Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2017 Mar 1.
Published in final edited form as: Appl Geogr. 2016 Mar;68:20–27. doi: 10.1016/j.apgeog.2016.01.004

Scale effects in food environment research: Implications from assessing socioeconomic dimensions of supermarket accessibility in an eight-county region of South Carolina

Timothy L Barnes a,b, Natalie Colabianchi c, James D Hibbert a, Dwayne E Porter d, Andrew B Lawson e, Angela D Liese a,*
PMCID: PMC4807632  NIHMSID: NIHMS753115  PMID: 27022204

Abstract

Choice of neighborhood scale affects associations between environmental attributes and health-related outcomes. This phenomenon, a part of the modifiable areal unit problem, has been described fully in geography but not as it relates to food environment research. Using two administrative-based geographic boundaries (census tracts and block groups), supermarket geographic measures (density, cumulative opportunity and distance to nearest) were created to examine differences by scale and associations between three common U.S. Census–based socioeconomic status (SES) characteristics (median household income, percentage of population living below poverty and percentage of population with at least a high school education) and a summary neighborhood SES z-score in an eight-county region of South Carolina. General linear mixed-models were used. Overall, both supermarket density and cumulative opportunity were higher when using census tract boundaries compared to block groups. In analytic models, higher median household income was significantly associated with lower neighborhood supermarket density and lower cumulative opportunity using either the census tract or block group boundaries, and neighborhood poverty was positively associated with supermarket density and cumulative opportunity. Both median household income and percent high school education were positively associated with distance to nearest supermarket using either boundary definition, whereas neighborhood poverty had an inverse association. Findings from this study support the premise that supermarket measures can differ by choice of geographic scale and can influence associations between measures. Researchers should consider the most appropriate geographic scale carefully when conducting food environment studies.

Keywords: Food environment, Geographic scale, Neighborhood boundaries, Socioeconomic characteristics, Supermarket

1. Introduction

Over the past decade, in efforts to combat food insecurity and the obesity epidemic, researchers and policymakers have been concerned with the influence of local food environments and disparities in food access (Larson et al., 2009). In investigating this issue, many studies have shown significant associations between the neighborhood food environment, diet and obesity (Caspi et al., 2012; Morland and Everson, 2009; Wang et al., 2007); however, many other have suggested that no significant relationship exists (Hattori et al., 2013). Inconsistencies between findings could perhaps be due to how neighborhood and local food environments have been defined (Liu et al., 2015).

Food environments have been characterized in many ways, with geographic information systems (GIS) being the most frequently used analytical tool (Caspi et al., 2012; Thornton et al., 2011; Charreire et al., 2010; Larson and Story, 2009; Larson et al., 2009). Using GIS, geographic-based measures of availability and accessibility of specific food retailers (e.g., supermarkets) and healthier or less-healthful foods have been created (Thornton et al., 2011; Kelly et al., 2011; Charreire et al., 2010; Larsen et al., 2008; Apparicio et al., 2007). Availability is typically defined as the presence or count of an attribute, e.g., supermarkets, in a defined area (Thornton et al., 2011; Charreire et al., 2010). Availability can also be represented as a density, e.g., the number of supermarkets per population or per geographic area (Thornton et al., 2011). Accessibility has been defined as ease of access to available supermarkets, taking factors such as travel distance, travel time and financial resources into consideration (Thornton et al., 2011). Accessibility has been extensively measured within the field of geography (Handy and Niemeier, 1997; Geurs and van Wee, 2004). The simplest measure of access—distance or proximity to the nearest supermarket—has been most commonly used in food environment research. However, accessibility has also been characterized by several other measures, including the cumulative opportunity index (Thornton et al., 2011; Van Meter et al., 2010, 2011).

Studies have pointed out several challenges when deriving geographic measures of the food environment (Fleischhacker et al., 2013; Liese et al., 2010, 2013). Problems with GIS include count, type and spatial inaccuracies when using secondary, commercial databases (Liese et al., 2013). These issues have generally led to under- or over-counting of food venues and misclassification of venue type and have risked mixed, diminished or overstated findings and effects of associations. To improve data quality and minimize measurement error, researchers are increasingly conducting primary data collection and field validation (Fleischhacker et al., 2013).

