38  0.74 0.85 0.34  0.49 0.69 0.28  As illustrated in this table, rather than capturing additional variation, larger buffer sizes have the effect of smoothing or averaging out variation that would be captured using smaller buffer sizes. Conversely, smaller buffer sizes may be viewed as exaggerating local characteristics. This effect can be further illustrated using some more specific examples. For instance, at a buffer size of 200 meters, the maximum net residential density detected is 157 units per acre. At 1600 meters, this drops to 93 units per acre. Similarly, the maximum intersection density at 200 meters is 433 intersections per square kilometer versus 84 intersections per square kilometer at 1600 meters. While the maximum values noted at 200 meter buffer sizes may appear dramatic, this is simply because of the small spatial unit they are measured over. For example, the net residential density of 157 units per acre corresponds to a single 30 plus storey tower development, and an intersection density of 433 intersections per square kilometer corresponds to 50 meter intersection spacing within the buffer, due to a diagonal road cutting across a tight square grid (as in Figure 20). Although such extreme values are thus representative of real features, the sensitivity of small buffer measures to highly localized features may reduce their explanatory power. That is, because local patches of high (or low) density may not be representative of their broader context, a measure based on a small buffer size may capture too much variation in built environment characteristics. In the case of density measures, for example, a high density neighbourhood as characterized by a 200 meter buffer size might consist of a single apartment complex. But the area immediately surrounding 141  the 200 meter buffer might be entirely low density residential development (such mixed housing forms are in fact common in Burnaby, the location of three of the study schools). In contrast, a high density neighbourhood as characterized by a 1600 meter buffer implies high density development over a large area. Insofar as density is hypothesized to influence physical activity patterns by shaping access to destinations and thus influencing mode choice, a sufficient area of high density is likely required to create the ‘critical mass’ of destinations necessary to influence behaviour. Thus, larger buffer sizes may be more consistent with such hypothesized mechanisms of influence. This line of reasoning is also consistent with another explanation for the importance of larger buffer sizes. It is possible that children have an extended range of independent mobility, and thus environmental characteristics on the scale of 800-1600 meters directly influence their behaviour. While there is little quantitative evidence to refute this, literature on parental perceptions and the role of parents as gatekeepers suggest that this is likely not the case (S. 2.3.4.2.1), particularly with younger children. Thus it is perhaps more likely that larger buffer sizes are important because they capture environments that influence the behaviour of children’s parents in ways that influence child physical activity. Higher densities and street connectivity  may result in an increased likelihood that parents will choose active modes of transportation rather than chauffeuring their children to particular destinations. Alternately, higher densities and street connectivity may shape parental perceptions in a way that positively influences child physical activity patterns. Higher densities may in this manner contribute towards perceptions of increased eyes on the street. However, this type of mechanism may be less likely than the former one because of the broad range of other factors that contribute towards perceptions of safety (Loukaitou-Sideris 2006).  142  3.6.3.3 Final Model Predicting Overall MVPA (Model 2) Instead of entering all significant variables identified above simultaneously when testing a final model, significant built environment measures were first assessed for multicollinearity. This is because as with regular OLS regression, Generalized Estimating Equations are premised upon an absence of high multicollinearity. All built environment measures were found to be moderately collinear, with a maximum Variance Inflation Factor (VIF) of 4.1, and an average of 2.9. As a result, a single built environment index was created following an approach developed by Frank and colleagues (2007b; see also Appendix 4), to create a walkability index. In this approach, individual measures are normalized as z-scores: _  x −x z-score = i s  where xi = the value of the measure for student i, _  x = the mean of the measure, and s = the standard deviation of the measure  The built environment index is then calculated as the sum of z-scores, with signs reflecting the directions of association determined via the single covariate models (Table 15 above). The decision was made to keep the z-scores for individual measures unweighted, as is appropriate in the absence of further information (DiStefano et al. 2009). The final built environment index included the following combination of four measures which best explained variation in MVPA: distance to school, net commercial density (1600 meters), intersection density (1600 meters), and distance to closest non-park recreation site. Combination of these measures into a single index is conceptually justifiable given that all four measures are measures of access (Figure 6) and are hypothesized to influence physical activity patterns through similar mechanisms (Table 6).  143  The results of a model incorporating all control variables and the built environment index are presented in Table 18. As illustrated in this table, in the final model, the built environment index accounts for 5.0% of variance in MVPA beyond the 24.0% in variance explained by the control variables. The sign of the GEE parameter estimate for the built environment index is as hypothesized, implying that shorter distances to school and non-park recreation sites, higher commercial density and higher intersection density are associated with more MVPA. Because the individual component measures in the built environment index are collinear, it is not possible to isolate their unique influence directly on MVPA. It is, however,  Table 18: Final Model Predicting Overall MVPA (Model 2) Marginal Parameter Estimate (B) – R2change Control Only Model (Model 1) Control Variables age gender=female ethnicity – North American/European as reference category East/Southeast Asian South Asian Mixed  Parameter Estimate (B) – Final Model (Model 2)  0.240 -19.28*** -16.82***  -19.94*** -17.01***  -25.71*** -8.50** -9.88  -37.96*** -20.15*** -10.98  Built Environment Index 0.050 *significant at p <.05, **significant at p <.01, ***significant at p <.001  3.49***  possible to isolate the unique contribution of each component towards explaining variance in MVPA, in relation to the overall explanatory power of the built environment index. These estimates can be derived by removing a single component from the index and re-estimating the model. Based on this approach, the unique contributions of each component, expressed as percent of variance in MVPA are estimated as: 0.79% for distance to school, 0.53% for  144  commercial density, 0.51% for shortest distance to non-park recreation sites, and 0.33% for intersection density. Finally, the importance of the built environment index in explaining MVPA can be illustrated by examining a range of predicted values using the final model. Based on this model, a child living in a neighbourhood characterized by the third quartile of the built environment index would obtain a modest 10 additional minutes of MVPA per day compared to a child of the same age, gender and ethnicity living in a neighbourhood characterized by the first quartile of the index. In the case of an 10 year old East/Southeast Asian girl, this would translate into 97 minutes versus 87 minutes. Considering a more extreme difference, a child living in a neighbourhood characterized by the ninth decile of the index would obtain approximately 23 more minutes of MVPA per day than a child living in an area characterized by the first decile.  3.7 Research Question 2 3.7.1 Research Question and Hypothesis Investigated This section summarizes results of models designed to investigate Research Question 2: Are objective measures of average daily MVPA outside of school significantly associated with objective measures of characteristics of built and social environments when controlling for child age, gender and ethnicity? To explore this question, two hypotheses are proposed: Hypothesis 2.1: Measures of characteristics relating to the school commute (Measures 1-3 in Table 7) will be significantly associated with average daily minutes of MVPA outside of school, on school days.  Hypothesis 2.2: Average daily minutes of MVPA on weekends will be significantly associated with measures of characteristics relating to the neighbourhood surrounding  145  the homes of children (Measures 4-16 in Table 7). For example, it is anticipated that distance to the closest park will be negatively associated with MVPA on weekends.  3.7.2 Analytical Framework and Methods Figure 23 illustrates the conceptual model for testing Hypotheses 2.1 and 2.2. This is identical to the model for testing Hypotheses 1.1 to 1.3 (Figure 22), with the exception of the dependent variable. In this case, two dependent variables were tested in separate models: average daily minutes of MVPA outside of school on school days (Hypothesis 2.1), and average daily minutes of MVPA on weekends (Hypothesis 2.2).  Figure 23: Conceptual Model for Research Question 2, Hypotheses 2.1, 2.2 BUILT ENVIRONMENT CHARACTERISTICS SOCIAL ENVIRONMENT CHARACTERISTICS  AGE  measures of physical activity outside of school (one model for each measure)  GENDER ETHNICITY  A similar process of model building was also applied, starting with a control only model and followed by single covariate models.  146  3.7.3 Results and Discussion 3.7.3.1 Control Variables Only Model (Models 3a and 3b) Table 19 presents the results of the control only model. Comparing the results of Models 3a and 3b to those of Model 1 (Table 14), reveals that ethnicity influences measures of MVPA outside of school similarly to its influence on overall average daily MVPA. Parameter estimates for age and gender influences on average daily MVPA on weekends are also similar to those estimates for overall average daily MVPA. However, parameter estimates for age and gender influences on MVPA outside of school on school days are substantially smaller in magnitude than their counterparts for overall average daily MVPA. Further, gender is insignificant in Model 3a. Nonetheless, given its theoretical importance as a correlate of physical activity in general, gender will be retained in all subsequent models.  Table 19: Control Only Models. Model 3a for Average Daily MVPA Outside of School on School Days, and Model 3b for Average Daily MVPA on Weekends  age gender = female ethnicity – North American/European as reference category East/Southeast Asian South Asian Mixed  Parameter Estimate (B) average daily MVPA average daily MVPA on outside of school on weekends (Model 3b) school days (Model 3a) -6.24* -21.93*** -6.56 -17.85***  Marginal R2 0.208  -19.37*** -7.90 -9.32*  -34.18*** -9.14* -9.54 0.208  *significant at p <.05, **significant at p <.01, ***significant at p <.001  147  3.7.3.2 Single Covariate models Tables 20 and 21 summarize the results of the single covariate models. As previously, parameter estimates for control variables are not shown for these models given space constraints. As illustrated in Table 20, average daily MVPA outside of school on school days has only one significant correlate in the single covariate models: distance to school. The direction of association indicates that, consistent with Model 2, longer distances to school are associated with less average daily MVPA, outside of school. This is also consistent with Hypothesis 2.1 in that characteristics relating to the school commute were anticipated to be predictors of average daily MVPA outside of school on school days. However, neither of the other two route characteristics examined were significant predictors. The results for the single covariate models for average daily weekend MVPA are more difficult to interpret (Table 21). Consistent with Hypothesis 2.2, multiple characteristics relating to the neighbourhood surrounding children’s homes were significantly associated with weekend MVPA. Notably, distance to closest non-park recreation sites was negatively associated with weekend MVPA as anticipated. Inconsistent with Hypothesis 2.2, two measures of school route characteristics were also found to significantly predict average daily MVPA on weekends: average intersection spacing and total number of four way intersections. While significant relationships between these variables and weekend MVPA are unanticipated because they relate specifically to the route between a child’s home and school, it is possible that they are proxies for broader neighbourhood characteristics and therefore influence active transportation patterns outside of school. The signs of association for both of these are as expected, with greater spacing between intersections and fewer four way intersections en route to school being negatively correlated with weekend MVPA.  148  Table 20: Summary of Model Results for Single Covariate Models Predicting Average Daily MVPA Outside of School on School Days Variable Route Between Home and School 1. distance to school+ 2. average intersection spacing++ 3. total number of four way intersections++  Home Neighbourhood 4. net commercial density 5. net residential density+ 6. land use mix 7. intersection density 8. cul-de-sac density++ 9. proportion of streets with posted speed limits of 30 kilometers per hour or less++ 11. number of parks in buffer 12. distance to closest park+ 13. distance to closest possible non park recreation site+ 14. population density of children aged 5-14++ 15. proportion of population made up by children aged 5-14 16. median household income  Parameter Estimate -3.25** -0.10 -2.64 Parameter Estimates by Buffer Size 200 meter 400 meter 3.18 -2.93 -0.95 -1.83 6.72 3.00 0.02 0.05 0.47 1.43 0.07 0.52  800 meter -2.52 -2.36 -0.36 -0.25 0.28 -0.77  0.56 0.37 -8.42  -2.54 -  -0.16  -0.17 -0.36  -3.19 0.54  -4.71 0.74  2.27E-5  4.01E-5  1600 meter 6.86 0.66 8.51 0.19 2.95 -0.52 -0.38  -  -2.90E-5  -  6.03 1.32 -2.53E-4  *significant at p <.05, **significant at p <.01, ***significant at p <.001; all significant results also bolded + logarithmic