R E S E A R CH AR T I C L E

The interaction between individual, social andenvironmental factors and their influence ondietary intake among adults in Toronto

Daniel Liadsky | Brian Ceh

Dept. of Geography, Ryerson University,

350 Victoria Street, Toronto, Ontario,

Canada

Correspondence

Brian Ceh, Dept. of Geography, Ryerson

University, 350 Victoria Street, Toronto,

Ontario, Canada.

Email: bceh@ryerson.ca

AbstractHealth outcomes related to vegetable and fruit consumption are

widely recognized in the literature. This study investigates how die-

tary intake is influenced by individual, social, and environmental

factors in the Toronto Census Metropolitan Area. The analysis and

findings are based on data from the Canadian Community Health Sur-

vey which provides self-reported vegetable and fruit intake from

6,513 adults in 2009-2010. Food environment measures were con-

structed from commercial databases using kernel density estimates

and network drive times. Spatial and multivariable techniques were

used to determine the associations between diet, the food environ-

ment, and other health and socioeconomic factors. Particular

emphasis was given to understanding the interaction between the

food environment and socioeconomic position. Unexpectedly, super-

market density was found to have an inverse association with

vegetable and fruit intake. Interaction terms for individuals with low

income and reduced mobility produced different responses in men

and women, confirming that the influence of the food environment is

not uniform for all subgroups.

K E YWORD S

Food environment, Geographic Information Systems, socioeconomic

position, vegetable and fruit intake

1 | INTRODUCTION

Diet has long been linked to a number of chronic disease outcomes and it continues to be investigated as a risk factor

to human health (WHO & FAO, 2003). In Canada, poor diet accounts for the greatest disease burden, above tobacco

use and high body mass index (Institute for Health Metrics & Evaluation, 2013). Despite increasing consumer aware-

ness and ever expanding food choices, the prevalence of diet-related conditions points to a shortcoming of the con-

ventional food system. In urban settings, much attention has focused on whether the modern food environment

adequately and equitability supports the nutritional needs of the population. In recent years, there have been numer-

ous attempts to parameterize these environments and measure the accessibility and availability of healthy food

Transactions in GIS. 2017;1–20. wileyonlinelibrary.com/journal/tgis VC 2017 JohnWiley & Sons Ltd | 1

DOI: 10.1111/tgis.12276

(Shearer et al., 2015; Crawford, Jilcott Pitts, McGuirt, Keyserling, & Ammerman, 2014; Bader, Purciel, Yousefzadeh, &

Neckerman, 2010). Similarly, there have been many efforts to measure dietary behaviours and correlated health out-

comes using a variety of health and diet recall surveys (Fung et al., 2009; Dansinger, Gleason, Griffith, Selker, &

Schaefer, 2005, Hu, van Dam, & Liu, 2001). Yet, few studies explore the connections between these twomodes of inquiry.

Previous approaches to nutrition research assumed that individuals act independently of their environment or soci-

oeconomic background (Nestle et al., 1998). More recently, researchers have embraced ecological frameworks which

recognize that a number of contexts and environmental factors influence behavior and that these interactions can take

place along different dimensions (e.g. physical, sociocultural, or economic) and at different scales (e.g. individual, organi-

zational, or regional; Tremblay & Richard, 2014; Richard, Gauvin, & Raine, 2011). Moreover, the interaction between

dimensions and/or scales may matter more than any one factor by itself (Sallis, Owen, & Fisher, 2008). For example, the

presence alone of stores selling fruit and vegetables does not guarantee increased vegetable and fruit intake. One of the

most commonly cited conceptual frameworks for modeling eating patterns, proposed by Glanz, Sallis, Saelens, and Frank

(2005), considers the influences of social, environmental, and policy variables on individual behaviors. The model further

deconstructs the food environment into four subtypes: the community nutrition environment, organizational nutrition

environment, consumer nutrition environment, and information environment (Glanz et al., 2005).

To date, most food environment research has focused on the community nutrition environment which is characterized

by the location, type, and accessibility of food outlets. These characteristics can be measured by examining the geographic

distribution of outlets and their attributes or by recording consumer perceptions of those outlets (Health Canada, 2013).

Typically, Geographic Information Systems (GIS) are used to derive objective measures for the accessibility and availability

of food by calculating the proximity, density, and diversity of food outlets (Black, Moon, & Baird, 2014). These measures

have formed the basis for food desert investigations which traditionally focus on identifying neighborhoods where poor

geographical access to food outlets coincides with high levels of poverty (Cummins &Macintyre, 2002). While food deserts

are thought to be prevalent in the United States, evidence of this phenomenon elsewhere is inconclusive (Beaulac, Krist-

jansson, & Cummins, 2009). In Canada, investigations in Montreal and Edmonton found limited evidence of food deserts,

even when smaller independent supermarkets and fruit and vegetable stores were excluded from analysis (Apparicio,

Cloutier, & Shearmur, 2007; Smoyer-Tomic, 2006). Similarly, the most deprived neighborhoods in Toronto have been

found to contain twice the expected number of healthy food retailers (Polsky, Moineddin, Glazier, Dunn, & Booth, 2014).

A notable limitation to the food desert literature is that findings are rarely connected to dietary or health outcomes

(Cummins, 2007). Where studies have made this link, vegetable and fruit intake has commonly been used as an outcome

measure (Caspi, Sorensen, Subramanian, & Kawachi, 2012). Recent research has produced mixed evidence that there is

a measurable relationship between food environments and vegetable and fruit intake. Of 42 research papers reviewed

by Black, Moon, and Baird (2014), only a quarter of them show that there is an association between the density of

healthy food stores and improved dietary outcomes and a fifth of the studies show an association between proximity to

healthy food stores and vegetable and fruit intake. These results are likely due, in part, to a lack of standardized meas-

ures and methodologies which make comparison between studies challenging (Larson & Story, 2009; Charreire et al.,

2010). Some researchers have begun to advocate for better calibrated GIS techniques to measure food availability and

accessibility, such as kernel density estimates and network drive times instead of buffered densities and Euclidean dis-

tances, respectively (Caspi et al., 2012; Charreire et al., 2010). In Canada, the relationship between the community nutri-

tion environment and dietary outcomes has received less attention and with mixed results. An extensive study in

Waterloo, Ontario found significant relationships between objective food environment measures and body mass index

(BMI), but the relationships between these measures and diet quality were not significant (Minaker et al., 2013). By com-

parison, a study using a national retailer database found a relationship between the percentage of healthy food outlets

and vegetable and fruit intake, but only amongmen in major urban areas (Clary, Ramos, Shareck, & Kestens, 2015). How-

ever, it is not unusual for secondary source databases to be representative of fewer than 80% of retailers in an area

(Clary & Kestens, 2013). Indeed, both proprietary and government-sourced databases have been shown to be prone to

inconsistent classifications, geospatial inaccuracies, and a bias towards undercounting food outlets (Fleischhacker, Even-

son, Sharkey, Pitts, & Rodriguez, 2013; Forsyth, Lytle, & Riper, 2010; Liese et al., 2010).

2 | LIADSKY AND CEH

In addition, the influence of the food environment on dietary outcomes is thought to be mediated by a number of

other influences. For instance, neighborhood-level socioeconomic status has been implicated in dietary outcomes (Ball,

Crawford, & Mishra, 2006; Winkler, Turrell, & Patterson, 2006) and Macintyre’s (2007) deprivation amplification

hypothesis proposes that disadvantaged neighbourhoods produce poorer quality environments which, in turn, embel-

lish individual disadvantages. While it is not explicit in the model proposed by Glanz et al. (2005), Lytle (2009) suggests

that the food environment interacts with individual and social factors such that a restricted food environment (i.e. lack-

ing in food availability or accessibility) may limit the expression of individual and social influences on diet (Figure 1). A

corollary to this model is that individuals who are more socioeconomically constrained may be more susceptible to the

influence of the food environment on their diet (Health Canada, 2013). For example, individuals with lower income or

reduced mobility may be more reliant on their local food environment than those who have additional time and resour-

ces to travel further to do their shopping. It follows that constrained groups may show a greater range in diet quality

across food environments. This hypothesis, along with the other conceptual models reviewed here, remain largely

untested in the literature.

