UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

The food environment surrounding Vancouver schools : associations of access to food outlets and children's… Daepp, Madeleine I. G. 2016

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


24-ubc_2016_september_daepp_madeleine.pdf [ 20.75MB ]
JSON: 24-1.0308712.json
JSON-LD: 24-1.0308712-ld.json
RDF/XML (Pretty): 24-1.0308712-rdf.xml
RDF/JSON: 24-1.0308712-rdf.json
Turtle: 24-1.0308712-turtle.txt
N-Triples: 24-1.0308712-rdf-ntriples.txt
Original Record: 24-1.0308712-source.json
Full Text

Full Text

The Food Environment SurroundingVancouver SchoolsAssociations of Access to Food Outlets and Children’sIntake of Minimally Nutritious Foods At or En-Route toSchoolbyMadeleine I. G. DaeppB.A., Washington University in St. Louis, 2013A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFMASTER OF SCIENCEinThe Faculty of Graduate and Postdoctoral Studies(Integrated Studies in Land & Food Systems)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)August 2016© Madeleine I. G. Daepp 2016AbstractBackground: Canada has seen a dramatic increase in the prevalence ofchildhood obesity in recent decades. Researchers have argued that this prob-lem could be addressed through improvements to the “food environment”—the food vendors comprised in the built environment. Children’s diets arepoorer in nutritional quality during school hours, suggesting that the foodenvironments surrounding schools may be an important area of inquiry.Objectives: This thesis sought (1) to identify the best available data set forassessing the distributions of food outlets in Vancouver, (2) to characterizethe food environments surrounding Vancouver public schools, testing fordemographic or socioeconomic disparities in access and (3) to examine theassociations between school food environments and the dietary intakes ofchildren and adolescents at- or en-route to school.Methods: Food outlet data were obtained from two municipal and twocommercial sources and validated against primary data on the food outletslocated within 800m of 26 schools. Outlet density and proximity to Van-couver schools (n=113) were evaluated with the best performing data set;negative binomial regression models examined whether disparities existediiAbstractin environments according to % aboriginal students, % English LanguageLearners, and school poverty, controlling for neighbourhood-level factors.Multilevel logistic regression analyses evaluated the associations of schoolfood environment measures and 950 children’s odds of daily consumption,at or en-route to schools (n=26), of minimally nutritious foods.Results & Conclusions: The City of Vancouver Business Licenses datahad the highest sensitivity (0.69) and positive predictive value (0.55). High-poverty schools had more convenience stores within 400m than low-povertyschools, even after controlling for commercial density and neighbourhoodsocioeconomic deprivation (IRR=1.74, 95% CI 1.003 - 3.032); no robuststatistically significant relationships were identified between school food en-vironments and school-level demographic factors. No consistent associationswere identified between school food environment measures and students’ in-takes of minimally nutritious foods. The findings do not support policies toreduce student access to food outlets near schools.iiiPrefaceThis thesis is my original work, completed under the supervision of Dr.Jennifer Black with the guidance of committee members Carol McAuslandand Nadine Schuurman. I designed and implemented the ground-truthingprotocol used in this study in collaboration with Dr. Cayley Velazquez, andboth Dr. Velazquez and Koharu Loulou Chayama assisted in the collection ofprimary food outlet location data. I analyzed the data and wrote the thesisindependently with guidance from Dr. Jennifer Black and the members ofmy supervisory committee.The project was a component of the Food Practices on School DaysStudy, which was overseen by Dr. Jennifer Black and Dr. Gwen Chapman.The third component of this thesis, the study of students’ dietary intakes,relied on data from classroom surveys designed and conducted by NaseamAhmadi, Teya Stephens, and Dr. Cayley Velazquez with the supervision ofDr. Jennifer Black and Dr. Gwen Chapman.The procedures for the Food Practices on School Days Study were ap-proved by the Behavioural Research Ethics Board of the University of BritishColumbia (certificate number H11-01369); in addition, sampling and datacollection within Vancouver public schools received permission from the Van-couver School Board.ivTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . vList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ixList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xivAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . xvDedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvi1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . 31.1.1 The Food Environment and Obesity . . . . . . . . . . 31.1.2 School Food Environments . . . . . . . . . . . . . . . 71.1.3 Gaps in the Existing Literature . . . . . . . . . . . . 121.2 Study Objectives . . . . . . . . . . . . . . . . . . . . . . . . . 152 The Validation of Food Environments Data . . . . . . . . . 16vTable of Contents2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 162.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182.3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . 242.3.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . 242.3.2 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 332.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352.4.1 Evaluation of Sensitivity, PPV, and Concordance . . 352.4.2 Assessment of Systematic Bias . . . . . . . . . . . . . 372.4.3 Comparison of Food Environment Measures . . . . . 382.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 412.5.1 Recommendations . . . . . . . . . . . . . . . . . . . . 442.5.2 Strengths and Limitations . . . . . . . . . . . . . . . 452.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 463 Associations of School Characteristics and the Food Envi-ronments Surrounding Schools . . . . . . . . . . . . . . . . . 483.1 Introduction & Background . . . . . . . . . . . . . . . . . . . 483.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523.2.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . 523.2.2 Food Environment Measures . . . . . . . . . . . . . . 563.2.3 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 583.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 603.3.1 Descriptive Statistics . . . . . . . . . . . . . . . . . . 603.3.2 Results from Negative Binomial Regression Models . 653.3.3 Sensitivity Analyses . . . . . . . . . . . . . . . . . . . 69viTable of Contents3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 733.4.1 Endogeneity Concerns in Food Environments Research 783.4.2 Strengths and Limitations . . . . . . . . . . . . . . . 793.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 814 Associations of School Food Environment Measures and Chil-dren’s School-Day Dietary Intakes . . . . . . . . . . . . . . . 834.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 834.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 874.2.1 Dependent Variables: Dietary Intake . . . . . . . . . 884.2.2 Independent Variables: Food Environment Measures . 904.2.3 Independent Variables: Controls . . . . . . . . . . . . 924.2.4 Data Analysis . . . . . . . . . . . . . . . . . . . . . . 954.2.5 Sensitivity Analyses . . . . . . . . . . . . . . . . . . . 974.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 984.3.1 Sensitivity Analyses . . . . . . . . . . . . . . . . . . . 1074.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1094.4.1 Limitations . . . . . . . . . . . . . . . . . . . . . . . . 1114.4.2 Strengths . . . . . . . . . . . . . . . . . . . . . . . . . 1124.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1135 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1145.1 Contributions and Significance . . . . . . . . . . . . . . . . . 1145.2 Strengths and Limitations of the Research . . . . . . . . . . 1165.3 Avenues for Future Research . . . . . . . . . . . . . . . . . . 1195.3.1 Understanding Adolescents’ Dietary Behaviours . . . 119viiTable of Contents5.3.2 Methodological Improvements . . . . . . . . . . . . . 1205.4 Policy Relevance and Implications . . . . . . . . . . . . . . . 122Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124AppendicesA Ground-Truthing Protocol and Classification Scheme . . . 158B Additional Tables for Chapter 2 . . . . . . . . . . . . . . . . . 173C Additional Tables for Chapter 3 . . . . . . . . . . . . . . . . . 178D Additional Tables for Chapter 4 . . . . . . . . . . . . . . . . . 184viiiList of Tables2.1 Measures of dataset validity . . . . . . . . . . . . . . . . . . . 192.2 Sources of data for food outlet locations in Vancouver, BC . . 252.3 Methods used to classify listings as limited-service outlets,convenience stores, and grocery stores . . . . . . . . . . . . . 302.4 Sensitivity, positive predictive value (PPV), and concordanceof two municipal and two commercial data sources in compar-ison with ground-truthed data for the locations of food outletsin Vancouver, BC . . . . . . . . . . . . . . . . . . . . . . . . . 372.5 Density of food outlets within 800 metres of Vancouver pub-lic schools (n=26), evaluated across data sources: Summarystatistics and correlations with measures constructed fromgold standard data . . . . . . . . . . . . . . . . . . . . . . . . 392.6 Proximity in metres of food outlets to Vancouver public schools(n=26), evaluated across data sources: Summary statisticsand correlations with measures constructed from gold stan-dard data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403.1 Limited-service outlets identified as “major chains” . . . . . . 57ixList of Tables3.2 Descriptive statistics for school and school neighbourhood char-acteristics for all public schools (n=113) in operation duringthe 2011/2012 academic year in Vancouver, BC . . . . . . . . 613.3 Descriptive profile of food environments around Vancouverschools (n=113) between March and June 2012 . . . . . . . . 643.4 Results from multivariate negative binomial regressions withfood outlet densities within 400m of Vancouver schools (n=113)as dependent variables and student socioedemographic factorsas independent variables . . . . . . . . . . . . . . . . . . . . . 653.5 Results from multivariate negative binomial regressions withfood outlet densities within 400m of Vancouver schools (n=113)as dependent variables and student socioedemographic fac-tors as independent variables, adjusted for school controls andneighbourhood factors . . . . . . . . . . . . . . . . . . . . . . 673.6 Results from multivariate ordinary least squares regressionswith food outlet proximities Vancouver schools (n=113) asdependent variables and student socioedemographic factorsas independent variables . . . . . . . . . . . . . . . . . . . . . 703.7 Results from multivariate ordinary least squares regressionswith food outlet proximities (metres) to Vancouver schools(n=113) as dependent variables and student socioedemographicfactors as independent variables, adjusted for school controlsand neighbourhood factors . . . . . . . . . . . . . . . . . . . . 71xList of Tables4.1 Minimally nutritious intake categories and their componentfood items . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 904.2 Individual Eating Assessment Tool: Food insecurity module . 934.3 Sample characteristics (n=950) . . . . . . . . . . . . . . . . . 994.4 Descriptive statistics for objective density, objective proxim-ity and perceived proximity of food outlets in association toVancouver Schools . . . . . . . . . . . . . . . . . . . . . . . . 1014.5 Bivariate associations of outlet proximity and students’ dailyintakes of minimally nutritious foods at or en-route to schoolfrom multilevel logistic regression models . . . . . . . . . . . . 1034.6 Multivariate adjusted associations from multilevel logistic mod-els of outlet proximity and students’ odds of daily intake ofminimally nutritious foods at or en-route to school . . . . . . 105B.1 Name-based classifications system applied to identify majorstore types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174B.2 Bivariate associations of commercial density or socioeconomicstatus and the odds of false positive listings in each secondarydata source . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176B.3 Bivariate associations of commercial density or socioeconomicstatus and the odds of false negative listings in each secondarydata source . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177xiList of TablesC.1 Results from negative binomial regressions with food outletdensities within 800m of Vancouver schools (n=113) as de-pendent variables and student socioedemographic factors asindependent variables . . . . . . . . . . . . . . . . . . . . . . 178C.2 Results from negative binomial regressions with food outletdensities within 160m of Vancouver schools (n=113) as de-pendent variables and student socioedemographic factors asindependent variables . . . . . . . . . . . . . . . . . . . . . . 179C.3 Results from negative binomial regressions with food outletdensities within 800m of Vancouver schools (n=113) as de-pendent variables and student socioedemographic factors asindependent variables, adjusted for school controls and neigh-bourhood factors . . . . . . . . . . . . . . . . . . . . . . . . . 180C.4 Results from negative binomial regressions with food outletdensities within 160m of Vancouver schools (n=113) as de-pendent variables and student socioedemographic factors asindependent variables, adjusted for school controls and neigh-bourhood factors . . . . . . . . . . . . . . . . . . . . . . . . . 181C.5 Results from negative binomial regressions with major chainlimited-service outlet densities surrounding Vancouver schools(n=113) as dependent variables and student socioedemographicfactors as independent variables . . . . . . . . . . . . . . . . . 182xiiList of TablesC.6 Results from negative binomial regressions with major chainlimited-service outlet densities surrounding Vancouver schools(n=113) as dependent variables and student socioedemographicfactors as independent variables, adjusted for school and neigh-bourhood factors . . . . . . . . . . . . . . . . . . . . . . . . . 183D.1 Bivariate associations of outlet density within 400m and stu-dents’ daily intakes of minimally nutritious foods at or en-route to school . . . . . . . . . . . . . . . . . . . . . . . . . . 185D.2 Bivariate associations of outlet density within 800m and stu-dents’ daily intakes of minimally nutritious foods at or en-route to school . . . . . . . . . . . . . . . . . . . . . . . . . . 186D.3 Bivariate associations of perceived outlet proximity and oddsof daily intake, at or en-route to school, of minimally nutri-tious foods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187D.4 Multivariate adjusted associations from multilevel logistic mod-els of outlet proximity and students’ odds of daily intake ofminimally nutritious foods at or en-route to school, completecase analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189xiiiList of Figures3.1 Flow chart of of 2012 Business License Data cleaning process 553.2 Socioeconomic Deprivation and Access to Grocery Stores orSupermarkets within 400m of Vancouver Public Schools . . . 62xivAcknowledgementsI am grateful to all of the people who supported me during this research. Ifeel so fortunate to have had a supervisor, Jennifer Black, who was endlesslysupportive in this process; thank you so much for your careful guidance.I would also like to thank my committee members, Carol McAusland andNadine Schuurman, for their thoughtful comments.I was fortunate to have the support of the Li Tze Fong Fellowship as wellas support from the UBC Faculty of Land & Food Systems. This projectalso received funding from the Canadian Institutes of Health Research. I amgrateful for the opportunities these resources made possible.Thank you to my parents, Pam and Ueli, for offering me the world, and tomy brother, Hannes, for finding the humor in stressful moments. To Charlie,for feeding me vegetables.To my ISLFS friends, my LFS 250 TA crew, and my PHUN family, thankyou so much for your support. I am grateful for the insights and laughter Ishared with my labmates Adrienne, Ellie, Cayley, and Claire.Finally, I was so glad to have spent these last two years with Susanna andMartina—the morning coffees at Kafka’s, the afternoon beers at Koerner’s,and evening potlucks we shared at Wreck beach were critical to this thesisprocess. I could not have asked for better ISLFS partners-in-crime.xvTo Pam and UelixviChapter 1IntroductionCanadian children are failing to meet the dietary intake guidelines ofthe 2007 Canada’s Food Guide. In 2004, the most recent iteration of thenationwide Canadian Community Health Survey, a majority of Canadianteenagers failed to meet the Guide’s minimum recommended daily numberof fruit and vegetable servings (Black and Billette, 2013). Among adolescentsages 14-18, 53% of boys and 35% of girls consumed a soft drink during theday prior to the survey (Garriguet, 2008), and over 80% of girls and 90% ofboys were estimated to have a daily sodium intake high enough to increasethe risks of hypertension and other health consequences (Garriguet, 2007).These dietary intake patterns likely contribute to Canada’s high prevalence ofdiet-related disease: nearly a third of 5- to 17-year-old children were classifiedas overweight (19.8%) or obese (11.7%) in the 2009-2011 Canadian HealthMeasures Survey (Roberts et al., 2012).Researchers and public health practitioners have argued that improve-ments to the “food environment”—the grocery stores, restaurants, and otherfood sources comprised in an area’s built environment—could help mitigatethe rise of obesity (Morland et al., 2002; Papas et al., 2007; Brownell andHorgen, 2004; Black and Macinko, 2008). While many factors likely con-1Chapter 1. Introductiontribute to diet-related disease, studies have reported desirable diet-relatedhealth outcomes for people living in census tracts with grocery stores or su-permarkets, as compared to those with more limited access (Morland et al.,2002, 2006; Morland and Evenson, 2009) and higher obesity rates as well aslower diet quality for people with high access to convenience stores or fastfood restaurants, as compared to those with less easy fast- and snack foodaccess (Morland and Evenson, 2009; Maddock, 2004; Rummo et al., 2014).In recent years, researchers have begun to examine the food environmentsurrounding schools. In the United States, 1 in 3 schools is located withinwalking distance (approximately 800 metres) of a convenience store or fastfood outlet (Zenk and Powell, 2008), and a majority of public schools inBritish Columbia are located within walking distance of a fast food out-let, snack food outlet, convenience store, or deli (Black and Day, 2012). Anumber of studies have found that higher access to fast food restaurants orconvenience stores at school is associated with creased dietary quality amongstudents (He et al., 2012b; Laska et al., 2010; Davis and Carpenter, 2009),but several studies have produced conflicting results (An and Sturm, 2012;Gebremariam et al., 2012; Van Hulst et al., 2014; Richmond et al., 2013).The Canadian school food environment may have a particularly strongimpact on students’ diets. Canada is the only G8 country without a federalschool lunch program: while provincial and municipal programs and ad-hoccharity efforts offer lunches in some Canadian schools, a dearth of subsi-dized school cafeterias leaves many students particularly susceptible to thewares of nearby food vendors. Despite this unique policy context, researchon the effects of the Canadian school food environment remains limited.21.1. Literature ReviewSeveral studies have examined the associations of Canadian students’ ac-cess to food retailers and obesity (Seliske et al., 2009a; Héroux et al., 2012;Leatherdale et al., 2011) or food purchasing behaviours (Seliske et al., 2013;He et al., 2012a; Héroux et al., 2012), but few researchers have looked at theassociations between food environment measures and Canadian children’sschool-day dietary behaviours (Laxer and Janssen, 2013; He et al., 2012b)This study sought to fill the gap in Canadian school food environmentsresearch through an examination relating school food environments and thedietary behaviours of schoolchildren in Vancouver, BC. The study examinedthe associations between food outlet locations near schools and the school-day food intake of 950 5th – 8th grade students across 26 public schools.The objectives of the study were (1) to validate commonly used data sourcesfor the school food environment, (2) to examine disparities in the food en-vironments surrounding schools according to school-level demographic andsocioeconomic characteristics, and (3) to assess the relations between mea-sures of the school food environment and students’ self-reported intake, at-or en-route to school, of minimally nutritious foods or beverages.1.1 Literature Review1.1.1 The Food Environment and ObesityDuke professor Kelly Brownell and clinical psychologist Katherine Horgenwere among the first researchers to address the role of the environment inthe rise of adult obesity. In their book Food Fight: The Inside Story of theFood Industry, America’s Obesity Crisis, and What We Can Do About It31.1. Literature Review(2004), Brownell and Horgen argue that the food environment has become“toxic”. That is, the disappearance of neighbourhood produce outlets coupledwith the proliferation of fast-food restaurants has, according to Brownell andHorgen, created a world in which it is far easier to consume fat- and calorie-laden happy meals than to maintain a healthful diet.Race, Poverty and Food AccessMuch of the empirical research relating food environments and dietaryintake has focused on the relationship of poverty or race and access to foodstores. Researchers Kimberly Morland, Steve Wing, and Ana Diez Roux con-ducted a pioneering study associating food frequency questionnaire (FFQ)responses from the Atherosclerosis Risk in Communities Study, stratified byrace, with counts of supermarkets, fast-food restaurants, and conveniencestores near study participants’ residences (Morland et al., 2002). The studyfound only weak relations between diet and food outlet access for whiteAmericans—but for black Americans, access to a supermarket predicted asignificant increase in fruit and vegetable intake and access to a full-servicerestaurant predicted a significant increase in saturated fat intake. In addi-tion, the researchers reported that black Americans faced significant con-straints to supermarket access, suggesting that the food environment maycontribute to the disparities in diet-related health commonly reported in U.S.health assessments (Braveman et al., 2010).In the decade following Morland, Wing, and Diez-Roux’s seminal work,more studies have uncovered racial and socioeconomic disparities in food ac-cess in the United States (Morland, 2015). In a systematic review of fast food41.1. Literature Reviewaccess studies, Fleischhacker et al. (2011) found strong evidence of a rela-tionship between the prevalence of fast food restaurants and neighbourhoodsocioeconomic status, with most studies finding more outlets in low-incomeareas. The researchers also found systematic evidence of a higher prevalenceof fast food restaurants in association with higher concentrations of non-white racial and ethnic groups in the United States. However, only limitedresearch has been conducted relating race or ethnicity and food access out-side of the United States (Fleischhacker et al., 2011), and existing studiesoffer equivocal results: In Edmonton, Canada, for example, a higher abo-riginal population was associated with a higher odds of fast food exposureat the neighbourhood level—but no significant associations were identifiedbetween visible minority or immigrant populations and fast food exposure(Smoyer-Tomic et al., 2008).The associations of socioeconomic status and food outlet access are sim-ilarly inconsistent outside of the United States. In particular, researcherssearching for socioeconomically disadvantaged neighbourhoods lacking su-permarkets and grocery stores (“food deserts”) were largely able to identifysuch areas in the United States, but studies conducted in other countries haveproduced mixed results (Beaulac et al., 2009). In Canada, one study foundfewer well-stocked stores in low-income versus high-income areas (Lathamand Moffat, 2007), but another study found mixed results (Smoyer-Tomicet al., 2006), and a Montreal study actually identified more supermarkets inlow-income areas (Apparicio et al., 2007).51.1. Literature ReviewCurrent Evidence Associating Food Environments and ObesityExamining associations of food access and obesity, researchers have re-ported lower obesity rates among groups living in census tracts with super-markets (Morland et al., 2006; Morland and Evenson, 2009) and individu-als who reported shopping for groceries within their census tracts (Inagamiet al., 2006), as compared with groups living in supermarket-free tracts orgoing elsewhere to shop. In contrast, likelihood of obesity has been found tocorrelate positively with access to convenience stores (Morland et al., 2006;Bodor et al., 2010) small grocery stores (Morland and Evenson, 2009; Gib-son, 2011), and fast food restaurants (Morland and Evenson, 2009; Maddock,2004; Dunn, 2010; Bodor et al., 2010; Dubowitz et al., 2012). A numberof studies, however, have reported inconsistent or no significant relation-ships between food environment measures and obesity rates (Burdette andWhitaker, 2004; Simmons et al., 2005; Sturm and Datar, 2005; Jeffery et al.,2006; Block et al., 2011; Li et al., 2008; Ford and Dzewaltowski, 2011; Hick-son et al., 2011), and a recent systematic review found a large proportionof null results in studies associating food environment measures and obesity,even after accounting for study quality (Cobb et al., 2015).Reviewing studies focused on diet, rather than obesity, Caspi et al. (2012)find “moderate” evidence that neighbourhood food environments influencedietary health. The researchers note, however, that more consistent associ-ations were found with perceived measures of food retailer access and foodavailability in comparison with measures from Geographic Information Sys-tems (GIS) of distance to retailers. At present, researchers have not attained61.1. Literature Reviewa consensus regarding the effect of the food environment on the diet-relatedhealth of adults (Caspi et al., 2012; Cobb et al., 2015).Results are similarly equivocal for studies focused on children and adoles-cents. Cobb et al. (2015), reviewing 21 studies of obesity in children and foodretailer access, find some evidence that increased convenience store access isassociated with increased obesity, but mixed results for associations betweenfast food outlet access and obesity in children as well as mostly null resultsfor associations between supermarket access and obesity in children. It ispossible that the association of obesity and the food environment is hiddenby confounding factors at the neighbourhood level: examining the associa-tions of the food environment and diet, rather than obesity, Engler-Stringeret al. (2014a) found “moderately strong evidence” that food environmentsplayed a role in children’s dietary behaviours.1.1.2 School Food EnvironmentsThe school food environment is of particular interest because the area sur-rounding schools demarcates a region where children may have more auton-omy to make their own dietary decisions. There is fairly consistent evidence,from studies in the U.S. and Canada, that a large proportion of schoolchil-dren have access to a food outlet near their schools: Zenk and Powell (2008)found that at least one in three U.S. schools was located within walking dis-tance of a fast food restaurant or convenience store. In British Columbia,over half of schools are estimated to be located within walking distance of atleast one fast food restaurant, convenience store, or similar limited-servicefood outlet (Black and Day, 2012).71.1. Literature ReviewSpatial clustering analyses conducted with data from Chicago and NewZealand have yielded evidence that the prevalence of restaurants near schoolsis unlikely to be the product of random chance (Austin et al., 2005; Day andPearce, 2011). However, researchers in Germany found that relaxing theassumption of a constant probability surface across the study area—that is,recognizing that food outlets are more likely to be located in some areas ofa city rather than others—led to no significant evidence of clustering aroundschools (Buck et al., 2013).Some researchers have also suggested that attributes of schools may cor-relate with student exposure to food outlets. Most research comes fromthe United States, where studies suggest that there are more “unhealthy”outlets—fast food restaurants, convenience stores, or other outlets sellingenergy-dense foods—within walking distance of high schools versus elemen-tary schools (Simon et al., 2008; Neckerman et al., 2010), larger versussmaller schools (Zenk and Powell, 2008), schools with higher versus lowerproportions of low-income students (Sturm, 2008; Neckerman et al., 2010),and schools with higher proportions of black or hispanic students in compar-ison with schools with more white students (Sturm, 2008; Kwate and Loh,2010; Neckerman et al., 2010). The handful of Canadian studies that haveexamined school food environments have mostly focused on associations withschool- or neighbourhood-level poverty, finding more fast food outlets or con-venience stores near low-income schools or schools in low-income neighbour-hoods (Black and Day, 2012; Robitaille et al., 2010; Engler-Stringer et al.,2014b; Kestens and Daniel, 2010); the lone study examining ethnicity didnot find significant associations between the school food environment and81.1. Literature Reviewneighbourhood demographic characteristics (Engler-Stringer et al., 2014b).However, area-level factors may confound differences in access betweenareas with and without schools: Neckerman et al. (2010), examining schoolfood environments in New York City, found that many of the significantassociations between school-level demographic or socioeconomic factors dis-appeared when built environment factors were included in the analyses. Sim-ilarly, Kestens and Daniel (2010), studying the distribution of food outletssurrounding schools in Montreal, found higher food outlet density in low-and middle income areas as compared with high income areas—but control-ling for commercial density accounted for much of the difference. In BritishColumbia, Black et al. (2011) found that a majority of the variation in foodoutlet density could be accounted for by urban planning factors.School Food Environments and Children’s DietsResearchers have begun to examine the potential of school food envi-ronments to contribute to rates of obesity, food purchasing behaviours, anddietary behaviours among schoolchildren. A California study found thatmiddle and high school students whose schools were located near a fastfood restaurant consumed fewer fruits and vegetables, drank more sugar-sweetened beverages, and had a higher likelihood of obesity than studentswithout easy store access (Davis and Carpenter, 2009); another Californiastudy found significant and positive associations between obesity and 9thgrade students’ access to fast food outlets (Currie et al., 2010); however, athird study in the state found no significant relationship (An and Sturm,2012).91.1. Literature ReviewTwo studies of open-campus policies, which allow students to leave groundsduring the school day, produced similarly conflicting results: Forsyth et al.(2012) found no significant association of open campus policies and studentfast food consumption, but Neumark-Sztainer et al. (2005) found that opencampus policies were associated with students eating more lunches at fastfood restaurants. Outside of the U.S., a study conducted in Australia foundno significant associations of food availability and children’s diets (Timperioet al., 2009), while a study from the United Kingdom found that proximityto outlets with takeaway food was predictive of less healthy diets (Smithet al., 2013). In a recent systematic review, Williams et al. (2014) foundlimited evidence for an association between the food environment surround-ing schools and students’ dietary intakes, although evidence was stronger foran association between the school food environment and body weight.In Canada, several published studies have examined the associations offood outlet locations near schools and childhood obesity or children’s dietarybehaviours (Seliske et al., 2009a, 2013; Laxer and Janssen, 2013; He et al.,2012b,a; Héroux et al., 2012; Leatherdale et al., 2011). In 2009, Seliske etal. found that food outlet density within 1 and 5 km of Canadian schoolswas not associated with increased levels of overweight or obesity in students.Revisiting the topic in 2013, the researchers found that food outlet densitywithin 1km of Canadian schools was positively correlated with student like-lihood of eating lunch at a food store or restaurant (Seliske et al., 2013).Laxer and Janssen (2013), in a national-level study looking just at studentswho lived within 1km of their school, similarly found a modest positive re-lationship of outlet density and the proportion of youth eating fast food two101.1. Literature Reviewor more times per week. Finally, Héroux et al. (2012) observed that thedensity of chain food outlets surrounding schools was positively associatedwith Canadian children’s odds of eating lunch at a food retailer.Studies at the municipal or provincial level have also identified associa-tions between the food environment and obesity, diet, and food purchasingbehaviours in Canadian children. Although Gilliland et al. (2012) found in-consistent associations between children’s BMIs and the number of fast foodoutlets or convenience stores within 500 or 1000 metre buffers of schools inLondon, Ontario, Leatherdale et al. (2011) found that Ontario students withmore fast food retailers or more grocery stores surrounding their schoolswere more likely to be overweight than comparable students with more lim-ited access to food outlets. Additionally, in two studies conducted with 7thand 8th grade students in London, Ontario, researchers found statisticallysignificant associations of both the home and school food environment andHealthy Eating Index (HEI) scores (He et al., 2012b) or fast food purchasing(He et al., 2012a). In the former study, students living more than 1 kmfrom a fast food restaurant had a 1.1 point increase in HEI scores, whilestudents attending school more than 1 km from a fast food restaurant had a2.6 point increase in HEI scores (both models controlled for gender, grade,and neighbourhood distress); in the latter, having 1-2 fast food restaurants,rather than zero, within 1km of a student’s residence was associated with 1.6times the purchasing of fast food at least once weekly, while having fast foodrestaurants within 1km of a student’s school was associated with an oddsratio of 1.4 for the same outcome. It follows that the effect of the schoolfood environment on children’s diets could be greater than that of the home111.1. Literature Reviewfood environment—but the increased ease of fast food purchasing may notbe the mechanism that amplifies this effect.1.1.3 Gaps in the Existing LiteratureCity-Level Studies in CanadaDespite the proliferation of food environments studies, there are manygaps in the existing literature. Most notably, no study has been conductedrelating school food environments and dietary outcomes in a major Cana-dian city, even though 35% of Canada’s residents live in Toronto, Montreal,or Vancouver (Statistics Canada, 2011). This lack of empirical attentionis particularly surprising given Canada’s lack of a federal school lunch pro-gram, which may leave students more susceptible to the effect of the foodenvironment.Concerns Regarding Data QualityPoor data quality may contribute to the conflicting results obtained infood environments research. Researchers commonly rely on data sources de-signed for commercial rather than academic purposes (Moore and Diez-Roux,2015; Cobb et al., 2015), and recent evidence suggests that poor specificityand moderate positive predictive value for such datasets (Clary and Kestens,2013; Fleischhacker et al., 2012; Han et al., 2012; Liese et al., 2013; Lu-can et al., 2013). Other food environment studies have used governmentaldata sources, but these often classify outlets incorrectly (Fleischhacker et al.,2012; Hosler and Dharssi, 2010; Toft et al., 2011). Misclassification could af-121.1. Literature Reviewfect results for researchers who filter the data to exclude non-food stores.