transformation, ++ square root transformation  149  Table 21: Summary of Model Results for Single Covariate Models Predicting Average Daily MVPA on Weekends Variable Route Between Home and School 1. distance to school+ 2. average intersection spacing++ 3. total number of four way intersections++  Home Neighbourhood 4. net commercial density 5. net residential density+ 6. land use mix 7. intersection density 8. cul-de-sac density++ 9. proportion of streets with posted speed limits of 30 kilometers per hour or less++ 11. number of parks in buffer 12. distance to closest park+ 13. distance to closest possible non park recreation site+ 14. population density of children aged 5-14++ 15. proportion of population made up by children aged 5-14 16. median household income  Parameter Estimate -6.01 2.10* -6.09*  Parameter Estimates by Buffer Size 200 meter 400 meter -8.49 -15.46** 5.87 -7.00* -12.87 -13.73 0.02 -0.01 1.65 2.18 -0.39 -0.47 -1.07 1.03 -11.22*  -11.92*** -0.38 3.01E-4  800 meter -4.02 -5.05 -9.89 0.13 1.90 -1.27  -0.44  0.97 -  0.43 -  -12.14* 0.26  -2.43 0.82  3.47E-4  2.95E-4  -  -12.04* -0.39 3.11E-4  1600 meter 2.24 -0.31 1.18 0.31 5.73 1.68  *significant at p <.05, **significant at p <.01, ***significant at p <.001; all significant results also bolded + logarithmic transformation, ++ square root transformation  150  Although three other neighbourhood characteristics measured at smaller buffer sizes were significantly associated with weekend MVPA, the directions of association are the opposite of what was anticipated: commercial density, residential density and population density of children are all negatively associated with weekend MVPA. These results are superficially at odds with the results for overall average daily MVPA, but a closer examination illustrates some consistency in findings. Specifically, the negative associations noted for weekend MVPA were also found for average daily MVPA, but were lesser in magnitude and not significant (Table 15). These findings suggest that commercial and residential density in close proximity to the homes of children may somehow inhibit physical activity, possibly by influencing parental perceptions of safety. High densities close to a child’s home may, for example, be associated with higher traffic levels. These findings also may reflect differences in where activity occurs at different times of the week. That is, the relative importance of smaller buffer sizes in explaining weekend MVPA may simply be a result of children engaging in more activities closer to home on the weekend than during the week. The finding with regards to population density of children is more difficult to interpret, particularly since it is negatively associated with MVPA across multiple buffer sizes (from 200800 meters). However, as noted above, this may be a result of measure definition, with an age range of 5-14 being too expansive. The presence of older children in particular may be inhibitive of physical activity due to fears of bullying. Ultimately, these results need to be interpreted with caution because the dependent variables used were based on between one and two days of measurement and thus do not reliably measure habitual physical activity (S. 3.2.1.2). Because of the uncertainty surrounding these findings, further model development based on the results of the single covariate models was not pursued.  151  3.8 Research Question 3 3.8.1 Research Question and Hypothesis Investigated This section summarizes results of models designed to investigate Research Question 3: Are relationships between objectively measured average daily minutes of MVPA, and objective measures of characteristics of built and social environments moderated by child age and gender? To explore this question, one hypothesis is proposed: Hypothesis 3.1: Both age and gender will moderate the relationship between measures of built and social environment characteristics and average daily MVPA, with younger children and girls having stronger associations than older children and boys. This hypothesis was made because younger children and girls are often viewed as more vulnerable populations, and parental controls may be more stringent for these populations (Carver et al. 2005, Korpela et al. 2002, Kytta 2002, Sener and Bhat 2007). As a result, environmental supports for physical activity may be more important for these groups.  3.8.2 Analytical Framework and Methods Figure 24 illustrates the conceptual model for testing Hypothesis 3.1. This represents an extension of the model for testing Hypotheses 1.1 to 1.3 (Figure 22), with the addition of age and gender moderating effects.  152  Figure 24: Conceptual Model for Research Question 3, Hypothesis 3.1 GENDER  AGE  BUILT ENVIRONMENT INDEX  average daily minutes of Moderate to Vigorous Physical Activity (MVPA)  AGE GENDER ETHNICITY  In this illustration:  solid arrows represent main effects  and dashed arrows represent moderating effects  A two step approach was used to explore moderating effects: 1. Starting with Model 2 which predicted overall MVPA based on age, gender, ethnicity and a built environment index as explanatory variables, three interaction terms were added as illustrated in Figure 24: (gender x built environment index), (age x built environment index), and (age x gender). 2. Based on the results from step 1, the sample was stratified by gender, and separate models exploring built and social environment correlates of MVPA were tested for each gender following the process outlined in S. 3.6.2. As previously, Generalized Estimating Equations were used to test all models.  3.8.3 Results and Discussion Table 22 outlines the results of the model with interaction terms (Model 4). Taken on their own, the parameter estimates for age, gender and the built environment index are not of interest because with the interaction terms in the model, they are conditional upon values of the  153  Table 22: Model 4 Predicting Overall MVPA with Interaction Terms for Age and Gender Parameter Estimate (B) – Final Model with Main Effects Only (Model 2) Control Variables age gender=female ethnicity – North American/European as reference category East/Southeast Asian South Asian Mixed Built Environment Index  Parameter Estimate (B) – Final Model with Main Effects and Interaction Terms (Model 4)  -19.94*** -17.01***  -25.45*** -16.77***  -37.96*** -20.15*** -10.98  -38.05*** -19.97*** -12.33*  3.49***  4.85***  Interaction Terms (female coded as 1, male as 0 for gender) gender x built environment index age x built environment index age x gender  -2.55** 1.53 10.45*  *significant at p <.05, **significant at p <.01, ***significant at p <.001  variables they are moderated by (Jaccard and Turrisi 2003). Thus, for example, the coefficient for age of -25.45 is conditional upon both corresponding moderating terms being set to 0 (gender = 0, the coding corresponding to male subjects, and built environment index = 0). In this model, the interaction terms are of more immediate interest as these provide insight into potential moderating effects. First, these results indicate that there is not a significant interaction between age and the built environment index, suggesting that built environment influences on MVPA are similar across ages. This result may be indicative of the truncated age range involved in the study. Although age has been found to be an important correlate of physical activity within this truncated range (S. 3.6.3.1), its influence may be due to factors unrelated to the built 154  environment. It is possible for instance that parental restrictions on independent mobility do not vary substantially with age, between the ages of 8 to 11. In contrast to the findings for the interaction between age and the built environment index, both interaction terms involving gender are significant. These findings indicate that first, the influence of gender on MVPA (whereby males engage in more MVPA than females), decreases with increasing age. This effect may be expressed as: Influence of being female on average daily minutes of MVPA = -16.77 minutes + (10.45 minutes)*age Where age was entered in the model as mean-centered, with the mean age of children as 9.94 years. This represents a substantial interaction effect whereby the influence of being a female on MVPA changes from -22 minutes for a younger child (defined as the first quartile of age) to -11 minutes for an older child (defined by the third quartile of age). The second significant interaction is that between the built environment index and gender. The negative sign of this interaction term indicates that built environment influences on MVPA are greater for males than females, the opposite of what was anticipated (Hypothesis 3.1). This interaction can also be expressed as a formula: Built Environment influence on average daily minutes of MVPA = [4.85 minutes - (2.55 minutes)*gender] per unit of Built Environment Index Where gender was coded as 1 for female and 0 for male. Thus, the influence of built environment characteristics as represented by the built environment index is 2.30 minutes for female students, and 4.85 minutes for male students, per unit of the built environment index. This translates into an influence on boys of approximately 2.1 times that for girls, as illustrated in Table 23.  155  Table 23: Predicted Changes in Overall MVPA Associated with Changes in Built Environment Index Values, Based on Model 4 Difference between Built Environment Index Values lowest and highest quartiles lowest and highest deciles  Change in Minutes of MVPA corresponding to difference between Built Environment Index Values Male Female 14 7 32 15  The reasons for this finding are not clear, because youth at high risk for physical inactivity, including girls, may rely more on external supports for physical activity (Davison and Schmalz 2006), and thus it was hypothesized that girls would respond more to differences in built environment characteristics. One possibility is that destinations of interest to girls may be different than those for boys, with girls favoring indoor and commercial spaces more than boys (S. 2.3.4.1.1). The built environment index, however, includes both measures of access to outdoor spaces (e.g. school sites and some non-park recreation sites), and commercial spaces (as gauged by commercial density), so the substantial differences in effect noted do not appear to be consistent with this explanation. To further explore gender differences in built and social environment influences on average daily MVPA, the sample was stratified by gender, and separate models developed for each gender. This approach was taken to investigate whether environmental correlates of MVPA vary by gender. The results of single covariate models are highlighted for male and female children in Tables 24 and 25, respectively. A superficial examination of these tables indicates very clearly that correlates of MVPA vary by gender. In general, a wider range of built and social environment characteristics are significant predictors of MVPA for boys than for girls. Some characteristics appear to influence MVPA levels of boys and girls in the same direction, but the magnitude of their influence is greater for boys (e.g. the parameter estimates for commercial and residential density at 1600  156  Table 24: Summary of Model Results for Single Covariate Gender Stratified Models Predicting Overall MVPA for Male Children (n=174). Variable Route Between Home and School 1. distance to school+ 2. average intersection spacing++ 3. total number of four way intersections++  Home Neighbourhood 4. net commercial density 5. net residential density+ 6. land use mix 7. intersection density 8. cul-de-sac density++ 9. proportion of streets with posted speed limits of 30 kilometers per hour or less++ 11. number of parks in buffer 12. distance to closest park+ 13. distance to closest possible non park recreation site+ 14. population density of children aged 5-14++ 15. proportion of population made up by children aged 5-14 16. median household income  Parameter Estimate -10.56*** -0.60 -9.40** Parameter Estimates by Buffer Size 200 meter 400 meter 11.78 0.39 1.61 2.31 17.18 14.30 -0.02 0.23 -1.88 -1.57 0.20 1.92 1.08 -1.83 -28.20***  800 meter 1.29 3.52 12.05 0.18 -6.14* 1.14  1600 meter 35.97** 10.90* 53.34** 0.58* -6.89 -2.25  1.18 -  2.50  0.28  -2.67 -0.73  -0.38 -1.04  2.45 -0.72  17.63 -3.54*  -1.17E-4  -1.39E-4  -1.87E-4  -8.02E-4*  -  -  *significant at p <.05, **significant at p <.01, ***significant at p <.001; all significant results also bolded + logarithmic transformation, ++ square root transformation  157  Table 25: Summary of Model Results for Single Covariate Gender Stratified Models Predicting Overall MVPA for Female Children (n=192) Variable Route Between Home and School 1. distance to school+ 2. average intersection spacing++ 3. total number of four way intersections++  Home Neighbourhood 4. net commercial density 5. net residential density+ 6. land use mix 7. intersection density 8. cul-de-sac density++ 9. proportion of streets with posted speed limits of 30 kilometers per hour or less++ 11. number of parks in buffer 12. distance to closest park+ 13. distance to closest possible non park recreation site+ 14. population density of children aged 5-14++ 15. proportion of population made up by children aged 5-14 16. median household income  Parameter Estimate -1.84 0.43 -1.81 Parameter Estimates by Buffer Size 200 meter 400 meter -3.56 -11.60 -3.89 -2.53 -6.60 -18.94 0.11 0.07 2.44* 3.55** -0.09 -1.13  800 meter 5.40 1.33 -14.74 0.20 4.72* -0.47  1.08 -0.68 -1.23  1.18  2.50  -5.48 -0.21  -4.43 -0.34  -6.64 1.22  2.63E-4  3.59E-4*  4.63E-4*  -  1600 meter 10.10 2.06 -15.27 0.41 9.99*** 8.91* 0.28  -  -  5.76 0.85 2.63E-4  *significant at p <.05, **significant at p <.01, ***significant at p <.001; all significant results also bolded + logarithmic transformation, ++ square root transformation  158  meters have the same signs for both boys and girls but are only significant for boys). However in other instances, the direction of association is different. Notably, cul-de-sac density, measured at 800 meters is significantly negatively associated with MVPA for boys, but is significantly positively associated with MVPA for girls. Comparing these models to their counterparts for the unstratified sample (Table 15), indicates that the correlates of MVPA identified for boys largely parallel those for the entire sample, but the magnitude of influence is more pronounced for boys. Thus, as in the unstratified models, larger buffer sizes appear to better describe neighbourhood environment characteristics for boys, with the large majority of significant effects at 1600 meter buffer