The purpose of this study is to characterize the relationship between the food environment and dietary behavior

in a Canadian context. Using the Toronto Census Metropolitan Area (CMA) as the study area, this research considers

the influence of individual, social, and environmental factors on vegetable and fruit intake using self-reported data

from the Canadian Community Health Survey (Statistics Canada, 2011). Furthermore, this study seeks to test the cor-

ollary of Lytle’s (2009) conceptual model that the influence of the food environment on vegetable and fruit intake is

more pronounced among individuals who are socioeconomically constrained.

2 | METHODS

2.1 | Study area

The Toronto CMA, with just over 6 million people and spanning 5,905 km2, anchors a broader and fast growing urban-

ized area (called the Greater Golden Horseshoe) approaching nine million people (Statistics Canada, 2015a, b). As with

FIGURE 1 The relationship between individual, social, and environmental factors that influence eating behaviors (Lytle,2009)

LIADSKY AND CEH | 3

many large metropolitan areas, the Toronto CMA is ethnically and culturally diverse with visible minorities representing

47% of the population (Statistics Canada, 2013).

2.2 | Canadian community health survey

Vegetable and fruit intake along with other health-related and sociodemographic data were obtained from the 2009-

2010 Canadian Community Health Survey (CCHS) Share File, a cross-sectional survey administered by Statistics Can-

ada (2011). Conducted over a two-year cycle, the CCHS collects population-level information on health determinants,

health status, and health system utilization from 130,000 respondents across Canada. The survey has a multistage

stratified design and sampling is balanced over the duration of the year to account for seasonal variation. DMTI’s Spa-

tial CanMap Postal Code Suite (DMTI, 2011) was used to link postal codes (similar to US ZIP codes, but at a much finer

resolution, these are typically assigned to a single block face between intersections or a single apartment building) with

geographic precision for 6,513 respondents in the Toronto CMA. Respondents who were under the age of 20 were

excluded from this study because their dietary habits were more likely to be determined by the household maintainer.

Similarly, respondents with missing records on fruit and vegetable consumption, who were pregnant, or who lived out-

side of an urban area1 were also excluded. In addition, to account for possible edge effects in the spatial analysis, a

5 km inner buffer from the Toronto CMA boundary (excluding Lake Ontario) was created and respondents living within

this buffer were excluded from the analysis. Survey weights provided by Statistics Canada to produce estimates for

the entire population were normalized by dividing the weight for each respondent included in the study by the average

weight of the sample size. These weights were used throughout the analysis.

Since 2001, the CCHS has used survey questions from the Centers for Disease Control’s Behavioral Risk Factor

Surveillance System to determine vegetable and fruit intake (BRFSS; P�erez, 2002). The BRFSS fruit and vegetable mod-

ule provides a derived variable for total daily frequency of consumption; but since it does not take quantity into

account, estimates cannot be compared to the recommendations in the Canada Food Guide (P�erez, 2002). However,

while the BRFSS fruit and vegetable module tends to underreport vegetable and fruit consumption, it has been vali-

dated against other vegetable and fruit intake assessment methods (Smith-Warner, Elmer, Fosdick, Tharp, & Randall,

1997) and in a Canadian context using similar, but slightly different questions (Traynor, Holowaty, Reid, & Gray-

Donald, 2006).

Based on standards followed in the literature, a set of individual-level health, demographic, socioeconomic, and

lifestyle covariates were selected from the CCHS that have been shown to be potential determinants of vegetable and

fruit consumption. These included: age; marital status; mental health (measured as sense of belonging to a local com-

munity); incidence of chronic diseases (diabetes, cardiovascular disease, high blood pressure, and cancer); presence of a

disability that requires assistance preparing meals or shopping for groceries; daily energy expenditure on physical activ-

ity; body mass index; smoking habits; immigration status; adjusted household income (adjusted by household size and

grouped by decile); and driving habits in the last 12 months (Azagba & Sharaf, 2011; Dubowitz et al., 2008; P�erez,

2002).

2.5 | Spatial analysis

Since spatial data can display positive spatial autocorrelation which can bias ordinary least squares regression (OLS)

models (Anselin, 2001), the global Moran’s I statistic was used to test whether vegetable and fruit consumption was

more similar among neighboring respondents than would be expected under a spatially random distribution. Moran’s I

tends towards 0 when there is no spatial autocorrelation and takes positive values of less than 1 when there is cluster-

ing. The Getis-Ord Gi* statistic was then used to detect local hot spots and cold spots across the study area and the

spatial patterns were visually inspected to assist with identifying explanatory variables.

As an outcome of the spatial autocorrelation analysis, additional ethnocultural variables were included as covari-

ates. These variables were derived by grouping responses to mother language and ethnicity questions with considera-

tion for regional similarities in diet. Respondents whose mother language was Italian, Greek, or Portuguese were

4 | LIADSKY AND CEH

grouped to create a Mediterranean variable (other Mediterranean countries of origin are not reported in the CCHS).

Similarly, respondents who identified as Chinese, Korean, Japanese, Southeast Asian, or Filipino were grouped to cre-

ate an East and Southeast Asian variable. South Asian, Black, and Latin American ethnicities were each included as sep-

arate variables.

2.6 | Food retailer databases

Food retailer locations were obtained from two proprietary commercial databases: the Centre for the Study of

Commercial Activity’s retail store database and DMTI’s Enhanced Points of Interest database, both of Canadian

retailers (CSCA, 2010; DMTI, 2010). Both databases are compiled annually and include a list of business names,

addresses, geographic coordinates, and standard industry classification (SIC) codes. Food stores, full-service

department stores, and warehouse food retailers that were located in the Toronto CMA in 2010 were extracted

using SIC codes. The inclusion of stores within the 5 km inner buffer and exclusion of respondents in this area

ensured that all stores within 5 km of each respondent were included in the analysis, minimizing possible edge

effects. The databases were joined to improve data completeness and duplicate data were removed. Because

combining databases increases the risk of overcounting (due to errors and closed outlets; Liese et al., 2010), a

combination of local knowledge, business name internet searches, and searches through Google Street View’s

historical imagery were used to validate retail locations, help determine whether they were operating in 2010,

and identify whether they sold vegetables and fruit. Stores that were estimated to be over 5,000 ft2 were classi-

fied as supermarkets in this study. This cut-off was selected as a way to include small- to medium-sized urban-

format supermarkets, such as Sobey’s Express and Urban Fresh stores, which are prevalent in Toronto’s densely

populated neighborhoods.

To model the food environment, kernel density estimates were created using store locations from the

merged retailer databases. Similar to the methodology employed by Thornton, Pearce, Macdonald, Lamb, and Ell-

away (2012), a cell size of 100 m and kernel sizes of 0.5, 1, 2.5, and 5 km were used to produce kernel density

estimate surfaces for all stores selling vegetables and fruit and for supermarkets alone. For each of the eight ker-

nel density estimate surfaces produced, the value at each CCHS respondent’s location was recorded. In addition,

an accessibility measure was calculated using the ArcGIS Business Analyst extension (Esri, 2014) to find the net-

work drive time for each respondent to the nearest store selling vegetables and fruit and to the nearest

supermarket.