“Ground-Truthing”—the systematic exploration of a region to collect fieldobservations of store and restaurant locations—is considered the gold stan-dard in store location assessment (Hosler and Dharssi, 2010; Paquet et al.,2008; Powell et al., 2011), but the method requires a potentially prohibitivetime and monetary commitment.Outlet Distribution and School-Level Demographic orSocioeconomic FactorsThere has been limited research into the role that systematic differencesin school food environments may play in explaining disparities in children’sdiet-related health. In Canada, several researchers have reported associa-tions between the density of or proximity to fast food outlets in relation toschools according to neighbourhood socioeconomic status (Engler-Stringeret al., 2014b; Kestens and Daniel, 2010) or school-weighted measures of low-income populations (Black and Day, 2012; Robitaille et al., 2010), but onlyone study examined associations between measures of the school food en-vironment and neighbourhood demographic characteristics (Engler-Stringeret al., 2014b). No known study has used school-level measures of studentdemographic characteristics in such research in Canada, and research on so-cioeconomic factors and school food environments would benefit from school-level measures of poverty rather than the proxies of neighbourhood or school-weighted census measures currently used in Canadian literature.131.1. Literature ReviewDietary Behaviour or Body Mass Index?Although there has been substantial research focused on food environ-ments and obesity, fewer researchers have looked at the “food environment-diet relationship”—that is, the role food retailers play in facilitating un-healthy or healthy dietary behaviours (Caspi et al., 2012). The gap is cu-rious considering that the food environment-diet relationship is likely themain mechanism at play in determining whether an environment fostersunhealthy eating: the easier it is for a child to access to fast food, sugarsweetened beverages (SSBs), or packaged snacks, the more likely it may befor that child to consume such minimally nutritious foods. In contrast, themore commonly used dependent variable of obesity status is a distal outcomethat may be confounded by built environment factors (Cobb et al., 2015).Neighbourhoods with more convenience stores, for example, may also tend tobe more walkable (Saelens et al., 2003), allowing increased physical activitythat may counter the obesogenic effects of convenience store access on diet(Saelens et al., 2012). In the context of Canadian school food environmentsresearch, just two studies have examined associations of food retailer densitysurrounding or proximity to schools with children’s dietary intakes (Laxerand Janssen, 2013; He et al., 2012b). There is thus a need for studies fo-cused on children’s dietary behaviours in association with their school foodenvironments.141.2. Study Objectives1.2 Study ObjectivesThis thesis sought to help fill several of the gaps in the existing bodyof literature. The study was comprised of three components: (1) a fieldvalidation of food outlet locations around Vancouver schools; (2) an ecologicanalysis examining store distributions across the city and their associationswith school attributes and (3) a multilevel analysis assessing the relationsof food outlet proximity and density with students’ self-reported school-dayfood intake. The data validation component aimed to fill researchers’ needfor an assessment of existing store location data sets; in addition, it ensuredthat the subsequent analyses were conducted with accurate information onstore types and locations. Both the ecologic analysis and the dietary intakeassessment help to fill the gap in Canadian research by offering a study ofthe food environment in Vancouver, British Columbia; the ecologic analysisalso offered the first examination of school-level demographic disparities infood access conducted in Canada. Finally, the third component of this thesisadds to researchers’ understandings of how the food environment may affectchildren’s dietary intakes.15Chapter 2The Validation of FoodEnvironments Data2.1 IntroductionData quality poses a serious challenge for food environments research.Researchers commonly obtain store location data from one of three sources:(1) “ground truthing” or primary data collection, (2) commercial databaseproviders or (3) government sources (Morland, 2015). Each of these datasources is subject to varying levels of over- or undercounting due to classifica-tion errors, incorrect geocoding or inaccurate listings (Moore and Diez-Roux,2015). Some researchers believe that compromised data may help explaininconsistent findings regarding the contributions of food environments todiet-related health (Lucan, 2015).The current gold standard method for obtaining food environments datais ground-truthing, the systematic surveying of a region to identify and clas-sify food retailers (Lucan, 2015; Fleischhacker et al., 2013). Ground-truthingwith validated protocols can ensure high validity and reliability of the listingsidentified, but conducting surveys can require a prohibitive time investment.162.1. IntroductionFurthermore, ground-truthing is not possible for past years’ food retailers,and the depreciation in validity of ground-truthed data over time—as outletsclose and new outlets are opened—remains poorly understood.Commercial data sets require far less time to obtain, and many are avail-able for historical periods (e.g. DMTI Spatial, Inc. 2003, 2006, 2009). Suchdata sets can be expensive, however, and researchers have argued that majorcommercial providers’ business lists may not achieve the level of accuracynecessary to obtain valid results in the context of academic research (Mooreand Diez-Roux, 2015). Though some researchers have also relied on freelyavailable directories like Yellow Pages (Burdette and Whitaker, 2004; Jef-fery et al., 2006; Maddock, 2004), a recent review found that these publicsources generally perform less well in measures of validity than private dataproviders like InfoUSA (Fleischhacker et al., 2013).Municipal data sets offer an attractive alternative to commercial or pri-mary data. Business registries and inspections listings are generally inexpen-sive or free to obtain. They are also expected to have fewer missing listingsdue to the legal requirements associated with the data collection (Hosler andDharssi, 2010; Toft et al., 2011). However, government agencies also vary intheir efforts to maintain and update registries, leading Fleischhacker et al.(2013) to recommend that government registries be validated on a case-by-case basis before being used for research purposes.For food environments research to be conducted on a multi-city or na-tional scale, rather than at the smaller scale of municipalities, it will benecessary for researchers to identify the highest quality data sets. Thisstudy sought to address the methodological problem of data source selection172.2. Backgroundthrough a comparative evaluation of government, commercial, and ground-truthed food outlet data for the city of Vancouver B.C. The study’s objectiveswere threefold: (1) to assess the validity of two commercial and two munic-ipal data sources in comparison with ground-truthed data; (2) to test eachdata set for evidence of systematic bias in association with neighbourhoodsocioeconomic deprivation or commercial density; and (3) to compare foodenvironment measures constructed from each data source to estimate theeffect of over- or undercounting in outlet listings on research outcomes.2.2 BackgroundA data source is considered to have a high degree of “validity” if it mea-sures the concept it is intended to represent (Carmines and Zeller, 1979). Inthe case of food outlet listings, commercial and municipal data sources wouldbe considered valid if they offer accurate information on the locations andclasses of food retailers under examination (Fleischhacker et al., 2013). Datasource accuracy may be compromised, however, if the data sets undercountlistings, failing to include outlets that exist in the field, or if they over-countlistings, for example by including outlets that have closed. Misclassificationcan further compromise accuracy: if a data source tends to misclassify conve-nience stores, for example, as grocery stores, it will both over-count grocerystores and undercount convenience stores. Such errors could lead researchersto estimate research subjects’ exposure to the food environment incorrectly.It is thus important that researchers interested in the food environmentsensure that they use the data that best characterizes true outlet counts.182.2. BackgroundOver nineteen studies have characterized the validity of commonly-usedfood environment data sources (Fleischhacker et al., 2013). These stud-ies generally compare the data source of interest with data collected viaground-truthing; researchers then rely on validity measures including sensi-tivity, positive predictive value (PPV), and concordance (Table 2.1) to char-acterize levels of over- and undercounting. In a review of food environmentdata validation studies, Fleischhacker et al. (2013) found that governmentregisters of food outlets had higher levels of agreement with gold-standarddata than did other secondary data sources, while the commercial databaseprovider InfoUSA was among the highest performing data sources overall.Results vary widely, however, with researchers reporting aggregate sensitiv-ity estimates from 17% (Fleischhacker et al., 2012) to 85% (Rossen et al.,2012) and positive predictive values from 13% (Fleischhacker et al., 2012) to98% (Svastisalee et al., 2012).Table 2.1: Measures of dataset validityClassification DefinitionSensitivity Proportion of outlets observed during ground-truthing that were listed in the data setPositive Predictive Proportion of outlets listed in the data set thatValue (PPV) were observed during ground-truthingConcordance Proportion of outlets both listed in the dataset and observed on the ground in comparisonto the total number of observed or listed outletsIn addition to examining validity, a number of studies have examined192.2. Backgrounddifferent data sets for evidence of systematic error, which could lead to con-founding of the associations between neighbourhood-level factors (e.g. areaincome or racial makeup) and food environment measures (Powell et al.,2011; Burgoine and Harrison, 2013; Gustafson et al., 2012). Most evidence,however, suggests that the error is not systematic: Paquet et al. (2008), Cum-mins and Macintyre (2009), Bader et al. (2010), Lake et al. (2012), Rossenet al. (2012), Svastisalee et al. (2012), and Burgoine and Harrison (2013) re-ported no evidence of systematic bias according to neighbourhood socioeco-nomic status (SES) and Bader et al. (2010), Rossen et al. (2012), and Rummoet al. (2014) found no statistically significant differences in measures of va-lidity according neighbourhood racial demographics—although two studiesin the United States did find statistically significant differences in data setsensitivity or PPV across neighbourhoods according to levels of socioeco-nomic status or racial makeup (Powell et al., 2011; Liese et al., 2013). Thestrongest evidence for systematic bias is in relation to commercial or popula-tion density: at least four studies in the United States identified statisticallysignificant differences in validity levels1 according to neighbourhood com-mercial density (Bader et al., 2010; Longacre et al., 2011; Liese et al., 2010;Powell et al., 2011), while no significant associations were identified in twoUK studies (Lake et al., 2012; Burgoine and Harrison, 2013) .Although data set validity has been the subject of extensive research,Fleischhacker et al. (2013) note that the validation literature has not yet1It should be noted, however, that the associations identified were inconsistent: Baderet al. (2010) found a positive association between levels of error and commercial density,while Longacre et al. (2011), Liese et al. (2010), and Powell et al. (2011) obtained highervalidity scores in urban versus in rural areas.202.2. Backgroundresolved the question of which data sources should be used in academic re-search. Existing studies comparing validity levels across data sources havegenerally been small in scale and localized in geographic scope (Fleischhackeret al., 2013), and thus limited in the generalizability of their results. Fur-thermore, researchers have focused on calculating validity statistics, whileonly Ma et al. (2013) looked at the effect of data source choice on measuresof the food environment—the reason that data quality is of interest—and atpresent, no known study has assessed whether results regarding associationsof the food environment with BMI, dietary intake or other outcomes changeswith data source choice. Finally, most research has focused on locales in theUnited States; no study has been conducted assessing data quality in Van-couver, BC and only three known studies have been conducted in Canada(Paquet et al., 2008; Clary and Kestens, 2013; Seliske et al., 2012).In Montreal, Canada, Paquet et al. (2008) conducted a study examiningtwelve census tracts, in which the researchers validated both a commerciallist (Tamec Inc) for 2005 and a listing compiled from publicly available data(e.g. www.Canada411.ca and http://www.pagesjaunes.ca). Stores in theformer database were classified according to Standard Industrial Classifica-tion (SIC) codes—a classification system used by government agencies tocategorize businesses for legal and statistical purposes (Economic Classifica-tion Policy Committee, 1994)—as well as with a name-based classificationsystem; stores from the internet listings were classified according to prod-uct and business names. The commercial list had high sensitivity (81%)and high PPV (88%); the publicly available listings offered lower sensitivity(63%), but PPV remained high (93%). There was no evidence of systematic212.2. Backgrounddifferences in PPV by tract socioeconomic status.Revisiting the same twelve census tracts five years later, Clary and Kestens(2013) computed sensitivity and positive predictive values for the 2010 En-hanced Points of Interest from DMTI Spatial, Inc. The researchers foundthat just over half of outlets in the field were included in the database,while 64.4% of outlets in the database were found in the field. A fairlyhigh proportion of the error, however, was due to small discrepancies inname or geocoded location: when researchers assessed differences in the com-mercial database and field validation results due, exclusively, to differencesin a store’s operational status, existence, or classification (ignoring nameor geocoding errors), sensitivity jumped to 65.5% while PPV increased to77.3%. The researchers did not find evidence of systematic bias in validityscores by tract-level socioeconomic status.The final study conducted in Canada validated both the InfoCanada andYellow Pages listings for food outlets located within 1km of 34 schools inOntario, Canada (Seliske et al., 2012). The researchers did not, however,ground-truth each school buffer zone; instead they assessed only the exis-tence of the stores in their list, and thus could only report a data set’sPPV (77.1% for InfoCanada versus 88.1% for Yellow Pages). The study isnoteworthy, though, for its geographic scope as well as for its assessmentof “positional accuracy”: the researchers evaluated the Euclidean distancebetween primary data on store coordinates and the coordinates of stores inthe commercial database. Finally, the Ontario study is one of just 3 studiesto focus specifically on the food environments surrounding schools (Seliskeet al., 2012; Svastisalee et al., 2012; Rossen et al., 2012).222.2. BackgroundAll three Canadian studies observed noteworthy discrepancies betweencommercially available databases and ground-truthed data. It is not possible,however, to compare the different commercial database providers examined—Tamec Inc, DMTI Spatial Inc, and InfoCanada—across studies due to differ-ences both in approaches to store classification and in the geographic areasassessed. There is also still a need for a comparison of commercial and mu-nicipal data sources, as none of the Canadian studies examined the validityof municipal outlet registries as a data source for food environments research.Finally, the literature remains limited in geographic scope: localized researchhas been conducted in just two regions of Canada—the city of Montreal andthe province of Ontario—and results may not be generalizable for researchersseeking to study the food environment in Vancouver, BC.This study sought to fill gaps in the literature by offering a systematicvalidation of four data sources in Vancouver, BC. The research offered ex-aminations of two commercial database providers—DMTI Spatial Inc andPitney Bowes Software—which had not previously been compared directly.In addition, the study assessed the validity of two municipal data registries—Vancouver Coastal Health Inspection Records and City of Vancouver Busi-ness Licenses—and compared validity across time, looking at the change inthe quality of Business License data in 2015 versus in 2012. This studywas the first in Canada to assess whether data set error was associated withcommercial density, a critical gap considering the evidence of systematicbias according to commercial density in the United States (Bader et al.,2010; Longacre et al., 2011; Liese et al., 2010; Powell et al., 2011). In addi-tion, this chapter offers the first validation study for food environments data232.3. Methodologysources specific to Vancouver, BC.2.3 Methodology2.3.1 DataData were obtained from five sources: (1) the systematic ground-truthingof all streets within 800 metres of 26 Vancouver schools2, (2) Business Li-censes (City of Vancouver, 2016), (3) Vancouver Coastal Health inspectionslists (Vancouver Coastal Health, 2015), (4) the Canada Business Points (Pit-ney Bowes Software, 2012), and (5) the Enhanced Points of Interest (DMTISpatial, Inc., 2013b). An overview of these data sets can be found in Ta-ble 2.2.The ground-truthed data were obtained through systematic surveyingbetween June 29th and September 30th, 2015. Two researchers visited eachmajor commercial street located within an 800m line-based buffer surround-ing each school to identify, photograph, and classify all food outlets; a singleresearcher also examined each residential street included in the sample. Thesurveyors followed a surveying protocol developed for this study (see Ap-pendix A) according to the approach in Fleischhacker et al. (2012), usinga Garmin eTrex 20x Worldwide Handheld GPS Navigator to collect GPScoordinates for each outlet. One school buffer zone was visited twice bytwo separate surveying teams, and the results were compared with Cohen’sKappa to assess inter-rater reliability in surveyors’ store classifications.2The 26 schools sampled with the I-EAT survey, discussed in Chapter 4242.3. MethodologyTable 2.2: Sources of data for food outlet locations in Vancouver, BCData Source Classifiers YearGold Standard(1) Ground-Truthed Classification Scheme 2015Primary Data (see Appendix A)Municipal(2) City of Vancouver Business Type 2015Business Licences Business Sub-Type 2012(3) Vancouver Coastal Health Service Type 2015Inspections ListsCommercial(4) Pitney Bowes Software NAICS† codes 2012Canada Business Points SIC‡ codes(5) DMTI Spatial, Inc. NAICS† codes 2013Enhanced Points of Interest SIC‡ codes†North American Industry Classification System‡Standard Industrial ClassifcationThe two government data sources—Business Licenses and VancouverCoastal Health inspections lists—were obtained from Vancouver Open DataCatalogue and from the Vancouver Coastal Health Inspections website, re-spectively, in October 2015. Historical records were available from the Van-couver Open Data Catalogue, allowing this study to examine Business Li-censes from both 2015 and 2012. The Vancouver Coastal Health inspectionslists comprised food service establishments, food stores, and food processorsin the city of Vancouver, classified by “service type.” The Business Licensesdata were similar, though they offered a more fine-grained “business sub-type” classification system for identifying convenience stores, grocery stores,252.3. Methodologyand produce outlets.Up-to-date data for the commercial data sources were not available atthe time of this project. As a result, this study examined Canada BusinessPoints data from 2012 and Enhanced Points of Interest data for 2013. TheCanada Business Points included geographic locations, Standard IndustrialClassification (SIC) codes, and North American Industry Classification Sys-tem (NAICS) codes—a business establishment classification system that hasreplaced SIC codes for many government agencies in Canada, the UnitedStates, and Mexico (United States Census Bureau, 2016). The EnhancedPoints of Interests similarly included NAICS and SIC codes for classificationpurposes.All food outlet data sets were examined and outdated listings, dupli-cate listings, or listings without geographic information were deleted. Forthe Vancouver Coastal Health inspections lists, which did not include ge-ographic coordinates, an address locator from DMTI Spatial, Inc. (2013a)was used to geolocate outlets; unmatched listings were manually assignedto the closest match. After data cleaning, geographic coordinates were pro-jected to the NAD83 / UTM zone 10N coordinate system and mapped withArcGIS (ESRI, 2015). Studies of the food environment surrounding schoolsmost commonly look at the regions within 800m of schools (Williams et al.,2014), so 800 metre line-based buffers were created surrounding each of the26 schools of interest (Oliver et al., 2007), and food outlet data sets werelimited to the outlets located within at least one of the 26 buffers.Other geographic data used for this study included a cartographic bound-ary shapefile for the city (Statistics Canada, 2006a), a shapefile of school262.3. Methodologylocations from the Vancouver Open Data Catalogue (BC Ministry of Edu-cation, 2016) and a shapefile of Vancouver City streets (DMTI Spatial, Inc.,2013a). All geographic data were projected to the NAD83 / UTM zone 10Ncoordinate system. The Business License Data (City of Vancouver, 2016)were used to measure commercial density, defined as the total number ofbusinesses of any type located within the 800m buffer surrounding schools.Finally, the Vancouver Area Neighbourhood Deprivation Index (VANDIX)offered a dissemination area-level measure of relative socioeconomic depriva-tion (Bell et al., 2007; Bell and Hayes, 2012). The VANDIX is an area-basedindex of deprivation constructed from seven census variables—proportion ofthe population with less than a high school education, proportion with auniversity degree, the unemployment rate, proportion lone-parent families,average income3, proportion of home owners, and the labour force participa-tion rate—which were selected and weighted according to a survey of BritishColumbia medical health officers. The VANDIX has been used to identifysocial gradients in the frequency of assault injuries (Bell et al., 2009a), the ef-fects of severe burns (Bell et al., 2009b), and the relative risk of motor vehiclecollision mortality in rural British Columbia (Bell et al., 2012); furthermore,the VANDIX has been shown to perform comparably to other Canadian de-privation indices in identifying social gradients in the prevalence of fair orpoor self-rated health in Vancouver (Bell et al., 2007). For this study, theVANDIX was constructed at the dissemination-area level with variables fromthe 2006 Census of Canada4.3Average income was defined as average 2006 total income, in Canadian dollars, “amongpopulation 15 years and over by sex and presence of income” (Bell and Hayes, 2012)4While the mandatory long-form census in 2006 attained a response rate of 93.5%,272.3. MethodologyOutlet ClassificationThis study focuses on the comparison of three classes of outlets: (1)limited-service food outlets; (2) convenience stores; and (3) grocery stores orsupermarkets (see Table 2.3). The ground-truthed outlets were classified fol-lowing a modification of the flowchart used by Clary and Kestens (2013) withdefinitions from Fleischhacker et al. (2012); Han et al. (2012); Lucan et al.(2013); full details and a classification scheme can be found in Appendix A.For the 2015 and 2012 Business Licenses, “Business Type” and “BusinessSubtype” columns were used to classify listings. However, the “Facility Type”classification included in the Vancouver Coastal Health inspections lists wastoo coarse-grained to identify each of the three outlet classes. Similarly, al-though the NAICS codes provided in the two commercial data sources (theCanada Business Points and the Enhanced Points of Interest) are the stan-dard used by U.S. statistical agencies to identify business type DMTI Spatial,Inc. (2013b), these codes were only available for a subset of businesses. SICcodes were available for all listed outlets, but were inadequate for classifi-cation; many well-known fast food outlets (e.g. Mcdonald’s) were listed asfull-service restaurants, and the codes often failed to discriminate betweenconvenience stores and small grocery outlets. To address these concerns,following Clary and Kestens (2013) and Burgoine and Harrison (2013), the“Facility Type” and SIC/NAICS codes were supplemented with the applica-the long-form census in 2011—which was made optional—had a much lower responserate of 68.6% (Statistics Canada, 2015a,b). Although Smith (2015) argues that samplingadjustments mitigated the effect of non-response bias among off-reserve aboriginal people,Statistics Canada (2015b) recommended that researchers use caution when relying onvariables related to low-income. As a result, this thesis relies on data from the 2006Census.282.3. Methodologytion of a name-based classification scheme.First, overall facility type codes (in the Vancouver Coastal Health in-spections lists) and SIC codes (in the Canada Business Points and the En-hanced Points of Interest) were used to eliminate non-food outlets from thedata sets. Next, name frequencies were examined to identify national andregional chain outlets; names including words like “pub”, “bistro” or “wine”—indicative of specialty stores or full-service restaurants and pubs—were usedto identify outlets to be omitted. For all outlets retained, name frequencieswere tabulated to identify major chains (e.g. “Subway” for limited-servicefood outlets) and words indicative of each class (e.g. “Mart” for conveniencestores or “pizza” and “express” for limited-service food outlets). The lists ofsuch indicator names and words were applied and iteratively refined untilall remaining outlets were classified or deleted in each of the VCH, EPOIand PBS data sets. The final classifications were determined in the CanadaBusiness Points and Enhanced Points of Interest data by combining NAICScode classifications with name-based searches, and in the Vancouver CoastalHealth data by combining the name-based approach with the Facility Typelisting. Definitions and codings for each of the three final classification cat-egories can be found in Table 2.3; the detailed name-based classification isincluded in Table B.1.292.3.MethodologyTable 2.3: Methods used to classify listings as limited-service outlets, convenience stores, and grocery storesData Source Ground- Business Vancouver Canada EnhancedTruthed Licenses Coastal Business Points ofHealth Points InterestDefinition† Business Type Facility Type§ NAICS‡ Code§ NAICS‡ Code§Limited Outlets where - “Limited - “Food Service - 72251302 - 722211Service customers pay Service Food Establish- - 72251512 - 722213Outlets before eating Establish- ment 1” - 72251115 - 445299and order at ment” - 72251402a counter; in- - “Restaurant - 72251505cludes cafés Class 1” - 72251510(subtype - 72251518“w/o liquor”) - 44529905Convenience Stores selling - “Gasoline - “Retail Food - 4512001 - 44512Stores a variety of Station” Store” - 44611009 - 44611products in - “Retail dealer - 44719005 - 44719addition to - food” (sub-food; includes type “Smallmarts at gas Pharmacy”,stations and “Pharmacy”drugstores or “Conven-ience Store”)302.3.MethodologyData Source Ground- Business Vancouver Canada Enhanced(Continued) Truthed Licenses Coastal Business Points ofHealth Points InterestDefinition† Business Type Facility Type§ NAICS‡ Code§ NAICS‡ Code§Grocery Stores that in- - “Retail dealer - “Retail Food - 44511001 - 44511Stores clude all sec- - Grocery” Store” - 44529912 - 44523tors of a trad- - “Retail dealer - 44529918itional grocer - food” (sub-(produce, deli type “Retailbutcher, dairy, Food Store”,and bakery) or “Produce”)†Definitions were constructed following Clary and Kestens (2013), Fleischhacker et al. (2012),Han et al. (2012), and Lucan et al. (2013); see Appendix A for details.§Additional name-based classifications were applied to ensure all outlets were classified (Appendix B).‡North American Industrial Classification System312.3. MethodologyOutlet Matching ApproachTwo approaches were applied to match outlets in the secondary data setwith outlets in the primary data set. First, addresses in each data set werestandardized and two outlets were matched if the listings included identicalstreet names and house numbers. However, this approach left some storesunmatched due to small inconsistencies in addresses, so an algorithm wasencoded in R 3.2.4 (R Core Team, 2016) to match each store according toname and geographic location, following the approach of Auchincloss et al.