Copyright is owned by the Author of the thesis. Permission is given for a copy to be downloaded by an individual for the purpose of research and private study only. The thesis may not be reproduced elsewhere without the permission of the Author. DELINEATING NEIGHBOURHOOD AND EXPOSURE IN BUILT ENVIRONMENT AND PHYSICAL ACTIVITY RESEARCH A thesis presented in partial fulfilment of the requirements for the degree of PhD in Public Health at Massey University, New Zealand. Suzanne Mavoa 2015 i ABSTRACT Several decades worth of public health research has shown that characteristics of people’s environment are associated with health-related behaviours and outcomes. Much of this research has used the concept of a residential neighbourhood to delineate the relevant environment. However, there is no uniformity in the neighbourhood delineation methods used in the literature and little consideration is given to whether they adequately capture people’s exposure to the environmental characteristics under investigation, or whether the choice of delineation methods influences results. This dissertation has addressed these issues and suggested some methods researchers may use to delineate spatial context more precisely. The first part of the thesis used data from a study of neighbourhood environment and physical activity in adults to examine the impact of different methodological choices on modelling results. Both neighbourhood delineation method and scale were shown to determine whether significant associations were found between the built environment and physical activity. Modelling results also varied depending on the built environment and outcome measures used. A detailed exploration of different methods of operationalising the road network buffer demonstrated that, even for a single neighbourhood delineation method, methodological choices can alter the results. The second part of the thesis used GPS data from a study of children’s physical activity and independent mobility to examine how well a number of road network buffers and activity space delineation methods represented exposure to the environment. Results showed less than half of children’s seven-day activity was ii captured by residential road network buffers at a range of scales. Most activity space delineations were better representations of where children spent time than road network buffers. However, the measures of activity space commonly used in health research - the convex hull and standard deviation ellipse – were poor representations of exposure. Activity space delineations require detailed location data that is not always available. Therefore, there is a need for delineation methods that do not require this data. Five enhancements to standard road network buffers were proposed. One enhancement - including school and home in the buffer - was tested and shown to be an improvement on standard road network buffers. iii ACKNOWLEDGEMENTS This thesis would not have been possible without data from two Health Research Council of New Zealand funded projects - the URBAN study (grant: 07/356), and Kids in the City study (grant: 10/497) – and the Marsden funded Kids in the City study (grant: 21568 RSNZ). I would also like to acknowledge everyone who contributed to the successful completion of these projects: the URBAN, Neighbourhoods and Health, and Kids in the City research teams; the territorial authorities and spatial data providers; and study participants. My supervisors, Professor Karen Witten and Associate Professor David O’Sullivan, were invaluable; providing consistent feedback and support, while also allowing me the intellectual freedom to develop and pursue my own ideas. Thank you, Karen, for helping me get it done and teaching me to always step back and think about whether the research makes sense. Thank you, David, for your insights and birds eye view; they have allowed me to view my research in new ways. Numerous colleagues have provided support and input throughout my PhD journey. Thanks to colleagues at the SHORE & Whariki Research Centre at Massey University and Place, Health and Liveability research group at the University of Melbourne. The support of Karen Witten, Billie Giles-Corti, and Hannah Badland was instrumental in allowing me to finish this dissertation while working two jobs. Thanks are also due to Jan Sheeran, Lisa Morice, Stephanie Mackenzie and Caroline Lowe, who provided much appreciated assistance with all dissertation related administration, printing, and binding. iv My apologies to Chris, who, despite having no interest in neighbourhoods and health, has acted as my sounding board and has been forced to read multiple drafts of this dissertation. Also, special ‘thanks’ to Cymbal the cat, who has continually yelled at me to hurry up and finish so that I have more time to rub his belly. Last, but by no means least, thanks to my parents, Siti Mavoa and Helen Mavoa, and all my friends and family, for their support and encouragement. v TABLE OF CONTENTS Abstract ....................................................................................................................................................i Acknowledgements ............................................................................................................................... iii Table of Contents .................................................................................................................................... v Figures ....................................................................................................................................................xi Tables ................................................................................................................................................. xiii Abbreviations ..................................................................................................................................... xvii CHAPTER 1. INTRODUCTION ........................................................................................................ 1 1.1 Dissertation goals and aims ............................................................................................................... 2 1.2 Thesis Structure ................................................................................................................................. 4 1.3 General statement of candidate contributions ................................................................................... 5 CHAPTER 2. LITERATURE REVIEW: DELINEATING NEIGHBOURHOOD AND EXPOSURE ....................................................................................................................................... 7 2.1 Introduction ....................................................................................................................................... 7 2.2 Neighbourhood .................................................................................................................................. 