2.7 | Statistical analysis

Descriptive and multivariable analyses were conducted to assess the association between vegetable and fruit con-

sumption and individual, social, and environmental factors. Because the magnitude of the effect on diet of some cova-

riates differed between men and women, data and analyses were stratified by gender. The mean difference in

vegetable and fruit consumption was compared using t-tests or ANOVA for each of the independent variables. To

reduce the influence of any skewed distribution on the mean, vegetable and fruit intake was truncated for men and

women at three multiples of the standard deviation.

Multiple linear regression using OLS estimators was used to model the relationship between vegetable and

fruit intake against the independent variables. The dependent variable (untruncated) was transformed using the

natural log to normalize the distribution which was highly skewed to the right. The model was then run with each

of the eight kernel density estimates and the network drive time variables to determine which representation of

the food environment contributed the most explanatory power to the model. Diagnostics were run on the final

model in GeoDa (Anselin, 2015), a spatial data analysis software package, to test for spatial dependence between

the error terms (spatial error) or between the dependent variable observations (spatial lag). The regression was

run again using the model specified by the spatial diagnostic tests to improve its fit. Finally, to test Lytle’s con-

ceptual model that there is an interaction between socioeconomic position and the food environment, two

LIADSKY AND CEH | 5

dummy variables were created from CCHS survey questions to represent a low-income group (respondents in

the bottom three deciles of adjusted household income) and a reduced mobility group (respondents who had not

driven a vehicle in the last 12 months). Interaction terms for each of these groups were created by taking the

product of the dummy variable and the food environment variable.

3 | RESULTS

Males respondents (n53,199) consumed 4.6 (SD52.3) vegetables and fruit per day on average and had a mean age

of 45.1 years (not shown). Female respondents (n53,314) consumed 5.2 (SD525) vegetables and fruit per day on

average and had a mean age of 46.1 years (not shown). The mean value of vegetable and fruit consumption for each

independent variable is given in Table 1.

TABLE 1 Descriptive statistics and mean daily vegetables and fruit consumption for each independent variable,stratified by sex

Male (n53,199) Female (n53,314)

Independent variable % Mean (SD) % Mean (SD)

Age group

20-24 10.8 4.5 (2.3)a 9.4 5.0 (2.6)a,b

25-34 19.7 4.4 (2.4)a 18.5 5.0 (2.3)a

35-44 21.1 4.5 (2.2)a 21.8 5.1 (2.3)a,b

45-54 19.6 4.4 (2.3)a 20.9 5.2 (2.5)a,b

55-64 16.0 4.6 (2.3)a 14.8 5.4 (2.7)b

651 12.8 5.1 (2.3)b 14.7 5.2 (2.3)b

Marital status

Married/Common-law 65.6 4.7 (2.3)a 59.5 5.4 (2.4)a

Divorced/Separated/Widowed 8.2 4.1 (2.4)a 16.6 5.0 (2.6)b

Single 26.0 4.5 (2.4)b 23.7 4.8 (2.5)b

Children present in household

Yes 31.6 4.5 (2.3)a 34.8 5.1 (2.4)a

No 68.4 4.6 (2.3)a 65.2 5.2 (2.5)a

Sense of belonging

Very strong 16.1 4.9 (2.4)a 15.3 5.8 (2.5)a

Somewhat strong 44.8 4.9 (2.4)a 49.9 5.3 (2.4)b

Somewhat weak 28.6 4.1 (2.1)b 22.5 4.8 (2.3)c

Very weak 10.0 4.1 (2.2)b 11.6 4.5 (2.7)c

Has chronic health condition

Yes 23.4 4.6 (2.3)a 23.2 5.0 (2.5)a

No 76.6 4.6 (2.3)a 76.8 5.2 (2.4)b

Needs help preparing meals or shopping

Yes 4.3 5.1 (2.8)a 8.4 4.8 (2.5)a

No 95.7 4.6 (2.3)b 91.6 5.2 (2.4)b

BMI

Underweight 1.5 3.2 (2.1)a 4.7 5.2 (2.6)a,b

Normal weight 43.3 4.5 (2.2)b 54.7 5.3 (2.4)a

Overweight 38.5 4.7 (2.3)b 24.4 5.2 (2.5)a

Obese 15.9 4.6 (2.5)b 13.2 4.7 (2.4)b

(Continues)

6 | LIADSKY AND CEH

In total, 1,011 retail locations that sold fruit and vegetables were identified between the CSCA (2010) and DMTI

(2010) datasets. Forty-nine percent were classified as supermarkets (Figure 2) and, of these, 82% were operated by

major chains (Table 2). The distribution of Toronto’s retail food stores by size and brand affiliation can be found in

TABLE 1 (Continued)

Male (n53,199) Female (n53,314)

Independent variable % Mean (SD) % Mean (SD)

Physical activity

Active 25.5 5.2 (2.4)a 16.4 5.9 (2.5)a

Moderately active 23.0 4.6 (2.2)b 22.7 5.5 (2.3)b

Inactive 50.6 4.3 (2.3)c 60.5 4.9 (2.4)c

Type of smoker

Daily 15.9 4.1 (2.2)a 9.4 4.5 (2.5)a

Occasional 6.5 4.4 (2.2)a,b 3.4 4.5 (2.0)a

Non-smoker 76.5 4.7 (2.3)b 86.6 5.3 (2.5)b

Highest Education Achieved

Post-secondary 66.2 4.6 (2.2)a,b 62.0 5.3 (2.5)a

Other post-secondary 9.0 4.8 (2.7)a 10.4 5.1 (2.4)a

High school 15.4 4.4 (2.6)a,b 17.0 5.0 (2.3)a

Some high school 6.1 4.3 (2.0)b 7.8 5.0 (2.3)a

Income

Higher (8-10 deciles) 25.0 4.8 (2.2)a 16.9 5.3 (2.5)a

Middle (4-7 deciles) 31.6 4.5 (2.4)b 31.2 5.3 (2.5)a

Lower (1-3 decile) 25.3 4.4 (2.3)b 30.1 5.1 (2.5)a

Immigrant

Yes 52.2 4.6 (2.3)a 56.0 5.3 (2.5)a

No 44.6 4.6 (2.3)a 41.1 5.0 (2.4)b

East and Southeast Asian

Yes 28.6 4.3 (2.3)a 28.2 5.4 (2.4)a

No 71.4 4.7 (2.3)b 71.8 5.1 (2.5)b

Black

Yes 6.1 4.8 (2.9)a 8.0 4.9 (2.6)a

No 90.4 4.6 (2.3)a 88.8 5.2 (2.5)b

Latin American

Yes 2.3 4.3 (2.3)a 3.4 4.6 (2.5)a

No 94.2 4.6 (2.3)a 93.4 5.2 (2.5)b

Mediterranean

Yes 10.2 5.0 (2.5)a 11.5 5.5 (2.4)a

No 89.8 4.5 (2.3)b 88.5 5.1 (2.5)b

South Asian

Yes 13.4 4.6 (2.3)a 13.1 5.4 (2.4)a

No 83.1 4.6 (2.3)a 83.7 5.1 (2.5)b

Drove vehicle in last 12 months .

Yes 84.2 4.6 (2.3)a 66.2 5.3 (2.5)a

No 13.8 4.3 (2.5)b 32.3 5.0 (2.3)b

Based on daily vegetable and fruit intake with outliers reassigned to three times the standard deviation.a,b,c Values with unlike superscript letters denote significant differences between group means (p<0.05).For ANOVA, Tukey’s HSD post hoc test was run for groups where assumptions of homogeneity of variance were met.When assumptions of homogeneity of variance were violated, the Brown-Forsythe test was used and the Games-Howellpost hoc test was run to determine significance between groups.