(2012) and Hoehner and Schootman (2010). For each store in the gold stan-dard data set, geographic coordinates were used to identify all stores in thesecondary data set located within 100 metres of the ground-truthed store.The Levenshtein similarity, a similarity function based on the Levenshteindistance, or the minimum number of edits necessary for one store name tobecome identical to the other (Winkler, 1990), was calculated for all potentialmatches within 100m with the RecordLinkage Package (Sariyar and Borg,2010); the ground-truthed store was then matched with the outlet with thehighest Levenshtein similarity score. The results from the two approacheswere then compared and, for ground-truthed outlets with different resultsacross the two approaches, the best match was determined manually. Forthe Canada Business Points, which did not include addresses, the algorithmwas applied twice and each entry was reviewed and, if necessary, matchedmanually.322.3. Methodology2.3.2 AnalysisValidity measures (Table 2.1) were calculated both for all stores and foreach of the three classes of stores. A matched store was considered a truepositive (TP) if it was listed in both the secondary source and the ground-truthed data with the same classification, a false positive (FP) if the storewas listed in the secondary source but not in the ground-truthed data, anda false negative (FN) if the store was listed in the ground-truthed data butnot in the secondary source. If a store was listed in both data sets but theclassifications differed, the listing was considered both an FP and an FN.The resulting values were summed to evaluate the sensitivity (TP/(TP +FN)), positive predictive value (TP/(TP + FP)) and concordance (TP/(TP+ FP + FN)) of each secondary data source. The approach allowed a listingto be considered a TP even if it had a different name in the secondary sourcefrom that in the gold standard data, so long as the two listings includedidentical addresses and classifications; as a sensitivity analysis, “strict” TP’swere calculated omitting stores with highly dissimilar names.To assess the secondary data sources for systematic bias, logistic regres-sion was applied to examine associations between each data set’s sensitivityor PPV and measures of socioeconomic deprivation and commercial den-sity5. Two sets of logistic regressions were applied for each secondary data5Most studies validating on food environments data have relied on Fisher’s Exact Test,applied to contingency tables, to assess systematic bias in levels of sensitivity or PPV(Burgoine and Harrison, 2013; Liese et al., 2013; Cummins and Macintyre, 2009; Paquetet al., 2008; Powell et al., 2011). As Clary and Kestens (2013) point out, this approachis not ideal due to a lack of prior knowledge of the row and columns sums. Furthermore,Fisher’s Exact Test generally has lower power than exact unconditional tests (Lydersenet al., 2009). This study instead uses logistic regression models to test for associationsbetween sensitivity or PPV and neighbourhood characteristics (Bader et al., 2010).332.3. Methodologyset. For the analysis of sensitivity, regressions were run for all stores in theground-truthed data set with the outcome equal to 1 if the store was a falsenegative and 0 if the outlet was an true positive; the PPV analyses were runfor all stores in the secondary data set with the outcome equal to 1 if thestore was a false positive and 0 if the store was an true positive.Each model was fitted with either VANDIX score tertile or commercialdensity, in units of 100 outlets, as independent variables. These independentvariables were assigned to each store according to its school buffer; schoolswere assigned a “high”, “medium” or “low” VANDIX tertile based on theVANDIX scores of the dissemination area directly surrounding the school,while commercial density was calculated as the total number of stores of anytype, as listed in the 2015 Business Licenses, located within the 800 metrebuffer zone of the school. A cutoff of p < 0.05 was used for determiningstatistical significance.Finally, ArcGIS (ESRI, 2015) was used to create measures of the foodenvironment for each school with each of the classified data sets. Densitywas calculated as the total number of outlets located within each 800m line-based school buffer and proximity was measured as the shortest street-baseddistance from each school to a food outlet. food environment measures wereconstructed for outlets in each of the three categories (Table 2.3) as well asfor the aggregate food outlet data, and evaluated with summary statistics. Inaddition, similarity in the density and proximity measures calculated fromthe ground-truthed data and those obtained from each of the secondarydata sources, ranked across schools, were evaluated through the calculationof Kendall’s Tau, a non-parametric measure of correlation (Newson, 2002).342.4. Results2.4 ResultsThe ground-truthing protocol identified 267 limited-service food outlets,124 convenience stores, and 64 grocery stores or supermarkets. For thesubset classified by two surveyors, percent agreement was 93% and Cohen’sKappa was 0.883, indicating strong inter-rater reliability (McHugh, 2012).The Vancouver Coastal Health inspections lists, which included 225 limited-service outlets, 138 convenience stores, and 42 grocery/supermarket stores,was geocoded with 98% accuracy, and manual matches were identified forthe remaining 2% of outlets. After store classification, the 2015 BusinessLicense data included 375 limited-service outlets, 156 convenience stores,and 36 grocery/supermarket stores. The 2012 Business License data weresimilar, comprising 361 limited-service outlets, 153 convenience stores and38 grocery/supermarket stores. In contrast, the two commercial data setslisted fewer food outlets: the Canada Business Points included 197 limited-service outlets, 148 convenience stores, and 81 grocery or supermarket stores,and the Enhanced Points of Interest included 264 limited-service outlets, 174convenience stores, and 35 grocery or supermarket stores.2.4.1 Evaluation of Sensitivity, PPV, and ConcordanceThe 2015 Business Licenses had the highest overall scores for sensitivity,identifying 69% of the ground-truthed stores. The data set’s sensitivity washighest for convenience stores (0.75) and limited-service outlets (0.72); itssensitivity for grocery stores was lower (0.42) but remained the highest va-lidity reported for that class of stores in any of the secondary data sources352.4. Resultsexamined. The Vancouver Coastal Health inspections lists, in contrast, hadthe highest PPV: of the outlets listed in the Vancouver Coastal Health dataset, 60% were also listed in the ground-truthing results. Across all measures,the 2012 Business License data had lower validity than the 2015 Business Li-cense data; it also performed more poorly than the Vancouver Coastal Healthinspections lists in terms of overall PPV and concordance. The overall PPV,sensitivity and concordance estimates obtained for each of the municipal datasets—both for 2015 and for 2012—were higher than those obtained for eitherof the two commercial data sets. Detailed results for the validity measurescan be found in Table 2.4.With strict name matching, the 2015 Business License data lost 28 out-let matches, leading its sensitivity to drops to 0.58 while PPV decreasedto 0.49. The 2012 Business License data lost 34 matches (sensitivity=0.48,PPV= 0.41), the Vancouver Coastal Health data lost 15 matches (sensitiv-ity=0.48, PPV=0.55), and the Enhanced Points of Interest lost 27 matches(sensitivity=0.33, PPV=0.31). The Canada Business Points had the fewestmatched outlets with different names, with just 7 outlets failing the strictername-based standard; with strict matching, its sensitivity was equal to 0.37while PPV was 0.36. Although sensitivity, PPV and concordance decreaseacross all data sets when the stricter matching standards are applied, themunicipal data sets remained the highest performers in terms of overall sen-sitivity and PPV.362.4. ResultsTable 2.4: Sensitivity, positive predictive value (PPV), and concordance oftwo municipal and two commercial data sources in comparison with ground-truthed data for the locations of food outlets in Vancouver, BCBusiness Vancouver Canada EnhancedLicenses Coastal Business Points2015 2012 Health Points of InterestSensitivity 0.69 0.59 0.54 0.39 0.41Ltd. Service 0.72 0.62 0.55 0.38 0.40Convenience 0.75 0.65 0.60 0.48 0.46Grocery 0.42 0.31 0.34 0.25 0.36PPV 0.55 0.48 0.60 0.37 0.44Ltd. Service 0.51 0.46 0.66 0.38 0.54Convenience 0.60 0.53 0.54 0.35 0.39Grocery 0.75 0.53 0.52 0.46 0.28Concordance 0.44 0.36 0.40 0.23 0.27Ltd. Service 0.43 0.36 0.43 0.23 0.30Convenience 0.50 0.41 0.39 0.25 0.27Grocery 0.37 0.24 0.26 0.19 Assessment of Systematic BiasSystematic associations were observed between commercial density andthe proportion of false negative versus true positive listings identified in theVancouver Coastal Health inspections lists, the Enhanced Points of Interest,and the Canada Business Points: every 100 additional stores in a school’sbuffer zone were associated with an increase, in the odds that a store inthe ground-truthed data would be missing from the secondary data set, of1.07 in the Vancouver Coastal Health inspections lists (95% CI 1.01 - 1.14),1.11 in the Canada Business Points (95% CI: 1.04 - 1.18), and 1.08 in theEnhanced Points of Interest (95% CI 1.02 - 1.15). No statistically signifi-372.4. Resultscant associations were identified between the odds of false positive listingsand commercial density. Finally, no consistent significant associations wereidentified between the odds of listings being false positives or false negativesand neighbourhood socioeconomic deprivation in the data sources examined.Full results can be found in Tables B.2 and B. Comparison of Food Environment MeasuresAll associations of the measure of density constructed from the gold stan-dard data set and density constructed from secondary data (Table 2.5) weresignificantly different from zero (p<0.01). In terms of similarity to the goldstandard measure, commercial data sets performed slightly better for theconstruction of both density and proximity than did the municipal data sets:the mean, median and standard deviation for the gold standard measure ofdensity were most similar to those obtained for measures constructed fromthe Enhanced Points of Interest and the Canada Business Points; similarly,the range and mean of the gold standard proximity measure—calculatedacross all stores—were most similar to those of the proximity measure con-structed from the Canada Business Points. For density calculations acrossall stores, the Canada Business Points measure was 94% more likely to agreethan to disagree with the gold standard measure on its rankings of schoolsby store densities (95% CI 86.3% - 100%) while the 2012 Business Licensedata and the Vancouver Coastal Health inspections lists measures of densitywere 87% more likely to agree than to disagree with the gold standard den-sity measure on school rankings (95% CIs 78.5% - 95.9% and 75.6ˆ- 98.9%,respectively). The remaining secondary data sets both had 90% likelihoods382.4. Resultsfor agreement versus disagreement, with a 95% CI of 81.2% - 98.1% for the2015 Business Licenses measure and 82.8% - 98.1% for the measure con-structed from the Enhanced Points of Interest.Table 2.5: Density of food outlets within 800 metres of Vancouver publicschools (n=26), evaluated across data sources: Summary statistics and cor-relations with measures constructed from gold standard dataDensity† Ground- Business Vancouver Canada EnhancedTruthed Licenses Coastal Business Points2015 2012 Health Points of InterestSummary StatisticsMinimum 0.0 0.0 0.0 0.0 0.0 0.0Median 20.0 23.0 24.0 16.0 20.0 19.0Mean 24.6 30.4 30.0 16.2 23.7 25.9Std Dev 19.4 23.1 22.5 16.2 17.9 19.7Maximum 73.0 84 80.0 65.0 62.0 66.0Kendall’s TauOverall 1.00 0.90 0.87 0.87 0.94 0.90Ltd. Service 1.00 0.87 0.84 0.83 0.86 0.91Convenience 1.00 0.72 0.70 0.57 0.64 0.76Grocery 1.00 0.80 0.78 0.74 0.56 0.51†Count of outlets located within 800m line-based buffers around schoolsThere were more noteworthy differences in Kendall’s Tau statistics bystore type: although the Enhanced Points of Interest performed compara-bly to other data sets in evaluating density for limited-service outlets andconvenience stores, it was just 51% more likely to agree than disagree withgold standard data on the rankings of grocery store densities across schools(95% CI 31.6% - 69.8%); the Canada Business Points performed similarlypoorly in estimating grocery store densities. Measures of density constructed392.4. Resultsfrom the 2015 and 2012 Business License data sets, in contrast, had moreconsistent results across store types, with Kendall’s Tau ≥ 0.70 for all threestore types.Table 2.6: Proximity in metres of food outlets to Vancouver public schools(n=26), evaluated across data sources: Summary statistics and correlationswith measures constructed from gold standard dataProximity† Ground- Business Vancouver Canada EnhancedTruthed Licenses Coastal Business Points2015 2012 Health Points of InterestSummary StatisticsMinimum 131.5 98.8 98.8 107.6 133.7 155.6Median 333.3 322.0 332.0 307.4 347.7 348.1Mean 364.2 331.3 346.0 338.3 363.1 363.9Std Dev 174.7 159.6 176.5 177.6 167.1 158.0Maximum 793.6 744.9 750.3 798.8 798.5 750.3Kendall’s TauOverall 1.00 0.72 0.61 0.70 0.74 0.73Ltd. Service 1.00 0.58 0.57 0.71 0.72 0.63Convenience 1.00 0.63 0.61 0.68 0.59 0.67Grocery 1.00 0.54 0.38 0.39 0.31 0.39†Shortest street-network distance (metres) from a school to an outletExamining similarities in proximity measures, all estimates for Kendall’sTau were statistically significant at the 5% level. For proximity calculationsconducted across all stores, the 2012 Business License data were the overallpoorest performer, with Kendall’s Tau equal to 0.61 (95% CI 0.37 - 0.84%),while the 2015 Business Licenses offered higher results across store types(Table 2.6). The proximity measure constructed from the Canada Busi-ness Points again performed best, with Kendall’s Tau=0.74 (95% CI 0.50402.5. Discussion- 0.98) despite having the lowest Kendall’s Tau for grocery store proximity(0.31, 95% CI 0.20 - 0.69). This result can be interpreted to mean that theproximity measure constructed from the Canada Business Points data were74% more likely to agree than to disagree with the ground-truthed data onrankings of schools according to proximity of any of the three types of foodoutlets, but just 31% more likely to agree than to disagree on school rankingsaccording to grocery store proximity.2.5 DiscussionThe objective of this chapter was to assess the validity of two municipaland two commercial food outlet location data sources in comparison with thegold standard of ground-truthed food outlet data in Vancouver, BC. Thisstudy assessed the sensitivity, positive predictive value, and concordance foreach secondary data set, finding that all data sets were subject to high levelsof error: data sets both (1) failed to include at least 20% of outlets observedin the field and (2) consisted at minimum of 25% listings not found in thefield. Although no consistent evidence was observed of associations betweenthe odds of false negative or false positive listings and school neighbourhoodsocioeconomic deprivation across data sets, significant associations betweenthe odds of false negatives and commercial density were observed in theVancouver Coastal Health inspections lists as well as in the two commercialdata sets. Despite this evidence of poor validity across data sources, foodenvironment measures constructed from the secondary data sources weresimilar in overall distribution to food environment measures constructed from412.5. Discussionthe gold standard data.The results for measures of validity obtained in this study were, for themunicipal data sets, comparable with those obtained in previous studies.The 2015 Business License data and the Vancouver Coastal Health data hadsensitivity and PPV values in the range of 0.54 - 0.69 (for all food out-lets), which is similar to the results Fleischhacker et al. (2012) obtained forlocal health department listing sensitivity (0.66) and PPV (0.49) in NorthCarolina, U.S., as well as for the sensitivity estimate (0.66) obtained byLake et al. (2010) for city council data in Newcastle, U.K. The municipaldata sources’ PPV scores were lower, however, than those observed by Lakeet al. (2010) for Newcastle city council data PPV (0.92) and by Liese et al.(2010) for South Carolina Department of Health and Environmental Controldata PPV (0.89). These differences in findings offer support for the recom-mendation that researchers evaluate the validity of government data on acase-by-case basis (Fleischhacker et al., 2013).The commercial data sources had lower sensitivity, PPV and concordancemeasures than those observed elsewhere. Examining food outlets in the UKPoints of Interest, Burgoine and Harrison (2013) obtained a sensitivity valueof 0.60 and PPV of 0.75, significantly higher than the values of 0.41 and 0.44,respectively, observed in this study; Clary and Kestens (2013) similarly ob-tained higher PPV and sensitivity estimates (0.64 and 0.55, respectively)for their examination of the Enhanced Points of Interest data in Montreal.Both researchers, however, had a smaller temporal difference between thelast update of the secondary data source and their collection of gold stan-dard data in comparison with this study, suggesting that the difference in422.5. Discussionresults may be explained by the depreciation of data quality over time (seeSection 2.5.2).Most data sets examined in this study had higher PPV, sensitivity andconcordance values for limited-service outlets and convenience stores in con-trast with those obtained for grocery stores. These results contrast withthe findings obtained by Fleischhacker et al. (2012), who observed highersensitivity and PPV estimates for their examination of 37 grocery stores incomparison with lower estimates for 277 convenience stores. However, Fleis-chhacker et al. (2012) also included a classification for “specialty markets andshops” within which, across data sets examined, validity measures were quitelow. It is thus possible that the classification scheme used in Section 2.3.1failed to eliminate specialty shops (e.g. seafood vendors or butcher shops)from the grocery store class.A recent systematic review of food outlet data validation studies foundevidence of systematic differences in validity between rural or urban areasand urban versus suburban areas, but reported “little” evidence of system-atic biases according to neighbourhood socioeconomic status (Fleischhackeret al., 2013). The results obtained in this study are thus consistent with theprevious literature: the statistically significant associations of commercialdensity and sensitivity for the Vancouver Coastal Health inspections lists,the Canada Business Points, and the Enhanced Points of Interest are alignedwith findings from Bader et al. (2010) and Powell et al. (2011); the absenceof consistent significant associations identified between measures of data setvalidity and neighbourhood socioeconomic deprivation was in keeping withPaquet et al. (2008), Cummins and Macintyre (2009), Rossen et al. (2012),432.5. DiscussionBurgoine and Harrison (2013), and Clary and Kestens (2013).When secondary data sets were used to construct food environment mea-sures, summary statistics and nonparametric measures of correlation sug-gested that the measures’ distributions were similar to those of measuresconstructed from gold standard data sets. This observation is consistentwith the findings of the only other known study examining the effect of dataset validity on food environment measures: Ma et al. (2013) found that fooddesert measures created from two commercial data sets (InfoUSA and Dun& Bradstreet) had 87.6% – 93.5% concordance with comparable measuresobtained from the United States Department of Agriculture and the Centersfor Disease Control and Prevention. Low validity scores do not necessar-ily translate into low validity for food environment measures, suggestingthat a reliance on evaluations of sensitivity, positive predictive value, andconcordance may be leading researchers to underestimate the usefulness ofsecondary data sets for food environments research.2.5.1 RecommendationsGiven the high performance of the 2015 Business Licenses data set inmeasures of sensitivity, PPV and concordance, its lack of systematic error inassociation with socioeconomic deprivation or commercial density, and thehigh correlations between density and proximity measures constructed fromthe Business Licenses and those constructed from gold standard data, thisevaluation suggests that the Vancouver Business Licenses may be the bestavailable data set for school food environments research in Vancouver, BC.Although Vancouver Coastal Health inspections listings outperformed442.5. Discussionthe Business License data in PPV, there was evidence of systematic errorin association with with commercial density in the former data set; associ-ations between the odds of including false negative listings and commercialdensity were also observed in the commercial data sets examined in thisstudy. Researchers using such data sets should thus be cautious when con-ducting research in neighbourhoods with a range of commercial densities,because systematic error may obscure the true associations of commercialdensity with food outlet access.2.5.2 Strengths and LimitationsDepreciation of data quality over time may contribute to the lower va-lidity scores of the commercial data sets in comparison with the munici-pal data sets; while the former reported data for 2012 and 2013, the latterwere obtained immediately after the completion of ground-truthing in 2015.However, the inclusion of both current (2015) and historical (2012) BusinessLicense data suggests that deprecation explains only part of the differencein validity: in comparison with the more temporally similar 2012 BusinessLicenses, the two commercial data sets performed between 5 and 10 percent-age points worse in PPV and nearly 20 percentage points worse in sensitivityscores.This study relied on a name-based classification scheme to augmentthe codes provided in the Vancouver Coastal Health Inspections Lists, theCanada Business Points, and the Enhanced Points of Interest. Although thename-based classifications were necessary to identify major chain fast foodoutlets in the Canada Business Points and Enhanced Points of Interest, as452.6. Conclusionwell as to distinguish store and restaurant types in the coarse-grained Van-couver Coastal Health inspections lists, it may have introduced new sourcesof error by, for example, failing to account for independent retailers with lessrecognizable names. Further research is necessary to understand the effectof classification on data set quality and to identify the optimal means ofclassification for these data sources.Finally, this chapter does not predict the effect of data inaccuracy onthe measure ultimately of interest, the association of the food environmentand diet. However, the study does offer an examination of the correlation offood environment measures calculated from gold standard data with mea-sures constructed from secondary data sources in an effort to bring foodenvironments researchers a step closer to understanding the impact of over-and undercounting in common sources of food outlet location data.2.6 ConclusionThis research examined the validity of two commercial and two govern-ment data sources for the city of Vancouver B.C. The study is one of justtwo studies examining the validity of data sources for Canadian food environ-ments surrounding schools, and it is one of the most comprehensive validationstudies conducted both in terms of types of secondary data sources assessedand in the evaluation of both listings and food environment measures con-structed from different data sets. Furthermore, the results offer guidance forfuture research, suggesting that the City of Vancouver’s Business Licensesoffer the best source of food environments data currently available for re-462.6. Conclusionsearch in Vancouver, BC. For researchers planning to use commercial dataproviders, this chapter suggests that researchers should be wary of systematicerror in areas with varying commercial density; the high levels of over- andunder-counting observed in commercial food outlet data, however, do notseem to lead to large changes in proximity or density measures constructedfrom such secondary data sources.47Chapter 3Associations of SchoolCharacteristics and the FoodEnvironments SurroundingSchools3.1 Introduction & BackgroundCanadian children and youth living in socioeconomically deprived neigh-bourhoods are more likely to be overweight or obese than their peers fromless socioeconomically deprived neighbourhoods (Oliver and Hayes, 2005).Disparities in the food environments to which children in low- versus high-income neighbourhoods and from high- versus low- SES households are ex-posed could help explain these differences in diet-related health. Canadianchildren may be particularly susceptible to the food vendors they encounteren route to school or during their lunch breaks: Tugault-Lafleur et al. (2016)find that Canadian children’s diets during school hours are of poorer nutri-tional quality than their pre- or post- school-hour diets, and Héroux et al.483.1. Introduction & Background(2012) report that Canadian children are more frequent school-day patronsof food retailers than are American children.In the United States, where Black and Mexican-American children aresignificantly more likely to be overweight or obese than other children (NCHS,2012; Wang and Beydoun, 2007; Ogden et al., 2002), several studies haveidentified disparities in the food environments surrounding schools accordingto student racial or ethnic demographics (Sturm, 2008; Kwate and Loh, 2010;Neckerman et al., 2010). However, fewer studies have examined sociodemo-graphic disparities in Canadian children’s diet-related health, and the studiesthat have been conducted are less conclusive: researchers found no evidenceof disparities in Canadian children’s diets according to visible minority status(Riediger et al., 2007) or aboriginal ethnicity at the individual level (Tayloret al., 2007; Garriguet, 2009), while recent immigrants to Canada generallyhave fewer diet-related health conditions and maintain healthier dietary be-haviours than long-term residents (Sanou et al., 2014). In Canada, then, onemight expect to see socioeconomic—but not demographic—disparities in thefood environments to which children are exposed at and en-route to school.Indeed, Canadian studies have observed consistent associations betweenneighbourhood socioeconomic status (SES) and school food environments.Studies using school-weighted census measures of income have observed thatlow and medium income schools, in comparison with high income schools,had easier access—as measured by the density or proximity of stores in rela-tion to schools—to fast food outlets or convenience stores in British Columbia(Black and Day, 2012), Saskatoon, Saskatchewan (Engler-Stringer et al.,2014b), and Quebec (Robitaille et al., 2010). In Montreal, a study found493.1. Introduction & Backgroundthat schools in low-income neighbourhoods, versus high-income neighbour-hoods, were more likely to have access to fast food outlets even after the re-searchers controlled for commercial density(Kestens and Daniel, 2010). Onenationwide study offered contradictory results, finding that higher SES inthe neighbourhood surrounding a school was associated with increased foodretailer density, but the findings may be due to the researchers’ failure tocontrol for the effects of commercial density (Seliske et al., 2009b). Engler-Stringer et al. (2014b) additionally examined the associations of aboriginalor immigrant status and school food access, finding no association betweenthe proportion of aboriginal residents or recent immigrants in the neighbour-hoods surrounding schools and distance from schools to food outlets.While Canadian school food environments research offers growing evi-dence of disparities in school food environments according to neighbourhoodsocioeconomic status, there are significant gaps in the literature. Many stud-ies have relied on school-weighted census measures of income (Black andDay, 2012; Robitaille et al., 2010) or neighbourhood measures of income(Kestens and Daniel, 2010; Engler-Stringer et al., 2014b) and socioeconomicdeprivation (Seliske et al., 2009b) as proxies for student SES, rather thanusing student-specific measures of poverty6. Because neighbourhood fac-tors are more likely to contribute to the neighbourhood food environment(e.g. through residents’ shopping behaviours) than school factors, the use ofneighbourhood variables may lead researchers to overestimate associations.6In the United States, researchers commonly use student eligibility for free and reduced-price lunch as a measure of poverty specific to students within a school (Sturm, 2008;Neckerman et al., 2010; Currie et al., 2010); however, such measures are not available forresearchers in Canada due to the absence of a comparable federal school lunch program.503.1. Introduction & BackgroundSimilarly, the only study that examined disparities in access according to eth-nicity relied on neighbourhood census measures rather than on school-levelmeasures (Engler-Stringer et al., 2014b). No known studies have examinedthe associations of school food environments and student ethnicity or immi-grant status in a Canadian city, a significant research gap given the evidenceof racial and ethnic disparities in school food environments in the UnitedStates (Black, 2015).There is also evidence, from the United States, that built environmentfactors confound the associations of school-level demographic or socioeco-nomic characteristics and food environment measures (Neckerman et al.,2010), but only one Canadian study controlled for commercial density (Kestensand Daniel, 2010)7. Finally, no study has examined disparities in the food en-vironments surrounding schools in Vancouver, BC. Though 35% of Canada’sresidents live in one of three major cities—Toronto, Montreal, and Vancouver(Statistics Canada, 2011)—only the food environments surrounding schoolsin and around Montreal have been examined in association with neighbour-hood socioeconomic status (Black, 2015; Kestens and Daniel, 2010). Thisstudy thus sought to offer the first city-specific study of disparities in theschool food environments for Vancouver, one of Canada’s three largest mu-nicipalities.This study sought to address the current research gaps through an ex-amination of the food environments surrounding public schools in Vancou-ver BC. The research had three objectives: (1) to offer a descriptive pro-7Several studies did control for residential density (Black and Day, 2012; Seliske et al.,2009b) or urban versus rural status (Robitaille et al., 2010), but these measures are lesslikely to capture the urban planning factors at play (Black et al., 2011).513.2. Methodsfile of Vancouver school food environments, (2) to evaluate differences inaccess to food outlets according to school-level demographic and socioeco-nomic characteristics, and (3) to assess whether disparities in access couldbe explained by neighbourhood characteristics such as commercial densityor neighbourhood-level socioeconomic deprivation. The hypothesis of thisstudy was that schools with high levels of student poverty would have in-creased access to food retailers in comparison with low-poverty schools, butthat differences in access would be explained by neighbourhood factors. Incontrast, food environment measures were not expected to relate with theproportion of aboriginal students or recent immigrants to Canada enrolledin Vancouver schools.3.2 Methods3.2.1 DataSchool-Level Demographic and Socioeconomic CharacteristicsThis study examined schools located in Vancouver, BC in operation dur-ing the 2011/2012 academic year (n=113). Data on school locations andattributes were obtained from the British Columbia Ministry of Educationvia the BC open data catalogue (DataBC, 2016). School demographic char-acteristics examined included the proportion of enrolled students who wereEnglish Language Learners (ELL) and the proportion aboriginal students.ELL status, a designation referring to students whose primary languageor language spoken at home is a language other than English, served as523.2. Methodsa proxy measure of immigrant status; proportion aboriginal was included toassess whether children’s exposure to food retailers would parallel evidenceof shorter proximities to unhealthy food retailers from census disseminationblocks with higher aboriginal populations (Engler-Stringer et al., 2014b)School-level poverty was assessed with a binary variable identifying schoolsin the Vancouver School Board’s Inner City Schools Project (ICP) in 2012(Vancouver Board of Education, 2009). The Inner City Schools Project iden-tified schools with a high number of students living in poverty; these schoolsthen received additional staffing and discretionary funding8. After a 2009review, 14 elementary schools and 4 annexes were recommended for the pro-gram based on the numbers of vulnerable children attending as well as theschools’ Ministry of Education Social Services Indices (Vancouver Board ofEducation, 2009). One school and one annex9 were also identified as transi-tional schools10; analyses were conducted both including and excluding thesetwo transitional schools as ICP schools to test the robustness of the resultsto the measurement of school poverty.8ICP schools received additional staffing and discretionary funding as well as a breakfastprogram and access to a universal school meal program. The program aimed to reducethe stigma associated with food insecurity by asking parents to make confidential monthlycontributions of any amount according to their self-assessed ability to pay (VancouverBoard of Education, 2009). The Inner City Schools Project was replaced by a tieredsystem of funding provisioning in 2014 (CBC Radio-Canada, 2014); in 2012, however, itwould still have been operating in these schools.9Annexes, in Vancouver, are smaller schools usually serving students in grades K - 3.10These schools were transitional in the sense that they were seeing declines in thenumber of enrolled students living in poverty, and thus were selected to be slowly phasedout of ICP program533.2. MethodsSchool-Level Control VariablesIn addition to the demographic and socioeconomic characteristics of aschool’s student body, school level and size may be associated with studentaccess to food outlets (Simon et al., 2008; Day and Pearce, 2011; Robitailleet al., 2010; Black and Day, 2012). For this study, school size was measuredas the total number of students enrolled in each school in the 2011 - 2012academic year. School level—elementary versus secondary—was included tocompare access between secondary schools, where students are often affordedmore autonomy through open-campus policies, which allow students to leavethe school for lunch, and elementary schools where students are more likely tostay on campus all day and to be accompanied by adults on their commutesto and from school. School level also served to compare access between olderand younger students given the lower quality of adolescents’ versus youngerchildren’s diets (Garriguet, 2009). Following Black and Day (2012), schoolsoffering grades 8 to 12 were considered secondary schools; schools with lowergrades were categorized as elementary schools11.Neighbourhood FactorsThe VANDIX, discussed in Section 2.3.1, was used to assess the socioe-conomic deprivation of each school neighbourhood. As in Section 2.3.2, eachschool was assigned the VANDIX score of its surrounding dissemination area.The final school-level VANDIX scores were split into tertiles for the analy-sis; “low” indicates the least deprived tertile while “high” denotes the most11Ecole Secondaire Jules-Verne offered grades 7 to 12; it was included in the secondaryschool category.543.2. Methodsdeprived tertile.Figure 3.1: Flow chart of of 2012 Business License Data cleaning processCommercial density was obtained from the 2012 City of Vancouver Busi-ness Licenses (City of Vancouver, 2016). As described in Figure 3.1, thebusiness licenses were filtered to identify all businesses located in the cities553.2. Methodsof Vancouver or neighbouring Burnaby, BC. The data were then limited tostores with license expiration dates later than March 1, 2012 and issuancedates before June 30, 2012 to (1) ensure that all stores included were openand in operation at the end of the 2011/2012 academic year, when schoolattributes would be known to outlet owners, and (2) to limit stores to ashort time period in order to prevent overcounting in neighbourhoods withhigh outlet turnover12. ArcMap 10.3.1 (ESRI, 2015) was then used to limitthe data to outlets located in Vancouver or, to guard against the impact ofan arbitrary geographic boundary, within 800m of the city’s Eastern border.Food retailers (see Section 3.2.2) were excluded from the measure to reducecollinearity between the measures of commercial density and food retailerdensity. Finally, commercial density was calculated as the total number ofretail or commercial outlets, excluding food retailers, located within a 160m,400m and 800m line-based street network buffer of the school.3.2.2 Food Environment MeasuresFood outlets were identified from the final set of business licenses follow-ing the protocol outlined in Section 2.3.1. For the purposes of this study, foodoutlets are defined as any store meeting the classification of limited-servicefood outlet, convenience store, or supermarket/grocery store, defined follow-ing the flowchart given in Appendix A. In addition, business and trade nameswere standardized to identify the most prevalent limited-service outlets. Fol-lowing Currie et al. (2010), these outlets were then used to construct a clas-12The specific time period was chosen to ensure comparability with the Individual EatingAssessment Tool data (Section 4.2), which was collected between March and June 2012.563.2. Methodssification of major-chain fast food restaurants (Table 3.1). For each outlettype, ArcMap 10.3.1 was used to assess proximity—the shortest street-baseddistance from the school to a food outlet—and density, the total number ofoutlets included within a 800m line-based street network buffer surroundingthe school. In addition to the calculation of density within 800m, used inChapter 2, density was assessed within 400m and 160m of each school. Theshorter distances were included in keeping with the possibility that studentsface highly nonlinear transportation costs, and thus that outlets within a 1-2minute walk (160m) may have a stronger association with children’s habitsthan outlets within a 5 minute walk (400m) or a 10 minute walk (800m)(Currie et al., 2010; Pikora et al., 2002).Table 3.1: Limited-service outlets identified as “major chains”Standardized Outlet Name Frequency*1. Starbucks Coffee 892. Subway Sandwiches & Salads 543. Blenz Coffee 284. Tim Horton’s 255. Freshslice Pizza 176. A & W Restaurant 157. McDonald’s Restaurant 148. Dairy Queen 149. Jugo Juice 1410. Quizno’s Classic Subs 11*Number of occurences within study area573.2. Methods3.2.3 AnalysisData cleaning and measure construction were conducted with R 3.2.4(R Core Team, 2016) and ArcMap 10.3.1 (ESRI, 2015); all statistical anal-yses were conducted with STATA 14 (StataCorp, 2015). First, descriptivestatistics—including means and standard deviations for continuous variablesand frequencies for categorical variables—were calculated across schools.Next, proximities and densities were calculated for all stores and across storetypes. Summary statistics, including means and standard deviations, charac-terized the general nature of the food environments surrounding Vancouverschools. Finally, multivariate regression