sizes. In contrast, none of the characteristics with significant effects identified for girls were significant in the unstratified models, and no clear generalizations can be made with regards to buffer sizes that best explain neighbourhood environment characteristics for girls. As with the unstratified model, measures of access to parks are not significant for either boys or girls. Focusing on the female models, two built environment characteristics and one social environment characteristic are significantly associated with MVPA. Cul-de-sac density stands out because it is significantly positively associated with MVPA at all buffer sizes examined. The proportion of streets with posted speed limits of 30km/h or less is also significantly positively associated with MVPA, but only at a 1600 meter buffer size. Both of these characteristics are related in the sense of their hypothesized mechanism of influence on MVPA. That is, both are associated with perceptions of safety, suggesting that such perceptions are important for either female children or their parents (or both). Further, these perceptions may play a more important role than for male children. With regards to the social environment, median household income was found to be significantly positively associated with MVPA for girls, at 400 meter and 800 meter buffer sizes. The direction of this association is consistent 159  with what was hypothesized, and may reflect unmeasured aspects of design of recreational facilities. Higher neighbourhood income levels may, for instance, correspond to better quality parks (Crawford et al. 2008). Alternately, neighbourhood income may also be associated with perceptions of neighbourhood safety, with parents perceiving low income neighbourhoods to be less safe than their higher income counterparts (Gielen et al. 2004). Taken together then, the significant associations noted are consistent with perceptions of safety being particularly important for girls. As noted above, the results of the male models share many similarities with the single covariate models for the broader sample, including, for instance, highly significant positive associations between average daily MVPA and commercial density, residential density and intersection density at 1600 meter buffer sizes. Distance to school, total number of four way intersections en route to school and distance to closest non-park recreation site are also significantly negatively associated with MVPA. Beyond these common results, the models for male students yielded the following significant associations, all at 1600 meter buffer sizes: land use mix (+), proportion of population made up by children aged 5-14 (-), and median household income (-). The finding for land use mix is consistent with the findings regarding commercial, residential and intersection density, further highlighting the importance of physical access of a variety of destinations in shaping physical activity patterns. However, the other additional correlates identified indicate associations in the opposite direction of what was expected. The finding that median household income is negatively associated with overall MVPA is difficult to explain in light of existing evidence, and contrasts with the findings for girls. It is possible that the unmeasured environmental characteristics associated with higher income neighbourhoods that are appealing to girls or their parents are not appealing to boys (and vice versa). For instance boys may perceive certain environments as exciting or intriguing, whereas these may be perceived of as unsafe by girls. Similarly, the proportion of population made up by children 160  aged 5-14 was negatively associated with MVPA for boys. As discussed previously, this finding may be due in part to the measure definition which encompasses a wide age range of children. Final models were developed for both girls and boys, based on the results outlined in Tables 24 and 25. The first step in developing a final female model was to implement a model with the following built and social environment variables entered, in addition to the control variables age and ethnicity: cul-de-sac density (1600 meters), proportion of streets with posted speed limits of 30 km/h or less (1600 meters), and median household income (400 meters). For characteristics significantly associated with MVPA at multiple buffer sizes in the single covariate models, the buffer size chosen for the final model was that with the most significant association with MVPA. All variables were entered separately, rather than as part of an index, because collinearity was not found to be problematic (VIFs ranging from 1.1 to 1.4). In this intermediate model, the parameter estimate for median household income was found to be nonsignificant. Thus, a final model was created retaining only significant explanatory variables. Results of this model are summarized in Table 26. As illustrated in this table, the final model confirms the findings of the single covariate models that both cul-de-sac density and proportion of streets with posted speed limits of 30 km/h or less are positively correlated with average daily MVPA for girls. In this model, the two built environment measures account for 6.2% of the variation in MVPA, beyond the 20.6% accounted for by the control variables. A final male model was then developed following a similar procedure used to create the final female model. First, variables were assessed for collinearity. In contrast to the built environment characteristics entered in the female model, those entered in the male model were moderately collinear (average VIF of 3.7). Based on the findings from the single covariate models, eight built environment variables were incorporated in a single built environment index,  161  Table 26: Final Gender Stratified Female Model (Model 5) Predicting Overall MVPA (n = 192). Marginal Parameter Estimate (B) – Final Model R2change Control Variables age ethnicity – North American/European as reference category East/Southeast Asian South Asian Mixed  0.206 -13.86***  -19.77*** 1.87 -7.41  Built Environment Variables 0.062 cul-de-sac-density 9.61*** proportion of streets with 7.89** posted speed limits of 30 km/h or less *significant at p <.05, **significant at p <.01, ***significant at p <.001  using z-scores as outlined above. The variables incorporated in the index represent all those identified as significant in the single covariate models, with the exception of measure three, the total number of four way intersections en route to school. This measure was not included in the index because: a) it was only weakly collinear with other built environment variables and so could be entered separately, and b) the hypothesized mechanism of influence of this characteristic on physical activity differs from that of the other measures. The eight measures incorporated in the index are primarily hypothesized to influence physical activity patterns by shaping access to destinations (Table 6). In contrast, the underlying mechanism of influence corresponding to measure three is related to safety, with more intersections en route to school translating into greater perceived danger on the part of children or their parents. Thus, an intermediate model was created which included control variables and the following covariates: total number of four way intersections, the built environment index, and the two social environment measures found significant in the single covariate models (measures 15 and 16). 162  The only significant parameter estimates in this model were for the control variables, and the built environment index. Retaining only significant covariates yielded the final model summarized in Table 27 (Model 6).  Table 27: Final Gender Stratified Male Model (Model 6) Predicting Overall MVPA (n = 174). Marginal R2change Control Variables age ethnicity – North American/European as reference category East/Southeast Asian South Asian Mixed  0.231  Built Environment Index  0.106  Parameter Estimate (B) – Final Model  -25.78***  -41.50*** -18.88*** -17.57 3.11***  *significant at p <.05, **significant at p <.01, ***significant at p <.001  In sum, these models demonstrate markedly different influences on MVPA by gender, highlighting the need in future research to further explore such differences. Built environment characteristics incorporated in both final models for female and male children explained a higher proportion of variation in MVPA (Marginal R2change = 6.2% and 10.6%, respectively) than the corresponding final model for the entire sample (Marginal R2change = 5.0%). However, due to the small sample sizes involved, subsequent models in the present study focus on the entire sample of students, male and female. This is in part because many of these analyses are based on a smaller sample (sample 1B, introduced in Table 1; ntotal = 255, nmale =120 and nfemale =135), and simultaneously incorporate additional explanatory variables, creating the potential for model over-fitting. Certain methods used for subsequent analyses also would require larger  163  sample sizes than available (for example use of structural equation modeling as described below is not recommended for use with sample sizes of less than 200 cases; Golob 2003).  3.9 Research Question 4 3.9.1 Research Question and Hypotheses Investigated This section summarizes results of models designed to investigate Research Question 4: Are objective measures of characteristics of built and social environments significantly associated with percentage of time spent in: sedentary activity, light activity, moderate activity and vigorous activity, when controlling for child age, gender and ethnicity? Rather than addressing this question by examining associations with individual built and social environment characteristics, this question is explored by building on the findings of Model 2, and specifically exploring whether the built environment index developed for this model is associated with sedentary, light, moderate and vigorous physical activity. Five hypotheses are proposed, Hypotheses 4.1-4.5: It is hypothesized that the built environment index used in Model 2, will be significantly positively associated with percentage of time spent in: light physical activity  (Hypothesis 4.1), moderate physical activity (Hypothesis 4.2), and vigorous physical activity (Hypothesis 4.3). It is also hypothesized that the strength of association will be greatest for moderate physical activity, followed by vigorous physical activity, followed by light physical activity (Hypothesis 4.4). Finally, it is hypothesized that the built environment index will be significantly negatively associated with percentage of time spent in sedentary activity (Hypothesis 4.5). Taken together, these hypotheses imply that increasing built environment index values (representing shorter distances to school and non park recreation sites, higher residential and  164  commercial density and higher intersection density) will be associated with the substitution of light, moderate and vigorous physical activity for sedentary activity. Moderate physical activity encompasses a range of activities that are likely to be influenced by built environment characteristics, including walking briskly, biking and outdoor play. For this reason, MPA is anticipated to be strongly positively associated with the built environment index. Vigorous physical activity is anticipated to be positively associated with the built environment index, but to a lesser degree than MPA because vigorous activities may be influenced less by environmental characteristics, and more by other factors such as participation in organized sports. Light physical activity is anticipated to be less strongly associated with the index than moderate or vigorous activity because, as noted above, many activities influenced by built environment characteristics are likely to be of at least moderate intensity. However, some low intensity activities such as slow walking may be influenced by built environment characteristics reflected in the index, suggesting at least a weak association between LPA and the index. Finally, although sedentary activities have some different correlates than MVPA (Roemmich et al. 2007), it is anticipated that built environment characteristics that encourage physical activity  might be expected to also discourage sedentary behaviour through displacement.  3.9.2 Analytical Framework and Methods Figure 25 illustrates the conceptual model for testing Hypotheses 4.1-4.5. In contrast to previous models, the dependent variables in these models are not measures of minutes per day of sedentary, light, moderate or vigorous activity, but rather are expressed as the percentage of time spent in each of these intensities of activity. This approach was taken specifically because exploratory data analysis revealed a strong correlation between wear time (i.e. the length of time per day for which children wore their accelerometers) and time spent in sedentary activity  165  Figure 25: Conceptual Model for Research Question 4, Hypotheses 4.1-4.5. BUILT ENVIRONMENT INDEX*  AGE GENDER  proportion of time spent in sedentary, light, moderate or vigorous activity (one model for each of four dependent variables)  ETHNICITY * index developed in Model 2  per day. Regressing minutes of sedentary activity per day on number of wear hours produced a significant R2 of 0.407, and a slope of 43.7, translating into 43.7 minutes of sedentary activity per hour of wear time. This strong dependence of sedentary activity on wear time might reflect real differences between study participants in total time spent in sedentary activities (i.e. children who wore their accelerometers longer also engaged in more sedentary activity), but it might also simply reflect differences in when children removed their accelerometers. That is, children with shorter wear hours may have simply removed their accelerometers earlier in the evening, after which they too engaged in sedentary activity. Given this possibility, the decision was made to normalize sedentary activity by wear hours, which removed the dependency on wear time. For consistency and comparability, measures of other intensities of activity were also normalized by wear time. As previously, Generalized Estimating Equations were used. One model was tested per level of intensity of physical activity.  