7 2.3 Towards improved conceptualisation of context: From neighbourhood to exposure ........................ 9 2.4 Operationalising context ................................................................................................................. 10 2.4.1 Territorial units ................................................................................................................. 12 2.4.2 Ego-centric delineations .................................................................................................... 17 2.4.3 Location-centric ................................................................................................................ 21 2.4.4 Activity spaces .................................................................................................................. 22 2.4.5 Other delineation methods ................................................................................................ 25 vi 2.5 Does the choice of delineation method affect research results? ...................................................... 27 2.6 Theoretical considerations .............................................................................................................. 34 2.6.1 Scale ............................................................................................................................... 34 2.6.2 Fuzzy vs clear-cut ............................................................................................................. 37 2.6.3 Oriented vs isotropic ......................................................................................................... 37 2.6.4 Time ............................................................................................................................... 38 2.6.5 The modifiable areal unit problem and the uncertain geographic context problem .......... 39 2.7 Summary ......................................................................................................................................... 39 CHAPTER 3. THE INFLUENCE OF METHODOLOGICAL CHOICES ON NEIGHBOURHOOD DELINEATION AND RELATIONSHIPS BETWEEN THE BUILT ENVIRONMENT AND PHYSICAL ACTIVITY ............................................................................ 41 3.1 Introduction ..................................................................................................................................... 41 3.2 URBAN Study methods .................................................................................................................. 44 3.2.1 Overview of the URBAN Study ....................................................................................... 44 3.2.2 Study area and neighbourhood selection........................................................................... 44 3.2.3 Participant sampling strategy ............................................................................................ 47 3.2.4 Participants ....................................................................................................................... 48 3.2.5 Demographics, neighbourhood preference, and neighbourhood deprivation.................... 48 3.2.6 Physical activity measures ................................................................................................ 48 3.2.7 URBAN dataset and spatial data sources .......................................................................... 49 3.2.8 Candidate contributions to the URBAN study .................................................................. 52 3.3 Do different neighbourhood delineations change the results of models of the relationship between the built environment and physical activity? ......................................................................................... 53 3.3.1 Methods ............................................................................................................................ 53 vii 3.3.2 Results ............................................................................................................................... 57 3.3.3 Discussion ......................................................................................................................... 67 3.4 Do different buffering algorithms change the neighbourhood definitions?..................................... 72 3.4.1 Methods ............................................................................................................................ 72 3.4.2 Results: .............................................................................................................................. 75 3.4.3 Discussion ......................................................................................................................... 80 3.5 Methodological commentary ........................................................................................................... 82 3.5.1 Selecting units for the built environment measures .......................................................... 82 3.5.2 Measuring street connectivity ........................................................................................... 83 3.6 Discussion and conclusion .............................................................................................................. 87 CHAPTER 4. KIDS IN THE CITY STUDY METHODS ............................................................... 89 4.1 Introduction ..................................................................................................................................... 89 4.2 School selection .............................................................................................................................. 90 4.3 Participant recruitment .................................................................................................................... 92 4.4 Data collection ................................................................................................................................ 92 4.5 GIS, GPS, and accelerometer data processing ................................................................................ 93 4.5.1 GPS ................................................................................................................................ 93 4.5.2 GIS ................................................................................................................................ 94 4.6 Candidate contributions to the KITC study ..................................................................................... 94 CHAPTER 5. GPS INCLUSION CRITERIA .................................................................................. 97 5.1 Introduction ..................................................................................................................................... 97 5.2 Methods – creation of the three GPS datasets ................................................................................. 99 5.2.1 Creating the complete GPS dataset ................................................................................... 