LIADSKY AND CEH | 7

Table 2. The average kernel density estimates for each respondent are given in Table 3 and the kernel density surfaces

are shown in Figure 3. Because most small stores were within close proximity to a supermarket, the coverage of the

kernel density surfaces for all stores and supermarkets were nearly identical (Table 3). At a kernel bandwidth of 5 km,

99.7% of respondents had a kernel density estimate greater than zero (Table 3). Kernel density estimates were also

correlated with population density (r50.259 to 0.583, p<0.001 from the smallest to largest bandwidths; Statistics

FIGURE 2 Locations of stores selling fruit and vegetables

TABLE 2 Summary of food retailers by store size and brand affiliation

Store size in m2 (ft2) Independent Major Chain Total

Other stores selling fruit and vegetables, under 465 (5,000) 499 7 506

0 to 464 (0 to 4,999) 290 3 293Not provided (estimated)* 209 4 213

Supermarkets, 465 (5,000) and over 92 413 505

465 to 1,393 (5,000 to 14,999) 40 32 721,394 to 2,322 (15,000 to 24,999) 31 62 932,323 to 4,645 (25,000 to 49,999) 15 158 1734,645 to 9,290 (50,000 to 99,999) 2 76 789,290 (100,000) and over 0 48 48Not provided (estimated)* 4 37 41

Total 591 420 1,011

*When a store size was not provided in the retailer database, the store was estimated to be less than 465 m2 or greaterthan 465 m2 using Google Street View.

8 | LIADSKY AND CEH

Canada, 2012). Respondents were located an average of 659 meters from the nearest store selling fruit and vegetables

(including supermarkets) and 817 m from the nearest supermarket.

Vegetable and fruit consumption for male and female respondents was positively, spatially autocorrelated across

the study area (men: I50.0108, p<0.002; women: I50.0079, p<0.001; not shown). A 5 km distance threshold had

the greatest statistical power. Due to provisions set by Statistics Canada to limit the risk of re-identifying CCHS

respondents, results from the hot spot analysis for each gender cannot be presented; however, a combined hot spot

map aggregated to hexagonal cells (cell sizes smaller than five have been suppressed) is shown in Figure 4 for illustra-

tive purposes. Local Gi* clusters of hot spots and cold spots were similar for both men and women. Hot spots can be

seen in Oakville, the east side of Brampton, Woodbridge, and southern Scarborough (Figure 4). A central cold spot

extends from the downtown core up to mid-town, from the northern part of Scarborough into Markham, and smaller

cold spots can be found in the west end of Brampton, Ajax, and Keswick (Figure 4). These local areas of spatial auto-

correlation show similarities to the distribution of ethnocultural communities in the Toronto CMA (Figure 5).

In the multiple linear regression model, the kernel density estimate for supermarkets with a bandwidth of 5 km

had the most explanatory power over the other kernel density surfaces. The two proximity variables (network drive

time to all stores selling fruit and vegetables and network drive time to supermarkets) did not make a significant contri-

bution to the model and were excluded from analysis.

Table 4 shows the initial OLS model, spatial regression results, and models with interaction terms for income,

mobility, and income and mobility combined for men and women. Since the dependent variable has been log trans-

formed, the estimated coefficient represents the percentage change in vegetable and fruit consumption with each unit

increase of that independent variable. For both genders, there was an inverse relationship between supermarket den-

sity and fruit and vegetable consumption (Table 4). There were also gender-specific differences with respect to varia-

bles that explain fruit and vegetable consumption. For example, highest education attainment and immigrant status

were significant for men, but for women the presence of children in the household, chronic health conditions, and

body mass index were significant (Table 4).

Of the ethnocultural variables analyzed, East and Southeast Asian and South Asian backgrounds were significant

for men, while Mediterranean backgrounds were significant for women (Table 4).

Spatial diagnostics on the OLS regression models for men and women indicated non-stationarity in the coefficients

(Breusch-Pagan test, p<0.05; not shown). For men, the residuals were not spatially autocorrelated and for women

there was spatial error dependence (Lagrange Multiplier test, p<0.05; not shown). Using a spatial error regression

model for the latter improved model fit (not shown), but did not entirely remove spatial dependence (maximum likeli-

hood estimation<0.05). The spatial error regression models also produced notable shifts in the coefficient values com-

pared to the OLS model suggesting model specification issues (not shown).

TABLE 3 Summary of average kernel density estimates

Store type (bandwidth)Average kerneldensity estimate mean (SD)

Percentage of respondents withkernel density estimate greater than 0

All stores (0.5 km) 1.38 (2.78) 47.3

All stores (1 km) 1.21 (1.65) 84.7

All stores (2.5 km) 0.98 (1.10) 98.7

All stores (5km) 0.84 (0.84) 99.7

Supermarkets (0.5 km) 0.52 (1.11) 47.3

Supermarkets (1 km) 0.50 (0.58) 84.7

Supermarkets (2.5 km) 0.39 (0.29) 98.7

Supermarkets (5 km) 0.34 (0.21) 99.7

Kernel density estimates are in units of stores per square kilometre.

LIADSKY AND CEH | 9

FIGURE 3 Kernel density estimate surfaces for all stores selling fruit and vegetables and supermarkets at 0.5, 2.5, and5 kmbandwidths. Red hues indicate themaximum density estimate, therefore the color scale is different for each cell.The 5 kmbandwidth kernel density estimate for supermarkets (framed in black) was selected for analysis

10 | LIADSKY AND CEH

For men, both the interaction terms for the low-income group (Table 4, Model 2) and reduced mobility group

(Table 4, Model 3) produced a more negative slope and a more positive intercept (Figure 6). These results suggest that

men in socioeconomically constrained groups are likely to consume more vegetables and fruit than unconstrained indi-

viduals in food environments with fewer supermarkets; but the steeper slope suggests that the influence of the food

environment on vegetable and fruit intake is stronger for the constrained group. Therefore, as supermarket density

increases, men in the constrained group are likely to consume similar or fewer vegetables and fruit than men in the

unconstrained group. For women, the near opposite was observed: the slope for the constrained group became slightly

positive and the intercept decreased (Figure 6). This suggests that women in socioeconomically constrained groups are

likely to consume fewer vegetables and fruit than unconstrained individuals in food environments with fewer super-

markets; however, as supermarket density increases, the fruit and vegetable intake of women in the constrained group

did not change considerably. Therefore, it is apparent from the shallower slope that women in the constrained group

are not as strongly influenced by the food environment as women in the unconstrained group. Combining the two

interaction terms did not lead to significant results for men or women (not shown).

4 | DISCUSSION

This study investigated individual, social, and environmental factors that explain vegetable and fruit consumption

among respondents to the 2009-2010 CCHS in Toronto, Canada (Statistics Canada, 2011). It was found that super-

market density had an inverse relationship with vegetable and fruit consumption, that proximity to food retailers did

not explain dietary intake, and that a variety of socioeconomic variables, including ethnocultural factors, mediated

FIGURE 4 Gi* clusters of high and low daily vegetable and fruit consumption (men andwomen) in the Toronto CMA

LIADSKY AND CEH | 11

these relationships. An interaction between socioeconomic position and the food environment was observed, although

to opposite effect in men compared to women.

This first finding is unexpected as other studies have largely uncovered either no relationship or a positive relation-

ship between supermarket density and vegetable and fruit intake (Black et al., 2014). Two exceptions, Powell et al.

(2009) and Zenk et al. (2009), found an inverse relationship between smaller grocery stores and vegetable and fruit

intake; however, this distinction is difficult to disambiguate in the Toronto CMA where there is a crowded continuum of

food stores from small, urban-format chain stores to larger superstores. In the present study, supermarket density was

strongly correlated with population density, so it follows that the density of other food retailers, such as convenience

stores, restaurants, and fast food outlets, also increase with population density. The availability of other food choices

could be confounding expected results, especially since prepared and convenience foods may be more attractive than

fruit and vegetables for some consumers; therefore, in future studies a relative measure that expresses vegetable and

fruit retailers as a ratio to stores that sell less healthy foodstuffs might provide clearer insight. Recent findings from Clary

et al. (2015) also came to this conclusion using CCHS data in a national-scale study. Furthermore, in-store food environ-

ments (as characterized by the relative cost and shelf space of vegetables and fruit) may need to be considered as many

retailers can be a source of both healthy and unhealthy food products. For example, a study in North Carolina found

that women who lived near superstores tended to weigh more than those who did not (Gustafson et al., 2011).