analysis was applied to examine theassociations between school food environments and school-level demographicor socioeconomic characteristics13.Negative binomial regressions were fitted with food outlet density (forlimited-service food outlets, convenience stores, and supermarkets or gro-cery stores) at 400m as the dependent variables. Negative binomial modelswere preferable to Poisson models due to evidence of overdispersion in theoutcome variables (Table 3.3); likelihood ratio tests confirmed that the addi-tional parameter offered a significant improvement in fit. Models were firstfitted with school-level demographic and socioeconomic characteristics as theindependent variables. School controls (school level and size) and neighbour-hood characteristics (VANDIX tertiles and commercial density) were then13Technically, this data is from a census of Vancouver schools rather than a randomsample, making statistical tests inappropriate. However, Sturm (2008) argues that thereis constant turnover in both school enrolment and business. The results from this study,then, can be interpreted as a sample of school/business observations over time. Statisticaltests, would distinguish effects from random variation in business and school enrolmentnumbers/characteristics.583.2. Methodsincluded in the models, and the two nested models were compared with like-lihood ratio tests. In addition, likelihood ratio tests were used to comparemodels with just the neighbourhood variables in comparison with modelsthat included both neighbourhood and school characteristics as explanatoryvariables. Models were additionally fitted with food outlet density at 160mand 800m following Currie et al. (2010) and Sturm (2008).A number of sensitivity analyses were run to examine the robustness ofthe model results. First, models were fitted with a broader definition of InnerCity Program (ICP) schools that included the two transitional schools. Thereis no consensus on the optimal means of measuring the school food environ-ment (Lytle, 2009; Lucan, 2015; Feng et al., 2010), so ordinary least squaresregressions were additionally fitted with food outlet proximity to schools asthe dependent variables. Again models were first fitted with school-leveldemographic and socioeconomic attributes and then with both school andneighbourhood characteristics; partial F-tests compared the two models aswell as models with just neighbourhood attributes in comparison with fullmodels. Finally, this study fitted models counting just the “major chain” out-lets identified in Section 3.2.1 as limited-service food outlets given evidenceof the difficulty studies face in classifying store types (Moore and Diez-Roux,2015; Lucan, 2015).593.3. Results3.3 Results3.3.1 Descriptive StatisticsThere were 113 schools and 2,223 food outlets located within the studyregion. The food outlets included 1,615 limited-service food outlets, 462convenience stores, and 146 grocery stores or supermarkets. Of the limited-service food outlets, 281 were outlets included in the major chain list. Intotal, the food retailers represented 8.7% of the 25,442 commercial outletsof any type identified within the study region.Descriptive statistics for school characteristics can be found in Table 3.2.On average, 31.2% of students enrolled in Vancuver schools were EnglishLanguage Learners, while an average of 5.7% of students were aboriginal.The eighteen ICP schools represented 15.9% of schools overall. There were94 elementary schools and 19 secondary schools in operation in Vancouverduring the 2011 - 2012 academic year; each school had an average of 474students (SD=427.0) enrolled. Schools were located in neighbourhoods witha range of socioeconomic deprivation scores, as measured by the VANDIX,though there was a geographic divide in socioeconomic deprivation: the east-ern side of the city had many highly deprived dissemination areas, while mostdissemination areas in the West were in the least deprived categories (Fig-ure 3.2). Commercial density also varied around schools: the number ofnon-food outlets located within 800m of a schools ranged from 1 to 1,636outlets. Although all schools had at least one non-food business outlet within400m, twelve schools (10.6%) had no non-food outlets within 160m.603.3. ResultsTable 3.2: Descriptive statistics for school and school neighbourhood char-acteristics for all public schools (n=113) in operation during the 2011/2012academic year in Vancouver, BCSchool-Level Demographic &Socioeconomic Characteristics Mean±SD Min MaxStudents% Aboriginal 5.7±11.0 0 65.3% English Language Learners 31.2±19.6 0 83.8N Schools PercentSchoolsInner City Project (ICP) 18 15.9% N/ASchool Controls Mean±SD Min MaxSchool SizeTotal Enrolment 473.5±427.0 57 2110N Schools PercentSchool LevelElementary 94 83.2% N/ASecondary 19 16.8% N/ANeighbourhood Factors N Schools PercentVANDIX tertile‡low 38 33.6% N/Amedium 38 33.6% N/Ahigh 37 32.7% N/AMean±SD Min MaxCommercial density‡160m 8.6±11.2 0 52400m 58.0±57.5 1 303800m 250.7±227.8 1 1,636†School neighbourhood VANDIX scores were ranked and split into tertileswhere “low” denotes the least deprived neighbourhoods and “high” refersto the most deprived neighbourhoods. ‡Commercial density is reported asthe total number of stores located within 160, 400 or 800 metres of eachschool; food outlets were excluded from the measure to avoid collinearity.613.3.ResultsFigure 3.2: Socioeconomic Deprivation and Access to Grocery Stores or Supermarkets within 400mof Vancouver Public Schools. The map depicts all public schools in Vancouver, BC in the 2011/2012 academicyear (n=113); Schools with 1 or more supermarkets or grocery stores within 400m are highlighted in yellow.Dissemination areas are categorized by socioeconomic deprivation—as measured by the VANDIX—with the mostdeprived areas indicated in red. Light grey indicates dissemination areas in which data were suppressed.623.3. ResultsMost schools (95.6%) had at least one food outlet located within the 800mline-based buffer surrounding the school. Of the 113 schools examined, 90.2%were located within 800m of a limited-service food outlet—though just 67.2%were within 800m of a major chain—90.2% were located within 800m of aconvenience store, and 68.1% were located within 800m of a supermarketor grocery store. At 400m, 69% of schools had access to any food outlet.A majority (58.4%) of schools had at least one limited-service food outlet,54.9% had at least one convenience store, 25.7% had a major chain limited-service outlet, and 25.7% had at least one supermarket or grocery store.Access tapers off significantly with smaller boundaries: though 29.2% ofschools had some type of food outlet within 160m, 22.1% had a limited-service food outlet, 16.8% had a convenience store, 5.3% had a supermarketor grocery store, and just 3.5% had a major chain limited-service outlet.Schools were, on average, closest to limited-service food outlets (medianproximity=396m) and farthest from supermarkets or grocery stores (medianproximity=653m). Full summary statistics for the proximity and densityof the food environments surrounding Vancouver schools can be found inTable 3.3.633.3.ResultsTable 3.3: Descriptive profile of food environments around Vancouver schools (n=113) between March and June2012All Limited Convenience Grocery/ MajorOutlets Service Store Supermarket ChainN Stores 2,060 1,498 428 134 281160m Density†Mean±SD 0.9±2.0 0.6±1.6 0.3±0.7 0.1±0.2 0.1±0.5Median 0 0 0 0 0Range (0 - 12) (0 - 10) (0 - 3) (0 - 1) (0 - 5)400m Density†Mean±SD 6.6±9.4 4.5±7.0 1.7±2.5 0.4±0.9 0.5±1.3Median 2 1 1 0 0Range (0 - 50) (0 - 38) (0 - 16) (0 - 5) (0 - 7)800m Density†Mean±SD 26.1±26.9 17.7±20.2 6.2±6.0 2.1±2.3 2.7±4.6Median 18 11 5 1 2Range (0 - 184) (0 - 149) (0 - 37) (0 - 12) (0 - 40)Proximity‡Mean±SD 411.1±349.8 450.5±366.9 534.3±389.7 774.1±470.6 759.3±437.5Median 340.0 395.7 448.3 652.8 679.8Range (1.1 - 2,569.4) (1.1 - 2,589.2) (86.0 - 2,704.6) (104.5 - 2,664.9) (115.1 - 2,589.2)†Density was measured as the count of outlets within a line-based buffer of schools.‡Proximity was the shortest street-network based distance from a school to an outlet.643.3. Results3.3.2 Results from Negative Binomial Regression ModelsResults for negative binomial regressions with food outlet density at 400mas the dependent variables and school-level demographic and socioeconomicfactors as the independent variables can be found in Table 3.4. Neither thepercent of aboriginal students nor the percent of ELL students in a schoolwas significantly associated with limited-service food outlet density, conve-nience store density, or supermarket/grocery store density within 400m ofschools. Furthermore, no significant associations of the demographic vari-ables were observed with food outlet densities within 800m (Table C.1) or160m (Table C.2) of schools.Table 3.4: Results from multivariate negative binomial regressions with foodoutlet densities within 400m of Vancouver schools (n=113) as dependentvariables and student socioedemographic factors as independent variables(1) (2) (3)400m Density Limited Convenience Grocery/Service Store Supermarket% Aboriginal 1.03 1.02 1.01(0.99 - 1.07) (0.99 - 1.05) (0.97 - 1.04)% English Language 0.99 0.99 0.99Learners (ELL) (0.97 - 1.00) (0.98 - 1.01) (0.97 - 1.01)Inner City 1.12 2.74∗ 2.06Project (ICP) (0.40 - 3.08) (1.27 - 5.89) (0.65 - 6.51)McFadden’s Pseudo R2 0.01 0.04 0.02Incidence rate ratios with 95% confidence intervals in parentheses∗significant at 0.05; ∗∗significant at 0.01; ∗∗∗significant at 0.001653.3. ResultsSchool-level poverty was associated with convenience store density. Hold-ing school demographic characteristics constant, ICP schools had approxi-mately 2.74 times the number of convenience stores located within the 400mline-based buffer surrounding the schools in comparison with non-ICP schools(95% CI 1.27 - 5.89). The association of convenience store density and schoolICP status remained statistically significant and positive, after adjusting fordemographic factors, when 800m line-based buffers were used (IRR=1.88,95% CI 1.13 - 3.14); ICP schools also had a significantly higher prevalence ofsupermarkets or grocery stores within the larger buffer (IRR=2.08, 95% CI1.09 - 3.95) in adjusted models. No associations were statistically significantat 160m.After controlling for school size, school level, neighbourhood commer-cial density, and socioeconomic deprivation, no significant associations wereobserved between school demographic factors and the densities of limited-service food outlets or supermarkets/grocery stores within 400m of Vancou-ver schools (Table 3.5). School-level poverty, as measured by ICP status,remained a statistically significant predictor of convenience store density at400m (IRR=1.74, 95% CI 1.00 - 3.03) and of supermarket/grocery storedensity at 800m (IRR=1.82, 95% CI 1.05 - 3.14). Neither school level norschool size were significantly associated with food retailer density surround-ing schools in any of the models.663.3. ResultsTable 3.5: Results from multivariate negative binomial regressions with foodoutlet densities within 400m of Vancouver schools (n=113) as dependentvariables and student socioedemographic factors as independent variables,adjusted for school controls and neighbourhood factors(1) (2) (3)400m Density Limited Convenience Grocery/Service Store SupermarketStudents% Aboriginal 1.02 1.01 1.00(0.99 - 1.04) (0.99 - 1.02) (0.97 - 1.02)% English Language 0.99 1.00 1.00Learners (ELL) (0.98 - 1.01) (0.99 - 1.02) (0.98 - 1.02)Inner City 0.85 1.74∗ 1.24Project (ICP) (0.41 - 1.75) (1.00 - 3.03) (0.52 - 3.00)SchoolsTotal Enrolment † 1.04 1.02 1.05(0.93 - 1.16) (0.94 - 1.11) (0.93 - 1.18)School LevelElementary – – –Secondary 1.09 1.27 1.16(0.33 - 3.62) (0.52 - 3.09) (0.37 - 3.63)NeighbourhoodsCommercial 1.24∗∗∗ 1.13∗∗∗ 1.15∗∗∗density (400m)‡ (1.18 - 1.31) (1.10 - 1.17) (1.10 - 1.20)VANDIX tertile§low – – –medium 1.51 1.46 2.63∗(0.85 - 2.70) (0.83 - 2.58) (1.05 - 6.60)high 1.12 1.64 2.68∗(0.59 - 2.15) (0.91 - 2.98) (1.01 - 7.12)McFadden’s Pseudo R2 0.15 0.19 0.21Incidence rate ratios with 95% confidence intervals in parentheses∗significant at 0.05; ∗∗significant at 0.01; ∗∗∗significant at 0.001†per 100 students; ‡per 10 non-food outlets within 400m§“high” refers to the most deprived neighbourhoods.673.3. ResultsAs expected, neighbourhood commercial density was significantly asso-ciated with the density of food retailers surrounding schools. For every 10additional non-food outlets located within 400m, schools had an average of1.24 times more limited-service food outlets (95% CI 1.18 - 1.31), 1.13 timesmore convenience stores (95% CI 1.10 - 1.17) and 1.15 times more supermar-kets or grocery stores (95% CI 1.10 - 1.20). The associations of commercialdensity and food retailer density were robust to the size of the buffer zonesurveyed, remaining statistically significant at 800m (Table C.3) and 160m(Table C.4).Schools with “high” VANDIX scores—indicating higher levels of depri-vation and thus lower neighbourhood SES—had 2.68 times the number ofsupermarkets or grocery stores within 400m observed, on average, aroundcomparable schools in the least deprived neighbourhoods (95% CI 1.01 -7.12), after adjusting for school-level factors and neighbourhood commer-cial density (see Figure 3.2). At 800m, the most socioeconomically deprivedneighbourhoods had more convenience stores (IRR=1.74, 95% CI 1.28 - 2.35)than did schools in the least deprived neighbourhoods, though no significantassociations were observed for supermarkets/grocery stores in the adjustedmodels. No significant associations between retailer density and neighbour-hood socioeconomic deprivation were observed within 160m of schools.Likelihood ratio tests comparing adjusted and unadjusted models con-firmed that including school controls and neighbourhood factors significantlyimproved model fit (p < 0.001 for all models). Additional likelihood ratiotests comparing models with just school controls and neighbourhood factorsnested within the full models were used to assess whether the simultaneous683.3. Resultsinclusion of percent aboriginal, percent ELL, and ICP status significantlyimproved model fit. The likelihood ratio tests examining models with con-venience store density as the dependent variables approached statistical sig-nificance (p=0.09 for 800m density, p=0.07 for 400m density), but acrossmodels, tests failed to reject the null hypothesis of no significant improve-ment in fit at the 5% significance level. Furthermore, likelihood ratio testsfailed to reject the hypothesis that full models offered a significantly betterfit than models including just commercial density and VANDIX tertiles asexplanatory variables.3.3.3 Sensitivity AnalysesThe associations of school-level poverty and food outlet density were rea-sonably robust to the inclusion of the transitional schools as ICP schools.When the two transitional schools were included, ICP status was signifi-cantly associated with convenience store density at 800m (IRR=2.53, 95%CI 1.61 - 3.96), 400m (IRR=2.75, 95% CI 1.34 - 5.65) and 160m (IRR=4.65,95% CI 1.50 - 14.43) and with supermarket/grocery store density at 400m(IRR=3.79, 95% CI 1.39 - 10.33) and 800m (IRR=3.13, 95% CI 1.81 -5.42). However, the significant associations were likely the product of highcommercial density surrounding the transitional ICP schools—after control-ling for neighbourhood characteristics, only the association of ICP statuswith supermarket/retailer density at 800m remained statistically significant(IRR=1.85, 95% CI 1.08 - 3.17).693.3. ResultsTable 3.6: Results from multivariate ordinary least squares regressions withfood outlet proximities Vancouver schools (n=113) as dependent variablesand student socioedemographic factors as independent variables(1) (2) (3)Proximity Limited Convenience Grocery/Service Store Supermarket% Aboriginal -2.7 -4.6 0.2(-10.1 - 4.8) (-12.5 - 3.3) (-9.2 - 9.5)% English Language 3.7∗ 1.1 5.4∗Learners (ELL) (0.1 - 7.3) (-2.8 - 4.9) (0.8 - 9.9)Inner City -132.5 -149.3 -360.5∗Project (ICP) (-361.7 - 96.7) (-392.9 - 94.3) (-648.5 - -72.4)R2 0.05 0.05 0.09Adjusted R2 0.03 0.03 0.07Root MSE 361.7 384.4 454.5Coefficients with 95% confidence intervals in parentheses∗significant at 0.05; ∗∗significant at 0.01; ∗∗∗significant at 0.001Proximity to food retailers was examined with OLS regressions (Ta-ble 3.6). In regressions with just school demographic variables as explanatoryvariables, schools were 3.7 metres farther, on average, from a limited-servicefood outlet (95% CI 0.1 - 7.3) and 5.4 metres farther from a supermarketor grocery store (95% CI 0.8 - 9.9) for every additional percentage point ofenrolled students considered English Language Learners. Associations of thedependent variable and ICP status with convenience store or limited-serviceoutlet proximity were not statistically significant at the 5% level. Grocerystores or supermarkets were, on average and controlling for demographicfactors, significantly closer to ICP schools than to non-ICP schools.703.3. ResultsTable 3.7: Results from multivariate ordinary least squares regressions withfood outlet proximities (metres) to Vancouver schools (n=113) as dependentvariables and student socioedemographic factors as independent variables,adjusted for school controls and neighbourhood factors(1) (2) (3)Proximity Limited Convenience Grocery/Service Store SupermarketStudents% Aboriginal -0.7 -1.9 2.4(-7.8 - 6.5) (-9.2 - 5.5) (-6.9 - 11.6)% English Language 3.0 0.3 4.7Learners (ELL) (-0.9 - 6.9) (-3.7 - 4.3) (-0.4 - 9.7)Inner City 3.2 13.3 -236.5Project (ICP) (-216.1 - 222.6) (-211.7 - 238.3) (-519.4 - 46.3)SchoolsTotal Enrolment† -17.2 -16.4 -0.7(-43.8 - 9.4) (-43.8 - 10.9) (-35.2 - 33.7)School LevelElementary – – –Secondary 194.9 167.7 122.8(-106.6 - 496.4) (-141.5 - 476.9) (-265.9 - 511.5)NeighbourhoodsCommercial -24.8∗∗∗ -28.3∗∗∗ -30.1∗∗∗density (800m)‡ (-36.8 - -12.8) (-40.7 - -16.0) (-45.6 - -14.6)VANDIX tertile§low – – –medium -110.2 -182.2∗ -18.9(-270.3 - 49.9) (-346.5 - -18.0) (-225.3 - 187.6)high -133.9 -207.7∗ 25.3(-304.0 - 36.1) (-382.1 - -33.3) (-194.0 - 244.5)R2 0.23 0.28 0.22Adjusted R2 0.17 0.23 0.16Root MSE 333.6 342.2 430.2Coefficients with 95% confidence intervals in parentheses∗significant at 0.05; ∗∗significant at 0.01; ∗∗∗significant at 0.001†per 100 students; ‡per 10 non-food outlets within 400m§“high” refers to the most deprived neighbourhoods713.3. ResultsCoefficients of the school-level demographic and socioeconomic character-istics were no longer statistically significant after school controls and neigh-bourhood factors were included in the models (Table 3.7). Associations ofneighbourhood commercial density and socioeconomic deprivation with foodoutlet proximity were comparable to those observed in models of density.Partial F-tests confirmed that the inclusion of neighbourhood characteris-tics significantly improved fit for all proximity models at the 1% significancelevel. Comparing full models with models lacking the three school-level de-mographic and socioeconomic variables, partial F-tests failed to reject thenull hypothesis of no significant improvement in model fit.Finally, negative binomial models were fitted with the density of ma-jor chain limited-service outlets (Table C.5). Controlling for school leveldemographic and socioeconomic factors, schools with higher proportions ofEnglish Language Learners did have a marginally lower density of majorchains within 400m (IRR=0.97, 95% CI 0.95 - 1.00), but associations werenot significant for density within 800m or 160m. After adjusting for neigh-bourhood commercial density and VANDIX tertile, only the association ofICP status and major chain density within 800m remained statistically sig-nificant (IRR=1.77, 95% CI 1.05 - 2.97; Table C.6). Likelihood ratio testsconfirmed that neighbourhood factors significantly improved model fit (p <0.001 for all models). The tests rejected the null hypothesis of no improve-ment in fit with the inclusion of school-level demographic or socioeconomicvariables in the case of major chain density at 800m (p=0.02) but not at400m (p=0.60) or 160m (p=0.31).723.4. Discussion3.4 DiscussionThis study offered a descriptive profile of the school food environments forpublic schools in Vancouver, BC, examining whether disparities in schoolchil-dren’s access to food retailers existed according to school enrolment of abo-riginal students, English Language Learners, or students living in poverty.The study additionally examined whether disparities could be explained byneighbourhood commercial density or neighbourhood socioeconomic depri-vation. As hypothesized, no consistent disparities in food outlet access wereobserved in association with student demographic characteristics, but school-level poverty was significantly and positively associated with density, at 400mand 800m, of convenience stores, and proximity to or density, at 800m, of su-permarkets or grocery stores. Associations were attenuated through adjust-ment for neighbourhood socioeconomic deprivation and commercial density,though the association of school-level poverty with the density of conveniencestores within 400m as well as with density of supermarkets/grocery storeswithin 800m of schools remained statistically significant in adjusted models.The density of food outlets observed in this study is relatively high for aCanadian city: Over ninety percent of Vancouver public schools were locatedwithin 800m of at least one food retailer, and a majority (58.4%) had at leastone limited-service food outlet within a 400m line-based buffer surroundingthe school. Similarly, a majority (54.9%) were located within 400m of atleast one convenience store. Across outlet types, the densities observed in thisstudy are considerably higher than those obtained by researchers studying allschools in the province of British Columbia (Black and Day, 2012), Quebec733.4. Discussion(Morin et al., 2015), Saskatoon, Saskatchewan (Engler-Stringer et al., 2014b),and the Montreal Urban Community (Kestens and Daniel, 2010). It shouldbe noted, however, that Vancouver is unique among Canadian municipalitiesin its high population density (Statistics Canada, 2016). In similarly denseBoston and more dense New York City, similarly high retailer densities wereobserved in the areas surrounding schools14 (Walker et al., 2013; Neckermanet al., 2010; Kwate and Loh, 2010).This study offered the first examination of school-level demographic char-acteristics in association with school food outlet access in Canada. No con-sistent associations were observed between the percent aboriginal studentsor proportion English Language Learners enrolled in Vancouver schools andthe food environments surrounding schools. This result is consistent with thefindings of Engler-Stringer et al. (2014b) of no associations between neigh-bourhood census demographics (aboriginal population or recent immigrants)and the proximities of food outlets to schools, but it diverges from studiesidentifying racial disparities in students’ access in the United States (Sturm,2008; Kwate and Loh, 2010; Neckerman et al., 2010). This research wasconducted, however, in a Canadian city with an ethnic and socioeconomiccomposition very different from that encountered in research in the UnitedStates (Census Bureau, 2005; Statistics Canada, 2006b).This study also examined the associations between school-level poverty,as measured by Vancouver’s Inner City Program, and the density and prox-imity of food outlets surrounding schools. Associations of ICP status and14Walker et al. (2013) obtain slightly lower mean and median proximity estimates,consistent with their use of straight-line rather than street-network based measures ofdistance.743.4. Discussionthe school food environment, in adjusted models, were inconsistent: at 400m,ICP schools had significantly more convenience stores than non-ICP schools;at 800m, positive and significant associations were observed between ICP sta-tus and and supermarket or grocery store density. Positive associations withconvenience store density are similar to those observed in previous studies(Engler-Stringer et al., 2014b; Robitaille et al., 2010), but the positive associ-ation of school poverty and supermarket or grocery store access is surprisinggiven evidence, from the United States, of gaps in grocery and supermarketaccess (“food deserts”) in low-income areas (Beaulac et al., 2009). However,the finding is consistent with research reporting better supermarket acces-sibility for low-income neighbourhoods in Montreal (Apparicio et al., 2007)and British Columbia (Black et al., 2011).This study did not find that school enrolment or secondary versus ele-mentary status were statistically significant predictors of the density of foodoutlets within 160, 400 or 800 metres of Vancouver schools. This finding is inkeeping with the observations of Neckerman et al. (2010) in New York City,but the result contradicts other studies associating school size or level andfood outlet density (Sturm, 2008; Black and Day, 2012). The differences inresults may be a product of different model specifications: while both thisstudy and the Neckerman et al. (2010) study controlled for neighbourhoodcommercial density, Sturm (2008) controlled only for whether a school waslocated in a rural, urban or suburban location and Black and Day (2012)included population/km2 as the measure of density15. The differences may15When the density models from this study were fitted with only school level covariates,total enrolment was a statistically significant predictor of the density of limited-serviceoutlets within 800m of Vancouver schools753.4. Discussionalso be geographic: this study focused on a large, dense municipality, whereoutlets have many sources of demand, while both the Sturm and the Blackand Day studies examined larger regions with many levels of population andcommercial density. It is also possible that the effect size was too small todetect given the power of the analysis in this study.Adjustment for neighbourhood factors attenuated the few associationsobserved between school-level demographic or socioeconomic characteristicsand the food environments surrounding schools. Proportion English Lan-guage Learners, which was significantly associated with limited-service out-let density at 160m, limited-service outlet and supermarket/grocery storeproximity, and major chain density at 400m, was no longer significantly as-sociated with any dependent variables in models adjusted for neighbourhodfactors. The association of ICP status and convenience store density within400m dropped in magnitude in adjusted models, but remained statisticallysignificant. The weakened associations, a finding similar to that observed byNeckerman et al. (2010), suggests that neighbourhood factors explain some ofthe associations observed between student demographic and socioeconomiccharacteristics and the food environments surrounding schools.As in Kestens and Daniel (2010), neighbourhood socioeconomic depri-vation was a significant predictor of outlet density even after adjusting forcommercial density. The most highly deprived neighbourhoods had morethan twice as many supermarkets or grocery stores within 400m and signifi-cantly more convenience stores within 800 metres than did the least deprivedneighbourhoods; a socioeconomic gradient was also observed in proximity toconvenience stores. The finding of high supermarket or grocery store den-763.4. Discussionsities in the most deprived school neighbourhoods (Figure 3.2) again alignswith studies suggesting that socioeconomic inequities in access to grocerystores (the “food desert” problem) is of less relevance in Canadian cities(Apparicio et al., 2007; Smoyer-Tomic et al., 2006; Black et al., 2011).Commercial density was statistically significant across models, with schoolsin more commercially dense areas having more access (either through higherdensity or shorter proximity) to food outlets of all types. This is in keepingwith previous findings (Kestens and Daniel, 2010; Neckerman et al., 2010;Day and Pearce, 2011). Likelihood ratio tests and partial F-tests confirmedthe importance of including neighbourhood factors; significance tests furtherfailed, across models, to reject the null hypothesis of no significant improve-ment in fit with the inclusion of school-level demographic or socioeconomicvariables in comparison with models including only commercial density andneighbourhood socioeconomic deprivation as explanatory variables. Givenincreasingly strong empirical evidence for the association of commercial den-sity and food outlet access, as well as the evidence presented in this study aswell as by Neckerman et al. (2010) and (Kestens and Daniel, 2010) that com-mercial density may confound the associations of school-level characteristicsand the food environment surrounding schools, this result suggests that re-searchers should control for commercial density in future food environmentsresearch.773.4. Discussion3.4.1 Endogeneity Concerns in Food EnvironmentsResearchStudies of disparities in the food environments surrounding schools oftenserve as a preliminary step for researchers seeking to examine disparities inchildren’s diet-related health or dietary behaviours in associations with thefood environments surrounding their schools. Some researchers have raisedconcerns that endogeneity may compromise the results of such studies. En-dogeneity refers to the correlation of an independent variable with the errorterm in a regression analysis, violating the assumption that E[|X] = 0, orthat the errors have conditional mean zero. Such a violation could be lead-ing researchers to biased or inconsistent results (Verbeek, 2012). Though theconcept encompasses a number of more specific problems, the cause of endo-geneity most relevant to food environments researchers is simultaneity—-theidea that just as the independent variable causes the dependent variable, thedependent variable also causally affects the independent variable16 (Verbeek,2012). As Currie et al. (2010) and Sturm (2008) point out, the location de-cision of a fast food outlet is not random: stores choose to locate in areas16Other problems encompassed by endogeneity include omitted variable bias, whichoccurs when researchers fail to account for a variable that is both related with the depen-dent variable and correlated with other explanatory variables, unobserved heterogeneity,or unmeasured variance across inviduals, and self-selection—the idea that students whoconsume higher intakes of snacks, sugar-sweetened beverages, or fast food meals wouldchoose to go to schools with easier access to convenience stores or fast food outlets. Omit-ted variable bias can be mitigated through the measurement and inclusion of the relevantfactors as well as through strong theoretical grounding (Clarke, 2005), unobserved hetero-geneity at the school level can be accounted for at the group-level residual in multilevelmodels (Dieleman and Templin, 2014), and self-selection is unlikely: Ries and Somerville(2010) offer evidence that Vancouver public schools accept few applications from studentsoutside of their catchment areas. For those students who do leave their catchment areas,factors like school quality and special program offerings are probably more important thanfood environments.783.4. Discussionwhere they expect high demand. Certainly easy access to food outlets couldlead students to purchase more fast food, but a population of students whoare particularly eager purchasers of fast food may also be appealing to a fastfood vendor who is choosing a new location.Fundamentally, the detection of simultaneity requires that researchersexplore the direction of the association between food outlet locations andchildren’s dietary choices. No observational study can answer such a ques-tion of causality, but the absence of correlations between outlet density orproximity in relation to schools and school characteristics would weaken thehypothesis of simultaneity. Applying likelihood ratio tests and partial Ftests to compare models with and without school characteristics, this studyfound no evidence of such associations for student demographic characteris-tics, school-level poverty, school size or school level and food outlet densityor proximity in Vancouver, BC., a result in keeping with that observed byCurrie et al. (2010). This study thus helps to assuage researchers’ concerns,though improved study designs and statistical approaches informed by causalinference remain necessary to ensure that endogeneity does not compromiseresults.3.4.2 Strengths and LimitationsThis study focused on the associations of student demographic character-istics and school-level poverty with the density or proximity of food outletssurrounding schools. Disparities may still exist in association with otherstudent characteristics such as student health; however, this study utilizedthe best available data to offer insight on demographic and socioeconomic793.4. Discussiondisparities in Vancouver school food environments. Similarly, this study ex-amined only the neighbourhood factors of socioeconomic deprivation andcommercial density; the contribution of transit stops or other built envi-ronment factors to the school food environment would be worthy of furtherexamination given their relevance for policy.Measurement error could also compromise the results of this research. Indiagnostic examinations of the OLS regressions, two schools (University HillElementary and University Hill Secondary) were high leverage observations.The two schools are located on the University of British Columbia campus,on the far Western edge of the city, outside of official municipal boundaries,and thus may be outside the bounds of the business license data; indeed,though there is evidence that University Hill Secondary School would havehad access to a supermarket within approximately 450m in 2012 (Cooper-smith, 2012), no outlet is found in the data set for that time. Coefficients andstandard errors in models including just the three school factors were robustto the exclusion of the two schools from the analysis. In models control-ling for neighbourhood factors, the association of limited service food outletdensity at 800m with medium VANDIX scores was no longer statistically sig-nificant; at 400m, the association of supermarket/grocery store density andhigh VANDIX scores is no longer statistically significant, although grocerystores density remains significantly higher near schools with medium versuslow VANDIX scores. Finally, the associations of the VANDIX and conve-nience store proximity are similarly attenuated, and no longer statisticallysignificant, in OLS models. Likelihood ratio tests and partial F test results,however, are robust to the omission of the University Hill schools from the803.5. Conclusionsmodels.The results of this study are not conclusive: both small sample size andmulticollinearity could inflate standard errors and thus obscure the statisticalsignificance of school-level predictors (Woolridge, 2009). However, varianceinflation factors were low (below 10) for all explanatory variables (Woolridge,2009); furthermore, the use of likelihood ratio tests and incremental F-teststo examine the simultaneous contribution of multiple variables to the modelreduces the potential impact of multicollinearity. While the study’s samplesize may have been too small to detect significant effects at the school level,the number of schools examined was the maximum possible given the studyregion. Multiple sensitivity analyses supported the robustness of findings toerrors in the assessment of school-level