Less discussed is the choice of appropriate “neighborhood” boundaries or scale and the geographic context in which to operationalize food environment data (Liu et al., 2015; Fan et al., 2014; Larson et al., 2009). Many geographical boundaries have been used to define neighborhood food environments, ranging from egocentric buffer distances of 100 meters to 2.5 kilometers (~1.6 miles) around individual residential, worksite or school addresses or using administrative-based units, e.g., census tracts or block groups (Liu et al., 2015; Fan et al., 2014; Caspi et al., 2012; Charreire et al., 2010). However, there is no consistent methodology with which food environment researchers have agreed to construct geographic measures. Therefore, when comparing findings across studies, the measurements of neighborhood food exposure can vary depending on the geographical units selected. In geography, this effect is attributed to the modifiable areal unit problem (MAUP) (Flowerdew et al., 2008; Schuurman et al., 2007; Haynes et al., 2007; Openshaw, 1983; Fotheringham and Wong, 1991).

The MAUP is composed of two aspects, a scale effect and zonation effect, which through their tandem relationship can have significant influence on characterizing and modeling associations between the environment and health-related outcomes (Flowerdew et al., 2008; Haynes et al., 2007; Jackson, Davies & Leyland, 2010; Reijneveld, Verheij & de Bakker, 2000; Oliver & Hayes, 2007; Martikainen, Kauppinen & Valkonen, 2003; Ross, Tremblay & Graham, 2004; Oliver & Hayes, 2007). The scale effect causes analytical differences based on the size and number of geographic units used. Thus, associations will vary based on how refined and robust the measures are for these different geographic units (Oliver & Hayes, 2007; Parenteau & Sawada, 2011). Understanding the scale effect and the associated MAUP is particularly important for many geographic-related analyses (Kwan and Weber, 2008).

To the best of our knowledge, scale effects have not been well explored in food environment research, although one of the most probable contributors to the many mixed findings relating neighborhood food environments to diet and/or weight is the choice of geographic scale (Liu et al., 2015; Fan et al., 2014). To date, only one study has explicitly examined the effect of geographic scale on detecting relationships using a neighborhood food environment measure (Fan et al., 2014). In cross-sectional analyses comparing four different scales, Fan and colleagues found that the choice of neighborhood geographic scale did affect the estimated significance of the association between neighborhood food environments and individual obesity risk. Specifically, if the relevant neighborhood is defined as too large, (i.e., larger than a census tract for a convenience store or full-service restaurant) or too small (smaller than a census tract for limited-service restaurants), then the statistical relationship became insignificant (Fan et al., 2014). However, this study had several limitations, including the use of a secondary database to define the food environment and using only store count as the measurement criteria.

The objective of this study, therefore, was to evaluate the influence of two commonly used administrative-based geographic definitions (i.e., census tract and block group) on the relationships between GIS-derived supermarket measures and common U.S. Census–based socioeconomic characteristics using data from a 2010 South Carolina food environment study (Liese et al., 2010, 2013; Van Meter et al. 2010, 2011). Supermarkets were selected for this analysis because compared with other food retailers, they provide access to a greater quantity, variety and quality of food items (Franco et al., 2008; Block and Kouba, 2006) and have been used as a major criterion of the quality of the food environment in many studies (KriŸan et al., 2014; Larson et al., 2009; Bader et al., 2010; U.S. Department of Agriculture, 2009). The association of supermarket measures and neighborhood socioeconomic characteristics was used because previous studies have shown significant associations between these attributes (Larson et al., 2009; Walker, Keane and Burke, 2010; Beaulac, Kristjansson and Cummins, 2009; Sharkey and Horel, 2008; Powell et al., 2007a, 2007b; Moore and Diez-Roux, 2006; Zenk et al., 2005; Morland et al., 2002; Morland and Filomena, 2007).

2. Methods

2.1 Study Design

This analysis was part of a large methodological study of the food environment in South Carolina (Liese et al., 2010, 2013; Van Meter et al., 2010, 2011). The study area consisted of a contiguous geographical area and encompassed eight counties in the Midlands region of the state. The project’s efforts established a spatially and temporally verified database comprising 2,208 food outlets in South Carolina, including the global positioning system coordinates of all retail food outlets (Liese et al., 2010, 2013). This study was reviewed and deemed exempt by the Institutional Review Board of the University of South Carolina.