The second finding that proximity measures do not explain vegetable and fruit intake is consistent with some literature

(Williams, Ball, & Crawford, 2010; Pearce, Hiscock, Blakely, & Witten, 2008; Pearson, Russell, Campbell, & Barker, 2005)

and not surprising in Toronto’s retail environment where consumers have a lot of choice and the average distance to the

FIGURE 5 Census tracts with top decile concentrations of East and Southeast Asian,Mediterranean, and South Asianethnocultural groups

12 | LIADSKY AND CEH

TABLE4

Summaryofmultiplelin

earregressionresultsde

scribing

relationshipbe

twee

nlog-tran

sform

edda

ilyve

getablean

dfruitco

nsum

ptionan

dinde

penden

tvariab

les

stratified

bysex

Men

Women

Variable

Mode

l1

(Initialmode

l)Mode

l2(Low

inco

meinteraction)

Mode

l3(Red

uced

mobilityinteraction)

Mode

l4

(Initialmode

l)Mode

l5(Low

inco

meinteraction)

Model

6(Red

uce

dmobility

interaction)

Intercep

t1.26(0.11)***

1.12(0.11)***

1.24(0.11)***

1.32(0.09)***

1.39(0.10)***

1.35(0.09)***

Age

0.00(0.00)

0.00(0.00)

0.00(0.00)

0.00(0.00)***

0.00(0.00)***

0.00(0.00)***

Marital

Status

Married

/Common-Law

20.01(0.04)

20.02(0.04)

20.02(0.04)

0.10(0.03)**

0.10(0.03)**

0.10(0.03)**

Divorced

/Sep

arated

/Wid.

20.14(0.05)**

20.15(0.05)**

20.16(0.05)**

0.02(0.04)

0.02(0.04)

0.01(0.04)

Sing

le†

––

––

––

Childrenpresen

tin

househ

old

20.02(0.03)

20.04(0.03)

20.04(0.03)

20.06(0.03)**

20.06(0.03)**

20.05(0.03)*

SenseofBelong

ing

Verystrong

0.23(0.05)***

0.23(0.05)***

0.24(0.05)***

0.31(0.04)***

0.30(0.04)***

0.31(0.04)***

Somew

hatstrong

0.22(0.04)***

0.22(0.04)***

0.22(0.04)***

0.23(0.04)***

0.22(0.04)***

0.23(0.04)***

Somew

hatwea

k0.10(0.04)*

0.10(0.04)*

0.10(0.04)*

0.14(0.04)***

0.14(0.04)***

0.14(0.04)***

Verywea

k†–

––

––

–

Has

chroniche

alth

cond

ition

20.05(0.03)

20.06(0.03)

20.06(0.03)

20.07(0.03)*

20.07(0.03)*

20.06(0.03)*

Nee

dshe

lpwithshopp

ing

0.14(0.06)*

0.13(0.06)*

0.13(0.06)*

20.12(0.04)**

20.12(0.04)**

20.11(0.04)*

Body

massinde

x0.00(0.00)

0.00(0.00)

0.01(0.03)

20.01(0.00)***

20.01(0.00)***

20.01(0.00)***

Phy

sicalactivity

0.04(0.01)***

0.04(0.01)***

0.04(0.01)***

0.06(0.01)***

0.06(0.01)***

0.06(0.01)***

Typ

eofSm

oke

r

Daily

20.11(0.03)***

20.11(0.03)***

20.10(0.03)***

20.13(0.04)***

20.14(0.04)***

20.14(0.04)***

Occasiona

l20.01(0.05)

20.01(0.05)

20.01(0.05)

20.13(0.06)*

20.13(0.06)*

20.12(0.06)*

Non-sm

oke

r†–

––

––

–

Highe

stEdu

cationAch

ieve

d

Post-second

ary

20.05(0.04)

20.04(0.04)

20.05(0.04)

20.01(0.04)

20.02(0.04)

20.03(0.04)

Other

post-second

ary

20.14(0.06)*

20.13(0.06)*

20.15(0.06)*

0.02(0.05)

0.02(0.05)

0.02(0.05)

Highscho

ol

20.13(0.05)*

20.11(0.05)*

20.13(0.05)**

20.04(0.05)

20.05(0.05)

20.05(0.05)

Somehigh

scho

ol†

––

––

––

(Continues)

LIADSKY AND CEH | 13

TABLE4

(Continue

d)

Men

Women

Variable

Mode

l1

(Initialmode

l)Mode

l2(Low

inco

meinteraction)

Mode

l3(Red

uced

mobilityinteraction)

Mode

l4

(Initialmode

l)Mode

l5(Low

inco

meinteraction)

Model

6(Red

uce

dmobility

interaction)

Adjustedho

useh

old

inco

me

0.01(0.01)

0.02(0.01)**

0.01(0.01)

20.00(0.01)

20.01(0.00)

20.01(0.01)

Immigrant

0.08(0.03)***

0.08(0.03)**

0.08(0.03)**

0.03(0.03)

0.03(0.03)

0.04(0.03)

Ethno

cultural

group

Eastan

dSo

uthe

astAsian

20.16(0.04)***

20.15(0.04)***

20.16(0.04)***

0.04(0.04)

0.04(0.04)

0.04(0.04)

Black

20.20(0.05)

20.09(0.05)

20.10(0.05)

20.04(0.04)

20.04(0.04)

20.04(0.04)

LatinAmerican

20.14(0.08)

20.15(0.08)

20.14(0.08)

20.11(0.06)

20.11(0.06)

20.10(0.06)

Med

iterrane

an0.01(0.04)

0.02(0.04)

0.01(0.04)

0.12(0.04)***

0.12(0.04)***

0.12(0.04)***

SouthAsian

0.20(0.04)*

0.08(0.04)

0.09(0.04)*

0.03(0.04)

0.04(0.04)

0.04(0.04)

FoodEnv

ironm

ent

Supe

rmarke

tde

nsity

20.29(0.05)***

20.20(0.06)***

20.21(0.06)***

20.18(0.05)***

20.26(0.07)***

20.26(0.07)***

Socioecono

mic

restriction

Low

inco

megroup

–0.24(0.06)***

––

20.13(0.05)**

–Non-driver

group

––

0.17(0.08)**

––

20.15(0.05)**

Interactionterm

s

Low

inco

me3

supe

rmarke

tde

nsity

–20.38(0.12)***

––

0.26(0.11)**

–Non-driver

3supe

rmarke

tde

nsity

––

20.47(0.15)***

––

0.29(0.11)**

*p<0.05,**

p<0.01,***

p<0.001.

†Referen

cecatego

ry.

Allmode

lsusestan

dardized

survey

weigh

ts.