99 viii 5.2.2 Creating the subset GPS dataset 1................................................................................... 100 5.2.3 Creating the subset GPS dataset 2................................................................................... 102 5.3 Descriptive statistics for the three GPS datasets ........................................................................... 103 5.3.1 GPS descriptive statistics ................................................................................................ 105 5.4 Discussion and conclusion ............................................................................................................ 107 CHAPTER 6. HOW WELL DO ROAD NETWORK BUFFERS REPRESENT WHERE CHILDREN SPEND TIME? ........................................................................................................... 109 6.1 Introduction ................................................................................................................................... 109 6.2 Methods......................................................................................................................................... 111 6.2.1 Estimating distance travelled from home ........................................................................ 112 6.2.2 Calculating road network buffers around the residential address ................................... 112 6.2.3 Delineating places children went using daily path areas ................................................. 112 6.2.4 Comparing GPS daily path areas and road network buffers ........................................... 114 6.2.5 Missing GPS data ............................................................................................................ 117 6.2.6 Statistical analysis ........................................................................................................... 117 6.3 Results ........................................................................................................................................... 117 6.3.1 Distance from home ........................................................................................................ 117 6.3.2 Area of buffers ................................................................................................................ 120 6.3.3 Overlap of buffers ........................................................................................................... 121 6.3.4 GPS points/time within buffers ....................................................................................... 127 6.3.5 Is there an optimal road network buffer scale? ............................................................... 133 6.4 Discussion ..................................................................................................................................... 134 6.4.1 Limitations of road network buffers ............................................................................... 137 ix 6.4.2 Limitations of analyses ................................................................................................... 146 6.5 Conclusion .................................................................................................................................... 146 CHAPTER 7. HOW WELL DO ACTIVITY SPACE MEASURES REPRESENT WHERE CHILDREN SPEND TIME? ............................................................................................................ 149 7.1 Introduction ................................................................................................................................... 149 7.2 Methods ......................................................................................................................................... 150 7.2.1 Dataset ............................................................................................................................ 150 7.2.2 Creation of activity spaces .............................................................................................. 151 7.2.3 Comparing GPS data and activity spaces ........................................................................ 156 7.2.4 Statistical analysis ........................................................................................................... 156 7.3 Results ........................................................................................................................................... 157 7.4 Discussion and conclusion ............................................................................................................ 161 CHAPTER 8. HOME, SCHOOL AND IN BETWEEN: ENHANCING ROAD NETWORK BUFFER TO BETTER REPRESENT NEIGHBOURHOODS AND EXPOSURE .................... 165 8.1 Introduction ................................................................................................................................... 165 8.2 Potential enhancements to the standard road network buffers ...................................................... 166 8.3 Methods ......................................................................................................................................... 168 8.3.1 Data .............................................................................................................................. 168 8.3.2 Buffer creation ................................................................................................................ 168 8.3.3 Comparing the GPS data and enhanced road network buffers ........................................ 170 8.3.4 Statistical analyses .......................................................................................................... 171 8.4 Results ........................................................................................................................................... 171 8.5 Discussion ..................................................................................................................................... 177 x 8.6 Conclusion .................................................................................................................................... 183 CHAPTER 9. DISCUSSION AND CONCLUSIONS .................................................................... 185 9.1 Summary of findings ..................................................................................................................... 185 9.2 Considerations for future research ................................................................................................ 188 9.2.1 The challenge of identifying an optimal delineation method .......................................... 188 9.2.2 Neighbourhood vs Exposure; Potential vs Actual.......................................................... 189 9.2.3 Measuring ‘true exposure’ .............................................................................................. 190 9.2.4 Whose exposure are we measuring? ............................................................................... 