2.2 Neighborhood Geographic Boundaries

Two geographic units were selected for analysis. Data were based on both the census tract and block group administratively defined geographical boundaries obtained from the 2000 U.S. Census and the range of populations therein (U.S. Census Bureau, 2000). In our analyses, the selection of these units enabled us to examine changes in the relationship between supermarket accessibility measures and socioeconomic characteristics because of the differently sized and shaped spatial units. Illustrations of the geographic boundaries for the study region are shown in Figure 1.

Fig 1.

Fig 1

Open in a new tab

Neighborhood geographic boundaries

2.3 Neighborhood Socioeconomic Characteristics

Analyses utilized demographic, geographic and socioeconomic status (SES) data at the level of census tracts and block groups from the 2000 U.S. Census. Census tracts cover, on average, a population of 4,000 individuals. The block group is a smaller geographical unit for which the U.S. Census Bureau publishes data (U.S. Census Bureau, 2000). In total, 150 census tracts and 484 eligible block groups lie within the eight-county study region. Data utilized in the analyses included population size, area size (e.g., square miles), urban classification (i.e., urbanicity), percent minority population (percentage of non-white residents), median household income, percentage of the population living below the federally defined poverty level and percentage of the population (≥25 years of age) with at least a high school education. Individual variables were selected for the analyses because they represented the socioeconomic variables that typically serve as a proxy for neighborhood SES. However, in addition to individual measures, a summary score of area socioeconomic characteristics previously developed by Diez Roux and colleagues (2001) was also constructed by summing the z-scores of six variables (Diez-Roux et al., 2001a, 2001b; Unger et al., 2015). The variables were log median household income; percentage households with interest, dividend or rental income; log median value of housing units; percentage of persons 25 years or over with complete high school education; percentage of persons 25 years or over with complete college education; and percentage of persons in executive, managerial or professional specialty occupations. A higher score indicated better SES conditions compared to a lower (or negative) score.

2.4 Supermarket Availability and Accessibility Measures

We focused on supermarkets as a defining attribute of the food environment because compared to other types of food venues, supermarkets, supercenters and grocery stores typically provide access to healthy food in greater variety and of higher quality (Bilková and KriŸan, 2015; Krizan et al., 2014; Franco et al., 2008; Block & Kouba, 2006), and this classification was used previously by the U.S. Centers for Disease Control and Prevention in their 2013 State Indicator Report on Fruits and Vegetables (Centers for Disease Control and Prevention (CDC), 2013). Supermarkets were defined as any food outlet classified as a supermarket, supercenter, grocery store or warehouse club in our validated eight-county food environment database (Liese et al., 2010, 2013).

Availability and accessibility measures were determined by considering the count and distribution of supermarkets within the eight-county study region for both neighborhood geographic boundary measures. The availability and accessibility measures utilized in this analysis were supermarket density, cumulative opportunity and distance to nearest supermarket. Supermarket density is an availability measure and is defined as the number of food outlets adjusted for the geographic area of the specific geographic unit, i.e., census tract or block group (Van Meter et al., 2010, 2011; Thornton et al., 2011). Density is useful when explaining the distribution of features across areas (Thornton et al., 2011). The supermarket cumulative opportunity (also known as the cumulative opportunity index) was used as a measure of local accessibility (Guy, 1983; Van Meter et al., 2011). Similar to supermarket count or density, opportunity represents the number of food outlets within a certain area but gives less weight to food venues that are farther away. Thus, cumulative opportunity can be defined as Cp(s) where A is a predefined area within which the distances are measured and s represents the location points considered (Van Meter et al., 2010, 2011). The distance is measured to all outlets within the area A. For example, for an indexed location (i), the cumulative opportunity can be calculated using: Cpi=∑j∈A1dij. This measure provides cumulative evidence for accessibility at a spatial location and can be calculated for special cases such as Cp to the nearest outlet, Cp for a specified distance buffer and Cp total (calculated over the entire study region) (Van Meter et al., 2010, 2011). In our study, cumulative opportunity was determined for each census tract or block group’s population-weighted centroid (Thornton et al., 2011; Van Meter et al., 2011), giving less weight to food outlets that were farther away over the entire eight-county study region. Higher values of cumulative opportunity indicated higher supermarket access in that neighborhood boundary. The distance to nearest measure represented the closest food outlet determined by the shortest road and street network distance (i.e., drive time) to each census tract or block group’s population-weighted centroid (Thornton et al., 2011; Van Meter et al., 2011).