14 | LIADSKY AND CEH

nearest store is relatively small. Furthermore, it is reasonable to expect that the principal commercial establishment that one

shops for groceries is not necessarily the closest (Drewnowski, Aggarwal, Hurvitz, Monsivais, & Moudon, 2012). To address

this, researchers have been turning to approaches that incorporate analysis of GPS-tracked activities or travel survey data,

recognizing that most human activity is not restricted to predetermined geographies (Marquez & Guo, 2001; Sherman,

Spencer, Preisser, Gesler, & Arcury, 2005; Vall�ee, Cadot, Roustit, Parizot, & Chauvin, 2011; Sadler & Gilliland, 2015). In a

food environment context, these “activity spaces” have been found to be not only larger than residential neighborhoods,

but weakly associated with them, suggesting that food environment exposure from daily travel patterns may bemore impor-

tant than residential exposure (Zenk et al., 2011; Gustafson, Christian, Lewis, Moore, & Jilcott, 2013). However, while one’s

residential food environment may not resemble one’s activity space food environment, Kestens et al. (2012) showed that

individual-level and residential area characteristics are nevertheless good predictors of food environment exposure and

Zenk et al. (2011) have noted the difficulty of disambiguating a selection bias among individuals who seek out specific food

environments from those who are influenced by the environmental features in their “regular” activity spaces.

Though the relationship between ethnicity and vegetable and fruit consumption has previously been observed in

Canada (Quadir & Akhtar-Danesh, 2010), the results from the stratified analysis in this study suggest a possible interac-

tion between gender and ethnicity. From a public health perspective, it is important to understand how diet relates to

cultural norms and, with respect to gender, if there are within-cultural differences. It is also not known if ethnocultural

diets are mediated by the presence of ethnic markets or culturally-specific in-store products. In multicultural cities like

Toronto, further investigation along these lines could provide insight into the relationship between the food environ-

ment and the healthy immigrant effect, a phenomenon that describes the decline in the health of immigrants following

their arrival (Beiser, 2005; Newbold, 2006).

This study also found that vegetable and fruit intake remained practically unchanged among socioeconomically

constrained women across food environments. Since this restriction did not produce a response in both genders, this

study lends support to Lytle’s conceptual model for men, but not women. Because much of the food environment

FIGURE 6 Slope and intercepts of the interactions of income andmobilitywith the food environment for men (green)andwomen (blue). Due to data transformations, the units of the dependent variable are not easily interpretable and havebeen omitted

LIADSKY AND CEH | 15

literature does not stratify by gender, it is difficult to make comparisons with other studies; nevertheless, these results

suggest that gender-specific explanatory factors merit further study. Clary et al. (2015) made a similar observation, not-

ing that the effects of the food environment on diet are not universal for all population sub-groups. In this study, the

diagnostics for the OLS regression for women suggested model misspecification, possibly due to the omission of a crit-

ical variable. Perceptual and attitudinal differences between men and women towards diet may offer an explanation.

For instance, Emanuel, McCully, Gallagher, and Updegraff (2012) found that women reported greater perceived behav-

ioral control than men with respect to fruit and vegetable intake; for example, in the presence of junk food, women are

more confident that they can eat healthfully. In addition, women are shown to score higher than men on an eating-

related, self-determination index (SDI) and that eating-related SDI is significantly associated with dietary intake

(Leblanc, B�egin, Corneau, Dodin, & Lemieux, 2015). While these studies do not directly relate eating behaviors to the

food environment, they identify gender-specific differences that suggest women are more likely to be influenced by

individual factors and that their diets are not as strongly influenced by their surroundings.

For policy makers, Lytle’s (2009) model can also provide a framework for understanding the potential outcomes of

public health interventions targeted at individual, social, or environmental factors. Perhaps as a consequence of the

rise of food environment research, interventions that seek to enhance the food environment (or make it less restrictive)

have been increasingly proposed by public health advocates. For example, nine out of 10 strategies put forward by the

Centers for Disease Control and Prevention (2011) to encourage vegetable and fruit consumption focus on improving

access and availability. Similarly, Toronto’s public health unit has initiated a number of projects that attempt to increase

the availability of vegetables and fruit in public spaces (City of Toronto, n.d.). However, this study finds that there is an

inverse association between the availability of vegetables and fruit and their consumption. This does not suggest that a

new farmers’ market or healthy corner store will precipitate a decrease in vegetable and fruit intake; but, it is clear that

these types of interventions do not directly address these findings, particularly if exposure to unhealthy food products

is an underlying factor. Furthermore, given that this study found that women with reduced mobility tended to consume

fewer vegetables and fruit in neighborhoods with low supermarket density, a social intervention such as increasing

access to transportation in these neighborhoods may be a more effective measure, especially since this could yield

additional benefits to wellbeing, the economy, and the environment.

A number of limitations in this study should be recognized. Given the cross-sectional nature, causal relationships could

not be investigated. Therefore, the influence of consumers on the food environment or the food environment on migration

(i.e. market forces) cannot be discounted. Given the voluntary nature of the CCHS, the survey is not representative of all

sub-population groups, as evidenced by response rates for married and common-law couples and individuals with post-

secondary education that exceeded Toronto CMA population averages. In addition, the self-reported nature of the study is

susceptible to inflation or underestimation, especially for height and weight which are used to calculate BMI (Akhtar-

Danesh, Dehghan, Merchant, & Rainey, 2008). Similarly, the BRFSS food frequency questions are subject to reporting error

and tend to produce lower estimates of vegetable and fruit intake than more extensive dietary assessments (Smith-Warner,

et al., 1997). It is also possible that the small differences between the geocoded postal code and respondent address could

influence results, especially for respondents in less dense, suburban areas which tend to have longer block faces. As already

mentioned, the study makes assumptions that most shopping activity takes place within the retail food environment imme-

diately around the home; similarly, the study does not consider other effect modifiers of exposure such as modes of trans-

portation. While the food environment measures take into account geographical availability and accessibility, other factors

such as quality, price, and the amount of shelf space devoted to fruit and vegetables have not been investigated. Finally, by

limiting the analysis to the individual record level, the study avoids some issues encountered due to aggregation bias and

theModifiable Areal Unit Problem, but it does not take into consideration potential multilevel effects.

5 | CONCLUSIONS

This study identifies individual, social, and environmental factors that explain fruit and vegetable consumption in the

Toronto CMA. While the availability of food as measured by supermarket density was shown to influence dietary

16 | LIADSKY AND CEH

behaviors in men and women, this association was found to be inverse and unexpected. This study also lends support

to other findings that advocate for parameterizing the food environment using relative measures of healthy to unheal-

thy food outlets over absolute measures. In addition, the presence of gender-specific model differences suggests that

interaction effects require further attention in the food environment literature. Finally, while we acknowledge that con-

tinued methodological testing is needed to refine and standardize the tools and measures used in food environment

research, this study makes an attempt to introduce hypothesis testing using Lytle’s (2009) conceptual model as a start-

ing point. The mixed results between men and women suggest further investigation of Lytle’s model is warranted.

NOTE1 Urban status is determined in the 2009-2010 CCHS using Statistics Canada’s classification which defines urban areas ashaving a population greater than 1,000 and a population density of 400 or more people per square kilometre (StatisticsCanada, 2013).

REFERENCES

Akhtar-Danesh, N., Dehghan, M., Merchant, A. T., & Rainey, J. A. (2008). Validity of self-reported height and weight formeasuring prevalence of obesity. Open Medicine, 2(3): e83–88.

Anselin, L. (2001). Spatial econometrics. In B. H. Baltagi (Ed.), A companion to theoretical econometrics (pp. 310–330).Oxford, UK: Blackwell.

Anselin, L. (2015). GeoDa, (Version 1.6.6) [Software]. Tempe, AZ: Arizona State University.

Apparicio, P., Cloutier, M.-S., & Shearmur, R. (2007). The case of Montr�eal’s missing food deserts: Evaluation of accessibil-ity to food supermarkets. International Journal of Health Geographics, 6, 4.

Azagba, S., & Sharaf, M. F. (2011). Disparities in the frequency of fruit and vegetable consumption by socio-demographicand lifestyle characteristics in Canada. Nutrition Journal, 10, 118.

Bader, M. D., Purciel, M., Yousefzadeh, P., & Neckerman. K. M. (2010). Disparities in neighborood food environments:Implications of measurement strategies. Economic Geography, 86(4), 409–430.