192 9.2.4 Delineation methods need to be considered in combination with methods of representing and measuring the built environment .............................................................................. 192 9.2.5 The importance of time ................................................................................................... 193 9.3 Conclusion .................................................................................................................................... 194 REFERENCES .................................................................................................................................. 195 APPENDIX A. STATEMENT OF CONTRIBUTION TO DATASETS AND PUBLISHED PAPERS ................................................................................................................................... 211 APPENDIX B. BUFFER ALGORITHM COMPARISON RESULTS ........................................ 219 APPENDIX C. BUFFER ALGORITHM MODELLING RESULTS .......................................... 231 xi FIGURES Figure 1. Examples of delineation methods for a theoretical individual. .............................................. 13 Figure 2. Neighbourhood boundaries for an example participant. ........................................................ 55 Figure 3. Illustration of a sausage buffer with a 50 m buffer radius. .................................................... 74 Figure 4. Example park represented as a point and polygon. ................................................................ 81 Figure 5. An intersection that borders four meshblocks is assigned to only one meshblock (e.g., Meshblock 3). ....................................................................................................................... 84 Figure 6. When the meshblock boundary is buffered an intersection that borders four meshblocks is assigned to all four meshblocks. ................................................................... 85 Figure 7. When buffering neighbourhoods, intersections on the borders of neighbourhoods can be counted multiple times..................................................................................................... 86 Figure 8 Boxplot of the distributions of the three GPS datasets created with different inclusion criteria. ............................................................................................................................... 106 Figure 9. Example of a daily path area. ............................................................................................... 114 Figure 10. Illustration of the different measures of geographic overlap. ............................................ 115 Figure 11. Cumulative percentage of time spent at different distances from home using the complete GPS dataset (n = 236). ........................................................................................ 118 Figure 12. Cumulative percentage of time spent at different distances from home using the GPS dataset with inclusion criteria applied (n = 85). ......................................................... 119 Figure 13. Median overlap, commission error, and omission error areas at different road network buffer distances. Complete dataset (n = 236). ...................................................... 122 Figure 14. Median overlap, commission error, and omission error areas at different road network buffer distances. Subset GPS dataset (n = 85). ..................................................... 126 Figure 15. Participant with no GPS data in the 400 m and 600 m road network buffers. ................... 132 Figure 16. Additional road network buffer limitations – part 1. ......................................................... 142 Figure 17. Additional road network buffer limitations – part 2. ......................................................... 142 Figure 18. Additional road network buffer limitations – part 3. ......................................................... 143 Figure 19. Examples of limitations of road network buffers. .............................................................. 145 Figure 20. Example buffers and activity spaces for a single participant compared to the 50 m GPS buffer daily path area: a) road network buffers at 400, 800, 1600 m, b) convex xii hull based activity spaces (CH, TLCH50, TLCH75, TCH95), and c) activity spaces (SDE, KDE12, KDE09, KDE07, KDE05) ......................................................................... 154 Figure 21. Example buffers and activity spaces for a single participant compared to the 50 m GPS daily path area: a) road network buffers at 400, 800, 1600 m, b) convex hull based activity spaces (CH, TLCH50, TLCH75, TCH95), and c) activity spaces (SDE, KDE12, KDE09, KDE07, KDE05) ......................................................................... 155 Figure 22. Example enhanced road network buffer. ........................................................................... 170 Figure 23. Implementing an oriented road network buffer step 1/3. ................................................... 180 Figure 24. Implementing an oriented road network buffer step 2/3. ................................................... 181 Figure 25. Implementing an oriented road network buffer part 3/3. ................................................... 182 xiii TABLES Table 1. Main types of delineation methods. ........................................................................................ 11 Table 2. Research comparing the effect of different delineation methods. ........................................... 29 Table 3. Relevant items from the URBAN dataset. .............................................................................. 50 Table 4. URBAN study spatial data and sources. ................................................................................. 52 Table 5. Neighbourhood boundary size descriptive statistics. .............................................................. 58 Table 6. Built environment descriptive statistics for neighbourhood boundaries (n = 1989 adults) ................................................................................................................................... 59 Table 7. Descriptive statistics for the physical activity outcome measures. .......................................... 61 Table 8. Percentage change (95% CI) and R2 (marginal/conditional) in fully adjusted models of physical activity, for a 1 unit change in the built environment measures across the seven neighbourhood boundaries. ........................................................................................ 63 Table 9. ArcGIS service area types. ...................................................................................................... 73 Table 10. Comparison of median street connectivity measures (intersections/km2) for buffered and non-buffered meshblocks. ............................................................................................. 