2.5 Statistical Methods

Descriptive statistics calculated included the mean, standard deviation (SD), median and minimum and maximum values for all demographic, geographic, socioeconomic and supermarket availability and accessibility measures (Table 1).

Table 1.

Descriptive characteristics of the eight-county South Carolina food environment study area by neighborhood boundaries

Census tract N=150 Block group N=484
Mean SD Median Min Max Mean SD Median Min Max
Neighborhood demographic and geographic characteristics
 Geographic area (square miles) 37.2 49.8 8.2 0.1 219 11.5 18.1 1.4 0.04 133
 Population size 4,210 2,151 3,872 346 11,655 1,290 994 1,103 23 11,300
 Population density 1,267 1,413 621 14.6 4,440 1,420 1,571 792 8 5,026
 Minority (%) 50.5 25.2 49.8 2.1 99.5 51.3 28.5 48.2 1.2 100
Neighborhood socioeconomic characteristics
 Median household income (U.S. dollars per year) 34,399 13,190 33,250 0 81,524 35,845 15,701 33,477 0 14,455
 Percentage of the population with at least a high school education (%) 76.0 12.2 73.1 49.9 100 76.1 13.5 75.0 31.2 100
 Poverty (%) 17.7 11.7 16.4 0 61.9 17.6 13.2 14.7 0 80.7
 Neighborhood SES z-score 0 1 −0.3 −1.3 4.0 0 1 −0.2 −2.1 4.5
Availability measures
 Supermarket density 0.23 0.45 0.01 0 1.6 0.04 0.13 0 0 0.54
Accessibility measures
 Supermarket cumulative opportunity 10.5 5.6 9.2 3.3 20.9 0.3 0.7 0 0 2.3
 Distance to nearest supermarket (miles) 2.8 2.5 1.7 0.1 9.2 2.3 2.2 1.5 0.1 8
Open in a new tab

SES, Socioeconomic

General linear models were utilized to examine the association between the four neighborhood socioeconomic variables (median household income, percentage of the population below the poverty level, percentage of the population with at least a high school education and neighborhood SES z-score) and the three supermarket measures (density, cumulative opportunity and distance to nearest) in separate models. Median household income and distance to nearest supermarket were both log transformed to account for skewed distributions. Additionally, the distributions of supermarket density and cumulative opportunity were also slightly skewed because of outliers and were Winsorized at the 95th percentile. After adjustments, no violations of the assumptions of independent observations, linearity, homoscedasticity or normality were further noted. All linear models were adjusted by percent minority population and urbanicity, assuming a normal distribution. Linear models for supermarket cumulative opportunity and distance to nearest were also adjusted for population. No violations of multicollinearity between the demographic and socioeconomic measures were found (all variance inflation factors ≤10).

3. Results

Descriptive characteristics by census tract (N=150) and block group (N=484) are displayed in Table 1. The median sizes of the census tract and block group geographic areas were 8.2 and 1.4 square miles, respectively. Additionally, as expected based on U.S. Census Bureau definitions, the median population size for census tracts was much larger (median=3,872 people) than for block groups (median=1,104 people). The corresponding median population densities and percent minorities are also reported, along with the means and standard deviations of all variables. Not shown are the percentages of census tracts and block groups classified as urban. In the study region, 48% of census tracts and 43.8% of block groups were classified as urban based on the 2000 U.S. Census definitions.

When examining the availability and accessibility measures by neighborhood geographic boundaries, the mean density of supermarkets was quite low. These values reflect the fact that many geographic units did not have a single supermarket present. However, 86.6% of census tracts and 41.3% of block groups contained a supermarket. More importantly, the mean and median density of supermarkets were higher when using census tracts compared to block groups. This pattern continued when examining supermarket cumulative opportunity measures. However, there was little difference in the mean and median distances to the nearest supermarket between the two units. The median distances to the closet supermarkets were 1.7 and 1.5 miles in census tracts and block groups, respectively. The neighborhood SES z-score ranged from −1.3 to 4 and −2.1 to 4.5 in census tracts and block groups, respectively.