166  3.9.3 Results and Discussion Table 28 summarizes the results of the four models tested. As indicated below, findings are consistent with Hypotheses 4.2, 4.3 and 4.5. That is, the built environment index is significantly positively associated with both moderate and vigorous physical activity, and significantly negatively associated with sedentary activity. The findings are also consistent with Hypothesis 4.4 as each unit change in the built environment index is associated with a 0.26% percent increase in time spent in MPA (p < .001), a 0.12% increase in time spent in VPA (p < .001), and a 0.04% increase in time spent in LPA (non significant). Despite this, a larger proportion of VPA is explained by the built environment index (R2 = 4.7%) than the proportion of MPA (3.3%). The large proportion of variance in VPA explained by the built environment index may reflect the fact that the index incorporates measures of access to possible sites for vigorous activities (school and non-park recreation sites). This may also partly reflect the cut points used (S. 3.2.1.2). That is, because the Trost et al. (2002) cut points are lower than other cut points, some activities classified as moderate using more conservative cut points will have been classified as vigorous in the present analysis. This finding is important because it suggests that significant associations with neighbourhood environment characteristics could still be observed with more conservative cut points. For example, an alternative, more conservative cut point for MVPA established by Puyau et al. (2002) is roughly comparable to the cut point used in this study for VPA (it falls between 3 and 20% below the cut point for VPA used in the present analysis depending on the age of child). Finally, Hypothesis 4.1 was rejected as the proportion of time spent in LPA is not significantly associated with the built environment index. Table 29 presents a further interpretation of the results, by examining differences in sedentary activities, MPA and VPA corresponding to differences in built environment index values, as predicted by Models 7, 9 and 10. As illustrated in this table, increasing built environment index values corresponds to a shifting of sedentary activities to MPA and VPA. 167  Table 28: Association of Built Environment Index with Percentage of Time Spent in Sedentary Activity and Light, Moderate and Vigorous Physical Activity (Models 7-10, respectively). Significant results are bolded.  Sedentary (Model 7) -1.82*** 3.34***  Parameter Estimates (B) Light Moderate (Model 8) (Model 9) -0.26 1.36*** -0.65** -2.04***  gender=male Age ethnicity – North American/ European as reference category East/Southeast Asian 6.68*** -1.45*** -3.75*** South Asian 0.36 2.97*** -1.96*** Mixed -0.86 -1.29 2.74* Marginal R2 (controls) 21.6% 8.8% 23.9% built environment index 0.04 -0.41*** 0.26*** developed using Model 2 Marginal R2 (built 2.6% 0.1% 3.3% environment index) *significant at p <.05, **significant at p <.01, ***significant at p <.001  Vigorous (Model 10) 0.66** -0.64***  -1.17*** -0.45*** -0.38 19.7% 0.12*** 4.7%  Table 29: Predicted Changes in Percentage of Time Spent in Sedentary Activity, MPA and VPA Associated with Changes in Built Environment Index Values (based on Models 7-10) Difference between Built Environment Index Values  highest quartile – lowest quartile highest decile – lowest decile  Change in Percentage of Time Spent in Sedentary Activity, MPA or VPA corresponding to difference between Built Environment Index Values Sedentary MPA VPA -1.2% +0.8% +0.4% -2.7% +1.7% +0.8%  168  4. SAMPLE 1B: SURVEY DATA COMBINED WITH OBJECTIVE BUILT ENVIRONMENT AND PHYSICAL ACTIVITY DATA 4.1 Overview This chapter presents methodology and results relating to sample 1B as described in the introduction (S. 1.1), consisting of a subsample of 255 students from sample 1, for whom additional survey data on travel behaviour and neighbourhood perceptions were available. This chapter opens with an outline of survey data collection, variable selection and data cleaning processes. Following this, sample characteristics are described, with an emphasis on comparing sample 1 and sample 1B. This is followed by descriptive analyses of built environment characteristics, survey data and physical activity data. Finally, analytical methods and results relating to Research Questions 5 through 7 are discussed. These analyses build on Model 2 as developed in S. 3.6.3.3, by incorporating additional control variables (car ownership and household income), measures of child travel behaviour, and measures of parent and child perceptions of their neighbourhoods. Relationships between Research Questions 5 through 7 and earlier research questions were previously illustrated in Figure 2. With one exception, noted below, all analyses use overall average daily minutes of MVPA as the dependent variable. The power to detect a range of effect sizes using this sample was estimated using G*Power (Fraul 2009). Specifically, power was estimated based on the following parameters: n=255, alpha=0.05, 10 predictors. Given these parameters, power was estimated at 0.61 to detect a small effect size (f2=0.02), 0.99 for a medium effect size (f2=0.15), and 1.0 for a large effect size (f2=0.35).  169  4.2 Survey Data Collection, Variable Selection and Data Cleaning 4.2.1 Survey Design and Sampling Process As part of the AS!BC project outlined in S. 3.2.1.1, Dr. Lawrence Frank led the implementation of paired parent-child surveys on school travel behaviour and perceptions of neighbourhood characteristics. Both surveys are included as Appendix 5. These surveys were created through a collaborative effort by Jennifer Niece, a Masters Student at the School of Community and Regional Planning (SCARP), together with Dr. Frank, and Dr. James Sallis at San Diego University (Niece 2006). They were based on previous survey instruments developed and tested as part of the U.S. Neighbourhood Quality of Life Study and the Ontario Walkability Survey. The surveys were administered to students at all 13 participating AS!BC schools in the Lower Mainland, including the nine schools selected for objective physical activity measurement. As with the objective physical activity measurement, all grade four and five students at participating schools were invited to participate. Written informed consent was obtained from 839 students. Survey packages were distributed to these children by AS!BC research assistants between December 2005 and February 2006. In February 2006, nonresponding participants were mailed a duplicate package with a cover letter encouraging their participation. By April 2006, 498 children’s surveys and 500 parent’s surveys had been returned. Subsequent analyses incorporated measures from both parent and child surveys, and were therefore based on a sample of 494 students for whom both parent and child surveys were completed. The final sample 1B consists of a subset of these 494 students, for whom accelerometer data was also collected, n = 255 (Figure 26). This study was approved by the human research ethics board at the University of British Columbia (certificate number B050505). 170  Figure 26: Path from Initial Survey Distribution to Final Sample 1B  Total Number of Survey Packages Distributed n=839  Did Not Respond n=345  Total Number of Survey Respondents n=494 Physical Activity Data Unavailable Unavailable because school not included in accelerometry portion of study School included, but valid physical activity data unavailable (Figure 8)  n=80 n=159 n=239  Total Number of Participants Included in Sample 1B (with Valid Physical Activity, Survey and Objective Urban Form Data) n=255  4.2.2 Variable Selection The paired parent-child surveys address a variety of subjects relating to active transportation to school, including travel behaviour, perceptions of neighbourhood characteristics, mode choice preferences and household characteristics. Many of the questions posed could potentially be used to derive variables to incorporate in complex models linking objective measures of built environment characteristics to physical activity. As a result, careful attention was required to develop parsimonious models and conversely avoid overfitting (Babyak 2004). An important preliminary step was therefore to identify a theoretically relevant 171  subset of survey questions to create variables for the investigation of the proposed research questions. Table 30 outlines the final selection of variables and the rationale for their selection. These include measures of two household characteristics previously demonstrated to be important correlates of physical activity: household income and car ownership (S.2.3.4.2). A single measure of travel behaviour, child mode of commute to or from school, was selected as it was the primary focus of the survey and a potentially important correlate of MVPA (Cooper et al. 2005). Selection of the remaining variables reflects an emphasis on measures of perceptions of environmental characteristics given their theoretical relevance and documented importance in past empirical studies (S. 2.3.4). Particular emphasis was given to measures of perceptions relating to safety, as these may be critical for younger children as in the present study. Both measures of parent and child perceptions were included so that their influences on MVPA could be simultaneously assessed. The smaller number of measures of child perception variables reflects the limited number of questions on the topic included in the child survey.  172  Table 30: Survey Variables Selected for Sample 1B Models Survey Item  Question(s) in Survey  Theoretical Rationale  Household Characteristics S1. household income  parent’s #17  Household income may influence physical activity behaviour through a number of mechanisms. Children in low income households may for instance have less access to certain recreational facilities or activities relative to their counterparts in higher income households (S. 2.3.4.2.4).  S2. car ownership  parent’s #5  Car ownership may shape mode choice and travel habits of families, thus moderating associations between built environment characteristics and physical activity patterns (S. 2.3.4.2.3). Children in families with few or no cars may, for example, be less sensitive to built environment characteristics insofar as they have to engage in high levels of active transportation regardless of the environmental characteristics of their neighbourhoods.  Travel Behaviour S3. child mode of commute to or from school  parent’s #8 and #9  Active transportation to school is a potentially important mechanism through which children may engage in higher levels of habitual physical activity (Cooper et al. 2005). Data from parent’s surveys were chosen to measure child mode of commute because self-reported assessments of physical activity for younger children have been found to be of relatively low validity (Trost 2007)*.  Parent Perceptions  All questions addressing parental perceptions were worded in the original survey specifically in relation to children’s home-school commute. While active commuting to school is only one possible mechanism through which built environment characteristics may shape physical activity patterns, it may be reasonable to assume that parental perceptions of the environments their children are exposed to en route to school relate more generally to broader neighbourhood characteristics.  173  Table 30(continued): Survey Variables Selected for Sample 1B Models Survey Item  Question(s) in Survey S4. “There is too much traffic along parent’s – the route [between home and school]” barriers #6 S5. “There is one or more dangerous parent’s – crossings [between home and school]” barriers #7 S6. “My child is safe from traffic parent’s - #15b while walking to school or waiting for the school bus/public transit” S7. “[it is difficult for my child to parent’s – walk to school because] it is unsafe barriers #14 because of crime (strangers, gangs, drugs)” S8. “[it is difficult for my child to parent’s – walk to school because] my child gets barriers #15 bullied / teased / harassed” S9. “[it is difficult for my child to parent’s – walk to school because] there are stray barriers #17 dogs” Child Perceptions  S10. “[When I walk or bike in my neighbourhood,] I feel safe from cars”  children’s – #12a  S11. “[When I walk or bike in my neighbourhood,] I feel safe from strangers and bullies” S12. “[When I walk or bike in my neighbourhood,] I feel safe walking by myself”  children’s – #12b  Theoretical Rationale S4-S6 all assess parental perceptions of the traffic environment, which as outlined in S. 2.3.4.2.1 is an important aspect of parental perceptions of safety.  S7-S9 all assess parental perceptions of the social environment, which is a second important aspect of parental perceptions of safety.  In contrast to the above noted measures of parental perceptions, the survey questions corresponding to child perceptions explicitly refer to walking and biking in a child’s neighbourhood, and thus address broader activity patterns rather than the commute to school. S10-S12 assess different aspects of child perceptions of safety in their neighbourhood. S10 and S11 relate to the traffic and social environment, respectively, while S12 assesses perceptions of safety more broadly.  children’s – #12e 174  *The Parent’s surveys questions #8 and 9 were also chosen to create this measure because they are worded to most explicitly address habitual behaviour. That is, they query “How does your child usually get to [or home from] school”, and specify that “ “Usually” means 3 or more times per week”. Parent and child measures of physical activity are, however, highly correlated. When coded the same way (as a dichotomous variable discriminating between active transportation and other modes, as described in the following section), the correlation between parent and child measures of mode choice is r=0.845(p<.001). 4.2.3 Data Cleaning and Coding For variables selected for final analysis, missing data were imputed using one of two methods. For most variables, hot deck imputation was used, whereby missing data are replaced with randomly selected values from a relevant sub-sample of students (Andridge and Little 2010). Specifically, this method was used to impute one missing case for car ownership (S2), by randomly selecting a vehicle ownership category from respondents in the same income group. This approach was chosen because car ownership and household income are moderately correlated, spearman’s rho = 0.486, p < .001. For perception variables (S4-S12), missing values were replaced with a random sample from that child’s school. This approach is based on the assumption that parents of children attending the same school share similar perceptions about their neighbourhoods. A descriptive analysis of perceptions indicates moderate variations in the distributions of perceptions between schools (S. 4.4.2.3 below). This method was used to replace between 2 and 11 missing cases for each variable. Finally, household income was missing for 43 students, and was imputed by multiple regression with median neighbourhood household income (Variable 16 in Table 7) and car ownership as predictors (R=0.506, p < .001). With one exception, the chosen variables were coded as ordinal variables. This includes all measures of perceptions which were based on survey items gauging level of agreement or disagreement using an ordered scale (Likert Items). Mode of commute to school was, in contrast, coded as a dichotomous variable. Table 31 summarizes the coding used for all variables.  