87 Table 11. Characteristics of participating schools. Source: Ministry of Education 2010 (NZ Ministry of Education 2010). ............................................................................................... 91 Table 12. Characteristics of the three GPS datasets. ........................................................................... 103 Table 13. GPS data descriptive statistics. ............................................................................................ 106 Table 14. Descriptive statistics of the areas (km2) of GPS daily paths and road network buffers (RNBs) at different scales. Complete GPS dataset (n = 236). ............................................ 120 Table 15. Descriptive statistics of the areas (km2) of GPS daily paths and road network buffers (RNBs) at different scales. GPS dataset with inclusion criteria applied (n = 85). .............. 121 Table 16. Comparing the spatial extent of GPS daily paths with road network buffers (RNB): overlap and errors of commission and omission. Complete GPS dataset (n = 236). .......... 123 Table 17. Comparing the spatial extent of GPS daily paths with road network buffers (RNB): overlap and errors of commission and omission. Subset GPS dataset (n = 85). ................. 124 Table 18. Hours of non-vehicle GPS data recorded inside and outside the road network buffers (RNB). Complete GPS dataset (n = 236). .......................................................................... 129 Table 19. Hours of non-vehicle GPS data recorded inside and outside the road network buffers (RNB). Subset GPS dataset (n= 85). .................................................................................. 130 xiv Table 20. Combined overlap measures for the complete GPS dataset (n = 236) and the subset GPS dataset (n = 85). ......................................................................................................... 134 Table 21. Activity spaces methods. ..................................................................................................... 151 Table 22. Descriptive statistics of the areas (km2) of GPS daily path areas and different activity spaces. Subset GPS dataset (n = 85). ................................................................................. 157 Table 23. Comparing the spatial extent of GPS daily paths with activity spaces: overlap and errors of commission and omission. Subset GPS dataset (n = 85). .................................... 159 Table 24. Composite measures of activity space overlap with daily path areas. Subset GPS dataset (n = 85). .................................................................................................................. 160 Table 25. Hours and percentage of non-vehicle GPS data recorded within the activity spaces. Subset GPS dataset (n= 85). ............................................................................................... 163 Table 26. Enhanced road network buffer scale combinations. ............................................................ 169 Table 27. Comparison of the spatial extent of GPS daily paths with enhanced road network buffers: overlap and errors of commission and omission. Subset GPS dataset (n = 85). ..................................................................................................................................... 173 Table 28. Composite measures of road network buffer, activity space, and enhanced road network buffer overlap with daily path areas. Subset GPS dataset (n = 85). ..................... 174 Table 29. Descriptive statistics for the area (km2) of the different types of buffer at different scales. ................................................................................................................................. 219 Table 30. Spearman rank correlation coefficients (2 dp, α = 5%, p < 0.001) comparing the area of different buffer types at a range of scales. ..................................................................... 221 Table 31. Spearman rank correlation coefficients (2 dp, α = 5%, p < 0.001) comparing intersection counts (Cnt) and intersection densities (Dns) for different buffer types at a range of scales. ................................................................................................................ 223 Table 32. Spearman rank correlation coefficients (2 dp, α = 5%, p < 0.001) comparing bus stop count for different buffer types at a range of scales. .......................................................... 225 Table 33. Spearman rank correlation coefficients (2 dp, α = 5%, p < 0.001) comparing dwelling count for different buffer types at a range of scales. .......................................................... 227 Table 34. Spearman rank correlation coefficients (2 dp, α = 5%, p < 0.001) comparing park area and % park area for different buffer types at a range of scales. .................................. 229 Table 35. Percentage change in physical activity for a 1dph increase in dwelling density across a range of road network buffer types and scales. Bold text indicates a significant association. ......................................................................................................................... 231 Table 36. Percentage change in physical activity for a 1 bus stop increase for a range of road network buffers and scales. Bold text indicates a significant association. ......................... 233 xv Table 37. Percentage change in physical activity for a 1 intersection per square kilometre increase in street connectivity for a range of road network buffers and scales. Bold text indicates a significant association. .............................................................................. 235 Table 38. Percentage change in physical activity for a 1 ha increase in park area for a range of road network buffers and scales. Bold text indicates a significant association. ................. 237 Table 39. Association between percentage park area and physical activity for a range of road network buffers and scales. Bold text indicates a significant association. ......................... 239 xvi xvii ABBREVIATIONS BMI Body Mass Index CATI Computer-Aided Telephone Interview CAU Census Area Unit CBD Central Business District CCD Census Collection District CVD Cardiovascular Disease DA Dissemination Area DPA Daily Path Area dph Dwellings per hectare ED Enumeration District GIS Geographic Information Systems GPS Global Positioning System Ha Hectare IPEN International Physical Activity and Environment Network KDE Kernel Density Estimation KITC Kids In The City km Kilometres m Metres xviii MAUP Modifiable Areal Unit Problem MB Meshblock mi Mile MVPA Moderate-Vigorous Physical Activity NA Not Applicable NDAI Neighbourhood Destination Accessibility Index NO2 Nitrogen Dioxide RNB Road Network Buffer s Seconds SD Standard Deviation SDE Standard Deviation Ellipse SES Socio-Economic Status UGCoP Uncertain Geographic Context Problem UK United Kingdom URBAN Understanding the Relationship Between physical Activity and Neighbourhood