poverty, the classification scheme forfood outlet types, and the measurement of the food environment surroundingVancouver schools.3.5 ConclusionsUltimately, this study does not find consistent evidence that the char-acteristics of Vancouver public schools are associated with students’ accessto limited-service food outlets, convenience stores, or supermarkets. Thisstudy focused on demographic and socioeconomic characteristics of studentswithin schools; further research is needed to examine whether other factorssuch as school food policies (e.g. open versus closed campuses) or programs(e.g. universal breakfast or school lunch offerings) are also associated withthe food environment surrounding schools. Nevertheless, the results of this813.5. Conclusionsstudy suggest that while some disparities may exist in students’ access to con-venience stores according to school poverty, neighbourhood characteristics—particularly commercial density—play a more important role than studentdemographic or socioeconomic characteristics in predicting food outlet den-sity and proximity.82Chapter 4Associations of School FoodEnvironment Measures andChildren’s School-Day DietaryIntakes4.1 IntroductionCanadian adolescents have the poorest quality diets of any Canadianage group (Garriguet, 2009). According to the most recent iteration of theCanadian Community Health Survey: Nutrition (2004), Canadian childrenand youth consumed more sugar—and more sugar from soft drinks—thanother Canadians (Langlois and Garriguet, 2011). Furthermore, over 80%of Canadian adolescents exceeded the Institute of Medicine’s recommendedupper limits for sodium consumption (Garriguet, 2007). Excessive sugar con-sumption, particularly consumption of added sugars from sugar-sweetenedbeverages, is a risk factor for dental caries, obesity, and type 2 diabetesmellitus (The Lancet Diabetes Endocrinology, 2015; Malik et al., 2010); ex-834.1. Introductioncessive sodium consumption is associated with hypertension and cardiovas-cular disease (World Health Organization, 2012). Canadian children con-sume approximately one third of their weekday calories during school hours(Tugault-Lafleur et al., 2016), making schools a potentially critical leveragepoint for the amelioration of children’s diets.Policy efforts aimed at improving Canadian children’s school-day dietshave largely focused on nutrition education and food access within schools(Leo, 2007; Lassard, 2006). In British Columbia, for example, foods soldwithin schools are required to meet the Guidelines for Food & Beverage Salesin BC Schools (Ministry of Health and Ministry of Education, 2013). Van-couver schools can also participate in nutrition education programs such asAction Schools! BC, Farm to School BC, Sip Smart! BC, and the BritishColumbia School Fruit and Vegetable Nutritional Program (Romses andLam, 2015). The efficacy of such within-school interventions may be lim-ited, however, if interventions lead more students to purchase food fromoff-campus sources.A number of studies have sought to elucidate the impacts of children’saccess to off-campus food sources on diet and ultimately on the prevalenceof diet-related disease (Engler-Stringer et al., 2014a; Williams et al., 2014).Most studies of the public health impacts of the school food environmenthave focused on associations with obesity (Caspi et al., 2012). Results havebeen mixed: several researchers found positive associations between foodoutlet density or proximity and body mass index (BMI) (Gilliland et al.,2012; Leatherdale et al., 2011; Howard et al., 2011; Currie et al., 2010; Grierand Davis, 2013; Alviola et al., 2014), but other researchers reported null or844.1. Introductioninconsistent results (Seliske et al., 2009a; Griffiths et al., 2014; Langellier,2012; Harris et al., 2011). Obesity is a distal outcome, however, producedby many complex factors interacting over time; small, cross-sectional studiesmay be inadequate to separate the effects of food access on caloric intakefrom other long-term contributors to BMI such as neighbourhood walkability(Saelens et al., 2003; Cobb et al., 2015).Fewer studies have examined the more proximal outcome of dietary qual-ity in association with school food environment measures, though the re-search that has been conducted has uncovered statistically and clinicallysignificant results. In London, Ontario, He et al. (2012b) found that chil-dren’s Healthy Eating Index scores were higher (better) for students with nofast food outlets within 1km of their schools in contrast with students whodid have a fast food outlet within 1km. Two U.S. studies similarly foundstatistically significant associations of fast food outlet proximity to schoolsand students’ sugar-sweetened beverage intake (Laska et al., 2010; Davisand Carpenter, 2009). However, the evidence remains inconsistent: severalstudies failed to find robust significant associations between dietary intakeand the food environments surrounding schools (Gebremariam et al., 2012;Van Hulst et al., 2014; Richmond et al., 2013; An and Sturm, 2012).In addition to the limited number of studies examining children’s school-day dietary intakes, there are persistent gaps in the literature. Foremost,Canadian studies associating the food environment and children’s diet-relatedhealth have either been national in scale (Seliske et al., 2009a, 2013; Hérouxet al., 2012; Laxer and Janssen, 2013) or have been conducted in London,854.1. IntroductionOntario (Gilliland et al., 2012; He et al., 2012a,b)17. No study of the ef-fect of the school food environment on children’s diets has been conductedin one of Canada’s three biggest cities—Toronto, Montreal, or Vancouver—despite evidence that Vancouver students may be exposed to far more densefood environments from students elsewhere in Canada (see Chapter 3). Anadditional gap is an exclusive reliance on objective measures of the food envi-ronment: all of the aforementioned studies used GIS to measure school foodenvironments, but a recent review finds a need for further examination ofmeasures of the perceived food environment (Williams et al., 2014). Finally,the majority of Canadian school food environments studies obtained food re-tailer locations from Yellow Pages (Seliske et al., 2009a; Laxer and Janssen,2013; Héroux et al., 2012; Seliske et al., 2013), despite evidence that thesedata sources perform less well than government data sources (Fleischhackeret al., 2012; Lake et al., 2010) and may substantially misrepresent spatialdistributions of food retailers (Longacre et al., 2011). There is thus a needfor a study with high-quality data, using both perceived and objective foodenvironment measures, to be conducted in one of Canada’s more populouscities.This study seeks to contribute to the school food environments literaturethrough an examination of the associations between Vancouver 5th - 8thgrade students’ school-day consumption of sugar-sweetened beverages, fastfoods, or packaged snacks and the proximity or density of food retailerssurrounding their schools. The primary research objective was to examine17One additional study was conducted in the province of Ontario more generally(Leatherdale et al., 2011).864.2. Methodswhether elementary and secondary students’ dietary intakes at or en-route toschools were associated with access to fast food outlets, convenience stores, orsupermarket/grocery stores surrounding schools; a secondary objective wasto evaluate the potential of survey-based measures of the food environmentto serve as an alternative to objective measures through an assessment of theagreement between students’ perceived proximity and objective proximity toeach type of food outlet.4.2 MethodsStudents’ dietary intakes at or en-route to school were assessed with theIndividual Eating Assessment Tool (I-EAT) as part of the Food Practices onSchool Days Study. Questions and protocols were adapted from survey toolspreviously developed and validated for the study of eating behaviours inelementary and secondary school students (Birnbaum et al., 2002; Hanninget al., 2009; Pawlak and Malinauskas, 2008). The protocol was pilot-testedwith 10 content experts as well as with 54 students in grades 7 - 12; a revisedprotocol was further field tested with an additional class of grade 6 and 7students. Researchers visited each participating class between March andJune 2012 to facilitate completion of computerized surveys18.The sampling approach was similar to a two-stage cluster sample. First,schools were recruited from each of the six geographic sectors of the Van-couver School Board, ensuring that schools in neighbourhoods with differing18The development and design of the survey was part of the thesis work of NaseamAhmadi and Teya Stephens; further details can thus be obtained from their publications(Ahmadi et al., 2015; Stephens et al., 2016)874.2. Methodslevels of socioeconomic deprivation and commercial density would be rep-resented in the study (Vancouver School Board, 2012). Next, teachers andadministrators were invited to the study; all students in a sampled classparticipated unless a parent dissented, a student dissented, or a teacher re-quested that a student be excluded19. The final sample included 964 students(student-level participation rate: 81%) from twenty elementary schools andsix secondary schools (School-level participation rate: 74%). Fourteen sur-veys were excluded due to inappropriate answers, allowing for a final samplesize of 950 students in 26 schools. The Behavioural Research Ethics Board atthe University of British Columbia and the Vancouver Board of Educationapproved all protocols.4.2.1 Dependent Variables: Dietary IntakeDietary intake measures served as the dependent variables in this study.Student dietary intake was assessed with a modified food frequency ques-tionnaire adapted from the Student Health Action Planning & EvaluationSystem (SHAPES) Healthy Eating Module (University of Waterloo, 2008).Food frequency questionnaires are a practical and cost-effective means ofassessing dietary intake (Willett, 2012); furthermore, several studies havedocumented acceptable levels of validity and reliability for similar food fre-quency questionnaires assessing food group intake in older children (Specket al., 2001; Wong et al., 2012). However, all self-reported measures are sub-ject to error. In the case of food frequency questionnaires, several studies19Teachers excluded students based on behavioural or learning challenges and englishlanguage proficency, among other reasons.884.2. Methodsoffer evidence that subjects misreport total intake (Deschamps et al., 2009;Perks et al., 2000; Watson et al., 2009). As a result, Lietz et al. (2002)suggest that researchers should not rely on FFQs to evaluate children’s ab-solute dietary intakes, but that the approach remains reliable and valid foruse in comparison through the ranking of children’s intakes. This study thusused daily intake measures to compare students who were frequent versusinfrequent consumers of particular categories of foods.For each food item, students were asked whether they consumed eachitem “never”, “once a month or less”, “2 - 3 times a month”, “once a week”,“2 - 4 times a week”, “once a day”, or “2 or more times a day”. Responseswere then summed across intake categories of fruits, vegetables, whole grains,and low-fat milks (all considered “more nutritious” foods) or sugar-sweetenedbeverages, fast food, and processed snacks (“minimally nutritious” foods, Ta-ble 4.1), with groupings adapted from Canada’s Food Guide (2011) and theBritish Columbia Ministry of Education Food and Beverage Sales Guidelines(2013). Combined responses were split into a binary variable equal to 1 ifreported intake was “daily”—defined as a summed response ≥ 20 times permonth—and 0 otherwise. For this chapter, results are presented for students’daily consumption of foods in each of the three minimally nutritious intakecategories given their theoretical relevance to the study of food retailers20.20Analyses were additionally tested with daily intake of “more nutritious” foodcategories—fruits, vegetables, whole grains, and low fat milk—but no statistically sig-nificant associations were observed with school food environment measures in adjustedmodels with daily versus less-than-daily intake of vegetables or whole grains. While asso-ciations were identified for low-fat milk intake and daily fruit intake, access to conveniencestores and fast food outlets was more relevant in the context of the minimally nutritiousfood items—which are easily purchased at such stores—rather than in potentially lessaccessible food items like whole grains and low-fat milk (Glanz et al., 2007; Saelens et al.,2007). This thesis chapter thus focuses on minimally nutritious foods.894.2. MethodsTable 4.1: Minimally nutritious intake categories and their component fooditemsIntake Category DefinitionSugar-Sweetened Fruit-flavoured drinksBeverages Regular pop or soft drinks†Iced tea‡Sports drinksEnergy drinksSlurpees®, slushees, or snow conesFast Foods PizzaHotdogsHamburgers or cheeseburgersBreaded or fried chicken/fishFries or other fried potatoesTacos or nachosFrozen packaged dinnersPackaged Snacks Frozen dessertsBaked sweetsCandy or chocolate barsSalty packaged snacksList of items included in the food frequency questionnairein the Individual Eating Assessment Tool (I-EAT).†Not including diet drinks; ‡sugar-sweetened4.2.2 Independent Variables: Food Environment MeasuresThere is no consensus on the optimal measurement of the food environ-ment surrounding schools (Black, 2015). This study thus used several setsof measures: objective food outlet density, objective food outlet proxim-ity, and students’ perceived proximity to food outlets. The two objectivemeasures were constructed from the City of Vancouver (2016) Business Li-904.2. Methodscenses data validated in Chapter 2, which was limited to retailers in opera-tion when the IEAT surveys were conducted (see Figure 3.1), and classifiedas limited-service food outlets, convenience stores, or supermarket/grocerystores following the approach used in Section 2.3.1. Following the descrip-tion in Section 2.3.2, objective proximity was evaluated as the shortest streetnetwork-based distance from a school to a food outlet. Objective density wasdefined as the total count of food outlets located within a line-based buffersurrounding each school. For the main analyses in this model, objectivedensity was evaluated within 400 metre line-based buffers of each school—adistance at which 69% of all Vancouver schools had at least one food outletduring the study period (Section 3.3)—but associations were also tested forline-based buffers within 800m, the distance most commonly used in the foodenvironments literature (Williams et al., 2014)21.Although most studies associating food environments and adolescenthealth rely on objective measures of access (Williams et al., 2014), recentresearch suggests that perceived measures may be more strongly associatedwith adolescent intake of minimally nutritious foods (Hearst et al., 2012;Svastisalee et al., 2015). A perceived measure of proximity was thus derivedfrom a module in the I-EAT survey. Students were asked to estimate howlong it would take them to obtain a variety of foods including fast food,salty packaged snacks, and fruits or vegetables; possible responses were “lessthan 5 minutes”, “5–10 minutes”, “10–15 minutes”, “more than 15 minutes”and “I don’t know”. A five minute walk is approximately equivalent to a21Currie et al. (2010) recommends evaluating density within 160m, a 1-2 minute, butjust two schools in the I-EAT survey had food outlets within such a short distance.914.2. Methods400m distance (Pikora et al., 2002), so responses were recoded into threecategories—<5 minutes, 5–10 minutes, and ≥10 minutes—to be comparableto the measure of objective proximity. Each student’s perceived proximityto limited-service outlets was assessed as the minimum of that student’s re-ported distance from a source of fast food or French fries; proximity to conve-nience stores was the minimum reported distance to candy or salty packagedsnacks; and distance to a grocery store was the minimum reported distancea student would need to walk to obtain fruits or vegetables. Although thesedefinitions are slightly different from those used in classifying stores, theywere the classifications most similar to those used in Appendix A, given thelimitations of the I-EAT survey.4.2.3 Independent Variables: ControlsThis study additionally included student-level measures of gender, child-hood food security, acculturation, bringing lunch from home, and spendingmoney, as well as school-level median family income and school level (ele-mentary versus secondary) given evidence that these variables associate withchildren’s dietary behaviours (Svastisalee et al., 2015; Velazquez et al., 2015;Kirkpatrick et al., 2015; Sanou et al., 2014; Hanson and Chen, 2007) as wellas—for the student-level variables—their inclusion in the I-EAT tool.Self-reported gender was included based on previous evidence observingstronger associations between perceived access to food retailers and children’sfast food intake (Svastisalee et al., 2015). Childhood food security—definedby Coleman-Jensen et al. (2014) as access to adequate food to be healthy andactive—was additionally included given evidence of poorer dietary quality924.2. Methodsin food insecure children in comparison with food secure children (Velazquezet al., 2015). The concept was measured with five questions from a tool de-veloped by the United States Department of Agriculture (Economic ResearchService, 2012), included in Table 4.2. Children were considered food inse-cure if they responded “sometimes” or “a lot” to two or more of the questionsincluded in the module.Table 4.2: Individual Eating Assessment Tool: Food insecurity moduleIn the past 12 months: Never Some- Atimes LotDid the food that your family bought run out, ◦ ◦ ◦and you didn’t have money to get more?Were you not able to eat a balanced meal be- ◦ ◦ ◦cause your family didn’t have enough money?Have you skipped a meal or has the size of ◦ ◦ ◦your meals been cut because your familydidn’t have enough money for food?Did you have to eat less because your family ◦ ◦ ◦didn’t have enough money to buy food?Were you hungry but didn’t eat because your ◦ ◦ ◦family didn’t have enough food?The Vancouver School Board serves a diverse student body, includingmany students who have recently arrived in Canada (Vancouver SchoolBoard, 2016) and thus may have dietary practices informed by their spe-cific cultural backgrounds. Velazquez et al. (2015) thus developed a measureof “acculturation” as a proxy for factors related to immigrant status. Studentacculturation was considered “high” for students who reported speaking En-934.2. Methodsglish at home, who reported being born in Canada, and who reported thattheir parents or guardians were born in Canada. Acculturation was consid-ered “low” for students who reported speaking a language other English athome, who were born outside Canada, and whose parents or guardians wereborn outside Canada. “Medium” acculturation comprised students with amixed set of responses to the three acculturation questions.Students who report bringing lunch from home (1 if students bring foodon approximately a daily basis, 0 otherwise) may be less likely to make pur-chases at food vendors. Spending money, split into four categories (“none”,$0 - $10, $10 - $20, and >$20), was similarly expected to affect the ability ofstudents to make purchases at nearby food retailers. School level (secondaryversus elementary) was included due to increased autonomy of secondaryschool students, who are more likely to attend schools with open-campuspolicies where students are free to leave school grounds during lunchtime.Finally, school-level median family income was included in adjusted mod-els following Velazquez et al. (2015). Several studies have reported associ-ations between measures of neighbourhood income or socioeconomic statusand children’s dietary behaviours (Minaker et al., 2006; Velazquez et al.,2015), and previous research has shown a socioeconomic gradient in foodoutlet proximity to or density surrounding schools (Morin et al., 2015; Dayand Pearce, 2011; Zenk and Powell, 2008) as observed in Chapter 3. Themeasure of school-level median income used in the study, obtained from theBC Ministry of Education, was constructed from 2006 Canadian Census dis-semination area-level measures and weighted according to the proportion ofstudents enrolled in the school residing in each dissemination area.944.2. Methods4.2.4 Data AnalysisDescriptive StatisticsDescriptive statistics were calculated for all variables included in models.Response frequencies were tabulated for categorical variables; descriptivestatistics calculated for perceived proximity included frequencies of each re-sponse. For continuous variables, summary statistics included means andstandard deviations. Finally, missing observations were tabulated for eachvariable constructed from the I-EAT survey.Reliability of Perceived Measures of ProximityThe reliability of students’ perceptions of proximity was evaluated incomparison with objective proximity. Objective proximity for each outlettype was split into three categories (0 –400 metres, 400–800 metres, or ≥800metres) comparable to the categories of perceived proximity (<5 minutes,5–10 minutes, or ≥10 minutes), and agreement was evaluated with Cohen’sKappa.Multilevel ModelsBivariate and multivariate regression models were then applied to ex-amine associations between food environment measures and dietary intakeoutcomes; multilevel logistic models with random intercepts were fitted toaccount for the correlation of errors between students attending the sameschool (Diez Roux, 2004; Singer and Willett, 2003). The varying-interceptmultilevel model comprised two levels: Eq. 4.1 predicting individual out-954.2. Methodscomes and Equation 4.2 predicting group-level intercepts. For student i inschool j,log(piij1− piij ) = β0j + β1xij (4.1)β0j = γ00 + γ01zj + u0j (4.2)where piij is the student’s probability of daily consumption of the dietaryintake of interest. The level 1 model includes a coefficient β1 for individual-level explanatory variable xij and an intercept β0j that consists of, at level 2,an overall intercept γ00, a coefficient γ01 for the contribution of school-levelvariable zj to differences between schools, and school-level residual error u0j(Singer and Willett, 2003).Several iterations of modeling were conducted with each of the dependentvariables. First, null models with varying intercepts were fitted and intraclasscorrelation coefficients (ICCs) were calculated to compare within-group andbetween-group variation. The ICC measures the percent of total varianceattributable to variation at the group level (Singer and Willett, 2003). Al-though ICCs were small across models (0.042 - 0.076), Wald tests found thatvariance was significantly different 0, meaning that there was unexplainedvariance at the school level. In addition, likelihood ratio tests comparingthe multilevel model with simple logistic regression further confirmed thatincluding random intercepts offered a significant improvement in fit.Next, bivariate varying-intercept models were fitted with each of the ob-jective measures of density and proximity as well as with the individual-levelmeasures of perceived proximity. Additional explanatory variables, discussed964.2. Methodsin Section 4.2.3, were then iteratively included in multivariate models. Fi-nally, varying-intercept, varying-slopes model were fitted. Including a ran-dom slope did not significantly improve model fit; thus only the more parsi-monious varying intercept model is reported here. For all multi-level models,odds ratios and 95% confidence intervals were evaluated as measures of thedirection and strength of associations.Data were missing from 0.2% - 33.7% of responses on variables con-structed from the I-EAT survey, so observations were imputed with multipleimputation by chained equations (10 data sets)22. Following Von Hippel(2007), this study used multiple imputation then deletion, including depen-dent variables in the imputation but omitting imputed dependent observa-tions in the analysis. All statistical analyses were completed with STATA14 (StataCorp, 2015).4.2.5 Sensitivity AnalysesAlthough multiple imputation was used to impute missing observationsin the final analyses, models were also fitted with listwise deletion of missingobservations. In addition, final models were fitted with cutoffs of weekly (≥4x per month) versus less-than-weekly consumption to assess the robustnessof the results to the particular cutoff used to create binary outcomes. Multi-level OLS regressions with students’ self-reported frequency of consumptionper month as continuous outcomes served to further test the effect of di-chotomizing the dependent variables.22Jean-Michel Billette wrote the original multiple imputation code used by Velazquezet al. (2015); my code was adapted from his template.974.3. Results4.3 ResultsThe final sample comprised 950 students across 20 elementary schoolsand 6 secondary schools. Almost all students (98.6%) in the final samplewere in grades 6 - 8; due to the inclusion of split classes, the study alsoincluded 13 grade 5 students23.The study included slightly more male (51.4%) than female students andsignificantly more elementary (74.7%) than secondary school students. Asubstantial number of students (n=131, 15.8%) reported at least some levelof household food insecurity and most students (81.9%) either were not bornin Canada, had parents or guardians who were not born in Canada, or spokea language other than English or French at home. A majority of students(57.9%) reported bringing lunch from food at home on a daily basis, andmany students (85.9%) reported access to at least some spending money.Full descriptive statistics can be found in Table 4.3.As can be seen in Table 4.3, many students reported frequent consump-tion of minimally nutritious foods at or en-route to school: 294 students(31.4%) were classified as daily consumers of sugar-sweetened beverages,162 students (17.2%) were daily consumers of fast food, and 192 students(20.3%) were daily consumers of packaged snacks.23Similarly, due to the inclusion of split classes, a small number of students were quiteyoung. Though 88.9% of students were age 12 or older, the sample also included 8911-year-old students and 8 10-year-old students.984.3. ResultsTable 4.3: Sample characteristics (n=950)Count % N Missing (%)Daily Consumers ofSugar-Sweetened Beverages 294 31.4 14 (1.5%)Fast Food 162 17.2 8 (0.8%)Packaged Snacks 192 20.3 2 (0.2%)Gender 2 (0.2%)Female 461 48.6Male 487 51.4Food Security Status 19 (2.0%)Food Secure 700 84.2Food insecure 131 15.8Acculturation 73 (7.7%)high 159 18.1medium 619 70.6low 99 11.3Brought Lunch From Home 11 (1.2%)Daily 544 57.9Less than daily 395 42.1Spending Money 320 (33.7%)None 89 14.1$0 - $10 234 37.1$10 - $20 146 23.2>$20 161 25.6School Level 0 (0%)Elementary 710 74.7Secondary 240 25.3Mean Std Dev RangeMedian Family Income† $60,393 $11,744 $33,928 - $82,823†School-level variable constructed by the BC Ministry of Education994.3. ResultsMost schools (69.2%) were located within 400 metres of some food out-let (Table 4.4.) Fifteen schools (57.7%) were located within 400m of atleast one limited-service food outlet, fourteen schools (53.9%) were locatedwithin 400m of at least one convenience store, and two schools (7.69%) werelocated within 400m of a supermarket or grocery store as measured by ob-jective proximity. Sixteen schools (61.5%) were located within 800 metresof a supermarket or grocery store, while 25 schools (96.2%) were within 800metres of a limited-service food outlet and 24 schools (92.3%) were within800 metres of a convenience store. Just two schools had access to food outletswithin 160m; in both cases, the outlets were convenience stores. On average,schools had almost 6 limited service outlets within 400 metres.Perceived proximities displayed noteworthy differences from objectiveproximities. Agreement between students’ perceived proximities and objec-tive proximity ranged from 42.0% for supermarket or grocery stores (Cohen’sKappa = 0.106, p<0.001) to 55.2% for convenience stores (Cohen’s Kappa= 0.181, p<0.001); agreement for limited-service food outlets was 42.8%(Cohen’s Kappa = 0.114, p<0.001). Although student responses were sig-nificantly different from the responses that would be expected if studentswere answering randomly, the strength of the Kappa statistic was less than0.2 in all cases, or “slight” according to the scale proposed by Landis andKoch (1977). In comparison with the gold standard of objective proximity,perceived measures of proximity thus do not seem to offer reliable estimatesof student access to food outlets.1004.3. ResultsTable 4.4: Descriptive statistics for objective density, objective proximityand perceived proximity of food outlets in association to Vancouver SchoolsMedian Mean Std. DeviationDensity within 800m (n=26)Limited-Service Outlets 18 18.8 13.5Convenience Stores 6 8.5 8.0Supermarket/Grocery Stores 1 2.1 2.2Density within 400m (n=26)Limited-Service Outlets 4 5.8 6.6Convenience Stores 1 2.6 3.6Supermarket/Grocery Stores 0 0.3 0.7Density within 160m (n=26)Limited-Service Outlets 0 0.2 0.7Convenience Stores 0 0.4 0.9Supermarket/Grocery Stores 0 0 0Proximity in metres (n=26)Limited-Service Outlets 341 391 174Convenience Stores 370 412 221Supermarket/Grocery Stores 648 776 330Count % N Missing (%)Perceived Proximity (n=950)Limited-Service Outlets 203 (21.4%)< 5 minutes 275 36.85 - 10 minutes 254 34.0>10 minutes 218 29.2Convenience Stores 177 (18.6%)< 5 minutes 467 60.45 - 10 minutes 213 27.6>10 minutes 93 12.0Supermarket/Grocery Stores 254 (26.7%)< 5 minutes 205 29.55 - 10 minutes 249 35.8>10 minutes 242 34.81014.3. ResultsThe null model with log-odds of daily intake of sugar-sweetened beveragesas the dependent variable had an intraclass correlation coefficient (ICC) of0.042 (p < 0.01), meaning that approximately 4% of the total variance in themodel was at the school level. The ICC was not noticeably decreased throughthe inclusion of proximity to limited-service outlets, proximity to conveniencestores, or proximity to supermarkets/grocery stores in the model, suggestingthat objective proximity could explain little of the between-school variance instudents’ odds of daily sugar-sweetened beverage intake. Including objectivedensity at 400m of convenience stores led to a decrease in ICC to 0.038,but likelihood ratio tests failed to reject a null hypothesis of no significantimprovement in model fit.Approximately 7.5% of the total variance in the null model for log-odds ofdaily intake of fast foods was at the school level (p=0.01). Neither includingobjective proximity variables or objective density (at 400 or 800 metres) ledto a notable decrease in ICC.Finally, 4.6% of the total variance in the null model for log-odds of dailyintake of packaged snacks was at the school level (p < 0.01). Includingobjective proximity to convenience stores led to a small decrease in ICC to0.040; however, a likelihood ratio test again failed to identify a significantimprovement in model fit with the inclusion of proximity variables, and noother food access variables led to a noticeable reduction in ICC.1024.3. ResultsTable 4.5: Bivariate associations of outlet proximity and students’ dailyintakes of minimally nutritious foodsa at or en-route to school from multilevellogistic regression modelsProximityb Sugar-Sweetened Beverages (n=936†)Ltd. Service Outlet 0.97(0.86 - 1.09)Conv. Store 0.97(0.88 - 1.07)Grocery Store 0.98(0.92 - 1.05)Fast Foods (n=942†)Ltd. Service Outlet 1.00(0.86 - 1.17)Conv. Store 0.95(0.84 - 1.08)Grocery Store 0.96(0.89 - 1.05)Packaged Snacks (n=948†)Ltd. Service Outlet 1.08(0.96 - 1.23)Conv. Store 1.04(0.93 - 1.15)Grocery Store 1.01(0.94 - 1.08)Results are from multilevel logistic models with school random interceptsOdds Ratios are reported; 95% confidence intervals are in parentheses.aDependent variables = 1 if a student reported daily consumptionbDistance to the nearest outlet, reported in increments of 100 metres†Cases with missing dependent variables were omitted from the analysisNo statistically significant associations were identified.1034.3. ResultsUnadjusted multilevel models examining the bivariate associations of ob-jective food outlet proximity and students’ odds of daily intake of sugar-sweetened beverages, fast foods, or packaged snacks did not show evidenceof odds ratios significantly different from 1 (Table 4.5). Similarly, coefficientswere not statistically significant for bivariate models examining the associa-tions of objective density within 400m (Table D.1) or 800m (Table D.2).In the case of perceived proximity (Table D.3), one association was sta-tistically significant: students who reported that their school was more thana ten-minute walk from a grocery store had significantly lower odds of dailysugar-sweetened beverage intake (OR=0.63, 95% CI 0.41 - 0.96). This is asingle significant association in 36 tested unadjusted associations, however,making a false positive the most likely explanation (Rothman, 1990).After adjusting for multiple factors (Table 4.6), no statistically significantassociations were observed between objective proximity of limited-servicefood outlets, convenience stores, and grocery stores at the 5% level. Adjustedmodels were also fitted with objective density within 400m and 800m andstudents’ perceived proximity to sources of fast food, sources of packagedsnacks, and sources of fruits or vegetables. Densities of convenience storesat 400m and at 800m were associated with students’ odds of daily snackintake, but the directions of the associations (OR=0.93, 95% CI 0.86 - 1.00and OR=0.96, 95% CI 0.93 - 0.99, respectively) were the inverse of theassociation that theory would expect. No other measures of density at 400mor 800m were significantly associated with students’ odds of daily minimallynutritious dietary intakes, and no associations with perceived proximity werestatistically significant in the adjusted models.1044.3.ResultsTable 4.6: Multivariate adjusted associations from multilevel logistic models of outlet proximity and students’ oddsof daily intake of minimally nutritious foods at or en-route to schoolSugar-Sweetened Beveragesa Fast Foodsa Packaged SnacksaProximitybLtd. Service 1.01 1.07 1.14Outlet (0.90 - 1.13) (0.92 - 1.25) (1.00 - 1.30)Convenience 1.00 0.99 1.05Store (0.91 - 1.10) (0.87 - 1.12) (0.94 - 1.18)Grocery Store 0.99 0.97 1.01(0.93 - 1.05) (0.90 - 1.05) (0.94 - 1.08)ControlsGenderFemale – – – – – – – – –Male 1.73∗∗∗ 1.73∗∗∗ 1.73∗∗∗ 2.41∗∗∗ 2.41∗∗∗ 2.41∗∗∗ 1.43∗ 1.44∗ 1.44∗(1.29 - 2.33) (1.29 - 2.33) (1.29 - 2.33) (1.62 - 3.57) (1.62 - 3.57) (1.62 - 3.58) (1.02 - 2.00) (1.03 - 2.01) (1.03 - 2.02)Food 1.55∗ 1.55∗ 1.55∗ 1.87∗ 1.86∗ 1.85∗ 0.88 0.87 0.87Insecured (1.02 - 2.35) (1.02 - 2.34) (1.02 - 2.34) (1.12 - 3.12) (1.12 - 3.09) (1.11 - 3.08) (0.54 - 1.43) (0.54 - 1.41) (0.54 - 1.41)Acculturationehigh – – – – – – – – –medium 1.07 1.07 1.07 2.14∗ 2.17∗ 2.17∗ 0.78 0.79 0.79(0.69 - 1.63) (0.69 - 1.64) (0.70 - 1.64) (1.12 - 4.07) (1.14 - 4.12) (1.14 - 4.13) (0.49 - 1.25) (0.50 - 1.27) (0.50 - 1.27)low 1.68 1.69 1.70 7.08∗∗∗ 7.29∗∗∗ 7.31∗∗∗ 1.46 1.50 1.52(0.96 - 2.96) (0.96 - 2.97) (0.97 - 2.98) (3.29 - 15.19) (3.38 - 15.69) (3.40 - 15.72) (0.78 - 2.71) (0.81 - 2.79) (0.82 - 2.82)Brought from 1.07 1.07 1.07 1.05 1.05 1.06 1.77∗∗ 1.74∗∗ 1.77∗∗home dailyc (0.78 - 1.47) (0.78 - 1.47) (0.78 - 1.47) (0.69 - 1.59) (0.69 - 1.60) (0.69 - 1.60) (1.23 - 2.54) (1.21 - 2.51) (1.23 - 2.55)1054.3.Results(Cont.) Sugar-Sweetened Beveragesa Fast Foodsa Packaged SnacksaSpendingMoneyNone – – – – – – – – –$0 - $10 0.91 0.91 0.91 1.85 1.85 1.86 0.99 0.99 0.99(0.53 - 1.56) (0.53 - 1.56) (0.54 - 1.56) (0.80 - 4.27) (0.80 - 4.28) (0.81 - 4.30) (0.55 - 1.79) (0.55 - 1.79) (0.55 - 1.79)$10 - $20 1.15 1.15 1.15 3.07∗ 3.08∗ 3.09∗ 1.45 1.45 1.46(0.65 - 2.02) (0.65 - 2.02) (0.65 - 2.03) (1.22 - 7.74) (1.22 - 7.77) (1.22 - 7.79) (0.74 - 2.82) (0.74 - 2.83) (0.75 - 2.84)>$20 1.91 1.91 1.91 5.29∗∗ 5.32∗ 5.31∗ 1.64 1.64 1.65(0.99 - 3.67) (1.00 - 3.67) (1.00 - 3.67) (2.06 - 13.61) (2.08 - 13.65) (2.07 - 13.62) (0.88 - 3.04) (0.88 - 3.05) (0.89 - 3.06)School LevelElementary – – – – – – – – –Secondary 1.32 1.31 1.30 1.81∗ 1.71 1.71 1.50 1.44 1.37(0.85 - 2.04) (0.85 - 2.03) (0.85 - 2.00) (1.04 - 3.15) (0.97 - 3.03) (0.98 - 2.98) (0.90 - 2.50) (0.84 - 2.50) (0.81 - 2.34)Median Family 0.83∗ 0.83∗ 0.84∗ 0.72∗∗ 0.75∗ 0.76∗ 0.75∗∗ 0.78∗ 0.80∗Income† (0.69 - 0.98) (0.70 - 0.99) (0.71 - 0.98) (0.58 - 0.91) (0.60 - 0.95) (0.61 - 0.95) (0.61 - 1.66) (0.63 - 0.97) (0.65 - 0.99)N 936‡ 936‡ 936‡ 942‡ 942‡ 942‡ 948‡ 948‡ 948‡ICC 0.02∗∗ 0.02∗∗ 0.02∗∗ 0.04∗ 0.04∗ 0.04∗∗ 0.03∗∗ 0.04∗∗ 0.04∗∗Each column reports a model with dietary intake as the dependent variable and objective proximity as independent variablesadjusted for gender, food insecurity, bringing lunch from home, acculturation, spending money and school median income.Coefficients are reported as odds ratios with 95% confidence intervals in parentheses.aDependent variables = 1 if consumed at least daily. bDistance to nearest outlet, reported in units of 100 metres.cBrought from home = 1 if a student reported bringing lunch daily; dReference level is food secure students†School-level variable constructed by the BC Ministry of Education; reported in $10,000 units‡Missing values on independent variables were imputed with Multiple Imputation with Chained Equations (MICE, 10 data sets)missing dependent observations were included in MICE omitted from models.