175  Table 31: Survey Variable Coding Measure Coding Household Characteristics S1. household income 1=under $19,999 2=$20,000-$29,999 … 10=$100,000 or greater S2. car ownership 0=none 1=1 2=2 3=3 4=more than 3 Travel Behaviour S3. active transportation to or from school  For descriptive purposes below, this variable was used in its disaggregated form, as on the survey (an 11 category variable). For inferential analyses, it was coded as a dichotomous variable, as 1 for students who commute to or from school via active modes (walking, biking, using roller blade, scooter or skateboard) and 0 for all other modes (driven, school bus, transit).  Parent Perceptions S4-S9 in Table 30  Coded as 4 category Likert items: 1=strongly agree 2=somewhat agree 3=somewhat disagree 4=strongly disagree  Child Perceptions S10-S12 in Table 30  Coded as 3 category Likert items: 1=agree a lot 2=agree a little 3=don’t agree  4.3 Sample Description and Regional Context Table 32 compares student age, gender and ethnicity for students in samples one and two. As illustrated in this table, sample characteristics are nearly identical. As with sample 1, sample 1B therefore likely is a reasonable representation of the ethnic diversity of the Lower Mainland (S. 3.3.1). Further, the school and school neighbourhood characteristics described in S. 3.3.2 also apply to sample 1B. 176  Table 32: Descriptive Statistics for Sample 1B Control Variables. Data for sample 1B compared to sample 1. Sample 1B, n=255 is a subsample of sample 1, n=366. Percent / Value in Sample 1B Gender Male Female Ethnicity European/North American East/Southeast Asian South Asian Mixed Other Age Mean quartile 1 quartile 2 (median) quartile 3  Percent / Value in Sample 1  46.5% 53.5%  47.6% 52.4%  46.1%  44.0%  28.3% 10.1% 9.7% 5.8%  29.3% 11.1% 7.6% 8.0%  9.94 9.46 9.90 10.47  9.94 9.43 9.94 10.46  Data on household income unavailable in sample 1 can be used to further situate sample 1B within the broader regional context. Descriptive statistics on household income are illustrated in Table 33. Table 33: Descriptive Statistics for Household Income. Data for Sample 1B, n=255. Data for Lower Mainland from 2006 Census (Statistics Canada 2008) Household Income Under $20,000 $20,000-$29,999 $30,000-$39,999 $40,000-$49,999 $50,000-$59,999 $60,000-$69,999 $70,000-$79,999 $80,000-$89,999 $90,000-$99,999 Greater than $100,000  Percentage of Households in Income Category Sample 1B Lower Mainland 9.7 16.1 10.1 9.5 8.5 10.2 13.2 9.5 8.5 8.6 17.1 7.7 6.2 6.7 8.1 5.6 7.0 4.7 11.6 21.6  177  As illustrated in this table, sample 1B has relatively low proportions of both low (under $20,000) and high (greater than $100,000) incomes relative to the broader Lower Mainland population. In contrast, it has relatively high proportions of moderate income households ($60,000-$69,000). Taken together, this suggests that household income influences on the outcome variables in subsequent analyses may not be fully captured because the range of household incomes included in the sample is effectively truncated. Household income varies moderately by school, with relatively high proportions of low to moderate income households (under $50,000) corresponding to Vancouver schools (1 and 2). Households with higher incomes (over $50,000) are distributed across the region (high proportions of these households correspond to schools 4, 5 and 8, located in North Vancouver, Burnaby and Mission respectively).  4.4 Descriptive Analysis 4.4.1 Objective Built and Social Environment Characteristics Table 34 summarizes descriptive statistics for built and social environment measures for samples one and two. Absolute differences between mean values of individual measures for each sample are generally small, with only two variables differing by more than 10%: the proportion of streets with posted speed limits of 30 km/h or less, and the population density of children aged 5-14, both of which are higher in sample 1B. The difference in mean values of the composite built environment index in contrast appears substantial (-0.22 in sample 1B and 0.00 in sample 1). However, viewed in relation to the standard deviation, differences in index values are minor, reflecting the minor differences in its component measures. In sum, the neighbourhood environments of students in sample 1 and sample 1B are very similar as gauged by the objective measures created for this study. 178  Table 34: Descriptive Statistics for Built and Social Environment Measures for Students in Sample 1B. Sample 1B data (n=255) compared to data for sample 1 (n=366). Values reported for home neighbourhood based measures correspond to 800 meter buffer size. Measure  Route Between Home and School 1. distance to school (m) 2. average intersection spacing (m) 3. total number of four way intersections Home / School Neighbourhood 4. net commercial density (dimensionless floor area ratio) 5. net residential density (dwellings/residential acre) 6. land use mix (dimensionless) 7. intersection density (intersections/km2) 8. cul-de-sac density (cul-de-sacs/km2) 9. proportion of streets with posted speed limits of 30 kilometers per hour or less (%) 11. number of parks in buffer 12. distance to closest park (m) 13. distance to closest possible non park recreation site (m) 14. population density of children aged 5-14 (population/net residential acre) 15. proportion of population made up by children aged 5-14 (%) 16. median household income ($) Built Environment Index  Mean (Standard Deviation) Sample 1B  Sample 1  1390(1950) 130(60) 3.7(4.7)  1280(1780) 120(50) 3.7(4.6)  0.61(0.35)  0.60(0.34)  12.41(10.26)  13.33(11.33)  0.33(0.23) 50.9(17.2) 8.71(6.82)  0.33(0.23) 51.52(17.40) 8.25(6.61)  11.65(19.20)  8.24(6.46)  3.03(1.69) 440(1000) 520(810)  3.00(1.63) 410(860) 485(690)  4.11(3.23)  3.58(1.99)  11.32(2.90)  11.41(2.78)  $59,000($14,000)  $58,000($14,000)  -0.22(2.99)  0.00(2.95)  To further consider the influence of slight variations in sample characteristics on model results, and to create a base model for subsequent analyses, Model 2 was retested using sample 1B, with results illustrated in Table 35. The results illustrate similar highly significant parameter estimates are produced for age and ethnicity, however the estimate for gender is less significant when using sample 1B. The built environment index in both cases is highly significant, with a  179  slightly higher parameter estimate produced for the index using sample 1B. As illustrated in Figure 27, 95% confidence intervals for the built environment index parameter estimate largely overlap. Table 35: Model 2, Predicting Overall MVPA, Tested Using Both Sample 1 (n=366) and Sample 1B (n=255). Parameter Estimate Parameter Estimate (B) – Sample 1B (B) – Sample 1 Control Variables age gender=female ethnicity – North American/European as reference category East/Southeast Asian South Asian Mixed Marginal R2 0.231  -19.72*** -13.32*  -19.94*** -17.01***  -35.88*** -21.07*** -11.22  -37.96*** -20.15*** -10.98  Built Environment Index Marginal R2 0.057  0.240 3.70***  3.49*** 0.050  *significant at p <.05, **significant at p <.01, ***significant at p <.001 Figure 27: Parameter Estimates and 95% Confidence Intervals for Built Environment Index, when Predicting Overall MVPA, Using Sample 1 and Sample 1B.  Parameter Estimate  6  5  4  3  2  1  0 1  2  Sample 180  4.4.2 Survey Data 4.4.2.1 Car Ownership Table 36 presents summary statistics for car ownership in sample 1B. The vast majority of households own either one or two cars, with small proportions of households owning no cars (4.3%) or more than three cars (5.8%). Table 36: Car Ownership for Households in Sample 1B Number of Cars  % of households in sample 0 1 2 3 >3  4.3% 35.4% 44.7% 9.7% 5.8%  Car ownership varies moderately by school, likely reflecting in part the varying built environments in which study participants live. School two in Vancouver stands out, with all students in sample 1B attending this school living in households with one or no cars. All other schools have high proportions of multiple car households, ranging from 36.0% (school one) to 87.1% (school eight).  4.4.2.2 Mode of Transportation To and From School Table 37 summarizes commute mode of children for the trip to and from school (survey item S3 above). On the trip to school, a small majority of children are driven, a large proportion use active transportation, and very small proportions take bus/transit or specified multiple modes. On the return trip from school, fewer children are driven, and the proportions of children using all other modes is increased. This pattern is consistent with previously observed patterns, likely reflecting the convenience of driving children to school on the way to work.  181  Table 37: Child Commute To and From School by Mode Commute Mode active transportation (walk, bike, roller blade /scooter /skateboard) driven take bus/transit multiple modes  Trip to School 44%  Return Trip from School 49%  54% <1% 2%  47% 1.6% 2.4%  The vast majority of children who use active transportation to school walk (98% of active commuters), either by themselves (17%), with a parent or other adult (49%), or with siblings or friends (27%). The remaining 2% of active commuters bike to school. Similar patterns hold for the return trip from school, with 99% of active commuters walking. However, on the return trip, children are more likely to walk by themselves (19% of active commuters) or with siblings or friends (34%) and less likely to walk with parents or other adults (43%) than on the trip to school. The remaining 1% of students actively commuting from school do so by bike. For subsequent analyses, the child’s mode of commute to or from school was dichotomized as ‘active’ (1) or ‘not-active’ (0). This is because active commuting represents a mechanism through which children may accumulate MVPA. If multiple modes were specified, these were coded as active if both modes selected were different forms of active transportation, and inactive if one of the modes was an inactive mode (e.g. driven to school). The latter decision rule was established because specification of both active and inactive modes was taken to imply that the active modes were not habitual. Next, both commuting to and from school variables were combined into a single variable so that a child was classified as an active commuter if their parents indicated that they usually actively commute to school, from school or both. Based on this measure, 52.7% of all children in sample 1B were classified as active commuters.  182  Child mode of commute to or from school varies considerably in relation to objectively measured distance to school. As illustrated by the upper line in Figure 28, approximately 50% of children who actively commute live within 400 meters of school, and 100% of active commuters live within 1600 meters of their school. As indicated by the lower line, all children living within 200 meters of their school actively commute. However large proportions of children who take inactive modes of transportation also live relatively short distances from school. For example, approximately 33% of children who take inactive modes of transportation to school live within 200 and 800 meters of their school.  Figure 28: Active and Non-active Commuters by Distance Between Home and School 100 90 80  Percent (%)  70 60  active commuters*  50  non-active commuters**  40 30 20 10 0 0  200  400  600  800 1000 1200 1400 1600  Distance to School (meters)  *cumulative percentage of active commuters living within specified distance of school **cumulative percentage of non-active commuters living within specified distance of school  Rates of active transportation vary substantially between schools, in part because of variation in distances between homes and school. While the schools involved in this study all have small catchment areas (approximately 1600 meters in diameter on average), many students 183  live outside of these catchment areas. Some students at schools seven and nine, located in Burnaby and Mission, respectively, have particularly long commute distances because these schools offer French Immersion programs. Less than 40% of children at these schools actively commute. At the other extreme, more than 70% of children actively commute in both Vancouver schools (schools one and two), and one of the Burnaby schools (school six). At the remaining schools, between 40 and 65% of children actively commute.  