Ball, K., Crawford, D., & Mishra, G. (2006). Socio-economic inequalities in women’s fruit and vegetable intakes: A multile-vel study of individual, social and environmental mediators. Public Health Nutrition, 9(5), 623–630.

Beaulac, J., Kristjansson, E., & Cummins, S. (2009). A systematic review of food deserts, 1966-2007. Preventing ChronicDisease, 6, 3.

Beiser, M. (2005). The health of immigrants and refugees in Canada. Canadian Journal of Public Health, 96(2 Suppl),S30–S44.

Black, C., Moon, G., & Baird, J. (2014). Dietary inequalities: What is the evidence for the effect of the neighbourhoodfood environment? Health & Place, 27, 229–242.

Caspi, C. E., Sorensen, G., Subramanian, S. V., & Kawachi, I. (2012). The local food environment and diet: A systematicreview. Health & Place, 18(5), 1172–1187.

Centre for the Study of Commercial Activity (CSCA). (2010). Greater Toronto Area Retail Store Database [Data file].Toronto, ON: Ryerson University.

Centers for Disease Control and Prevention (CDCP). (2011). Strategies to Prevent Obesity and Other Chronic Diseases: TheCDC Guide to Strategies to Increase the Consumption of Fruits and Vegetables. Atlanta, GA: U.S. Department of Healthand Human Services.

Charreire, H., Casey, R., Salze, P., Simon, C., Chaix, B., Banos, A., Badariotti, D., Weber. C., Oppert, J.-M. (2010). Meas-uring the food environment using geographical information systems: A methodological review. Public Health Nutrition,13(11), 1773–1785.

City of Toronto. (n.d.). Projects - Toronto Food Strategy - Nutrition & Food Access. Retrieved from http://www1.toronto.ca/wps/portal/contentonly?vgnextoid580ca044e17e32410VgnVCM10000071d60f89RCRD.

Clary, C. M., & Kestens, Y. (2013). Field validation of secondary data sources: A novel measure of representativity appliedto a Canadian food outlet database. International Journal of Behavioral Nutrition & Physical Activity, 10(1), 77.

Clary, C. M., Ramos, Y., Shareck, M., & Kestens, Y. (2015). Should we use absolute or relative measures when assessingfoodscape exposure in relation to fruit and vegetable intake? Evidence from a wide-scale Canadian study. PreventiveMedicine, 71, 83–87.

Crawford, T. W., Jilcott Pitts, S. B., McGuirt, J. T., Keyserling, T. C., & Ammerman, A. S. (2014). Conceptualizing and com-paring neighborhood and activity space measures for food environment research. Health & Place, 30, 215–225.

LIADSKY AND CEH | 17

Cummins, S. (2007). Neighbourhood food environment and diet: time for improved conceptual models? Preventive Medi-cine, 44(3), 196–197.

Cummins, S., & Macintyre, S. (2002). “Food deserts:” Evidence and assumption in health policy making. British MedicalJournal, 325(7361), 436–438.

Dansinger, M. L., Gleason, J., Griffith, J. L., Selker, H. P., & Schaefer, E. J. (2005). Comparison of the Atkins, Ornish,Weight Watchers, and Zone Diets for weight loss and heart disease risk reduction: A randomized trial. Journalof the American Medical Association, 293(1), 43–53.

DMTI Spatial Inc. (2010). Enhanced Points of Interest [Data file]. Markham, ON: DMTI Spatial Inc.

DMTI Spatial Inc. (2011). CanMap Multiple Enhanced Postal Code [Data file]. Markham, ON: DMTI Spatial Inc.

Drewnowski, A., Aggarwal, A., Hurvitz, P. M., Monsivais, P., & Moudon, A. V. (2012). Obesity and supermarket access:Proximity or price? American Journal of Public Health, 102(8), e74–e80.

Dubowitz, T., Heron, M., Bird, C. E., Lurie, N., Finch, B. K., Basurto-D�avila, R., Hale, L. Escarce, J. J. (2008). Neighborhoodsocioeconomic status and fruit and vegetable intake among Whites, Blacks, and Mexican-Americans in the UnitedStates. American Journal of Clinical Nutrition, 87(6), 1883–1891.

Emanuel, A. S., McCully, S. N., Gallagher, K. M., & Updegraff, J. A. (2012). Theory of planned behavior explains gender dif-ference in fruit and vegetable consumption. Appetite, 59(3), 693–697.

Fleischhacker, S. E., Evenson, K. R., Sharkey, J., Pitts, S. B. J., & Rodriguez, D. A. (2013). Validity of secondary retail foodoutlet data. American Journal of Preventive Medicine, 45(4), 462–473.

Forsyth, A., Lytle, L., & Riper, D. V. (2010). Finding food. Journal of Transport & Land Use, 3(1), 43–65.

Fung, T. T., Rexrode, K. M., Mantzoros, C. S., Manson, J. E., Willett, W. C., & Hu, F. B. (2009). Mediterranean diet andincidence of and mortality from coronary heart disease and stroke in women. Circulation, 119(8), 1093–1100.

Glanz, K., Sallis, J. F., Saelens, B. E., & Frank, L. D. (2005). Healthy nutrition environments: Concepts and measures. Ameri-can Journal of Health Promotion, 19(5), 330–333, ii.

Gustafson, A. A., Christian, J. W., Lewis, S., Moore, K., & Jilcott, S. (2013). Food venue choice, consumer food environ-ment, but not food venue availability within daily travel patterns are associated with dietary intake among adults,Lexington Kentucky 2011. Nutrition Journal, 12, 17.

Gustafson, A. A., Sharkey, J., Samuel-Hodge, C. D., Jones-Smith, J., Folds, M. C., Cai, J., & Ammerman, A. S. (2011).Perceived and objective measures of the food store environment and the association with weight and diet amonglow-income women in North Carolina. Public Health Nutrition, 14(6), 1032–1038.

Health Canada. (2013). Measuring the food environment in Canada. Retrieved from http://www.hcsc.gc.ca/fn-an/nutrition/pol/som-ex-sum-environ-eng.php.

Hu, F. B., van Dam, R. M., & Liu, S. (2001). Diet and risk of Type II diabetes: The role of types of fat and carbohydrate.Diabetologia, 44(7), 805–817.

Institute for Health Metrics and Evaluation (IHME). (2013). Global burden of disease study 2010 (GBD 2010) risk factorattribution results 1990-2010 - Country level. Seattle, WA: Institute for Health Metrics and Evaluation.

Kestens, Y., Lebel, A., Chaix, B., Clary, C., Daniel, M., Pampalon, R., . . ., & Subramanian, S.V. (2012). Association betweenactivity space exposure to food establishments and individual risk of overweight. PLoS journals.plos.org/plosone/arti-cle?id510.1371/journal.pone.004141

Larson, N., & Story, M. (2009). A review of environmental influences on food choices. Annals of Behavioral Medicine, 38(1), 56–73.

Leblanc, V., B�egin, C., Corneau, L., Dodin, S., & Lemieux, S. (2015). Gender differences in dietary intakes: What is the con-tribution of motivational variables? Journal of Human Nutrition & Dietetics, 28(1), 37–46.

Liese, A. D., Colabianchi, N., Lamichhane, A. P., Barnes, T. L., Hibbert, J. D., Porter, D. E., Nichols, M.D., & Lawson, A. B.(2010). Validation of 3 food outlet databases: Completeness and geospatial accuracy in rural and urban food environ-ments. American Journal of Epidemiology, 172(11), 1324–1333.

Lytle, L. A. (2009). Measuring the food environment: State of the science. American Journal of Preventive Medicine, 36(4Suppl), S134–S144.