∗significant at 0.05; ∗∗significant at 0.01; ∗∗∗significant at 0.0011064.3. ResultsAll other explanatory variables were significantly associated with at leastone of the three dietary intake outcomes. For all three dependent variables,male students had significantly higher odds of being daily consumers thandid female students. Food insecurity was associated with increased odds ofdaily sugar-sweetened beverage consumption and daily fast food consump-tion, though not with increased odds of packaged snack consumption amongstudents. Similarly, the least acculturated students had more than seventimes the odds of being daily fast food consumers in comparison with themost acculturated students in the sample. Though students who broughtlunch from home on a daily basis did not have odds significantly differentfrom 1 of daily versus less frequent sugar-sweetened beverage consumptionor fast food consumption, bringing lunch from home was associated withincreased odds of daily packaged snack consumption. Finally, students with$10 or more of spending money had increased odds of being daily consumersof fast foods. The inclusion of school level (elementary or secondary) andmedian family income reduced the amount of unexplained variance observedat the school level, but one-sided Wald Tests confirmed that ICC’s remainedsignificantly greater than zero at the 2.5% significance level.4.3.1 Sensitivity AnalysesResults from models with multiple imputation on missing observationswere similar to those obtained from models fitted with listwise deletion ofmissing observations (Table D.4). Models with listwise deletion did havelower ICC’s—and in several cases ICC’s that were not significantly differentfrom zero—but the result is likely due to the smaller sample size in the1074.3. Resultsmodels.As in models with daily versus less-than-daily cutoffs, there were no sta-tistically significant associations between measures of objective proximityand consumption of sugar-sweetened beverages in models comparing weeklywith less-than-weekly consumption. In adjusted models, significant associ-ations were observed between odds of weekly packaged snack consumptionand limited-service outlet density at 400m (OR=0.97, 95% CI 0.94 - 1.00),but the association was the inverse of the expected relationship. No asso-ciations were observed between weekly consumption of minimally nutritiousfoods and perceived proximity in adjusted or bivariate models with weeklyintake as the outcome variables.In OLS regressions with continuous outcomes, no significant associationswere observed monthly frequencies of intake of sugar-sweetened beverages,fast food, or packaged snacks and objective proximity. Associations betweenfood access measures and students’ self-reported frequency of consumptionper month were the opposite of theoretical expectations: students reporting a≥10 minute walk to convenience stores or fast food retailers reported consum-ing significantly more sugar-sweetened beverages, fast foods and packagedsnacks than comparable students reporting a <5 minute walk to such re-tailers. Although several significant associations were identified, in adjustedmodels, between frequency of consumption and density of convenience storesat 400 or 800 meters, associations were again in the opposite of the expecteddirections and most associations examined were null, in keeping with theresults observed in multilevel logistic regressions.1084.4. Discussion4.4 DiscussionThis study examined the relationships between children’s access to foodretailers and their consumption, at or en-route to school, of minimally nutri-tious foods. The analyses examined three main sets of explanatory variables:(1) objective proximity, (2) objective density within 400 or 800 metres, and(3) perceived proximity of limited-service food outlets, convenience stores,and supermarkets or grocery stores in relation to Vancouver schools. Thestudy did not observe evidence of meaningful associations between thesefood environment measures and children’s odds of being daily consumers ofsugar-sweetened beverages, fast foods, or packaged snacks.Only one bivariate association—sugar-sweetened beverage intake andperceived proximity to a grocery store—was statistically significant, and theassociation was no longer significant after controlling for any one of gender,acculturation, or spending money. Given the number of associations tested,it is plausible that this result is a false positive (Rothman, 1990). In adjustedmodels, convenience store density at 400m and 800m was associated with adecreased odds that children would be daily packaged snack consumers–anassociation that is the inverse of the expected relationship. The results ofthis study thus suggest that ease of access is not associated with frequentconsumption of minimally nutritious foods among Vancouver children andyouth. This finding contrasts with the associations of Healthy Eating Indexscores and fast food outlet access identified by He et al. (2012b) in London,Ontario as well as the relationships Davis and Carpenter (2009) observedfor soda consumption and fast food outlet access. However, the results are1094.4. Discussionin keeping with those obtained by An and Sturm (2012), who did not findassociations of daily servings of soda, high-sugar foods and fast foods amongCalifornia students and food retailer densities surrounding schools.The models fitted in this study included controls for school level (sec-ondary versus elementary) and whether students reported bringing lunchfrom home on a daily basis; food insecurity offered a measure for student-level socioeconomic status while school-level median family income was usedas a measure of socioeconomic status at the school level. Associations ofcontrol variables and students’ regular consumption of minimally nutritiousfoods are comparable to those obtained in a previous study with the I-EATdata (Velazquez et al., 2015).Finally, this study also examined the agreement between students’ per-ceptions of food outlet proximity and objective proximity, finding “slight”agreement according to the Landis scale (Landis and Koch, 1977). This re-sult may explain inconsistencies in previous research between associationswith perceived and objective food outlet measures; Svastisalee et al. (2015),for example, find that boys who reported perceived access to 2 or moreoutlets within 5 minutes had an increased odds of weekly fast food con-sumption in comparison with boys who reported a lower level of access, butfailed to find statistically significant associations of fast food consumptionand comparable objective measures of proximity. Some studies have reliedon principals’ perceptions of food access rather than on students’ reports(Gebremariam et al., 2012; Morin et al., 2015); further research is necessaryto examine whether administrators can offer reliable and valid estimates offood environment measures.1104.4. Discussion4.4.1 LimitationsAlthough the null findings obtained in this study are comparable to thoseobtained by other researchers, they may be due to limitations of the studydesign. With 26 schools, the study had limited power to detect statisticallysignificant associations of dependent variables and food environment mea-sures at the school level. In addition, the study included both elementaryand secondary school students; there is evidence that secondary studentsmay be more likely to frequent off-campus vendors (Velazquez et al., 2015)and thus a need remains for research focused exclusively on older students.In addition, the study was conducted in a commercially dense city. Themaximum distance from any school to at least one food outlet was 931 metres(Southlands Elementary); thus there may not be enough variance in theschool food environment to detect statistically significant effects.The findings could also be affected by measurement error: the dependentvariables used in this study were constructed from self-reported data andthus may be subject to the problems of over- or underreporting commonlyobserved in food frequency questionnaire data (Deschamps et al., 2009; Perkset al., 2000; Watson et al., 2009). However, the food frequency questionnaireremains the most appropriate means of assessing diets in the context of thisstudy given its scope and low burden to participants (Willett, 2012) as well asevidence of acceptable validity and reliability for the measurement of dietaryintake in older children and adolescents (Maruti et al., 2006).Finally, dietary intake outcomes for this study were dichotomized intobinary measures of daily- versus less-than-daily consumption, potentially in-1114.4. Discussiontroducing bias through the choice of cutoff (Royston et al., 2006). However,results were reasonably robust to the use of alternative dichotomizations(weekly versus less-than-weekly consumption) as well as to the use of multi-level OLS regressions with continuous outcomes.4.4.2 StrengthsAlthough limitations of study design may underlie the null results, themethods used in this study represented the best practices recommended byfood environments researchers. The study relied on validated food environ-ments data (Chapter 2). In addition, multiple measures of the food environ-ment were examined, including objective density at 400 and 800m, objectiveproximity, and students’ perceptions of proximity. Objective density wasevaluated with line-based road network buffers, as recommended by Oliveret al. (2007), while the evaluation of street network distance—as used inthis study—is considered the best available means of quantifying proximity(Thornton et al., 2011). Multilevel modeling allowed for the simultaneousstudy of both school- and student-level variables (Diez Roux, 2004), whileresults were robust to the use of listwise deletion or multiple imputation formissing observations. Finally, models were adjusted for gender, spendingmoney and school-level socioeconomic status. Although controlling for race,recommended by Cobb et al. (2015), was not appropriate in the Canadiancontext, the measure of “acculturation” allowed this study to approximate acontrol immigrant status.1124.5. Conclusions4.5 ConclusionsThis study did not find statistically significant associations between stu-dents’ intake of minimally nutritious foods and the density or proximity offast food retailers, convenience stores or grocery/supermarket stores sur-rounding those students’ schools. Furthermore, this study found that stu-dents’ perceptions of access were not reliable alternatives to researchers’measurements of street-based proximity; however, results were robust to theuse of objective or perceived measures of food outlet access. While the nullresults obtained in this study may be a product of low statistical power ora lack of variance in access at the school level, the methods used in thisstudy represent best practices from current food environments research. InVancouver, the evidence for restrictive zoning or other policy measures torestrict food vendor access near schools remains weak; further research isneeded to identify effective policies for improving the nutritional quality ofchildren’s school-day diets.113Chapter 5ConclusionThis thesis consisted of three connected studies of the food environmentssurrounding schools in Vancouver, BC. Chapter 2 examined the validity oftwo commercial and two municipal data sources for food retailers; the best-performing data set, the City of Vancouver Business License Lists, was thenused in Chapter 3 to search for socioeconomic or demographic disparities inVancouver school food environments. Finally, Chapter 4 assessed the rela-tionships between measures of the food environments surrounding a subset ofVancouver Schools and student consumption of sugar-sweetened beverages,fast foods, or packaged snacks. Ultimately, the study did not find evidenceof consistent associations between the density or proximity of food outletsin relation to Vancouver schools and students’ consumption, at or en-routeto school, of these minimally nutritious foods.5.1 Contributions and SignificanceThis thesis joins a growing literature suggesting that the food environ-ments surrounding schools have, at most, a weak association with students’diet-related health (Williams et al., 2014). In this study, significance testsfailed to reject the null hypothesis of no association for almost all relation-1145.1. Contributions and Significanceships examined between food environment measures and students’ intake ofsugar-sweetened beverages, fast food, or packaged snacks. The study addi-tionally included a comparison of objective and perceived food environmentmeasures, finding that perceived measures had poor reliability for the mea-surement of students’ access to food outlets.The comparative validation of two commercial and two municipal sourcesof data on food retailer locations (Chapter 2) offers methodological insight:for other researchers hoping to study food outlets in Vancouver, BC, the re-sults from Chapter 2 suggest that the City of Vancouver’s Business Licensesdata have higher sensitivity, PPV, and concordance than do other availabledata sets; furthermore, the Business Licenses data did not show evidence ofsystematic over- or undercounting in association with socioeconomic depri-vation or commercial density. The study offers the first comparative study ofthe validity of municipal and commercial data sets for food outlet locationsin Canada; in addition, it joins just one other known study (Ma et al., 2013)examining the effect of over- or undercounting in food outlet data sourceson measures of the food environment.This thesis also examined the associations between school demographicand socioeconomic characteristics and the food environments surroundingschools (Chapter 3). The study did not find significant associations be-tween the percent of aboriginal students or the percent of English LanguageLearners enrolled in Vancouver schools and food outlet density or proxim-ity, but higher convenience store densities were observed within 400m and800m regions of high-poverty schools in comparison with schools with fewerstudents living in poverty; including neighbourhood commercial density and1155.2. Strengths and Limitations of the Researchneighbourhood socioeconomic deprivation weakened these associations. Al-though this study begins to address an important research gap—only Engler-Stringer et al. (2014b) previously examined for demographic disparities inthe food environments surrounding schools in Canada, and no previous studyin Canada has used school-level measures of student demographics for suchresearch—the potential existence of ethnic disparities in food access and foodretail exposure remains an important and under-examined issue in Canada(Black, 2015).5.2 Strengths and Limitations of the ResearchIn a systematic review of 71 food environments studies, Cobb et al. (2015)identify common flaws in examinations of the associations between food re-tailer access and obesity. Among other concerns, the researchers argue thatfood outlet locations should be validated, that researchers should control forsocioeconomic and demographic factors, and that analyses should accountfor multilevel data structures. This study addressed such issues by vali-dating the data sets used (Chapter 2), including controls for demographicand socioeconomic factors (Chapters 3 and 4) and using multilevel modeling(Chapter 4).However, Cobb et al. (2015) also suggest that studies should use objective—rather than self-reported—outcome measures and that study designs shouldaccount for neighbourhood self-selection. Dependent variables for Chap-ter 4 were constructed from self-report data, and the cross-sectional researchdesign of both chapters 3 and 4 failed to account for the possibility of self-1165.2. Strengths and Limitations of the Researchselection—though the ecological analysis of food retailer locations (Chap-ter 3) served to assuage some fears regarding possible simultaneity in foodretailer access and students’ dietary behaviours.Some of the data sources used in this research were also subject to tempo-ral mismatch (Fleischhacker et al., 2011); in particular, the school-level mea-sure of socioeconomic deprivation used in Chapter 3 was constructed fromthe 2006 Census of Canada, but all other measures used in that chapter werefrom the 2011-2012 school year. Similarly, the commercial business locationdata validated in Chapter 3 was several years older than the municipal andgold standard data, though the validation of both 2015 and 2012 BusinessLicense data offered insight on the amount of over- or under-counting thatmight be attributable to temporal changes in outlet locations. Both surveydata collection and food outlet location data in Chapter 4, however, werelimited to data collected between March and June of 2012.The use of school-day dietary intakes, rather than obesity or other mea-sures of diet-related health, as outcomes in Chapter 4 reduced the potentialof walkability or other neighbourhood factors to confound associations—asmay occur in the study of the school food environment and BMI or obesitystatus (Cobb et al., 2015)—but its use introduced the potential for measure-ment error in the construction of a binary measure of “daily” versus “less-than-daily” consumption of each dietary intake category. The results wererobust, however, to the use of different cut-points (Section 4.3.1). Chapter 4may also have had low power to detect effects specific to secondary schoolstudents, due to the lower percentage of secondary schools versus elementaryschools (and secondary school students versus younger students) in the sam-1175.2. Strengths and Limitations of the Researchple, which could obscure associations that affect only more autonomous olderstudents. In addition, this thesis focused exclusively on students’ access tofood retailers, though the food environment includes other dimensions suchas affordability or acceptability (Caspi et al., 2012); further work remainsnecessary to examine whether the prices of items within food retailers or thequality of foods available to students affect dietary behaviours.This thesis also had important strengths in the high standard of qual-ity for methodological approaches used, given the diversity of measurementand statistical approaches used in current food environments research (Caspiet al., 2012; Williams et al., 2014; Moore and Diez-Roux, 2015). Food outletlocations were validated against ground-truthed data, the gold standard infood environments research (Lucan, 2015; Fleischhacker et al., 2013), andall analyses were conducted with the best performing data set. In contrastwith studies that have considered food environments exposure within neigh-bourhoods based on administrative units like census tracts (e.g. Maddock(2004), Sturm and Datar (2005), or Inagami et al. (2006)), which have lim-ited validity as a measure of student exposure (Holsten, 2009), this studyassessed density and proximity for outlets within walking distance of schools;furthermore, proximity was measured with street-network distances, consid-ered a more accurate measure of walkable areas than Euclidean (straight-line)measures (Sparks et al., 2010), and density was evaluated within line-basedbuffers following Oliver et al. (2007). Finally, this thesis tested robustness ofthe results to both the choice of distance considered “walking” distance andto the choice of food environments measure used, evaluating density within160, 400 and 800 meters from schools and conducting all analyses with both1185.3. Avenues for Future Researchdensity and proximity.5.3 Avenues for Future Research5.3.1 Understanding Adolescents’ Dietary BehavioursDespite the proliferation of food environments studies, the nature of chil-dren’s and adolescents’ interactions with food retailers remains poorly un-derstood. Although few studies of the food environment explicitly state themechanism by which food outlets are expected to be affecting students’ diets(Caspi et al., 2012), it is likely that most researchers see food purchases as acritical component of the causal pathway—that is, that researchers think easeof access to food retailers translates into increased consumption by makingit easier for students to purchase fast and snack foods. Advertising exposure,however, may offer another means by which outlet exposure affects students’diets: posters or other advertisements are often located on or near stores(Walton et al., 2009), so children who see multiple stores at or en-route toschools may be exposed to more unhealthy advertisements and may, as aresult, develop less healthful dietary preferences. Other mechanisms beyondadvertising may affect children’s choices; qualitative studies with schoolchil-dren could help researchers identify other relevant mechanisms.The associations between within-school food environments and the foodenvironments surrounding schools would also be worthy of inquiry. Recentchanges in policies regarding vending and school food availability might beleading more students to make purchases off campus (Mâsse et al., 2013);food retailers may also be more successful near schools without cafeterias,1195.3. Avenues for Future Researchwhere students are in need of alternative lunch options. Finally, the associ-ations of student ethnicity and food access remain understudied in Canada(Black, 2015); further research testing for disparities in food environmentsboth within and surrounding schools in association with student sociodemo-graphic characteristics is necessary fill this critical research gap.5.3.2 Methodological ImprovementsAlthough there have been a number of small-scale studies validating foodenvironments data sources, there is still no data source validated for nationalor multi-city inquiry in Canada. Canadian studies have largely focused onindividual cities; the few national food environments studies may be com-promised by systematic bias according to urban density (discussed in Sec-tion 3.4). But nationwide work remains critical given evidence of seriousdisparities in supermarket and grocery store access in rural Northern com-munities (Health Canada, 2013) as well as evidence in Section 3.4 of notewor-thy differences between municipalities and provinces in terms of food retailerdistributions.The field of food environments research also suffers from a lack of consen-sus regarding gold-standard measures of the food environment (Cobb et al.,2015; Caspi et al., 2012; Feng et al., 2010). In the case of food retailer den-sity, for example, researchers have measured the food environment withindistances as short as 160m from schools (Currie et al., 2010) and as far as3000m from schools (Laska et al., 2010). Although different measures arenecessary in the diverse of regions examined as well as in contexts of olderversus younder students, some measures are probably more valid than others1205.3. Avenues for Future Research(Cobb et al., 2015). Several studies have examined activity spaces—the setof locations people visit regularly throughout their daily activities—ratherthan focusing on buffers (Zenk et al., 2011; Christian, 2012; Stewart et al.,2015). Activity spaces should not replace buffer zones, because the formerintroduce endogeneity (students choose to include food outlets in their vis-ited locations (Chaix et al., 2012)), but activity space research could be usedto identify buffer sizes most representative of the distances schoolchildrentravel during lunchtime breaks.Finally, a recent review of studies associating obesity outcomes and schoolfood environments observed that almost all studies have been cross-sectionalin design (Williams et al., 2014). Smith et al. (2013) and Rossen et al. (2013)offered longitudinal examinations, which allow researchers to examine asso-ciations between the length of a students’ exposure and change in outcomesover time, offering more utility than cross-sectional studies for researchershoping to uncover cause and effect (Smith et al., 2013). But even in lon-gitudinal studies, the possibility of endogeneity prevents researchers frommaking causal statements. A need remains for study designs and analyti-cal approaches informed by causal inference, such as the identification andevaluation of natural experiments—a natural variation in the exposure ofinterest, for example due to a policy change (Petticrew et al., 2005)—or theuse of instrumental variables as in Alviola et al. (2014).1215.4. Policy Relevance and Implications5.4 Policy Relevance and ImplicationsSome local governments in the United States have restricted the ability offast food vendors to locate near schools. Detroit, Michigan has an ordinancerequiring a minimum of 500 feet between schools and fast food or drive-in restaurants, while Arden Hills, Minnesota requires that drive-in and fastfood restaurants be located at least 400m from schools (Mair et al., 2005).While these policies have not been evaluated, a systematic review of studiesassociating obesity outcomes and school food environments (Williams et al.,2014) finds that the empirical evidence does not, at present, offer strong sup-port for such policies from the perspective of public health advocacy. Giventhe present limitations of the empirical study of school food environmentsas well as the findings, in this thesis similarly, of no consistent evidence thatthe school food environment plays a significant role in Vancouver children’sdietary behaviours at or en-route to school, efforts to alter the food environ-ments surrounding Vancouver public schools would be premature.For public health practitioners and policymakers seeking to reduce chil-dren’s intakes of minimally nutritious foods, there are other interventionareas with more reliable results. Increasing within-school availability offruits and vegetables, for example, and decreasing within-school availabilityof sugar-sweetened beverages has been consistently associated with desirabledietary behaviours in children (Afshin et al., 2015). Nutrition standards forfood items available within schools have similarly been shown to encouragehealthier dietary intakes in students (McKenna, 2010), though an evaluationof Canadian school nutrition policies found that current standards were in1225.4. Policy Relevance and Implicationsneed of improvement (Leo, 2007).In the face an obesity epidemic of unprecedented global scale, incremen-tal interventions have been inadequate: Ng et al. (2014) found that despitesome evidence of slowdowns in the increasing prevalences of obesity in de-veloped countries, no country showed a significant decrease in obesity preva-lence. There were, according to the researchers, “no national success stories”in the 33-year time period examined. While methodological improvementsmay help food environments researchers uncover a role of food retailer ex-posure in the rise of childhood obesity and diabetes (Cobb et al., 2015), thisthesis research joins an increasing number of studies suggesting that publichealth practitioners will need to look elsewhere for high-impact approachesto obesity prevention.123BibliographyAfshin, A., Penalvo, J., Del Gobbo, L., Kashaf, M., Micha, R., Morrish,K., Pearson-Stuttard, J., Rehm, C., Shangguan, S., Smith, J. D., andMozaffarian, D. (2015). CVD prevention through policy: a review of massmedia, food/menu labeling, taxation/subsidies, built environment, schoolprocurement, worksite wellness, and marketing standards to improve diet.Curr Cardiol Rep, 17(11):98.Ahmadi, N., Black, J. L., Velazquez, C. E., Chapman, G. E., and Veenstra,G. (2015). Associations between socio-economic status and school-daydietary intake in a sample of grade 5-8 students in Vancouver, Canada.Public Health Nutr, 18(5):764–773.Alviola, P. A., Nayga, R. M., Thomsen, M. R., Danforth, D., and Smartt, J.(2014). The effect of fast-food restaurants on childhood obesity: a schoollevel analysis. Econ Hum Biol, 12:110–9.An, R. and Sturm, R. (2012). School and residential neighborhood foodenvironment and diet among California youth. Am J Prev Med, 42(2):129–35.Apparicio, P., Cloutier, M.-S. S., and Shearmur, R. (2007). The case of124BibliographyMontréal’s missing food deserts: evaluation of accessibility to food super-markets. Int J Health Geogr, 6:4.Auchincloss, A. H., Moore, K. A. B., Moore, L. V., and Diez Roux, A. V.(2012). Improving retrospective characterization of the food environmentfor a large region in the United States during a historic time period. HealthPlace, 18(6):1341–7.Austin, S. B., Melly, S. J., Sanchez, B. N., Patel, A., Buka, S., and Gort-maker, S. L. (2005). Clustering of fast-food restaurants around schools: anovel application of spatial statistics to the study of food environments.Am J Public Health, 95(9):1575.Bader, M. D. M., Ailshire, J. A., Morenoff, J. D., and House, J. S. (2010).Measurement of the local food environment: a comparison of existing datasources. Am J Epidemiol, 171(5):609–17.BCMinistry of Education (2016). Schools. British Columbia Data Catalogue,Vancouver, BC. Retrieved June 1, 2016, from https://catalogue.data.gov.bc.ca/dataset/bc-schools-school-locations.Beaulac, J., Kristjansson, E., and Cummins, S. (2009). Peer reviewed: Asystematic review of food deserts, 1966-2007. Prev Chronic Dis, 6(3).Bell, N. and Hayes, M. V. (2012). The Vancouver Area NeighbourhoodDeprivation Index (VANDIX): a census-based tool for assessing small-areavariations in health status. Can J Public Health, 103(8):S28–S32.Bell, N., Schuurman, N., and Hameed, S. M. (2009a). A multilevel analysis125Bibliographyof the socio-spatial pattern of assault injuries in greater Vancouver, BritishColumbia. Can J Public Health, 100(1):73–77.Bell, N., Schuurman, N., Oliver, L., and Hayes, M. V. (2007). Towards theconstruction of place-specific measures of deprivation: a case study fromthe Vancouver metropolitan area. Can Geogr-Geogr Can, 51(4):444–461.Bell, N., Simons, R. K., Lakha, N., and Hameed, S. M. (2012). Are wefailing our rural communities? Motor vehicle injury in British Columbia,Canada, 2001-2007. Injury, 43(11):1888–91.Bell, N. J., Schuurman, N., and Morad Hameed, S. (2009b). A small-area population analysis of socioeconomic status and incidence of severeburn/fire-related injury in British Columbia, Canada. Burns, 35(8):1133–41.Birnbaum, A. S., Lytle, L. A., Murray, D. M., Story, M., Perry, C. L., andBoutelle, K. N. (2002). Survey development for assessing correlates ofyoung adolescents’ eating. Am J Health Behav, 26(4):284–295.Black, J. L. (2015). Local food environments outside of the United States.a look to the North: Examining food environments in Canada. In Mor-land, K. B., editor, Local Food Environments: Food Access in America,chapter 8, pages 231–261. CRC Press, Boca Raton, FL.Black, J. L. and Billette, J.-M. (2013). Do Canadians meet Canada’s FoodGuide’s recommendations for fruits and vegetables? Appl Physiol NutrMetab, 38(3):234–242.126BibliographyBlack, J. L., Carpiano, R. M., Fleming, S., and Lauster, N. (2011). Explor-ing the distribution of food stores in British Columbia: associations withneighbourhood socio-demographic factors and urban form. Health Place,17(4):961–70.Black, J. L. and Day, M. (2012). Availability of limited service foodoutlets surrounding schools in British Columbia. Can J Public Health,103(4):e255–e259.Black, J. L. and Macinko, J. (2008). Neighborhoods and obesity. Nutr Rev,66(1):2–20.Block, J. P., Christakis, N. A., O’Malley, A. J., and Subramanian, S. V.(2011). Proximity to food establishments and body mass index in theFramingham Heart Study offspring cohort over 30 years. Am J Epidemiol,174(10):1108–14.Bodor, J. N., Rice, J. C., Farley, T. A., Swalm, C. M., and Rose, D. (2010).The association between obesity and urban food environments. J UrbanHealth, 87(5):771–81.Braveman, P. A., Cubbin, C., Egerter, S., Williams, D. R., and Pamuk, E.(2010). Socioeconomic disparities in health in the United States: what thepatterns tell us. Am J Public Health, 100(S1):S186–S196.Brownell, K. and Horgen, K. B. (2004). Food Fight: The Inside Story of theFood Industry, America’s Obesity Crisis, and What We Can Do About It.McGraw Hill.127BibliographyBuck, C., Börnhorst, C., Pohlabeln, H., Huybrechts, I., Pala, V., Reisch, L.,Pigeot, I., and I Family consortia (2013). Clustering of unhealthy foodaround German schools and its influence on dietary behavior in schoolchildren: a pilot study. Int J Behav Nutr Phys Act, 10:65.Burdette, H. L. and Whitaker, R. C. (2004). Neighborhood playgrounds,fast food restaurants, and crime: relationships to overweight in low-incomepreschool children. Prev Med, 38:57–63.Burgoine, T. and Harrison, F. (2013). Comparing the accuracy of two sec-ondary food environment data sources in the UK across socio-economicand urban/rural divides. Int J Health Geogr, 12:1–8.Carmines, E. G. and Zeller, R. A. (1979). Reliability and Validity Assessment.Quantitative Applications in the Social Sciences. SAGE Publications Ltd,London, England.Caspi, C. E., Sorensen, G., Subramanian, S. V., and Kawachi, I. (2012).The local food environment and diet: a systematic review. Health Place,18(5):1172–87.CBC Radio-Canada (2014). Vancouver School Board over-hauling inner city school funding. Retrieved June 1, 2016,from http://www.cbc.ca/news/canada/british-columbia/vancouver-school-board-overhauling-inner-city-school-funding-1.2533299.Census Bureau (2005). Statistical Abstract of the United States, 2006.128BibliographyGovernment Printing Office. Retrieved June 10, 2016, from https://www.census.gov/prod/2005pubs/06statab/educ.pdf.Chaix, B., Kestens, Y., Perchoux, C., Karusisi, N., Merlo, J., and Labadi, K.