4.4.2.3 Neighbourhood Perceptions Tables 38 and 39 summarize descriptive statistics on parent and child perceptions of their neighbourhoods, respectively. With regards to perceptions of traffic safety, the majority of parents somewhat agree or strongly agree that there is too much traffic (57.1%), and that there is one or more dangerous crossings (57.4%). Conversely, the majority of parents also somewhat or strongly agree that their child is safe from traffic while walking to school or waiting for the school bus/public transit (61.9%). Higher numbers of children (86.7%) agree either ‘a little’ or ‘a lot’ that they feel safe from cars when walking in their neighbourhood. While these scales and measures are not directly comparable, they appear to suggest that children are less concerned about traffic safety than their parents. Overall, parents indicate less concern regarding the social environment than the traffic environment, with more than 60% of parents disagreeing or strongly disagreeing with statements S6 to S9. The social environment measure with the highest proportion of parents agreeing or strongly agreeing is S7, suggesting that moderate numbers of parents are concerned about crime and strangers (39.1%). This is followed by S9, with 20.4% somewhat or strongly agreeing that stray dogs make it difficult for their child to walk to school. Relatively few parents somewhat or strongly agree that bullying or teasing makes active commuting difficult (S8, 10.9%). Responses to S11 suggest that less than one third of children are concerned about bullying or strangers, with 184  Table 38: Descriptive Statistics for Parent Perception Variables Survey Item  S4. “There is too much traffic along the route [between home and school]” S5. “There is one or more dangerous crossings [between home and school]” S6. “My child is safe from traffic while walking to school or waiting for the school bus/public transit” S7. “[it is difficult for my child to walk to school because] it is unsafe because of crime (strangers, gangs, drugs)” S8. “[it is difficult for my child to walk to school because] my child gets bullied / teased / harassed” S9. “[it is difficult for my child to walk to school because] there are stray dogs”  Strongly Disagree  Somewhat Somewhat Agree Disagree  Strongly Agree  23.6%  17.4%  27.1%  30.0%  24.1%  18.5%  22.9%  34.5%  14.1%  24.1%  43.4%  18.5%  38.7%  22.2%  22.2%  16.9%  66.8%  22.3%  8.1%  2.8%  53.9%  25.7%  15.1%  5.3%  Table 39: Descriptive Statistics for Child Perception Variables Survey Item S10. “[When I walk or bike in my neighbourhood,] I feel safe from cars” S11. “[When I walk or bike in my neighbourhood,] I feel safe from strangers and bullies” S12. “[When I walk or bike in my neighbourhood,] I feel safe walking by myself”  Don’t agree  Agree a little  Agree a lot  12.9%  42.0%  44.7%  28.5%  37.1%  34.4%  39.1%  38.3%  22.5%  185  28.5% indicating that they disagree with “When I walk or bike in my neighbourhood, I feel safe from strangers and bullies”. Finally, responses to a general question about perceived safety (i.e. not expressed in relation to specific elements of the physical or social environment), indicate that almost two thirds of children feel safe in their neighbourhood when walking on their own (S12). Measures of perceptions and, in particular, those measures relating to the traffic environment vary moderately by school. Parents and their children in Vancouver schools one and two, and Burnaby schools five and six are less likely to indicate agreement with statements indicating concerns with traffic safety (S4-S6 and S10) than those attending school seven in Burnaby or school nine in Mission. For example, 76% of parents of children attending school two in Vancouver indicate that they strongly or somewhat disagree with S4 (There is too much traffic along the route [between home and school]”), while 75% of parents of children attending school nine in Mission either somewhat or strongly agree with this statement. Parents and their children attending the remaining schools in North Vancouver (three and four) and Mission (eight) fall between these extremes. Similar but less pronounced differences are evident for measure S7 “[it is difficult for my child to walk to school because] it is unsafe because of crime (strangers, gangs, drugs)”, and the corresponding child survey measure S11 “When I walk or bike in my neighbourhood, I feel safe from strangers and bullies”. For example, approximately 40% of children attending school nine in Mission indicate that they don’t agree with S11, compared to 18% of children attending school two in Vancouver. The remaining measures, S8, 9 and 12, have relatively consistent responses across schools.  4.4.3 Physical Activity Data Students in sample 1B not only are characterized by a similar age, gender and ethnicity profile as those in sample 1, but also a very similar physical activity profile. Similarities 186  between the two samples are also apparent in the model results presented in Table 35. Thus, rather than repeating the full summary of physical activity descriptive statistics presented for sample 1 in S. 3.4.2, descriptive statistics relevant to subsequent models are presented in Table 40. Focusing on the primary dependent variable for subsequent analyses, average daily minutes of MVPA for all students, the difference in means between students in sample 1B and those excluded from sample 1B (students in sample 1 without survey data) is highly non significant at p = 0.417.  Table 40: Descriptive Statistics for Physical Activity Variables, Samples One and Two. Variable  total average daily MVPA outside of school MVPA average daily MVPA outside of school, on school days average daily MVPA on weekends total – average daily MVPA boys girls North American/European all other Ethnicities  Mean (Standard Deviation) minutes per day Sample 1B  Sample 1  126(37)  124(38)  76(29)  75(29)  120(45)  119(47)  132(39) 120(34)  134(39) 116(34)  133(39)  133(39)  119(34)  118(35)  187  4.5 Research Question 5 4.5.1 Research Question and Hypothesis Investigated This section summarizes results of models designed to investigate Research Question 5: How are previously observed relationships between built and/or social environment characteristics and average daily MVPA (Research Question 1) modified when car ownership and household income are controlled for, in addition to age, gender and ethnicity? To investigate this question, the following hypothesis was tested: Hypothesis 5.1: The built environment index developed in Model 2 will be significantly associated with average daily MVPA when controlling for car ownership and household income, but the association will be weaker than in models excluding these variables. This question builds on findings associated with Research Question 1, by incorporating additional control variables unavailable in sample 1. As noted above, both household income and car ownership are anticipated to be significant covariates of average daily MVPA. It is hypothesized that the observed influence of built environment characteristics on MVPA in Model 2 may in part be due to unmeasured variation in household income and car ownership. As a result, inclusion of these variables is anticipated to result in a smaller parameter estimate for the built environment index as a predictor of MVPA.  4.5.2 Analytical Framework and Methods The conceptual model used in this analysis is illustrated in Figure 29. As previously, this model was tested using GEE to account for the clustering of students within schools. Although household income was collected as an ordinal variable (Table 31), it was entered in the analysis as a continuous, ratio level variable. This approach is justified because income was  188  measured on ten categories, which with the exception of the first and last category are equally spaced, and has a symmetrical distribution (Leech et al. 2008). Car ownership was entered as a categorical variable, with coding as indicated in Table 31.  Figure 29: Conceptual Model for Research Question 5, Hypothesis 5.1 BUILT ENVIRONMENT INDEX  AGE  average daily minutes of Moderate to Vigorous Physical Activity (MVPA)  GENDER ETHNICITY HOUSEHOLD INCOME  dashed line indicates Model 2  CAR OWNERSHIP  4.5.3 Results and Discussion Results of the final model incorporating household income and car ownership are presented in Table 41. In this model, parameter and standard error estimates were largely unchanged for previously entered variables: age, gender, ethnicity, household income and the built environment index. Household income is significantly associated with MVPA, in the anticipated direction. That is, higher incomes are associated with more MVPA, possibly reflecting greater access to recreational resources or equipment. This is consistent with data from the 2005/2006 HBSC and 2005 CANPLAY surveys indicating that children in higher SES families have higher levels of physical activity than their lower SES peers (Active Healthy Kids Canada 2007). Having one or more cars was not significantly associated with MVPA relative to 189  not owning a car. While this result contrasts with findings of previous studies, it may be because Model 11 is predicting overall MVPA, whereas studies finding car ownership to be a significant predictor of physical activity related variables have generally focused on active transportation related variables as dependent variables (as in Kerr et al. 2007, Frank et al. 2007a and Timperio 2004). Car ownership might be expected to be more strongly associated with active transportation insofar as it directly influences travel behaviour and mode choice. The additional control variables accounted for an additional 2.3% of variance in average daily MVPA, beyond the influence of age, gender, ethnicity and the built environment index. Table 41: Model 11 Predicting Overall MVPA Using Sample 1B Data Marginal R2change  Original Control Variables age gender=female ethnicity – North American/European as reference category East/Southeast Asian South Asian Mixed  0.231  Built Environment Index  0.057  Sample 1B Control Variables household income car ownership – no cars as reference category 1 car 2 cars 3 cars > 3 cars  0.023  Parameter Estimate (B) – Model 2  Parameter Estimate (B) Model 11  -19.72*** -13.32*  -19.60*** -13.17*  -35.88*** -21.07*** -11.22  -33.07*** -18.70*** -11.96  3.70***  3.66***  -  2.08***  -  8.75 3.18 6.74 -4.22  *significant at p <.05, **significant at p <.01, ***significant at p <.001  190  4.6 Research Question 6 4.6.1 Research Question and Hypotheses Investigated This section summarizes results of models designed to investigate Research Question 6: When controlling for relevant socio-demographic covariates, are both objective measures of environmental characteristics, and measures of parent and child perceptions of environmental characteristics significantly associated with average daily MVPA? To investigate this question, the following hypotheses are examined: Hypothesis 6.1: The built environment index will remain a significant predictor of average daily MVPA in models incorporating measures of parent and child perceptions of environmental characteristics and controlling for child age, gender, ethnicity and household income. Hypothesis 6.2: Measures of both parent and child perceptions will also be significantly associated with average daily MVPA in these models, with higher levels of perceived safety significantly associated with higher levels of MVPA. Hypothesis 6.3: Measures of parent perceptions will be more strongly associated with average daily MVPA than measures of child perceptions. Hypothesis 6.4: Measures of parent and child perceptions will be significantly associated with child age and gender, with higher levels of perceived safety associated with boys and older children. Hypothesis 6.5: The built environment index will be significantly positively associated with parental perceptions of safety of the traffic environment (higher levels of the built environment index will be associated with greater perceived safety). The rationale for these hypotheses are also discussed below as they relate to the specification of the conceptual model.  191  4.6.2 Analytical Framework and Methods Figure 30 illustrates the conceptual model used in subsequent analyses. The hypothesized paths of influence directly from age, gender, the built environment index, household income and ethnicity to average daily MVPA illustrated in this figure were previously identified in the development of previous models (Figure 29). This model however introduces three perception variables, all of which are hypothesized to influence MVPA (Hypothesis 6.2), because high levels of perceived safety on the part of parents or their children may translate into increased independent mobility (and vice versa). Parental perceptions are hypothesized to be particularly important (Hypothesis 6.3) because of the role of parents as gatekeepers, especially for younger children (S. 2.3.4.2.1). All measures of perceptions are also hypothesized to all be influenced by age and gender (Hypothesis 6.4). Parental perceptions of the safety of their neighbourhood environment in particular may be shaped by the age and gender of their child, with younger children and girls more likely to be subject to restrictions on independent mobility (S. 2.3.4.2.1). Because of such restrictions or because of directions from their parents, younger children and girls might also perceive their environments differently. A path is also specified from the built environment index to parental perceptions of traffic environment safety (Hypothesis 6.5), although the measures incorporated in the built environment index relate primarily to access rather than safety. This linkage was specified in part because greater access to destinations may translate into reduced exposure to traffic simply due to the shorter distances that children have to travel, with higher built environment index values therefore associated with greater perceptions of safety of the traffic environment. In addition, environments with high degrees of access may have larger numbers of pedestrians and cyclists, thus contributing to perceptions of traffic safety.  192  Figure 30: Conceptual Model for Research Question 6. This illustrates the structural model tested using structural equation modeling (SEM) as described below. Following SEM conventions, rectangles represent observed variables and ovals represent latent variables (described below).  