In examining the association between socioeconomic characteristics and density of supermarkets, there were some significant associations observed in linear models (Table 2). First, as median household income increased, supermarket density decreased using both census tract and block group boundaries. This effect was stronger when using larger geographic boundaries, i.e., census tracts (CTs) compared to block groups (BGs) (BCT=−0.25, p-value=0.04 vs. BBG=−0.04, p-value=0.04). In addition, the model fit—indicated by R2—was much larger using census tract compared to block group boundaries (0.13 vs. 0.01, respectively). As the percent of the population below the poverty line increased, supermarket density increased significantly using census tract boundaries (BCT=0.72, p-value=0.05).

Table 2.

Regression analyses for the relationships between neighborhood socioeconomic characteristics and supermarket density and cumulative opportunity measures by neighborhood boundaries

Supermarket densitya Supermarket cumulative opportunityb
Census tract N=150 Block group N=484 Census tract N=150 Block group N=484
B (SE) p-value B (SE) p-value B (SE) p-value B (SE) p-value
Median household income (U.S. dollars per year)c −0.25 (0.12) 0.04 −0.04 (0.02) 0.04 −2.21 (0.76) <0.01 −1.41 (0.34) <0.01
 Population density -- -- -- -- 1.73 (0.20) <0.01 1.0 (0.12) <0.01
 Minority (%) −0.21 (0.17) 0.23 −0.02 (0.03) 0.45 −3.40 (1.07) <0.01 −2.0 (0.53) <0.01
 Urbanicity 0.32 (0.07) <0.01 0.02 (0.01) 0.05 4.74 (0.72) <0.01 0.6 (0.33) 0.09
Model R2 0.13 0.01 0.80 0.32
Percentage of population with at least a high school education (%) −0.40 (0.45) 0.37 −0.05 (0.06) 0.42 −0.67 (2.90) 0.82 −3.99 (1.32) <0.01
 Population density -- -- -- -- 1.72 (0.20) <0.01 1.09 (0.12) <0.01
 Minority (%) −0.11 (0.18) 0.54 −0.001 (0.03) 0.98 −2.0 (1.18) 0.09 −1.76 (0.56) <0.01
 Urbanicity 0.35 (0.10) <0.01 0.03 (0.02) 0.09 4.66 (0.82) <0.01 0.83 (0.38) 0.03
Model R2 0.13 0.01 0.77 0.31
Poverty (%) 0.72 (0.36) 0.04 0.07 (0.05) 0.22 8.23 (2.29) <0.01 4.80 (1.02) <0.01
 Population density -- -- -- -- 1.62 (0.20) <0.01 1.0 (0.12) <0.01
 Minority (%) −0.19 (0.17) 0.25 −0.01 (0.02) 0.83 −3.86 (1.05) <0.01 −1.93 (0.50) <0.01
 Urbanicity 0.29 (0.07) <0.01 0.02 (0.01) 0.11 4.86 (0.69) <0.01 0.41 (0.32) 0.20
Model R2 0.11 0.01 0.79 0.34
Neighborhood SES z-score 0.02 (0.06) 0.73 −0.01 (0.01) 0.40 0.97 (0.38) 0.01 −0.10 (0.05) 0.04
 Population density -- -- -- -- 1.61 (0.20) <0.01 0.08 (0.02) <0.01
 Minority (%) 0.04 (0.2) 0.84 −0.01 (0.03) 0.82 0.5 (1.30) 0.68 −0.21 (0.15) 0.18
 Urbanicity 0.27 (0.1) <0.01 0.03 (0.02) 0.09 3.79 (0.77) <0.01 0.01 (0.10) 0.91
Model R2 0.11 0.01 0.78 0.03
Open in a new tab