(2012). An interactive mapping tool to assess individual mobility patternsin neighborhood studies. Am J Prev Med, 43(4):440–50.Christian, W. J. (2012). Using geospatial technologies to explore activity-based retail food environments. Spat Spatiotemporal Epidemiol, 3(4):287–95.City of Vancouver (2016). Business Licences. Open Data Catalogue, Van-couver, BC. Retrieved October 20, 2015, from http://data.vancouver.ca/datacatalogue/businessLicence.htm.Clarke, K. (2005). The phantom menace: omitted variable bias in econo-metric research. Conflict Manag Peace Sci, 22(4):341–352.Clary, C. M. and Kestens, Y. (2013). Field validation of secondary datasources: a novel measure of representativity applied to a Canadian foodoutlet database. Int J Behav Nutr Phys Act, 10:77.Cobb, L. K., Appel, L. J., Franco, M., Jones-Smith, J. C., Nur, A., andAnderson, C. A. M. (2015). The relationship of the local food environmentwith obesity: A systematic review of methods, study quality, and results.Obesity, 23:1331–1344.Coleman-Jensen, A., Gregory, C., and Singh, A. (2014). Household foodsecurity in the United States in 2013. USDA-ERS Economic Research Re-129Bibliographyport. United States Department of Agriculture. Retrieved June 10, 2016,from http://www.ers.usda.gov/media/1565415/err173.pdf.Coopersmith, K. (2012). A tale of two villages: University Village de-velopment has exploded in the past 20 years, but it’s got a new chal-lenger. Retrieved July 11, 2016, from http://old.ubyssey.ca/features/two-villages685/.Cummins, S. and Macintyre, S. (2009). Are secondary data sources on theneighbourhood food environment accurate? Case-study in Glasgow, UK.Prev Med, 49(6):527–8.Currie, J., DellaVigna, S., Moretti, E., and Pathania, V. (2010). The effectof fast food restaurants on obesity and weight gain. Am Econ J: EconPolicy, 2(3):32–63.DataBC (2016). BC Data Catalogue. B.C. Government. Retrieved October20, 2016, from http://data.gov.bc.ca.Davis, B. and Carpenter, C. (2009). Proximity of fast-food restaurants toschools and adolescent obesity. Am J Public Health, 99(3):505.Day, P. L. and Pearce, J. (2011). Obesity-promoting food environments andthe spatial clustering of food outlets around schools. Am J Prev Med,40(2):113–121.Deschamps, V., De Lauzon-Guillain, B., Lafay, L., Borys, J.-M. . M.,Charles, M.-A. . A., and Romon, M. (2009). Reproducibility and rela-130Bibliographytive validity of a food-frequency questionnaire among French adults andadolescents. Eur J of Clin Nutr, 63(2):282–291.Dieleman, J. L. and Templin, T. (2014). Random-effects, fixed-effects andthe within-between specification for clustered data in observational healthstudies: a simulation study. PLoS One, 9(10):e110257.Diez Roux, A. V. (2004). The study of group-level factors in epidemiology:rethinking variables, study designs, and analytical approaches. EpidemiolRev, 26:104–11.DMTI Spatial, Inc. (2003). Enhanced Point of Interest layers [2003]. Re-trieved June 30, 2016, from http://hdl.handle.net.ezproxy.library.ubc.ca/11272/NBRIL.DMTI Spatial, Inc. (2006). Enhanced Point of Interest layers [2006]. Re-trieved June 30, 2016, from http://hdl.handle.net.ezproxy.library.ubc.ca/11272/KDY86.DMTI Spatial, Inc. (2009). Enhanced Point of Interest layers [v.2009.3]. Re-trieved June 30, 2016, from http://hdl.handle.net.ezproxy.library.ubc.ca/11272/JGQ3B.DMTI Spatial, Inc. (2013a). CanMap Streetfiles, v2013.3. Retrieved May15, 2015, from http://hdl.handle.net.ezproxy.library.ubc.ca/.DMTI Spatial, Inc. (2013b). EPOI v2013.3. Retrieved May 15, 2015, fromhttp://hdl.handle.net.ezproxy.library.ubc.ca/.131BibliographyDubowitz, T., Ghosh-Dastidar, M., Eibner, C., Slaughter, M. E., Fernan-des, M., Whitsel, E. A., Bird, C. E., Jewell, A., Margolis, K. L., andLi, W. (2012). The Women’s Health Initiative: the food environment,neighborhood socioeconomic status, BMI, and blood pressure. Obesity,20(4):862–871.Dunn, R. A. (2010). The effect of fast-food availability on obesity: an analysisby gender, race, and residential location. Am J Agr Econ, 92(4):1149–1164.Economic Classification Policy Committee (1994). Economic Concepts Incor-porated in the Standard Industrial Classification Industries of the UnitedStates. Bureau of Economic Analysis, U.S. Department of Commerce,Washington, DC. Retrieved June 30, 2016, from http://www.census.gov/eos/www/naics/history/docs/report_1.pdf.Economic Research Service (2012). U.S. Household Food Secu-rity Survey Module: Six-item short form. Retrieved June 10,2016, from http://www.ers.usda.gov/datafiles/Food_Security_in_the_United_States/Food_Security_Survey_Modules/short2012.pdf.Engler-Stringer, R., Le, H., Gerrard, A., and Muhajarine, N. (2014a). Thecommunity and consumer food environment and children’s diet: a system-atic review. BMC Public Health, 14:522.Engler-Stringer, R., Shah, T., Bell, S., and Muhajarine, N. (2014b). Geo-graphic access to healthy and unhealthy food sources for children in neigh-bourhoods and from elementary schools in a mid-sized Canadian city. SpatSpatiotemporal Epidemiol, 11:23–32.132BibliographyESRI (2015). ArcGIS Desktop: Release 10.3.1. Environmental SystemsResearch Institute, Redlands, CA.Feng, J., Glass, T. A., Curriero, F. C., Stewart, W. F., and Schwartz, B. S.(2010). The built environment and obesity: a systematic review of theepidemiologic evidence. Health Place, 16(2):175–90.Fleischhacker, S. E., Evenson, K. R., Rodriguez, D. A., and Ammerman,A. S. (2011). A systematic review of fast food access studies. Obes Rev,12(5):e460–71.Fleischhacker, S. E., Evenson, K. R., Sharkey, J., Pitts, S. B. J., and Ro-driguez, D. A. (2013). Validity of secondary retail food outlet data: asystematic review. Am J Prev Med, 45(4):462–73.Fleischhacker, S. E., Rodriguez, D. A., Evenson, K. R., Henley, A., Gizlice,Z., Soto, D., and Ramachandran, G. (2012). Evidence for validity offive secondary data sources for enumerating retail food outlets in sevenAmerican Indian communities in North Carolina. Int J Behav Nutr PhysAct, 9(1):137.Ford, P. B. and Dzewaltowski, D. A. (2011). Neighborhood deprivation,supermarket availability, and BMI in low-income women: a multilevelanalysis. J Community Health, 36(5):785–96.Forsyth, A., Wall, M., Larson, N., Story, M., and Neumark-Sztainer, D.(2012). Do adolescents who live or go to school near fast-food restaurantseat more frequently from fast-food restaurants? Health Place, 18(6):1261–9.133BibliographyGarriguet, D. (2007). Sodium consumption at all ages. Health Rep, 18(2):47–52.Garriguet, D. (2008). Beverage consumption of children and teens. HealthRep, 19(4):17–22.Garriguet, D. (2009). Diet quality in Canada. Health Rep, 20(3):41–52.Gebremariam, M. K., Andersen, L. F., Bjelland, M., Klepp, K.-I. I., Totland,T. H., Bergh, I. H., and Lien, N. (2012). Does the school food environmentinfluence the dietary behaviours of Norwegian 11-year-olds? The HEIAstudy. Scand J Public Health, 40(5):491–7.Gibson, D. M. (2011). The neighborhood food environment and adult weightstatus: estimates from longitudinal data. Am J Public Health, 101(1):71–78.Gilliland, J. A., Rangel, C. Y., Healy, M. A., Tucker, P., Loebach, J. E.,Hess, P. M., He, M., Irwin, J. D., and Wilk, P. (2012). Linking childhoodobesity to the built environment: a multi-level analysis of home and schoolneighbourhood factors associated with Body Mass Index. Can J PublicHealth, 103(9):15–21.Glanz, K., Sallis, J. F., Saelens, B. E., and Frank, L. D. (2007). Nutri-tion Environment Measures Survey in stores (NEMS-S): development andevaluation. Am J Prev Med, 32(4):282–9.Grier, S. and Davis, B. (2013). Are all proximity effects created equal? Fast134Bibliographyfood near schools and body weight among diverse adolescents. J PublicPolicy Mark, 32(1):116–128.Griffiths, C., Frearson, A., Taylor, A., Radley, D., and Cooke, C. (2014).A cross sectional study investigating the association between exposure tofood outlets and childhood obesity in Leeds, UK. Int J Behav Nutr PhysAct, 11:138.Gustafson, A. A., Lewis, S., Wilson, C., and Jilcott-Pitts, S. (2012). Val-idation of food store environment secondary data source and the role ofneighborhood deprivation in Appalachia, Kentucky. BMC Public Health,12:688.Han, E., Powell, L. M., Zenk, S. N., Rimkus, L., Ohri-Vachaspati, P., andChaloupka, F. J. (2012). Classification bias in commercial business listsfor retail food stores in the U.S. Int J Behav Nutr Phys Act, 9:46.Hanning, R. M., Royall, D., Toews, J. E., Blashill, L., Wegener, J., andDriezen, P. (2009). Web-based food behaviour questionnaire: validationwith grades six to eight students. Can J Diet Pract Res, 70(4):172–178.Hanson, M. D. and Chen, E. (2007). Socioeconomic status and health behav-iors in adolescence: a review of the literature. J Behav Med, 30(3):263–85.Harris, D. E., Blum, J. W., Bampton, M., O’Brien, L. M., Beaudoin, C. M.,Polacsek, M., and O’Rourke, K. A. (2011). Location of food stores nearschools does not predict the weight status of Maine high school students.J Nutr Educ Behav, 43(4):274–8.135BibliographyHe, M., Tucker, P., Gilliland, J., Irwin, J. D., Larsen, K., and Hess, P.(2012a). The influence of local food environments on adolescents’ foodpurchasing behaviors. Int J Environ Res Public Health, 9(4):1458–71.He, M., Tucker, P., Irwin, J. D., Gilliland, J., Larsen, K., and Hess,P. (2012b). Obesogenic neighbourhoods: the impact of neighbourhoodrestaurants and convenience stores on adolescents’ food consumption be-haviours. Public Health Nutr, 15(12):2331–9.Health Canada (2011). Eating well with Canada’s Food Guide.Retrieved June 6, 2016, from http://www.hc-sc.gc.ca/fn-an/food-guide-aliment/index-eng.php.Health Canada (2013). Measuring the Food Environment inCanada. Health Canada, Ottowa, ON. Retrieved June 6, 2016,from http://www.foodsecuritynews.com/resource-documents/MeasureFoodEnvironm_EN.pdf.Hearst, M. O., Pasch, K. E., and Laska, M. N. (2012). Urban v. subur-ban perceptions of the neighbourhood food environment as correlates ofadolescent food purchasing. Public Health Nutr, 15(2):299–306.Héroux, M., Iannotti, R. J., Currie, D., Pickett, W., and Janssen, I. (2012).The food retail environment in school neighborhoods and its relation tolunchtime eating behaviors in youth from three countries. Health Place,18(6):1240–7.Hickson, D. A., Diez Roux, A. V., Smith, A. E., Tucker, K. L., Gore, L. D.,Zhang, L., and Wyatt, S. B. (2011). Associations of fast food restaurant136Bibliographyavailability with dietary intake and weight among African Americans inthe Jackson Heart Study, 2000–2004. Am J Public Health, 101(S1):S301–S309.Hoehner, C. M. and Schootman, M. (2010). Concordance of commercial datasources for neighborhood-effects studies. J Urban Health, 87(4):713–25.Holsten, J. E. (2009). Obesity and the community food environment: asystematic review. Public Health Nutr, 12(3):397–405.Hosler, A. S. and Dharssi, A. (2010). Identifying retail food stores to evaluatethe food environment. Am J Prev Med, 39(1):41–4.Howard, P. H., Fitzpatrick, M., and Fulfrost, B. (2011). Proximity of foodretailers to schools and rates of overweight ninth grade students: an eco-logical study in California. BMC Public Health, 11:68.Inagami, S., Cohen, D. A., Finch, B. K., and Asch, S. M. (2006). You arewhere you shop: grocery store locations, weight, and neighborhoods. AmJ Prev Med, 31(1):10–7.Jeffery, R. W., Baxter, J., McGuire, M., and Linde, J. (2006). Are fast foodrestaurants an environmental risk factor for obesity? Int J Behav NutrPhys Act, 3:2.Kestens, Y. and Daniel, M. (2010). Social inequalities in food exposurearound schools in an urban area. Am J Prev Med, 39(1):33–40.Kirkpatrick, S. I., Dodd, K. W., Parsons, R., Ng, C., Garriguet, D., andTarasuk, V. (2015). Household food insecurity is a stronger marker of ad-137Bibliographyequacy of nutrient intakes among Canadian compared to American youthand adults. J Nutr, 145(7):1596–603.Kwate, N. O. A. and Loh, J. M. (2010). Separate and unequal: the influenceof neighborhood and school characteristics on spatial proximity betweenfast food and schools. Prev Med, 51(2):153–6.Lake, A. A., Burgoine, T., Greenhalgh, F., Stamp, E., and Tyrrell, R.(2010). The foodscape: classification and field validation of secondarydata sources. Health Place, 16(4):666–73.Lake, A. A., Burgoine, T., Stamp, E., and Grieve, R. (2012). The food-scape: classification and field validation of secondary data sources acrossurban/rural and socio-economic classifications in England. Int J BehavNutr Phys Act, 9:37.Landis, J. R. and Koch, G. G. (1977). The measurement of observer agree-ment for categorical data. Biometrics, pages 159–174.Langellier, B. (2012). The food environment and student weight status, LosAngeles County, 2008-2009. Prev Chronic Dis.Langlois, K. and Garriguet, D. (2011). Sugar consumption among Canadiansof all ages. Health Rep, 22(3):1–5.Laska, M. N., Hearst, M. O., Forsyth, A., Pasch, K. E., and Lytle, L. (2010).Neighbourhood food environments: are they associated with adolescentdietary intake, food purchases and weight status? Public Health Nutr,13(11):1757–63.138BibliographyLassard, R. (2006). Improving the Health of Canadians: Promoting HealthyWeights. Canadian Institute for Health Information, Ottawa. Re-trieved June 20, 2016, from https://secure.cihi.ca/free_products/healthyweights06_e.pdf.Latham, J. and Moffat, T. (2007). Determinants of variation in food costand availability in two socioeconomically contrasting neighbourhoods ofHamilton, Ontario, Canada. Health Place, 13(1):273–87.Laxer, R. E. and Janssen, I. (2013). The proportion of excessive fast-food consumption attributable to the neighbourhood food environmentamong youth living within 1 km of their school. Appl Physiol Nutr Metab,39(4):480–486.Leatherdale, S. T., Pouliou, T., Church, D., and Hobin, E. (2011). Theassociation between overweight and opportunity structures in the builtenvironment: a multi-level analysis among elementary school youth in thePLAY-ON study. Int J Public Health, 56(3):237–46.Leo, A. (2007). Are Schools Making the Grade?: School Nutrition PoliciesAcross Canada. Centre for Science in the Public Interest.Li, F., Harmer, P. A., Cardinal, B. J., Bosworth, M., Acock, A., Johnson-Shelton, D., and Moore, J. M. (2008). Built environment, adiposity, andphysical activity in adults aged 50–75. Am J Prev Med, 35(1):38–46.Liese, A. D., Barnes, T. L., Lamichhane, A. P., Hibbert, J. D., Colabianchi,N., and Lawson, A. B. (2013). Characterizing the food retail environment:139Bibliographyimpact of count, type, and geospatial error in 2 secondary data sources. JNutr Educ Behav, 45(5):435–442.Liese, A. D., Colabianchi, N., Lamichhane, A. P., Barnes, T. L., Hibbert,J. D., Porter, D. E., Nichols, M. D., and Lawson, A. B. (2010). Validationof 3 food outlet databases: completeness and geospatial accuracy in ruraland urban food environments. Am J Epidemiol, 172(11):1324–33.Lietz, G., Barton, K. L., Longbottom, P. J., and Anderson, A. S. (2002). Canthe EPIC food-frequency questionnaire be used in adolescent populations?Public Health Nutr, 5(6):783–9.Longacre, M. R., Primack, B. A., Owens, P. M., Gibson, L., Beauregard, S.,Mackenzie, T. A., and Dalton, M. A. (2011). Public directory data sourcesdo not accurately characterize the food environment in two predominantlyrural states. J Am Diet Assoc, 111(4):577–82.Lucan, S. C. (2015). Concerning limitations of food-environment research: anarrative review and commentary framed around obesity and diet-relateddiseases in youth. J Acad Nutr Diet, 2:205–212.Lucan, S. C., Maroko, A. R., Bumol, J., Torrens, L., Varona, M., and Berke,E. M. (2013). Business list vs ground observation for measuring a foodenvironment: saving time or waste of time (or worse)? J Acad Nutr Diet,113(10):1332–9.Lydersen, S., Fagerland, M. W., and Laake, P. (2009). Recommended testsfor association in 2 x 2 tables. Stat Med, 28(7):1159–75.140BibliographyLytle, L. A. (2009). Measuring the food environment: state of the science.Am J Prev Med, 36(4 Suppl):S134–44.Ma, X., Battersby, S. E., Bell, B. A., Hibbert, J. D., Barnes, T. L., andLiese, A. D. (2013). Variation in low food access areas due to data sourceinaccuracies. Appl Geogr, 45:131–137.Maddock, J. (2004). The relationship between obesity and the prevalence offast food restaurants: state-level analysis. Am J Health Promot, 19(2):137–143.Mair, J. S., Pierce, M. W., and Teret, S. P. (2005). The use of zoning torestrict fast food outlets: a potential strategy to combat obesity. The Centerfor Law and the Public’s Health at Johns Hopkins & Georgetown Univer-sities. Retrieved June 1, 2016, from http://www.publichealthlaw.net/Zoning%20Fast%20Food%20Outlets.pdf.Malik, V. S., Popkin, B. M., Bray, G. A., Després, J.-P. P., and Hu, F. B.(2010). Sugar-sweetened beverages, obesity, type 2 diabetes mellitus, andcardiovascular disease risk. Circulation, 121(11):1356–64.Maruti, S. S., Feskanich, D., Rockett, H. R., Colditz, G. A., Sampson, L. A.,and Willett, W. C. (2006). Validation of adolescent diet recalled by adults.Epidemiology, 17(2):226–9.Mâsse, L. C., Naiman, D., and Naylor, P.-J. J. (2013). From policy topractice: implementation of physical activity and food policies in schools.Int J Behav Nutr Phys Act, 10:71.141BibliographyMcHugh, M. L. (2012). Interrater reliability: the kappa statistic. BiochemMed, 22(3):276–282.McKenna, M. L. (2010). Policy options to support healthy eating in schools.Can J Public Health, 101:S14–S17.Minaker, L. M., McCargar, L., Lambraki, I., Jessup, L., Driezen, P., Calen-gor, K., and Hanning, R. M. (2006). School region socio-economic statusand geographic locale is associated with food behaviour of Ontario andAlberta adolescents. Can J Public Health, pages 357–361.Ministry of Health and Ministry of Education (2013). Guidelinesfor food and beverage sales in BC schools. Retrieved August 17,2016, from http://www.healthlinkbc.ca/healthyeating/everyone/schools-communities.html.Moore, L. V. and Diez-Roux, A. V. (2015). Measurement and analyticalissues involved in the estimation of the effects of local food environmentson health behaviors and health outcomes. In Morland, K. B., editor, LocalFood Environments: Food Access in America, chapter 7, pages 205–226.CRC Press, Boca Raton, FL.Morin, P., Demers, K., Robitaille, E., Lebel, A., and Bisset, S. (2015). Doschools in Quebec foster healthy eating? An overview of associations be-tween school food environment and socio-economic characteristics. PublicHealth Nutr, 18(9):1635–46.Morland, K., Diez Roux, A. V., and Wing, S. (2006). Supermarkets, other142Bibliographyfood stores, and obesity: the Atherosclerosis Risk in Communities Study.Am J Prev Med, 30(4):333–9.Morland, K., Wing, S., and Roux, A. D. (2002). The contextual effect ofthe local food environment on residents’ diets: the Atherosclerosis Risk inCommunities Study. Am J Public Health, 92(11):1761–1768.Morland, K. B. (2015). Geography of local food environments: People andplaces. In Morland, K. B., editor, Local Food Environments: Food Accessin America, chapter 4, pages 87–120. CRC Press, Boca Raton, FL.Morland, K. B. and Evenson, K. R. (2009). Obesity prevalence and the localfood environment. Health Place, 15(2):491–5.NCHS (2012). Health, United States, 2011: With special feature on socioe-conomic status and health. U.S. National Center for Health Statistics,Hyattsville, MD. Retrieved June 1, 2016, from http://www.cdc.gov/nchs/data/hus/hus11.pdf.Neckerman, K. M., Bader, M. D. M., Richards, C. A., Purciel, M., Quinn,J. W., Thomas, J. S., Warbelow, C., Weiss, C. C., Lovasi, G. S., andRundle, A. (2010). Disparities in the food environments of New York Citypublic schools. Am J Prev Med, 39(3):195–202.Neumark-Sztainer, D., French, S. A., Hannan, P. J., Story, M., and Fulker-son, J. A. (2005). School lunch and snacking patterns among high schoolstudents: associations with school food environment and policies. Int JBehav Nutr Phys Act, 2(1):14.143BibliographyNewson, R. (2002). Parameters behind “nonparametric” statistics: Kendall’stau, Somers’ D and median differences. The STATA Journal, 2(1):454–64.Ng, M., Fleming, T., Robinson, M., Thomson, B., Graetz, N., Margono, C.,Mullany, E. C., Biryukov, S., Abbafati, C., Abera, S. F., Abraham, J. P.,Abu-Rmeileh, N. M. E., Achoki, T., AlBuhairan, F. S., Alemu, Z. A.,Alfonso, R., Ali, M. K., Ali, R., Guzman, N. A., Ammar, W., Anwari, P.,Banerjee, A., Barquera, S., Basu, S., Bennett, D. A., Bhutta, Z., Blore,J., Cabral, N., Nonato, I. C., Chang, J.-C., Chowdhury, R., Courville,K. J., Criqui, M. H., Cundiff, D. K., Dabhadkar, K. C., Dandona, L.,Davis, A., Dayama, A., Dharmaratne, S. D., Ding, E. L., Durrani, A. M.,Esteghamati, A., Farzadfar, F., Fay, D. F. J., Feigin, V. L., Flaxman, A.,Forouzanfar, M. H., Goto, A., Green, M. A., Gupta, R., Hafezi-Nejad, N.,Hankey, G. J., Harewood, H. C., Havmoeller, R., Hay, S., Hernandez, L.,Husseini, A., Idrisov, B. T., Ikeda, N., Islami, F., Jahangir, E., Jassal,S. K., Jee, S. H., Jeffreys, M., Jonas, J. B., Kabagambe, E. K., Khalifa, S.E. A. H., Kengne, A. P., Khader, Y. S., Khang, Y.-H., Kim, D., Kimokoti,R. W., Kinge, J. M., Kokubo, Y., Kosen, S., Kwan, G., Lai, T., Leinsalu,M., Li, Y., Liang, X., Liu, S., Logroscino, G., Lotufo, P. A., Lu, Y.,Ma, J., Mainoo, N. K., Mensah, G. A., Merriman, T. R., Mokdad, A. H.,Moschandreas, J., Naghavi, M., Naheed, A., Nand, D., Narayan, K. M. V.,Nelson, E. L., Neuhouser, M. L., Nisar, M. I., Ohkubo, T., Oti, S. O.,Pedroza, A., Prabhakaran, D., Roy, N., Sampson, U., Seo, H., Sepanlou,S. G., Shibuya, K., Shiri, R., Shiue, I., Singh, G. M., Singh, J. A., Skirbekk,V., Stapelberg, N. J. C., Sturua, L., Sykes, B. L., Tobias, M., Tran, B. X.,144BibliographyTrasande, L., Toyoshima, H., van de Vijver, S., Vasankari, T. J., Veerman,J. L., Velasquez-Melendez, G., Vlassov, V. V., Vollset, S. E., Vos, T.,Wang, C., Wang, X., Weiderpass, E., Werdecker, A., Wright, J. L., Yang,Y. C., Yatsuya, H., Yoon, J., Yoon, S.-J., Zhao, Y., Zhou, M., Zhu, S.,Lopez, A. D., Murray, C. J. L., and Gakidou, E. (2014). Global, regional,and national prevalence of overweight and obesity in children and adultsduring 1980–2013: a systematic analysis for the Global Burden of DiseaseStudy 2013. The Lancet, 384(9945):766–781.Ogden, C. L., Flegal, K. M., Carroll, M. D., and Johnson, C. L. (2002).Prevalence and trends in overweight among US children and adolescents,1999-2000. JAMA, 288(14):1728–1732.Oliver, L. N. and Hayes, M. V. (2005). Neighbourhood socio-economic statusand the prevalence of overweight Canadian children and youth. Can JPublic Health, pages 415–420.Oliver, L. N., Schuurman, N., and Hall, A. W. (2007). Comparing circularand network buffers to examine the influence of land use on walking forleisure and errands. Int J Health Geogr, 6:41.Papas, M. A., Alberg, A. J., Ewing, R., Helzlsouer, K. J., Gary, T. L., andKlassen, A. C. (2007). The built environment and obesity. Epidemiol Rev,29:129–43.Paquet, C., Daniel, M., Kestens, Y., Léger, K., and Gauvin, L. (2008).Field validation of listings of food stores and commercial physical activityestablishments from secondary data. Int J Behav Nutr Phys Act, 5:58.145BibliographyPawlak, R. and Malinauskas, B. (2008). Predictors of intention to eat2.5 cups of vegetables among ninth-grade students attending public highschools in eastern North Carolina. J Nutr Educ Behav, 40(6):392–398.Perks, S. M., Roemmich, J. N., Sandow-Pajewski, M., Clark, P. A., Thomas,E., Weltman, A., Patrie, J., and Rogol, A. D. (2000). Alterations in growthand body composition during puberty. IV. Energy intake estimated by theyouth-adolescent food-frequency questionnaire: validation by the doublylabeled water method. Am J Clin Nutr, 72(6):1455–1460.Petticrew, M., Cummins, S., Ferrell, C., Findlay, A., Higgins, C., Hoy, C.,Kearns, A., and Sparks, L. (2005). Natural experiments: an underusedtool for public health? Public Health, 119(9):751–7.Pikora, T. J., Bull, F. C., Jamrozik, K., Knuiman, M., Giles-Corti, B., andDonovan, R. J. (2002). Developing a reliable audit instrument to measurethe physical environment for physical activity. Am J Prev Med, 23(3):187–194.Pitney Bowes Software (2012). Canada Business Data. Pitney Bowes Soft-ware Inc., Troy, New York.Powell, L. M., Han, E., Zenk, S. N., Khan, T., Quinn, C. M., Gibbs, K. P.,Pugach, O., Barker, D. C., Resnick, E. A., Myllyluoma, J., and Chaloupka,F. J. (2011). Field validation of secondary commercial data sources on theretail food outlet environment in the U.S. Health Place, 17(5):1122–31.R Core Team (2016). R: A Language and Environment for Statistical Com-puting. R Foundation for Statistical Computing, Vienna, Austria.146BibliographyRichmond, T. K., Spadano-Gasbarro, J. L., Walls, C. E., Austin, S. B.,Greaney, M. L., Wang, M. L., Mezegebu, S., and Peterson, K. E. (2013).Middle school food environments and racial/ethnic differences in sugar-sweetened beverage consumption: findings from the Healthy ChoicesStudy. Prev Med, 57(5):735–8.Riediger, N. D., Shooshtari, S., and Moghadasian, M. H. (2007). The influ-ence of sociodemographic factors on patterns of fruit and vegetable con-sumption in Canadian adolescents. J Am Diet Assoc, 107(9):1511–8.Ries, J. and Somerville, T. (2010). School quality and residential propertyvalues: evidence from Vancouver rezoning. Rev Econ Stat, 92(4):928–944.Roberts, K. C., Shields, M., de Groh, M., Aziz, A., and Gilbert, J.-A. A.(2012). Overweight and obesity in children and adolescents: results fromthe 2009 to 2011 Canadian Health Measures Survey. Health Rep, 23(3):37–41.Robitaille, E., Bergeron, P., and Lasnier, B. (2010). Geographical analysisof the accessibility of fast-food restaurants and convenience stores aroundpublic schools in Quebec. Institut national de santé publique du Québec.Romses, K. and Lam, V. (2015). Canadian Resources for Food & NutritionEducation. Vancouver Coastal Health, Vancouver, BC. Retrieved June 6,2016 from http://www.vch.ca/media/Canadian_Resources_for_Food%20_Nutrition_Education_Dec_2015_Vancouver.pdf.Rossen, L. M., Curriero, F. C., Cooley-Strickland, M., and Pollack, K. M.147Bibliography(2013). Food availability en route to school and anthropometric change inurban children. J Urban Health, 90(4):653–66.Rossen, L. M., Pollack, K. M., and Curriero, F. C. (2012). Verificationof retail food outlet location data from a local health department usingground-truthing and remote-sensing technology: assessing differences byneighborhood characteristics. Health Place, 18(5):956–62.Rothman, K. J. (1990). No adjustments are needed for multiple comparisons.Epidemiol, 1(1):43–46.Royston, P., Altman, D. G., and Sauerbrei, W. (2006). Dichotomizing contin-uous predictors in multiple regression: a bad idea. Stat Med, 25(1):127–41.Rummo, P. E., Gordon-Larsen, P., and Albrecht, S. S. (2014). Field val-idation of food outlet databases: the Latino food environment in NorthCarolina, USA. Public Health Nutr, pages 1–6.Saelens, B. E., Glanz, K., Sallis, J. F., and Frank, L. D. (2007). NutritionEnvironment Measures Study in restaurants (NEMS-R): development andevaluation. Am J Prev Med, 32(4):273–81.Saelens, B. E., Sallis, J. F., Black, J. B., and Chen, D. (2003). Neighborhood-based differences in physical activity: an environment scale evaluation. AmJ Public Health, 93(9):1552–1558.Saelens, B. E., Sallis, J. F., Frank, L. D., Couch, S. C., Zhou, C., Colburn, T.,Cain, K. L., Chapman, J., and Glanz, K. (2012). Obesogenic neighborhood148Bibliographyenvironments, child and parent obesity: the Neighborhood Impact on KidsStudy. Am J Prev Med, 42(5):e57–64.Sanou, D., O’Reilly, E., Ngnie-Teta, I., Batal, M., Mondain, N., Andrew,C., Newbold, B. K., and Bourgeault, I. L. (2014). Acculturation andnutritional health of immigrants in Canada: a scoping review. J ImmigrMinor Health, 16(1):24–34.Sariyar, M. and Borg, A. (2010). The RecordLinkage package: Detectingerrors in data. R J, 2(2):61–67.Seliske, L., Pickett, W., Bates, R., and Janssen, I. (2012). Field validationof food service listings: a comparison of commercial and online geographicinformation system databases. Int J Environ Res Public Health, 9(8):2601–7.Seliske, L., Pickett, W., Rosu, A., and Janssen, I. (2013). The numberand type of food retailers surrounding schools and their association withlunchtime eating behaviours in students. Int J Behav Nutr Phys Act, 10:19.Seliske, L. M., Pickett, W., Boyce, W. F., and Janssen, I. (2009a). As-sociation between the food retail environment surrounding schools andoverweight in Canadian youth. Public Health Nutr, 12(9):1384–91.Seliske, L. M., Pickett, W., Boyce, W. F., and Janssen, I. (2009b). Densityand type of food retailers surrounding Canadian schools: variations acrosssocioeconomic status. Health Place, 15(3):903–907.Simmons, D., McKenzie, A., Eaton, S., Cox, N., Khan, M. A., Shaw, J., and149BibliographyZimmet, P. (2005). Choice and availability of takeaway and restaurantfood is not related to the prevalence of adult obesity in rural communitiesin Australia. Int J Obes, 29(6):703–10.Simon, P. A., Kwan, D., Angelescu, A., Shih, M., and Fielding, J. E. (2008).Proximity of fast food restaurants to schools: do neighborhood incomeand type of school matter? Prev Med, 47(3):284–8.Singer, J. D. and Willett, J. B. (2003). Applied Longitudinal Analysis: Mod-eling Change and Event Occurence. Oxford University Press, New York,NY.Smith, D., Cummins, S., Clark, C., and Stansfeld, S. (2013). Does the localfood environment around schools affect diet? Longitudinal associationsin adolescents attending secondary schools in East London. BMC PublicHealth, 13:70.Smith, W. R. (2015). The 2011 National Household Survey—the completestatistical story, StatCan Blog. Retrieved May 12, 2016, from http://www.statcan.gc.ca/eng/blog-blogue/cs-sc/2011NHSstory.Smoyer-Tomic, K. E., Spence, J. C., and Amrhein, C. (2006). Food desertsin the prairies? Supermarket accessibility and neighborhood need in Ed-monton, Canada. Prof Geogr, 58(3):307–326.Smoyer-Tomic, K. E., Spence, J. C., Raine, K. D., Amrhein, C., Cameron,N., Yasenovskiy, V., Cutumisu, N., Hemphill, E., and Healy, J. (2008). Theassociation between neighborhood socioeconomic status and exposure tosupermarkets and fast food outlets. Health Place, 14(4):740–54.150BibliographySparks, A. L., Bania, N., and Leete, L. (2010). Comparative approachesto measuring food access in urban areas: the case of Portland, Oregon.Urban Stud, 48(8):1715–1737.Speck, B. J., Bradley, C. B., Harrell, J. S., and Belyea, M. J. (2001). A foodfrequency questionnaire for youth: psychometric analysis and summary ofeating habits in adolescents. J Adolesc Health, 28(1):16–25.StataCorp (2015). Stata Statistical Software: Release 14. StataCorp LP,College Station, TX.Statistics Canada (2006a). Cartographic Boundary File. StatisticsCanada. Retrieved June 20, 2016, from http://www12.statcan.gc.ca/census-recensement/2011/geo/bound-limit/bound-limit-2006-eng.cfm.Statistics Canada (2006b). Population estimates and Projections. Statis-tics Canada. Retrieved June 20, 2016, from http://www.statcan.gc.ca/tables-tableaux/sum-som/l01/ind01/l3_3867_3433-eng.htm.Statistics Canada (2011). The Canadian population in 2011: pop-ulation counts and growth. Retrieved August 17, 2016, fromhttp://www12.statcan.gc.ca/census-recensement/2011/as-sa/98-310-x/98-310-x2011001-eng.cfm.Statistics Canada (2015a). Final Report on 2016 Census Options:Proposed Content Determination Framework and Methodology Op-tions. Retrieved May 12, 2016, from http://www12.statcan.gc.ca/151Bibliographycensus-recensement/fc-rf/reports-rapports/r2_table-tableau_3-eng.cfm.Statistics Canada (2015b). National Household Survey: Final ResponseRates. Retrieved May 12, 2016, from http://www12.statcan.gc.ca/nhs-enm/2011/ref/about-apropos/nhs-enm_r012.cfm.Statistics Canada (2016). Population and Dwelling Count Highlight Ta-bles, 2011 Census. Statistics Canada. Retrieved June 19, 2016,from http://www12.statcan.gc.ca/census-recensement/2011/dp-pd/hlt-fst/pd-pl/Table-Tableau.cfm.Stephens, T. A., Black, J. L., Chapman, G. E., Velazquez, C. E., and Rojas,A. (2016). Participation in school food and nutrition activities amonggrade 6–8 students in Vancouver. Can J Diet Pract Res, 77(1):1–6.Stewart, T., Duncan, S., Chaix, B., Kestens, Y., Schipperijn, J., andSchofield, G. (2015). A novel assessment of adolescent mobility: a pilotstudy. Int J Behav Nutr Phys Act, 12:18.Sturm, R. (2008). Disparities in the food environment surrounding US middleand high schools. Public Health, 122(7):681–90.Sturm, R. and Datar, A. (2005). Body mass index in elementary schoolchildren, metropolitan area food prices and food outlet density. PublicHealth, 119(12):1059–68.Svastisalee, C., Pagh Pedersen, T., Schipperijn, J., Ellegaard Jørgensen, S.,Holstein, B. E., and Krølner, R. (2015). Fast-food intake and perceived152Bibliographyand objective measures of the local fast-food environment in adolescents.Public Health Nutr, pages 1–10.Svastisalee, C. M., Holstein, B. E., and Due, P. (2012). Validation of pres-ence of supermarkets and fast-food outlets in Copenhagen: case studycomparison of multiple sources of secondary data. Public Health Nutr,15(7):1228–31.Taylor, J. P., Timmons, V., Larsen, R., Walton, F., Bryanton, J., Critch-ley, K., and McCarthy, M. J. (2007). Nutritional concerns in aboriginalchildren are similar to those in non-aboriginal children in Prince EdwardIsland, Canada. J Am Diet Assoc, 107(6):951–5.The Lancet Diabetes Endocrinology (2015). Sugar intake: lowering the bar.Lancet Diabetes Endocrinol, 3(5):305.Thornton, L. E., Pearce, J. R., and Kavanagh, A. M. (2011). Using Ge-ographic Information Systems (GIS) to assess the role of the built envi-ronment in influencing obesity: a glossary. Int J Behav Nutr Phys Act,8:71.Timperio, A. F., Ball, K., Roberts, R., Andrianopoulos, N., and Crawford,D. A. (2009). Children’s takeaway and fast-food intakes: associations withthe neighbourhood food environment. Public Health Nutr, 12(10):1960–4.Toft, U., Erbs-Maibing, P., and Glümer, C. (2011). Identifying fast-foodrestaurants using a central register as a measure of the food environment.Scand J Public Health, 39(8):864–9.153BibliographyTugault-Lafleur, C., Black, J. L., and Barr, S. I. (2016). Examining temporaldifferences in school day dietary intakes and factors associated with school-hour dietary intakes in Canada. Appl Physiol Nutr Metab, 41(S41).United States Census Bureau (2016). Introduction to NAICS: North Amer-ican Industry Classification System. United States Census Bureau. Re-trieved June 30, 2016, from http://www.census.gov/eos/www/naics/.University of Waterloo (2008). School Health Action Planning and Evalua-tion (SHAPES): Healthy Eating Student Questionnaire. Propel Centre forPopulation Health Impact, University of Waterloo, Waterloo, ON.Van Hulst, A., Barnett, T. A., Gauvin, L., Daniel, M., Kestens, Y., Bird,M., Gray-Donald, K., and Lambert, M. (2014). Associations betweenchildrens diets and features of their residential and school neighbourhoodfood environments. Can J Public Health, 31:164–172.Vancouver Board of Education (2009). Inner City Schools ProjectReview. Vancouver Board of Education. Retrieved June 20, 2016,from http://www.vsb.bc.ca/sites/default/files/school-files/Resources/ICP_Recommendations_Report_May_2009.pdf.Vancouver Coastal Health (2015). Inspection Reports. Vancouver, BC. Re-trieved October 20, 2015, from http://www.vch.ca/your-environment/facility-licensing/residential-care/inspection-reports/.Vancouver School Board (2012). Vancouver School Board Sectoral Re-view: Our Schools, Our Programs, Our Future. Vancouver School Board.154BibliographyRetrieved June 1, 2016, from http://www.placespeak.com/uploads/assets/sectoral-review-mar30.pdf.Vancouver School Board (2016). Our District. Retrieved June 4, 2016, fromhttp://www.vsb.bc.ca/about-vsb.Velazquez, C. E., Black, J. L., Billette, J.