CHILD PERCEPTIONS OF SAFETY  PARENT PERCEPTIONS – SOCIAL ENVIRONMENT SAFETY  PARENT PERCEPTIONS – TRAFFIC ENVIRONMENT SAFETY  AGE  GENDER  BE INDEX*  Average Daily MVPA  HOUSEHOLD INCOME  ETHNICITY  *BE INDEX = BUILT ENVIRONMENT INDEX Paths are not shown from the built environment index to parental perceptions of social environment safety and child perceptions because causal linkages are not as clear. Specifically, while higher levels of access may be associated with more eyes on the street and thus greater perceived safety, it is also possible that they may be associated with more fears of strangerdanger, bullying, drugs or other elements contributing towards decreased perceptions of safety 193  of the social environment. Hypothesis 6.1 thus specifies that the built environment index will remain a significant predictor of average daily MVPA in models incorporating measures of parent and child perceptions, because the built environment index and measures of perceptions are associated with largely unique causal mechanisms. Because the simultaneous estimation of the complex series of causal pathways presented in Figure 30 would not be possible in a traditional regression framework (or using GEE), a method specifically designed for such analyses was used instead: structural equation modeling (SEM). A limitation of this approach is that clustering of students within schools could not be accounted for. Although a multilevel modeling approach to SEM has been developed, this requires substantially larger samples of clusters than in the present sample (n=9 schools; Hox et al. 2009). In addition to enabling the estimation of multiple relationships simultaneously, SEM has the benefit of being able to incorporate unobserved (latent) constructs (Ho 2006). These are variables that are not directly measured, but rather are measured by one or more indicator variables. The advantage of this approach is that measurement error associated with a single measure may be explicitly accounted for in the analysis, improving statistical estimation relative to methods which ignore such error (Meyers et al. 2006, Ho 2006). The present model incorporates two latent variables, distinguished from directly observed variables in Figure 30 by their depiction as ovals rather than rectangles. In both cases, multiple survey items are used as indicator variables (e.g. variables S4-S6 are used as indicators for parental perceptions of safety of the traffic environment). SEM is widely used in similar contexts, when multiple survey scale items may be linked to a smaller set of underlying constructs. While both measures of parental perceptions are thus modeled as latent variables, child perceptions of the environment are instead modeled as a single composite observed variable. This is because the Maximum Likelihood (ML) estimation procedure used to estimate the SEM model for this analysis requires that all endogenous variables are measured on an interval or 194  near interval scale (exogenous variables, such as gender and ethnicity may, however, be categorical; Meyers et al. 2006). Likert scaled items such as variables S4-S9 are widely incorporated in SEM analyses as approximating continuous variables (Byrne 2001), but four categories is considered the minimum number necessary to satisfy this approximation (Bentler and Chou 1987). In contrast to the questions on the parent surveys, which were based on four category likert scales, the measures of child perceptions from the child survey are based on three category likert scales, and therefore cannot reasonably be approximated as continuous. The three individual survey items used to assess child perceptions of the environment (S10 to S12) were instead summed to create a single composite measure or summative scale. The validity of combining nine individual survey items into three variables representing parent and child perceptions was assessed first by calculating a Cronbach Alpha as a measure of internal consistency reliability. The Alpha for the child perceptions variable was calculated at 0.65, which is below typical guidelines of 0.70 or higher, but considered reasonable according to less conservative guidelines specifying a minimum of 0.60 (Leech et al. 2008). Alpha values were above the 0.70 guideline for both parental perception variables (0.78 for perceptions of the traffic environment, and 0.72 for perceptions of the social environment). For the two latent variables, an additional step of first assessing a measurement model prior to assessing the full SEM model was taken, following common practice in SEM analyses (Meyers et al. 2006). A measurement model is a model used to assess the degree to which indicator variables adequately represent latent factors. The measurement model for the two latent factors is presented below in Figure 31. This figure illustrates factor loadings and correlations produced by implementing the model. This and subsequent SEM models described below were implemented using AMOS v.18.0.0 (Arbuckle 2009).  195  Figure 31: Measurement Model Corresponding to Structural Model Illustrated in Figure 30. In this illustration, factor loadings and correlations are shown as standardized estimates. All survey items load on their respective latent variable significantly, p < .001. Model degrees of freedom = 8.  0.89 PARENT PERCEPTIONS – SOCIAL ENVIRONMENT SAFETY  0.63 0.57  unsafe because of crime (S7) child gets bullied/ teased/ harassed (S8) stray dogs (S9)  0.46  PARENT PERCEPTIONS – TRAFFIC ENVIRONMENT SAFETY  0.88 0.83 0.63  too much traffic (S4) one or more dangerous crossings (S5) safe from traffic (S6)  AMOS output includes a wide array of measures of goodness of fit. Following the recommendations of Meyers et al. (2006), four of these measures are reported: chi square, the Comparative Fit Index (CFI), Normed Fit Index (NFI), and Root Mean Square Error of Approximation (RMSEA). These are presented in Table 42. Based on all measures of goodness of fit reported in Table 42, it is evident that the measurement model fits the data well. This model also satisfies the requirement of multivariate normality, with a Mardia’s normalized estimate of multivariate kurtosis < 1.96 (Gao et al. 2008).  4.6.3 Results and Discussion To ensure model parsimony, ethnicity was coded as a dichotomous variable with North American/European=1, and other ethnicities=0 in the final SEM model. This model was identified (df =47), with an adequate absolute sample size (greater than the recommended 196  Table 42: Fit Measures for Model 12 Measurement Model Goodness of Fit Measure  Chi square CFI NFI RMSEA  Reference Threshold(s)*  Measurement Model  greater than 0.05 greater than 0.95 greater than 0.95 0.05-0.08=good fit 0.08-0.10=moderate fit greater than 0.10=poor fit  0.055 0.985 0.970 0.061  *as recommended by Ho 2006 minimum of 200 cases; Golob 2003), and adequate sample size in relation to the number of parameters being estimated (5.5 cases per parameter, above the recommended threshold of 5 Bentler and Chou 1987). This model also satisfies the requirement of multivariate normality, with a Mardia’s normalized estimate of multivariate kurtosis of 0.59 (< 1.96). Significant path coefficients for the final model are illustrated in Figure 32. While covariances are not illustrated in the above figures, exogenous variables (age, gender, built environment index, household income and ethnicity) are highly non-collinear, with a maximum VIF of 1.2. As illustrated in Figure 32, all of the paths identified in previous models, linking the exogenous variables directly to average daily MVPA are significant and in the anticipated directions. As previously, age and being female is negatively associated with average daily MVPA. Conversely, higher incomes and students of North American/European ethnic origin correspond to higher average daily MVPA. In addition, the built environment index is significantly positively associated with average daily MVPA, supporting Hypothesis 6.1. While age and gender were hypothesized to influence all measures of perceptions (Hypothesis 6.4), only one such path was significant. That is, only being female is significantly  197  Figure 32: Significant Paths and Standardized Path Coefficients for Model 12 Final Structural Model. Significant paths are indicated by solid lines, non-significant paths by dashed lines. Standardized path coefficients are only shown for significant regression paths (covariances and non-significant regression paths are not illustrated for clarity).  CHILD PERCEPTIONS OF SAFETY  -0.23  PARENT PERCEPTIONS – TRAFFIC ENVIRONMENT SAFETY  AGE  GENDER  PARENT PERCEPTIONS – SOCIAL ENVIRONMENT SAFETY  0.32  -0.31 -0.17  BE INDEX*  +0.20  Average Daily MVPA  +0.15 HOUSEHOLD INCOME  +0.27 ETHNICITY  *BE INDEX = BUILT ENVIRONMENT INDEX  associated with lower perceptions of safety relative to being male. The non-significant results for gender influences on parental perceptions are difficult to explain, but may simply be a result of limited power with the available sample size. The non-significant path coefficients from age to the measures of perceptions, together with the significant path directly from age to average 198  daily MVPA, are consistent with earlier findings that while age has a significant main effect on MVPA, it does not moderate built environment influences on MVPA (S. 3.8.3). That is, the non-significant influences of age on perceptions may be explained by the truncated range of ages incorporated in the analysis, with parents similarly concerned about the safety of their children regardless of whether they are 8 or 11. In contrast to Hypothesis 6.4, Hypothesis 6.5 is supported by the model results, as the built environment index is significantly positively associated with parental perceptions of safety of the traffic environment. Thus, higher built environment index values correspond to greater perceived safety of the traffic environment on the part of parents. Contrary to Hypothesis 6.2, all paths from measures of perceptions to average daily MVPA are non significant, suggesting that when controlling for individual and household covariates (age, gender, ethnicity and household income) and objective measures of the built environment (the built environment index), measures of parent and child perceptions are not significantly associated with average daily MVPA. This also means that Hypothesis 6.3, suggesting that parent perceptions will be more strongly associated with average daily MVPA than child perceptions, is not relevant. Although the R2 for average daily MVPA in the final SEM model is 0.236, measures of goodness of fit suggest a poor overall model fit (with the exception of the RMSEA which suggests a moderate fit), likely reflecting the multiple non-significant paths. These measures of fit are presented in Table 43. Notably, if this model was modified by excluding variables that do not significantly predict average daily MVPA, it would reduce to Model 11. Because the results of this model suggest that perceptions of built and social environment characteristics do not significantly predict average daily MVPA, they appear to run counter to the findings of previous studies indicating that perceptions are associated with measures of physical activity (S.2.3.4). While the differences may be linked to wide ranging inconsistencies in methodology (e.g. different outcome variables, different measures of 199  Table 43: Fit Measures for Model 12, Predicting Overall MVPA. SEM Model Incorporating Measures of Parent and Child Perceptions.  Reference Threshold(s)*  Model  greater than 0.05 greater than 0.95 greater than 0.95 0.05-0.08=good fit 0.08-0.10=moderate fit greater than 0.10=poor fit  0.000 0.849 0.802 0.096  Goodness of Fit Measure  Chi square CFI NFI RMSEA  *as recommended by Ho 2006  perceptions etc.), they may also be due to a confounding effect of the built environment index with the measures of perceptions. That is, because many studies exploring the influence of perceptions on physical activity behaviour have focused solely on measures of perceptions (Carver et al. 2005, Hume et al. 2005, Salmon et al. 2007 and Timperio et al. 2004 among others), the associations observed may not have been evident if objective built environment measures were accounted for. To explore whether the objective built environment index used in the present study confounds associations between measures of perceptions and MVPA, a second model was fitted, excluding the index. The results of this model are illustrated in Figure 33 (a). As indicated in this figure, parental perceptions of traffic safety are significantly associated with average daily MVPA in this model. Treating this as an exploratory analysis for illustrative purposes, the model may be reduced by removing variables that do not significantly explain average daily MVPA, to the model illustrated in Figure 33 (b). Model fit statistics for this model presented in Table 44 indicate very good model fit. Given available data, it would therefore be possible to fit a model  200  Figure 33: Significant Paths and Standardized Path Coefficients for (a) Model Predicting Overall MVPA When Excluding the Built Environment Index, and (b) Reduced Model Retaining only Significant Predictors of MVPA. Significant paths are indicated by solid lines, non-significant paths by dashed lines. Standardized path coefficients are only shown for significant regression paths (covariances or non-significant regression paths are not illustrated for clarity) (a) initial model excluding built environment index (Model 13a) CHILD PERCEPTIONS OF SAFETY  -0.23  PARENT PERCEPTIONS – SOCIAL ENVIRONMENT SAFETY  AGE  PARENT PERCEPTIONS – TRAFFIC ENVIRONMENT SAFETY  GENDER  -0.30  0.15  -0.17 HOUSEHOLD INCOME  +0.1 3  Average Daily MVPA  +0.2 2  ETHNICITY  (b) reduced model retaining only significant predictors of MVPA (Model 13b) PARENT PERCEPTIONS – TRAFFIC ENVIRONMENT  0.18  AGE -0.30 GENDER  -0.15  HOUSEHOLD INCOME  +0.13  Average Daily MVPA  +0.24 ETHNICITY 201  Table 44: Fit Measures for Model 13b Goodness of Fit Measure  Chi square CFI NFI RMSEA  Reference Threshold(s)*  Model  greater than 0.05 greater than 0.95 greater than 0.95 0.05-0.08=good fit 0.08-0.10=moderate fit greater than 0.10=poor fit  0.581 1.000 0.978 0.000  *as recommended by Ho 2006  indicating that perceptions of environmental characteristics are significantly associated with average daily MVPA, but only when ignoring objective measures of built environment characteristics. This may partly explain the unanticipated discrepancy between this and studies  relying solely on measures of perceptions as explanatory variables. However, some other studies have found significant influences of perceptions even when simultaneously accounting for objective measures, including McDonald (2007b) and Kerr et al. (2006). A limitation of Model 12 that requires further consideration is that the variables assessing objective characteristics and perceptions of the neighbourhood environment did not directly assess the same features. That is, while objective measures focused on physical access, measures of perceptions were centered on parent and child perceptions of safety. While causal linkages were hypothesized between physical access as gauged by the built environment index and measures of perceived safety (S. 4.6.2), stronger evidence of the mediating influence of perceptions on built environment – physical activity relationships may have been demonstrated if measures of perceptions more directly corresponding to the built environment index were incorporated in the model. More generally, future research could benefit from such an approach. For example, objective measures of traffic safety such as measures of traffic volume, 202  speed and collision rates could be incorporated within a modeling framework similar to Model 12 together with measures of perceptions of traffic safety. Similarly, models incorporating objective measures of access could be coupled with measures of perceived access to neighbourhood amenities in this type of modeling framework. The results of such models could be used to more clearly assess the extent to which perceptions mediate built environment physical activity relationships. Another possible explanation for the inconsistency in findings is that the neighbourhood environments encompassed in this study may not be sufficiently different in terms of parental perceptions of safety to detect a significant influence of perceptions. This argument may hold for parental perceptions of the social environment (S7-S9), given that relatively small proportions of parents agreed or strongly agreed with statements expressing concerns about social safety (Table 38). However, substantial variation across response categories for all other measures suggests that this argument does not otherwise apply. In sum, differences in findings between the present and previous studies may be due to a variety of specific methodological differences (e.g. different measures used), or may reflect real differences due to unique characteristics of the specific populations studied.  