MacIntyre, S. (2007). Deprivation amplification revisited; or, is it always true that poorer places have poorer access toresources for healthy diets and physical activity? International Journal of Behavioral Nutrition & Physical Activity, 4, 32.

Marquez, L., & Guo, J. (2001). Activity patterns and pollution exposure: A case study of Melbourne. European Journal ofTransport & Infrastructure Research, 1(4), 371–390.

Minaker, L. M., Raine, K. D., Wild, T. C., Nykiforuk, C. I. J., Thompson, M. E., & Frank, L. D. (2013). Objective food envi-ronments and health outcomes. American Journal of Preventive Medicine, 45(3), 289–296.

18 | LIADSKY AND CEH

Nestle, M., Wing, R., Birch, L., DiSogra, L., Drewnowski, A., Middleton, S.-G., . . ., & Economos, C. (1998). Behavioral andsocial influences on food choice. Nutrition Reviews, 56(5 Pt 2), S50–64; Discussion S64–74.

Newbold, K. B. (2006). Chronic conditions and the healthy immigrant effect: Evidence from Canadian immigrants. Journalof Ethnic & Migration Studies, 32, 765–785.

Pearce, J., Hiscock, R., Blakely, T., & Witten, K. (2008). The contextual effects of neighbourhood access to supermarketsand convenience stores on individual fruit and vegetable consumption. Journal of Epidemiology & Community Health,62(3), 198–201.

Pearson, T., Russell, J., Campbell, M. J., & Barker, M. E. (2005). Do “food deserts” influence fruit and vegetable consump-tion? A cross-sectional study. Appetite, 45(2), 195–197.

P�erez, C. E. (2002). Fruit and vegetable consumption. Health Reports, 13(3), 23–31.

Polsky, J. Y., Moineddin, R., Glazier, R. H., Dunn, J. R., & Booth, G. L. (2014). Foodscapes of southern Ontario: Neigh-bourhood deprivation and access to healthy and unhealthy food retail. Canadian Journal of Public Health, 105(5),e369–e375.

Powell, L. M., Zhao, Z., & Wang, Y. (2009). Food prices and fruit and vegetable consumption among young Americanadults. Health & Place, 15(4), 1064–1070.

Quadir, T., & Akhtar-Danesh, N. (2010). Fruit and vegetable intake in Canadian ethnic populations. Canadian Journal ofDietetic Practice & Research, 71(1), 11–16.

Richard, L., Gauvin, L., & Raine, K. (2011). Ecological models revisited: Their uses and evolution in health promotion overtwo decades. Annual Review of Public Health, 32, 307–326.

Sadler, R. C., & Gilliland, J. A. (2015). Comparing children’s GPS tracks with geospatial proxies for exposure to junk food.Spatial & Spatio-temporal Epidemiology, 14, 1, 55–61.

Sallis, J., Owen, N., & Fisher, E. (2008). Ecological models of health behavior. In K. Glanz, B. Rimer, & K. Viswanath (Eds.),Health behavior and health education: Theory, research, and practice (4th Ed., pp. 465–482). San Francisco, CA: Jossey-Bass.

Shearer, C., Rainham, D., Blanchard, C., Dummer, T., Lyons, R., & Kirk, S. (2015). Measuring food availability and accessi-bility among adolescents: Moving beyond the neighbourhood boundary. Social Science & Medicine, 133, 322–30.

Sherman, J. E., Spencer, J., Preisser, J. S., Gesler, W. M., & Arcury, T. A. (2005). A suite of methods for representing activ-ity space in a healthcare accessibility study. International Journal of Health Geographics, 4, 24.

Smith-Warner, S. A., Elmer, P. J., Fosdick, L., Tharp, T. M., & Randall, B. (1997). Reliability and comparability of three die-tary assessment methods for estimating fruit and vegetable intakes. Epidemiology, 8(2), 196–201.

Smoyer-Tomic, K. E. (2006). Food deserts in the Prairies? Supermarket accessibility and neighborhood need in Edmonton,Canada. Professional Geographer, 58(3), 307–326.

Statistics Canada. (2011). Canadian Community Health Survey - Annual Component (CCHS): User guide, 2010 and 2009-2010 Microdata files. Retrieved from http://www23.statcan.gc.ca/imdb/p2SV.pl?Function5getSurvey&SurvId550653&InstaId581424&SDDS53226.

Statistics Canada. (2012). Census of Canada, 2011: Toronto (code 535), Ontario. Retrieved from http://www12.statcan.gc.ca/census-recensement/2011/dp-pd/prof/index.cfm?Lang5E.

Statistics Canada. (2013). Immigration and Ethnocultural Diversity in Canada: National Household Survey, 2011 (Catalogueno. 99-010-X2011001). Retrieved from http://www12.statcan.gc.ca/nhs-enm/2011/as-sa/99-010-x/99-010-x2011001-eng.pdf.

Statistics Canada. (2015a). Population of Census Metropolitan Areas. Retrieved from http://www.statcan.gc.ca/tables-tab-leaux/sum-som/l01/cst01/demo05a-eng.htm.

Statistics Canada. (2015b). Population of Census Metropolitan Areas. Retrieved from http://www.statcan.gc.ca/pub/16-201-x/2013000/t003-eng.htm.

Thornton, L. E., Pearce, J. R., Macdonald, L., Lamb, K. E., & Ellaway, A. (2012). Does the choice of neighbourhood super-market access measure influence associations with individual-level fruit and vegetable consumption? A case studyfrom Glasgow. International Journal of Health Geographics, 11(1), 29.

Traynor, M. M., Holowaty, P. H., Reid, D. J., & Gray-Donald, K. (2006). Vegetable and fruit food frequency questionnaireserves as a proxy for quantified intake. Canadian Journal of Public Health, 97(4), 286–290.

Tremblay, M-C., & Richard, L. (2014). Complexity: A potential paradigm for a health promotion discipline. Health Promo-tion International, 29(2), 378–388.

Vall�ee, J., Cadot, E., Roustit, C., Parizot, I., & Chauvin, P. (2011). The role of daily mobility in mental health inequalities:The interactive influence of activity space and neighbourhood of residence on depression. Social Science & Medicine,73(8), 1133–1144.

LIADSKY AND CEH | 19

WHO, & FAO. (2003). Diet, nutrition and the prevention of chronic diseases: Report of a joint WHO/FAO expert consultation.Retrieved from http://www.who.int/nutrition/publications/obesity/WHO_TRS_916/en/.

Williams, L., Ball, K., & Crawford, D. (2010). Why do some socioeconomically disadvantaged women eat better thanothers? An investigation of the personal, social and environmental correlates of fruit and vegetable consumption.Appetite, 55(3), 441–446.

Winkler, E., Turrell, G., & Patterson, C. (2006). Does living in a disadvantaged area entail limited opportunities to purchasefresh fruit and vegetables in terms of price, availability, and variety? Findings from the Brisbane Food Study. Health &Place, 12(4), 741–748.

Zenk, S. N., Schulz, A. J., Kannan, S., Lachance, L. L., Mentz, G., & Ridella, W. (2009). Neighborhood retail food environ-ment and fruit and vegetable intake in a multiethnic urban population. American Journal of Health Promotion, 23(4),255–264.

Zenk, S. N., Schulz, A. J., Matthews, S. A., Odoms-Young, A., Wilbur, J., Wegrzyn, L., . . ., & Stokes, C. (2011). Activityspace environment and dietary and physical activity behaviors: A pilot study. Health & Place, 17(5), 1150–1161.

How to cite this article: Liadsky D, Ceh B. The interaction between individual, social and environmental factors

and their influence on dietary intake among adults in Toronto. Transactions in GIS. 2017;00:1–20. https://doi.

org/10.1111/tgis.12276

20 | LIADSKY AND CEH