-M. M., Ahmadi, N., and Chapman,G. E. (2015). A comparison of dietary practices at or en route to schoolbetween elementary and secondary school students in Vancouver, Canada.J Acad Nutr Diet.Verbeek, M. (2012). A Guide to Modern Econometrics, 4th Edition. JohnWiley & Sons, Ltd, West Sussex, England.Von Hippel, P. T. (2007). Regression with missing Ys: an improved strategyfor analyzing multiply imputed data. Sociol Methodol, 37(1):83–117.Walker, R. E., Block, J., and Kawachi, I. (2013). The spatial accessibility offast food restaurants and convenience stores in relation to neighborhoodschools. Appl Spat Anal Policy, 7(2):169–182.Walton, M., Pearce, J., and Day, P. (2009). Examining the interactionbetween food outlets and outdoor food advertisements with primary schoolfood environments. Health Place, 15(3):811–8.Wang, Y. and Beydoun, M. A. (2007). The obesity epidemic in the UnitedStates—gender, age, socioeconomic, racial/ethnic, and geographic char-acteristics: a systematic review and meta-regression analysis. EpidemiolRev, 29:6–28.155BibliographyWatson, J. F., Collins, C. E., Sibbritt, D. W., Dibley, M. J., and Garg,M. L. (2009). Reproducibility and comparative validity of a food frequencyquestionnaire for Australian children and adolescents. Int J Behav NutrPhys Act, 6:62.Willett, W. (2012). Nutritional Epidemiology, 3rd Edition. Oxford UniversityPress, New York, NY.Williams, J., Scarborough, P., Matthews, A., Cowburn, G., Foster, C.,Roberts, N., and Rayner, M. (2014). A systematic review of the influenceof the retail food environment around schools on obesity-related outcomes.Obes Rev, 15(5):359–74.Winkler, W. E. (1990). String comparator metrics and enhanced decisionrules in the Fellegi-Sunter model of record linkage. Proceedings of theSection on Survey Research Methods, American Statistical Association,pages S. 354 – 369.Wong, J. E., Parnell, W. R., Black, K. E., and Skidmore, P. M. L. (2012).Reliability and relative validity of a food frequency questionnaire to assessfood group intakes in New Zealand adolescents. Nutr J, 11:65.Woolridge, J. M. (2009). Introductory Econometrics: A Modern Approach,4th Edition. South-Western Cengage Learning, Mason, OH.World Health Organization (2012). Sodium intake for adults and children.WHO Press, Geneva, Switzerland. Retrieved June 25, 2016, fromhttp://www.who.int/nutrition/publications/guidelines/sodium_intake/en/.156Zenk, S. N. and Powell, L. M. (2008). US secondary schools and food outlets.Health Place, 14(2):336–46.Zenk, S. N., Schulz, A. J., Matthews, S. A., Odoms-Young, A., Wilbur, J.,Wegrzyn, L., Gibbs, K., Braunschweig, C., and Stokes, C. (2011). Activityspace environment and dietary and physical activity behaviors: a pilotstudy. Health Place, 17(5):1150–61.157Appendix AGround-Truthing Protocol andClassification Scheme158			Ground	Truthing	Protocol	Data	Validation	Project			Checklist		ú Packet/binder	with	o Log	Sheet	o Store	Observation	Sheet	o Advertisement	Observation	Sheet	o Store	Classification	Guidelines	o Advertisement	Observation	Guidelines	o Overall	Map	o Individual	School	Map	o Official	Letter	ú Digital	camera	or	Camera	Phone	ú Mobile	GPS	Unit				 	159Strategy:		1. Record	start	date	and	time.	All	surveys	should	be	conducted	on	weekdays	between	9	a.m.	and	5	p.m.		2. For	each	school,	first	survey	both	sides	of	each	major	commercial	road	a. Then	start	at	the	north-most	point	on	the	individual	school	map.	i. 	Walk	each	east-west	road	(except	for	the	center	road)	first	on	the	north	side	and	then	on	the	south	side.	Take	the	most	central	road	to	move	from	north	to	south.	ii. Once	both	sides	of	each	east-west	road	have	been	examined,	apply	the	same	pattern	to	the	north-south	roads,	again	using	the	center	road	to	move	between	parallel	roads.	b. Now	examine	all	remaining	roads.		3. Upon	identifying	a	potential	food	vendor:	a. Assign	unique	id	number	representing	the	school,	number	representing	identification	order.	b. Photograph	site.	i. The	photo	should	be	recorded	with	coordinates	&	ID	number.	ii. At	least	one	photo	should	include	the	store	name.	c. Record	store	name	and	street	address.	d. Record	GPS	coordinates.	e. Follow	classification	chart	to	determine	classification.	4. Upon	identifying	a	potential	advertisement	or	signage:	a. Check	to	make	sure	the	object	is	visible	from	the	street	or	sidewalk.	b. Assign	a	unique	id	number	representing	the	school	and	the	identification	number.	c. Photograph	the	advertisement.	i. The	photo	should	be	recorded	with	coordinates	&	ID	number.	d. Record	the	advertisement	type,	description,	and	location	type	(e.g.	shop	window,	bus	station,	etc.).	e. Record	the	GPS	coordinates.	5. As	streets	are	visited,	record	on	individual	map.	Once	both	sides	of	each	street	have	been	examined,	record	end	time.	6. At	the	end	of	each	day,	download	photographs	to	the	project	computer.			Notes:		• If	you	encounter	someone	while	ground-truthing,	offer	the	attached	letter	to	describe	the	research	activities.		• If	a	potential	storefront	is	empty,	record	the	location	and	notes	on	what	may	have	been	there	previously;	similarly,	if	an	outlet	is	opening,	note	the	date.		 	160Log	Sheet		School	#1:	School	Name	________________________________________________________________________________	Date	visited	_________________________________________________________________________________	Start	Time	_______________		 End	Time	_______________		 Break	Periods	______________	Roads	examined	____________________________________________________________________________	_________________________________________________________________________________________________	_________________________________________________________________________________________________	No.	Stores	Identified	_______________________________________________________________________	No.	Advertisements	Identified	___________________________________________________________	Notes	_________________________________________________________________________________________	_________________________________________________________________________________________________	_________________________________________________________________________________________________		School	#2:	School	Name	________________________________________________________________________________	Date	visited	_________________________________________________________________________________	Start	Time	_______________		 End	Time	_______________		 Break	Periods	______________	Roads	examined	____________________________________________________________________________	_________________________________________________________________________________________________	_________________________________________________________________________________________________	No.	Stores	Identified	_______________________________________________________________________	No.	Advertisements	Identified	___________________________________________________________	Notes	_________________________________________________________________________________________	_________________________________________________________________________________________________	_________________________________________________________________________________________________		School	#3:	School	Name	________________________________________________________________________________	Date	visited	_________________________________________________________________________________	Start	Time	_______________		 End	Time	_______________		 Break	Periods	______________	Roads	examined	____________________________________________________________________________	_________________________________________________________________________________________________	_________________________________________________________________________________________________	No.	Advertisements	Identified	___________________________________________________________	No.	Stores	Identified	_______________________________________________________________________	Notes	_________________________________________________________________________________________	_________________________________________________________________________________________________	_________________________________________________________________________________________________		 	161Store	Observation	Sheet:	School	#2			Unique	ID		Name		Address	&	coordinates		Classification		Notes			2001		 	 N	___._________	W___._________	 	 		2002		 	 N	___._________	W___._________	 	 		2003		 	 N	___._________	W___._________	 	 		2004		 	 N	___._________	W___._________	 	 		2005		 	 N	___._________	W___._________	 	 		2006		 	 N	___._________	W___._________	 	 		2007		 	 N	___._________	W___._________	 	 		2008		 	 N	___._________	W___._________	 	 		2009		 	 N	___._________	W___._________	 	 		2010		 	 N	___._________	W___._________	 	 		2011		 	 N	___._________	W___._________	 	 		2012		 	 N	___._________	W___._________	 	 						162	Unique	ID		Name		Address	&	coordinates		Classification		Notes			2013		 	 N	___._________	W___._________	 	 		2014		 	 N	___._________	W___._________	 	 		2015		 	 N	___._________	W___._________	 	 		2016		 	 N	___._________	W___._________	 	 		2017		 	 N	___._________	W___._________	 	 		2018		 	 N	___._________	W___._________	 	 			2019		 	 N	___._________	W___._________	 	 		2020		 	 N	___._________	W___._________	 	 		2021		 	 N	___._________	W___._________	 	 		2022		 	 N	___._________	W___._________	 	 		2023		 	 N	___._________	W___._________	 	 		2024		 	 N	___._________	W___._________	 	 		2025		 	 N	___._________	W___._________	 	 				 	 163Advertisement	Observation	Sheet:	School	_______________________			Unique	ID		Category		Type		Location		Setting		Coordinates		2001		 Ad	Signage	 	 	 Main	Street	Residential	 N	___._________	W___._________	Content:								Food								Alcohol								Tobacco								Other	______________	Description	(include	size,	product	and	brand	name):		Notes:			2002	 Ad	Signage	 	 	 Main	Street	Residential	 N	___._________	W___._________	Content:								Food								Alcohol								Tobacco								Other	______________	Description	(include	size,	product	and	brand	name):		Notes:			2003		 Ad	Signage	 	 	 Main	Street	Residential	 N	___._________	W___._________	Content:								Food								Alcohol								Tobacco								Other	______________	Description	(include	size,	product	and	brand	name):		Notes:			2004		 Ad	Signage	 	 	 Main	Street	Residential	 N	___._________	W___._________	Content:								Food								Alcohol								Tobacco								Other	______________	Description	(include	size,	product	and	brand	name):		Notes:				 	164	Unique	ID		Category		Type		Location		Setting		Coordinates		2005		 Ad	Signage	 	 	 Main	Street	Residential	 N	___._________	W___._________	Content:								Food								Alcohol								Tobacco								Other	______________	Description	(include	size,	product	and	brand	name):		Notes:			2006	 Ad	Signage	 	 	 Main	Street	Residential	 N	___._________	W___._________	Content:								Food								Alcohol								Tobacco								Other	______________	Description	(include	size,	product	and	brand	name):		Notes:			2007		 Ad	Signage	 	 	 Main	Street	Residential	 N	___._________	W___._________	Content:								Food								Alcohol								Tobacco								Other	______________	Description	(include	size,	product	and	brand	name):		Notes:			2008		 Ad	Signage	 	 	 Main	Street	Residential	 N	___._________	W___._________	Content:								Food								Alcohol								Tobacco								Other	______________	Description	(include	size,	product	and	brand	name):		Notes:				165Classification	Guidelines		Store	Type	 Description	 Key	Questions	 Code	Drugstore		 A	retail	store	including	a	pharmacy	that	offers	snacks	or	beverages	1. Does	the	store	have	a	pharmacy?	 CvPh	Gas	station	convenience	store	 A	retail	store	attached	to	a	gas	station	offering	primarily	snacks	and	beverages		1. Is	the	store	connected	with	a	gas	station?	2. Do	snack	food	items	and	beverages	comprise	a	majority	of	the	goods	sold?	CvGa	Regular	convenience	store	 A	retail	store	offering	primarily	snack	foods	–	but	may	offer	a	variety	of	other	products;	open	18-24	hours	1. Do	snack	food	items	and	beverages	comprise	a	majority	of	the	goods	sold?	2. Does	the	store	have	fewer	than	three	cash	registers,	or	is	otherwise	smaller	than	a	traditional	grocery	store?	3. Is	the	store’s	stock	more	limited	than	what	would	be	available	in	a	grocery	store	or	supermarket?	Cv	Supermarket	 A	large	retail	store	with	all	of	the	departments	of	a	traditional	grocery	store	earning	over	$2mil/year	in	revenues	1. Does	the	store	have	all	of	the	departments	of	a	traditional	grocer	(dairy,	bakery,	produce,	butcher)?	2. Is	the	store	open	more	than	18	hours	per	day	or	7	days	per	week?	3. Does	the	store	have	more	than	two	cash	registers?	Sm	Grocery	store	 A	retail	store	with	all	the	depart-ments	of	a	traditional	grocery,	but	smaller	than	a	supermarket.	1. Does	the	store	have	dairy,	deli,	bakery,	butcher	and	produce	departments?		2. Is	the	store	closed	during	the	week	or	in	the	evening?		3. Is	the	store	smaller	than	a	conventional	supermarket?	4. Does	the	store	have	two	or	fewer	cash	registers?	SmGr				 	166Store	Type	 Description	 Key	Questions	 Code	Produce	Outlet	 A	retail	store	primarily	engaged	in	the	sale	of	fruits	and	vegetables.	1. Is	produce	displayed	prominently	outside	of	or	within	the	store?	2. Does	produce	comprise	a	majority	of	the	store’s	offerings?	SmPr	Other	specialty	food	store	 Any	retail	store	selling	food	or	beverages	that	does	not	qualify	in	the	above	categories.	1. Does	the	store	sell	mostly	one	type	of	food	item	to	be	prepared/eaten	at	home	(meat,	cheese,	etc.)?	2. Are	the	majority	of	the	store’s	food	items	associated	with	one	or	several	ethnic	groups?	SmSp	Fast	food	restaurant	 A	restaurant	offering	eat-in	or	takeaway	options	and	more	limited	service	than	that	of	a	traditional	restaurant	1. Does	the	outlet	provide	both	food	to	be	eaten	on	the	premises	and	takeaway	options?	2. Do	patrons	primarily	pay	before	consuming	foods	or	beverages?	ReFF	Coffee	shop	 A	restaurant	offering	eat-in	or	takeaway	options,	primarily	engaged	in	the	sale	of	beverages,	with	limited	service.	1. Does	the	outlet	offer	coffee	and	other	hot	beverages?	Are	these	items	a	majority	of	the	offerings	or	particularly	prominently	advertised	and	offered?	2. Do	patrons	primarily	pay	before	consuming	food	or	beverages?	ReCo	Other	Restaurant	 A	traditional	restaurant	offering	table	service,	where	eat-in	is	a	more	significant	portion	of	sales	than	takeaway	service	1. Does	the	outlet	provide	food	to	be	eaten	on	the	premises?	2. Do	patrons	primarily	pay	after	eating?	3. Are	orders	generally	taken	while	patrons	are	seated?	Re		 	167Classification	Choice	Flow	Diagram			 	168Advertisement	Recognition	Guidelines		In	addition	to	store	locations,	we	are	also	recording	the	locations	of	commercial	grade	outdoor	advertisements.	We	are	looking	for	two	types	of	marketing	materials			Advertisement:	a	sign	with	branded	information,	pictures,	or	logos.	Signage:		all	signs	unaccompanied	by	additional	branded	product	information		In	order	to	be	considered	for	this	study,	an	advertisement	must	be:	1. visible	from	the	street	or	sidewalk	a. e.g.	billboards,	bus	shelter	advertisements,	and	store	window	posters	2. Stationary	a. Hand-drawn	or	painted	advertisements	or	advertisements	on	buses	should	not	be	included.	3. Related	to	food	or	diet		Once	an	advertisement	is	identified,	the	category,	type,	location,	setting,	and	subject	should	be	recorded	in	the	advertisement	observation	sheet.	Possible	observations	include:		1. Category	• Advertisement	o e.g.	billboards/	posters,	event	advertising,	advertisements	on	outdoor	furniture,	building	signs	w/	branded	product	information.	• Signage	o signs	identifying	and	naming	sites/	buildings/	building	uses;	should	be	limited	to	symbols	or	words	only.	2. Type		&	Size	• Billboard	• Poster	• Freestanding	sign	• Neon	sign	• Electronic	boards	• Banners	• Bus	shelter	signs	• Other	______	Size:	• small:	≥21	cm	×	20	cm	but	<1.2	m	×	1.9	m	• medium	≥1.2	m	×	1.9	m	but	<2.0	m	×	2.5	m		• large:	≥2m×2.5m	3. Location	• Drugstore	• Gas	station	convenience	store	• Regular	convenience		• Supermarket	• Grocery	store	• Produce	outlet	• Other	specialty	food	store	• Fast	food	restaurant	• Coffee	shop	• Other	restaurant	• Other			________	4. Setting	• Main	street	• Residential	street	5. Subject	• Food	&	Beverage	• Alcohol	• Tobacco	• Other	____________			 	 169Maps			170Individual	School	#2:	David	Livingstone	Elementary				 	171					To	Whom	It	May	Concern:		During	the	summer	and	fall	of	2015,	researchers	at	the	University	of	British	Columbia	will	be	conducting	research	on	the	sources	of	food	available	to	secondary	and	elementary	school	students	in	the	city	of	Vancouver.			Researchers	will	be	examining	many	of	the	roads	within	1km	of	Vancouver	Public	schools		to	identify	store	locations	and	to	collect	basic	information	such	as	store	name	or	type.	We	do	not	work	for	the	city	or	provincial	governments,	and	our	findings	will	be	made	publicly	available	within	the	next	year,	but	will	not	identify	stores	by	name.			If	you	have	any	questions	about	the	research,	please	contact	Madeleine	Daepp	at	mdaepp@alumni.ubc.ca.	Thank	you	for	your	interest	in	this	study.		Sincerely,						Madeleine	Daepp	M.Sc.	Candidate	–	Integrated	Studies	in	Land	&	Food	Systems	University	of	British	Columbia,	Vancouver			172Appendix BAdditional Tables forChapter 2173AppendixB.AdditionalTablesforChapter2Table B.1: Name-based classifications system applied to identify major store typesVancouver Coastal Health Canada Business Points Enhanced Pointsof InterestLimited-Service “McDonald’s”, “Wendy’s”, “McDonald’s”, “Wendy’s”, “McDonald’s”, “Wendy’s”,Outlet “Subway”, “Quizno”, “Subway”, “Quizno”, “Subway”, “Quizno”,“freshslice”, “Church’s “freshslice”, “Church’s “freshslice”, “Church’sChicken”, “Vera’s”, Chicken”, “Vera’s”, Chicken”, “Vera’s”,“Kentucky Fried” “Kentucky Fried” “Kentucky Fried”“Panago”, “Al Basha”, “Panago”, “A & W” “Panago”, “A & W”“nando’s”, “Buddha’s “nando’s”, “Buddha’s “nando’s”, “Buddha’sOrient”, “Solly’s”, “creme”, Orient”, “Solly”, “creme”, Orient”, “Solly”, “creme”,“Freshii”, “Tim Hortons”, “Freshii”, “Tim Hortons”, “Freshii”, “Tim Hortons”,“Starbucks”, “Waffle Gone “Starbucks”, “Waffle Gone “Starbucks”, “Waffle GoneWild”, “Dairy Queen”, Wild”, “Dairy Queen”, Wild”, “Dairy Queen”,“shawarma”, “Pizza”, “shawarma”, “Pizza”, “shawarma”, “Pizza”,“Gelat”, “Bagel”, “Falafel”, “Gelat”, “Bagel”, “Falafel”, “Gelat”, “Bagel”, “Falafel”,“sandwich”, “burrito” “sandwich”, “burrito” “sandwich”, “burrito”“pizzeria”, “sweet”, “bur- “pizzeria”, “sweet”, “pizzeria”, “sweet”,ger” “donair”, “ice cream” “donair”, “ice cream” “donair”, “ice cream”“donut”, “Cafe”, “coffee”, “donut”, “blenz”, “coffee”, “donut”, “blenz”, “coffee”,“caffe”, “juice”, “bean”, “juice”, “tea”, “burger”, “juice”, “tea”, “burger”,“chai”, “cream”, “express” “chai”, “cream”, “express” “chai”, “cream”, “express”174AppendixB.AdditionalTablesforChapter2(Continued) Vancouver Coastal Health Canada Business Points Enhanced Pointsof InterestConvenience “Convenience”, “Mart” “Convenience”, “Mart” “ ‘Convenience”, “Mart”Stores “Shell” , “Chevron”, “Shell” , “Chevron”, “Esso”, “Shell” , “Chevron”, “Esso”,“Stop”, “Drug”, “Rx” “Food Stop”, “Drug”, “Rx” “Food Stop”, “Drug”, “Rx”“Gas”, “Store”, “food”, “Gas”, “Store”, “food”, “Gas”, “Store”, “food”,“Petro”, “Town Pantry”, “Petro”, “Town Pantry”, “Petro”, “Town Pantry”,“Husky”, “Pharmacy” “Husky”, “Pharmacy” “Husky”, “Pharmacy”“Rexall”, “Shoppers”, “Rexall”, “Shoppers”, “Rexall”, “Shoppers”,“7-Eleven”, “Medicine” “7-Eleven”, “Medicine”, “7-Eleven”, “Medicine”,“market”, “Esso”, “Pharmasave”, “market” “Pharmasave”, “market”Supermarket “Grocery”, “Supermarket”, “Grocery”, “Supermarket”, “Grocery”, “Supermarket”,or Grocery “Super Valu”, “Safeway”, “Super Valu”, “Safeway”, “Super Valu”, “Safeway”,Stores “Choices”, “Persia”, “Choices”, “Persia”, “Choices”, “Persia”,“Donald’s”, “Marketplace” “Donald’s”, “Marketplace” “Donald’s”, “Marketplace”“Famous Foods”, “Nesters”, “Famous Foods”, “Nesters”, “Famous Foods”, “Nesters”,“Co-op”, “Save-on”, “Co-op”, “Save-on”, “Co-op”, “Save-on”,“Farm Market”, “Price “Farm Market”, “Price “Farm Market”, “Pricesmart” smart”, “Grocer”, smart”, “Stop & Shop”“Stop & Shop”, “Loblaw”Relevant terms were identified with frequency tabulations and lists of terms were iteratively refined untilall food outlets were classified. Name-based classifications were not case sensitive.175Appendix B. Additional Tables for Chapter 2Table B.2: Bivariate associations of commercial density or socioeconomicstatus and the odds of false positive listings† in each secondary data sourceBusiness Vancouver Canada EnhancedLicenses Coastal Business Points2015 2012 Health Points of InterestCommercialDensityPer 100 0.95 0.96 1.02 1.05 1.06outlets (0.90 - 1.01) (0.91 - 1.02) (0.95 - 1.10) (0.98 - 1.12) (0.99 - 1.12)VANDIX§low – – – – –medium 1.05 0.97 0.86 0.70∗ 0.74(0.76 - 1.44) (0.70 - 1.33) (0.59 - 1.25) (0.50 - 0.99) (0.53 - 1.03)high 0.98 1.07 1.20 0.85 0.86(0.72 - 1.35) (0.79 - 1.47) (0.82 - 1.75) (0.60 - 1.21) (0.62 - 1.21)N 923 929 677 778 851Odds ratios with 95% confidence intervals in parentheses. †A listing was a false positiveif an outlet was listed in the secondary data but not identified or misclassified in theground-truthed data. §“high” refers to the most deprived neighbourhoods.∗significant at 0.05; ∗∗significant at 0.01; ∗∗∗significant at 0.001176Appendix B. Additional Tables for Chapter 2Table B.3: Bivariate associations of commercial density or socioeconomicstatus and the odds of false negative listings† in each secondary data sourceBusiness Vancouver Canada EnhancedLicenses Coastal Business Points2015 2012 Health Points of InterestCommercialDensityPer 100 0.95 0.97 1.07∗ 1.11∗∗ 1.08∗outlets (0.89 - 1.01) (0.91 - 1.03) (1.01 - 1.14) (1.04 - 1.18) (1.02 - 1.15)VANDIX§low – – – – –medium 1.11 1.26 0.95 0.67∗ 0.84(0.78 - 1.58) (0.89 - 1.77) (0.68 - 1.34) (0.47 - 0.95) (0.59 - 1.19)high 0.93 1.08 1.36 0.94 1.10(0.65 - 1.33) (0.76 - 1.53) (0.96 - 1.92) (0.66 - 1.33) (0.78 - 1.56)N 788 788 788 788 788Odds ratios with 95% confidence intervals in parentheses. †A listing was a false negativeif an outlet was identified while ground- but not identified or misclassified in thesecondary data source. §“high” refers to the most deprived neighbourhoods.∗significant at 0.05; ∗∗significant at 0.01; ∗∗∗significant at 0.001177Appendix CAdditional Tables forChapter 3Table C.1: Results from negative binomial regressions with food outlet den-sities within 800m of Vancouver schools (n=113) as dependent variables andstudent socioedemographic factors as independent variables(1) (2) (3)800m Density Limited Convenience Grocery/Service Store Supermarket% Aboriginal 1.01 1.01 1.01(0.99 - 1.04) (1.00 - 1.03) (0.99 - 1.03)% English Language 1.00 1.00 1.00Learners (ELL) (0.99 - 1.01) (0.99 - 1.01) (0.99 - 1.01)Inner City 1.35 1.88∗ 2.08∗Project (ICP) (0.69 - 2.61) (1.13 - 3.14) (1.09 - 3.95)McFadden’s Pseudo R2 0.01 0.03 0.02Incidence rate ratios with 95% confidence intervals in parentheses∗significant at 0.05; ∗∗significant at 0.01; ∗∗∗significant at 0.001178Appendix C. Additional Tables for Chapter 3Table C.2: Results from negative binomial regressions with food outlet den-sities within 160m of Vancouver schools (n=113) as dependent variables andstudent socioedemographic factors as independent variables(1) (2) (3)160m Density Limited Convenience Grocery/Service Store Supermarket§% Aboriginal 1.02 1.02(0.97 - 1.07) (0.98 - 1.06)% English Language 0.97∗ 0.99Learners (ELL) (0.97 - 1.00) (0.97 - 1.03)Inner City 1.63 2.99Project (ICP) (0.33 - 8.01) (0.82 - 10.88)McFadden’s Pseudo R2 0.01 0.06Incidence rate ratios with 95% confidence intervals in parentheses∗significant at 0.05; ∗∗significant at 0.01; ∗∗∗significant at 0.001§Model failed to converge; only six schools had any outlets within 160m.179Appendix C. Additional Tables for Chapter 3Table C.3: Results from negative binomial regressions with food outlet den-sities within 800m of Vancouver schools (n=113) as dependent variables andstudent socioedemographic factors as independent variables, adjusted forschool controls and neighbourhood factors(1) (2) (3)800m Density Limited Convenience Grocery/Service Store SupermarketStudents% Aboriginal 1.00 1.00 1.00(0.99 - 1.01) (0.99 - 1.01) (0.98 - 1.01)% English Language 1.00 1.00 1.00Learners (ELL) (0.99 - 1.00) (0.99 - 1.00) (0.98 - 1.01)Inner City 1.14 1.38 1.82∗Project (ICP) (0.76 - 1.72) (0.98 - 1.94) (1.05 - 3.14)SchoolsTotal Enrolment† 0.98 0.98 0.95(0.93 - 1.04) (0.94 - 1.03) (0.88 - 1.04)School LevelElementary – – –Secondary 1.21 1.18 1.34(0.65 - 2.27) (0.70 - 1.98) (0.58 - 3.13)NeighbourhoodsCommercial 1.04∗∗∗ 1.02∗∗∗ 1.02∗∗∗density (800m)‡ (1.03 - 1.05) (1.02 - 1.03) (1.01 - 1.03)VANDIX tertile§low – – –medium 1.38∗ 1.57∗∗ 1.42(1.01 - 1.91) (1.17 - 2.10) (0.88 - 2.30)high 1.39 1.74∗∗∗ 1.37(0.99 - 1.94) (1.28 - 2.35) (0.82 - 2.27)McFadden’s Pseudo R2 0.13 0.17 0.11Incidence rate ratios with 95% confidence intervals in parentheses∗significant at 0.05; ∗∗significant at 0.01; ∗∗∗significant at 0.001†per 100 students; ‡per 10 non-food outlets within 800m§“high” refers to the most deprived neighbourhoods.180Appendix C. Additional Tables for Chapter 3Table C.4: Results from negative binomial regressions with food outlet den-sities within 160m of Vancouver schools (n=113) as dependent variables andstudent socioedemographic factors as independent variables, adjusted forschool controls and neighbourhood factors(1) (2) (3)160m Density Limited Convenience Grocery/Service Store SupermarketaStudents% Aboriginal 1.01 1.01(0.97 - 1.05) (0.99 - 1.04)% English Language 0.98 1.01Learners (ELL) (0.95 - 1.01) (0.99 - 1.04)Inner City 1.06 1.19Project (ICP) (0.27 - 4.15) (0.37 - 3.82)SchoolsTotal Enrolment† 1.00 1.04(0.84 - 1.19) (0.87 - 1.25)School LevelElementary – – –Secondary 1.17 0.95(0.19 - 7.04) (0.14 - 6.41)NeighbourhoodsCommercial 2.71∗∗∗ 2.06∗∗∗density (160m)‡ (1.90 - 3.87) (1.59 - 2.67)VANDIX tertile§low – – –medium 1.01 1.59(0.38 - 2.70) (0.48 - 5.28)high 0.43 1.18(0.13 - 1.45) (0.32 - 4.29)McFadden’s Pseudo R2 0.22 0.24Incidence rate ratios with 95% confidence intervals in parentheses∗significant at 0.05; ∗∗significant at 0.01; ∗∗∗significant at 0.001aModel failed to converge; only six schools had any outlets within 160m.†per 100 students; ‡per 10 non-food outlets within 160m§“high” refers to the most deprived neighbourhoods. 181Appendix C. Additional Tables for Chapter 3Table C.5: Results from negative binomial regressions with major chainlimited-service outlet densities surrounding Vancouver schools (n=113) asdependent variables and student socioedemographic factors as independentvariables(1) (2) (3)Density Major Chains Major Chains Major Chains(160m) (400m) (800m)% Aboriginal 0.68 1.01 1.01(0.25 - 1.81) (0.97 - 1.05) (0.98 - 1.04)% English Language 0.95 0.97∗ 1.00Learners (ELL) (0.86 - 1.06) (0.95 - 1.00) (0.98 - 1.01)Inner City 0.00 2.24 1.42Project (ICP) (0 - .) (0.57 - 8.76) (0.62 - 3.26)McFadden’s Pseudo R2 0.07 0.03 0.01Incidence rate ratios with 95% confidence intervals in parentheses∗significant at 0.05; ∗∗significant at 0.01; ∗∗∗significant at 0.001182Appendix C. Additional Tables for Chapter 3Table C.6: Results from negative binomial regressions with major chainlimited-service outlet densities surrounding Vancouver schools (n=113) asdependent variables and student socioedemographic factors as independentvariables, adjusted for school and neighbourhood factors(1) (2) (3)Density Major Chains Major Chains Major Chains(160m) (400m) (800m)Students% Aboriginal 1.08 1.00 1.00(0.81 - 1.45) (0.97 - 1.03) (0.99 - 1.02)% English Language 1.05 1.00 0.99Learners (ELL) (0.99 - 1.12) (0.98 - 1.02) (0.98 - 1.00)Inner City 0.00 1.92 1.77∗Project (ICP) (0 - .) (0.69 - 5.37) (1.05 - 2.97)SchoolsTotal Enrolment† 0.63 1.02 1.00(0.15 - 2.69) (0.88 - 1.17) (0.93 - 1.07)School LevelElementary – – –Secondary 15.32 1.61 1.21(0.15 - 1528.4) (0.42 - 6.20) (0.58 - 2.53)NeighbourhoodsCommercial 5.46∗∗a 1.20∗∗∗b 1.03∗∗∗cdensity‡ (1.93 - 15.44) (1.14 - 1.26) (1.02 - 1.04)VANDIX tertile§low – – –medium 0.00 0.62 1.01(0 - .) (0.25 - 1.51) (0.67 - 1.52)high 0.05 0.55 0.70(0 - 3.65) (0.20 - 1.48) (0.44 - 1.11)McFadden’s Pseudo R2 0.64 0.25 0.19Incidence rate ratios with 95% confidence intervals in parentheses∗significant at 0.05; ∗∗significant at 0.01; ∗∗∗significant at 0.001†per 100 students; §“high” refers to the most deprived neighbourhoods.‡per 10 non-food outlets within a160m, b400m or c800m183Appendix DAdditional Tables forChapter 4184Appendix D. Additional Tables for Chapter 4Table D.1: Bivariate associations of outlet density within 400m and students’daily intakes of minimally nutritious foodsa at or en-route to schoolDensity (400m)b Sugar-Sweetened Beverages (n=936†)Ltd. Service Outlet 1.02(0.98 - 1.05)Conv. Store 1.04(0.98 - 1.09)Grocery Store 1.19(0.86 - 1.64)Fast Foods (n=942†)Ltd. Service Outlet 1.01(0.97 - 1.05)Conv. Store 1.03(0.95 - 1.10)Grocery Store 1.08(0.70 - 1.67)Packaged Snacks (n=948†)Ltd. Service Outlet 0.98(0.95 - 1.02)Conv. Store 0.96(0.90 - 1.02)Grocery Store 0.98(0.67 - 1.42)Results are from multilevel logistic models with school random interceptsOdds Ratios are reported; 95% confidence intervals are in parentheses.aDependent variables = 1 if a student reported daily consumptionbCount of outlets within 400m line-based buffers of each school†Cases with missing dependent variables were omitted from the analysis185Appendix D. Additional Tables for Chapter 4Table D.2: Bivariate associations of outlet density within 800m and students’daily intakes of minimally nutritious foodsa at or en-route to schoolDensity (800m)b Sugar-Sweetened Beverages (n=936†)Ltd. Service Outlet 1.00(0.99 - 1.02)Conv. Store 1.01(0.99 - 1.04)Grocery Store 1.04(0.95 - 1.13)Fast Foods (n=942†)Ltd. Service Outlet 1.01(0.99 - 1.03)Conv. Store 1.02(0.98 - 1.05)Grocery Store 1.08(0.97 - 1.21)Packaged Snacks (n=948†)Ltd. Service Outlet 0.99(0.98 - 1.01)Conv. Store 0.98(0.95 - 1.01)Grocery Store 0.99(0.90 - 1.10)Results are from multilevel logistic models with school random interceptsOdds Ratios are reported; 95% confidence intervals are in parentheses.aDependent variables = 1 if a student reported daily consumptionbCount of outlets within 800m line-based buffers of each school†Cases with missing dependent variables were omitted from the analysis186Appendix D. Additional Tables for Chapter 4Table D.3: Bivariate associations of perceived outlet proximity and odds ofdaily intake, at or en-route to school, of minimally nutritiousa foodsPerceived Proximity Sugar-Sweetened BeveragesLtd. Service Outlet (n=740†)< 5 minutes 1.005 - 10 minutes 0.98(0.66 - 1.44)> 10 minutes 0.90(0.60 - 1.37)Convenience Store (n=765†)< 5 minutes 1.005 - 10 minutes 0.81(0.56 - 1.19)> 10 minutes 1.43(0.88 - 2.34)Supermarket/Grocery Store (n=691†)< 5 minutes 1.005 - 10 minutes 0.75(0.50 - 1.12)> 10 minutes 0.63∗(0.41 - 0.96)Fast FoodsLtd. Service Outlet (n=743†)< 5 minutes 1.005 - 10 minutes 1.41(0.89 - 2.23)> 10 minutes 0.92(0.54 - 1.55)Convenience Store (n=769†)< 5 minutes 1.005 - 10 minutes 1.28(0.82 - 2.01)187Appendix D. Additional Tables for Chapter 4Perceived Proximity Fast Foods(Continued)> 10 minutes 1.80(0.99 - 3.27)Supermarket/Grocery Store (n=694†)< 5 minutes 1.005 - 10 minutes 0.87(0.54 - 1.39)> 10 minutes 0.60(0.36 - 1.00)Packaged SnacksLtd. Service Outlet (n=746†)< 5 minutes 1.005 - 10 minutes 1.11(0.71 - 1.72)> 10 minutes 1.17(0.73 - 1.86)Convenience Store (n=772†)< 5 minutes 1.005 - 10 minutes 1.10(0.72 - 1.67)> 10 minutes 1.40(0.81 - 2.42)Supermarket/Grocery Store (n=936†)< 5 minutes 1.005 - 10 minutes 1.06(0.67 - 1.66)> 10 minutes 0.75(0.46 - 1.21)Results are from multilevel logistic models with school random interceptsOdds Ratios are reported; 95% confidence intervals are in parentheses.aDependent variables = 1 if a student reported daily consumption†Cases with missing observations were omitted from the analysis∗significant at 0.05188AppendixD.AdditionalTablesforChapter4Table D.4: Multivariate adjusted associations from multilevel logistic modelsof outlet proximity and students’ odds of daily intake of minimally nutritiousfoods at or en-route to school, complete case analysisSugar-Sweetened Beveragesa Fast Foodsa Packaged SnacksaProximitybLtd. Service 1.00 1.00 1.00Outlet (0.99 - 1.00) (1.00 - 1.00) (1.00 - 1.00)Convenience 1.00 1.00 1.00Store (1.00 - 1.00) (1.00 - 1.00) (1.00 - 1.00)Grocery Store 1.00 1.00 1.00(1.00 - 1.00) (1.00 - 1.00) (1.00 - 1.00)ControlsGenderFemale – – – – – – – – –Male 1.57∗ 1.57∗ 1.58∗ 2.17∗∗ 2.19∗∗ 2.17∗∗ 1.58∗ 1.60∗ 1.60∗(1.08 - 2.29) (1.08 - 2.29) (1.08 - 2.30) (1.33 - 3.54) (1.34 - 3.59) (1.32 - 3.55) (1.02 - 2.47) (1.03 - 2.51) (1.02 - 2.51)Food 1.61∗ 1.59 1.60 2.20∗∗ 2.15∗∗ 2.19∗ 0.93 0.90 0.89Insecured (1.00 - 2.60) (0.99 - 2.57) (0.99 - 2.58) (1.25 - 3.87) (1.22 - 3.79) (1.19 - 3.69) (0.51 - 1.67) (0.50 - 1.63) (0.49 - 1.60)Acculturationehigh – – – – – – – – –medium 1.05 1.06 1.07 1.76 1.81 1.84 0.71 0.73 0.73(0.65 - 1.69) (0.66 - 1.71) (0.66 - 1.72) (0.88 - 3.53) (0.90 - 3.62) (0.92 - 3.69) (0.41 - 1.21) (0.42 - 1.25) (0.43 - 1.26)low 1.51 1.53 1.56 4.06∗∗ 4.17∗∗ 4.25∗∗ 1.37 1.43 1.45(0.73 - 3.11) (0.74 - 3.15) (0.76 - 3.22) (1.63 - 10.10) (1.67 - 10.44) (1.69 - 10.70) (0.61 - 3.08) (0.63 - 3.22) (0.64 - 3.26)Brought from 1.12 1.14 1.14 1.13 1.11 1.14 2.21∗∗ 2.21∗∗ 2.26∗∗home dailyc (0.76 - 1.66) (0.77 - 1.69) (0.77 - 1.69) (0.68 - 1.85) (0.67 - 1.83) (0.69 - 1.88) (1.39 - 3.52) (1.38 - 3.54) (1.41 - 3.61)189AppendixD.AdditionalTablesforChapter4(Continued) Sugar-Sweetened Beveragesa Fast Foodsa Packaged SnacksaSpendingMoneyNone – – – – – – – – –$0 - $10 0.70 0.70 0.70 1.68 1.70 1.73 0.82 0.83 0.83(0.40 - 1.23) (0.39 - 1.23) (0.40 - 1.24) (0.67 - 4.19) (0.68 - 4.25) (0.69 - 4.32) (0.40 - 1.68) (0.40 - 1.70) (0.40 - 1.71)$10 - $20 0.85 0.85 0.85 2.52 2.49 2.53 1.64 1.64 1.66(0.46 - 1.56) (0.46 - 1.56) (0.46 - 1.57) (0.99 - 6.43) (0.97 - 6.37) (0.99 - 6.48) (0.79 - 3.39) (0.79 - 3.41) (0.80 - 3.44)>$20 1.52 1.53 1.51 4.95∗∗∗ 4.97∗∗∗ 4.96∗∗∗ 1.58 1.58 1.59(0.85 - 2.73) (0.85 - 2.74) (0.85 - 2.71) (2.01 - 12.16) (2.02 - 12.25) (2.01 - 12.21) (0.77 - 3.26) (0.77 - 3.27) (0.77 - 3.28)School LevelElementary – – – – – – – – –Secondary 1.37 1.31 1.29 2.07∗∗ 1.99∗∗ 1.86∗ 1.58 1.50 1.41(0.84 - 2.22) (0.81 - 2.14) (0.81 - 2.05) (1.20 - 3.56) (1.11 - 3.56) (1.03 - 3.34) (0.92 - 2.72) (0.83 - 2.70) (0.79 - 2.52)Median Family 0.85∗ 0.88 0.88 0.66∗∗ 0.67∗∗ 0.71∗∗ 0.63∗∗∗ 0.66∗∗ 0.68∗∗Income† (0.69 - 1.04) (0.71 - 1.08) (0.73 - 1.06) (0.51 - 0.84) (0.52 - 0.87) (0.55 - 0.91) (0.50 - 0.80) (0.51 - 0.85) (0.54 - 0.87)N 570‡ 570‡ 570‡ 573‡ 573‡ 573‡ 575‡ 575‡ 575‡ICC 0.02 0.02∗ 0.01 0.01 0.02 0.03 0.01 0.03 0.03Each column reports a model with dietary intake as the dependent variable and objective proximity as independent variablesadjusted for gender, food insecurity, bringing lunch from home, acculturation, spending money and school median income.Coefficients are reported as odds ratios with 95% confidence intervals in parentheses.aDependent variables = 1 if consumed at least daily. bDistance to nearest outlet, reported in units of 100 metres.cBrought from home = 1 if a student reported bringing lunch daily; dReference level is food secure students†School-level variable constructed by the BC Ministry of Education; reported in $10,000 units‡Missing values were handled through complete case analysis as a sensitivity analysis to multiple imputation190


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items