4.7 Research Question 7 4.7.1 Research Question and Hypotheses Investigated This section summarizes results of models designed to investigate Research Question 7: Does a child’s mode of transportation to school mediate associations between built environment characteristics and: a) MVPA outside of school on school days, and b) average daily MVPA?  203  To investigate this question two hypotheses were investigated: Hypothesis 7.1: Mode of commute to school will strongly mediate associations between built environment characteristics and MVPA outside of school on school days, with children who actively commute engaging in significantly more MVPA outside of school than those who commute inactively. Hypothesis 7.2: Mode of commute to school will weakly mediate associations between built environment characteristics and average daily MVPA, with children who actively commute engaging in moderately higher levels of MVPA overall. Hypothesis 7.1 is premised upon active modes of commuting such as walking, biking and rollerblading incorporating MVPA and therefore contributing to MVPA levels outside of school. However, such increases in MVPA outside of school may be compensated for by reductions in MVPA during school. In addition, active transportation to school is only one mechanism through which the built environment may influence overall MVPA. Other specific types of activities which may be influenced by urban form and may in turn influence overall MVPA include active transportation to other destinations, unstructured play and structured play at specialized facilities. Given these considerations, mode of commute to school is anticipated to more weakly mediate average daily MVPA (Hypothesis 7.2).  4.7.2 Analytical Framework and Methods In contrast to the mediation models applied to research Question 6, Structural Equation Models based on Maximum Likelihood estimation are not suitable for the present analysis because the mediating variable is dichotomous (mode of commute to school, coded as active/inactive). Instead, mediation is assessed by adapting the four step approach proposed by Baron and Kenny (1986) to the present analysis. In this adaptation, GEE will be used to account for clustering of students within schools, in contrast to Baron and Kenny’s original 204  model based on the use of OLS regression. The conceptual model for the Baron and Kenny approach is illustrated in Figure 34. Figure 34: Mediation Models 14a Predicting MVPA Outside of School on School Days, and 14b Predicting Overall Average Daily MVPA BUILT ENVIRONMENT INDEX AGE, GENDER, ETHNICITY, HOUSEHOLD INCOME  a) MVPA outside of school on school days b) Overall average daily MVPA  C  A  B MODE OF TRANSPORTATION TO SCHOOL  According to Baron and Kenny, the following three steps are required to assess mediation: 1) Regress the mediator on the independent variable (path A). In the present case, this was done using GEE with a logistic link function, because mode of transportation to school is a dichotomous variable. 2) Regress the dependent variable on the independent variable (path C). 3) Regress the dependent variable on both the independent variable and the mediator (path B). Steps two and three above were implemented using GEE with an identity link function as with previous models. In all cases, relevant covariates are controlled for, including ethnicity entered as a dichotomous variable (1=North American/European, 0=Other).  205  4.7.3 Results and Discussion Following the approach described above, model results are outlined in Tables 45 and 46 corresponding to Hypothesis 7.1 and 7.2, respectively. Considering Hypothesis 7.1 first, the results for Step 1 indicate that mode of transportation is significantly predicted by the built environment index, with higher index values corresponding to an increased likelihood of active transportation (odds ratio 1.56). This is as expected, considering that all components of the built environment index gauge access, and in particular that one component of the index is distance to school. In contrast, none of age, gender, ethnicity or household income significantly predict mode of transportation. The results for Step 2 in turn indicate that the built environment index significantly predicts average daily MVPA outside of school, when controlling for age gender, ethnicity and household income. Finally, Step 3 results indicate that when controlling for mode of commute to school, the built environment index no longer significantly predicts average daily MVPA outside of school. Taken together, these are consistent with Hypothesis 7.1, that mode of commute to school substantially mediates associations between built environment characteristics and MVPA outside of school on school days. This mediation effect may be a direct result of physical activity engaged in during the commute, but may also possibly arise due to other forms of physical activity en route (e.g. if children visit friends or play in parks on the way home from school). The positive association between mode of commute to school and average daily MVPA outside of school (B=11.75*** in Step 3) further indicates that students who regularly commute to school using active modes also engage in more average daily MVPA outside of school. The reduction in magnitude of the regression parameter for the built environment index, and its loss in significance with the inclusion of the mode of commute to school (0.46, not significant in Step 3 compared with 1.53** in Step 2) suggests a moderate  206  Table 45: Mediation Model Predicting MVPA Outside of School on School Days Results (Model 14a) Parameter Parameter Estimates Step 1 (expressed as Step 2 Step 3 odds ratios for active transportation) built environment 1.56*** 1.53** 0.46 index mode of commute to 11.75*** school age 1.05 -7.91* -8.03* gender = female 0.79 -5.40 -4.93 ethnicity = North 0.75 13.82*** 15.62*** American/European household income 0.90 0.40 0.58 *significant at p <.05, **significant at p <.01, ***significant at p <.001  Table 46: Mediation Model Predicting Overall Average Daily MVPA Results (Model 14b) Parameter  built environment index mode of commute to school age gender = female ethnicity = North American/European household income  Parameter Estimates Step 1 (expressed as Step 2 odds ratios for active transportation) 1.56*** 2.92*** -  -  Step 3  2.87*** 1.16  1.05 0.79 0.75  -19.60*** -12.06* 20.74***  -19.60*** -11.93* 20.64***  0.90  2.32***  2.34***  *significant at p <.05, **significant at p <.01, ***significant at p <.001  mediating effect. However, mediation is not complete because the estimate for the built environment index is non-zero (Baron and Kenny 1986). Considering Hypothesis 7.2 (Table 46), the results for Step 1 are the same as previously (this path is identical in both mediation models), with the built environment index significantly predicting active transportation. Similarly, the built environment index is also found to be a 207  significant predictor of average daily MVPA when controlling for age, gender, ethnicity and household income (as shown in previous models). Finally, the results of Step 3 indicate that while very slightly reduced in magnitude, the parameter estimate for the built environment index remains highly significant following the inclusion of mode of commute to school (2.87*** in Step 2 compared to 2.92*** in Step 3). This result supports Hypothesis 7.2 because the slight reduction in parameter estimates suggests weak mediation.  208  5. DISCUSSION AND CONCLUSION 5.1 Overview This chapter begins with a summary and synthesis of the results of the previous two chapters. This summary highlights major findings and presents an interpretation of what the findings imply about causal mechanisms connecting neighbourhood characteristics to physical activity patterns. Next, the strengths and limitations of the study are considered. Limitations relating both to generalizability of the findings and broader methodological considerations are highlighted in this section. This discussion is followed by an overview of implications for future research, and then an overview of implications for planning policy and practice. The chapter closes with a brief concluding statement.  5.2 Summary and Synthesis of Results 5.2.1 Major Findings Tables 47 and 48 summarize results pertaining to sample 1 and two, respectively. Starting with sample 1, Research Question 1, relationships between overall average daily minutes of MVPA and built and social environment measures were investigated. Results indicate that overall MVPA is significantly predicted by age, gender and ethnicity, with younger children, boys and North American/European children engaging in higher levels of MVPA than their counterparts. In addition, in single covariate models, measures of a variety of built environment variables related to access (Figure 6) were found to be significantly associated with overall MVPA, in the directions hypothesized. These include measures corresponding to the route between home and school (e.g. distance to school), and the home neighbourhood environment (e.g. net commercial density). The strongest associations for the latter measures  209  Table 47: Summary of Results Based on Sample 1. GEE was used to investigate all research questions for this sample. Research Dependent Question Variable(s)  Independent Mediators/ Variables Moderators  Major Findings  1. overall average Primary daily minutes of Research MVPA Question  built and social environment measures; age, gender and ethnicity  none  2. MVPA outside of school: a) on school days, and b) on weekend days  built and social environment measures; age, gender and ethnicity  none  3. overall average  built environment index (derived via question 1); age, gender and ethnicity  age, gender as moderators  built environment index; age, gender and ethnicity  none  MVPA is significantly predicted by: o controls: age (-), gender (- for female) and ethnicity (+ for North American/European), which together account for 24.0% of its variation o measures of distance to school (-), distance to closest non-park recreation site (-), net commercial density (+) and intersection density (+), which account for 5.0% of its variation when entered as a built environment index, beyond what is explained by control variables. stronger associations are found for larger buffer size measures average daily MVPA outside of school on school days is significantly predicted by distance to school (-) average daily MVPA on weekends is significantly predicted by: o average intersection spacing (+) and total number of four way intersections (-) en route to school, and distance to non-park recreation sites (-), consistent with hypotheses o net commercial density (-), net residential density (-), and population density of children (-), contrary to hypotheses age does not moderate the relationship between the built environment index and MVPA gender moderates both the relationship between: o age and MVPA, such that the influence of being female on MVPA decreases with increasing age (differences in MVPA between older girls and boys are less than differences in MVPA for younger girls and boys) o the built environment index and MVPA, such that differences in the built environment index are associated with larger differences in MVPA for boys than for girls in gender stratified models, built environment characteristics account for 10.6% of variation in MVPA for boys and 6.2% for girls the built environment index is significantly associated with proportion of time spent in: sedentary activity (-), MPA (+), and VPA (+) the index is not significantly associated with proportion of time spent in LPA the index is most strongly associated with VPA (R2=4.7%)  daily minutes of MVPA  4. proportion of time spent in sedentary activity and LPA, MPA, VPA  210  Table 48: Summary of Results Based on Sample 1B. Research Methods Question  Dependent Independent Variable(s) Variables  Mediators/ Moderators none  previously estimated parameter estimates based on model without household income and car ownership (Research Question 1) are largely unchanged when the additional controls are included household income is significantly positively associated with overall MVPA car ownership is not significantly associated with overall MVPA  parent and child perceptions of safety as mediators  in a model excluding the built environment index, parent perceptions of the traffic environment were found to significantly predict MVPA, with greater perceived safety corresponding to more MVPA An SEM model incorporating both an objective built environment measure (the built environment index) and measures of parent and child perceptions produced very poor overall fit. In this model, only the built environment index and control variables significantly predict MVPA (perceptions do not). mode of transportation to school substantially mediates the relationship between the built environment index and MVPA outside of school on school days, with children who actively commute engaging in significantly more MVPA outside of school mode of transportation to school weakly mediates the relationship between the built environment index and overall MVPA, with children who actively commute to school engaging in slightly more MVPA outside of school (non significant)  5.  GEE  overall average daily minutes of MVPA  6.  SEM  overall average daily minutes of MVPA  7.  GEE with Baron/ Kenny mediation models  a) MVPA outside of school on school days b) overall average daily minutes of MVPA  built environment index; age, gender, ethnicity, household income, car ownership built environment index; age, gender, ethnicity, household income  built environment index; age, gender, ethnicity, household income  mode of transportation to school as mediator  Major Findings  211  were found at larger buffer sizes. When entered as a compound ‘built environment index’, these neighbourhood built environment measures were found to account for 5% of the variation in overall MVPA, beyond the variation explained by control variables. In contrast to these findings, certain other built environment measures were not found to be significantly associated with overall MVPA (e.g. cul-de-sac density and proportion of streets with posted speed limits of 30 km/h o