SES, Socioeconomic

a

Linear regression model adjusted for minority and urbanicity

b

Linear regression model adjusted for population density, minority and urbanicity

c

Median household income log transformed

Both median household income and percent high school education were significantly associated with supermarket cumulative opportunity (Table 2). Specifically, cumulative opportunity significantly decreased as median household income increased using the census tract boundaries (BCT=−2.21, p-value<0.01) and block group boundaries (BBG=−1.41, p-value<0.01) and decreased with increased high school education using the block group boundaries alone (BBG=−3.99, p-value<0.01). In addition, using block group boundaries, supermarket cumulative opportunity significantly increased with increasing percent below poverty (BBG=4.80, p-value<0.01). There was no significant association between supermarket cumulative opportunity and percent below the poverty line using the census tract boundary. When assessing the neighborhood SES z-score, there was a significant positive association using census tract boundaries (BCT=0.97, p-value=0.01) and a negative association using block group boundaries (BBG=−0.10, p-value=0.04). In all models, the model fit (R2) was better using census tract compared to block group boundaries. Additionally, the results suggest that based on the neighborhood characteristics included, a large percentage of the variation (30–80%) in models can be contributed to the cumulative opportunity measure, population density, percent minority, and urbanicity.

For distance to nearest supermarket, we found that as median household income or percent high school education increased, the distance to the nearest supermarket significantly increased using either census tract or block group boundaries (BCT=0.59, p-value<0.01 and BBG=0.40, p-value<0.01 for median household income; BCT=1.46, p-value=0.02 and BBG=0.88, p-value<0.01 for education) (Table 3). Distance to nearest supermarket decreased significantly with increased percent below the poverty level (BCT=−1.86, p-value<0.01 and BBG=−0.96, p-value<0.01). For all effects, the associations were stronger using the census tract compared to the block group boundaries.

Table 3.

Regression analyses for the relationships between neighborhood socioeconomic characteristics and distance to the nearest supermarket by neighborhood boundaries

Distance to nearest supermarket (miles)a
Census tract–based centroid Block group–based centroid
B (SE) p-value B (SE) p-value
Median household income (U.S. dollars per year)b 0.59 (0.17) <0.01 0.40 (0.08) <0.01
 Population density −0.40 (0.04) <0.01 −0.38 (0.02) <0.01
 Minority (%) 0.49 (0.24) 0.04 0.21 (0.12) 0.07
 Urbanicity −0.06 (0.16) 0.72 0.05 (0.08) 0.54
Model R2 0.64 0.58
Percentage of population with at least a high school education (%) 1.46 (0.63) 0.02 0.88 (0.29) <0.01
 Population density −0.41 (0.04) 0.01 −0.4 (0.02) <0.01
 Minority (%) 0.41 (0.26) 0.11 0.09 (0.12) 0.46
 Urbanicity −0.20 (0.18) 0.26 0.03 (0.09) 0.73
Model R2 0.61 0.57
Poverty (%) −1.86 (0.50) <0.01 −0.96 (0.25) <0.01
 Population density −0.38 (0.04) <0.01 −0.39 (0.02) <0.01
 Minority (%) 0.50 (0.23) 0.03 0.12 (0.12) 0.31
 Urbanicity −0.07 (0.15) 0.66 0.12 (0.08) 0.14
Model R2 0.63 0.57
Neighborhood SES z-score 0.11 (0.09) 0.20 0.07 (0.05) 0.10
 Population density −0.42 (0.04) <0.01 −0.40 (0.02) <0.01
 Minority (%) 0.31 (0.29) 0.29 0.05 (0.14) 0.72
 Urbanicity −0.09 (0.17) 0.59 0.08 (0.09) 0.35
 Model R2 0.60 0.56
Open in a new tab

SES, Socioeconomic

a

Linear regression model adjusted for population density, minority and urbanicity

b

Median household income log transformed

4. Discussion

Through our analyses, we explored whether using either of two common geographic boundaries would influence measures of supermarket accessibility and their relationship with U.S. Census–based neighborhood socioeconomic characteristics. We observed that supermarket density and cumulative opportunity were all higher when using census tract compared to block group definitions. However, distance to the nearest supermarket and distributions of supermarkets were quite similar between the two units. We also observed significant relationships of supermarket density and cumulative opportunity with socioeconomic characteristics, and larger effect sizes and improved model fit were observed when using census tract compared to block group boundaries. Additionally, significant associations between all three individual socioeconomic characteristics and distance to the nearest supermarket were observed. Specifically, median household income and percentage of the population with at least a high school education were both positively related (p-value<0.05) to supermarket distance, and poverty was inversely related (p-value<0.05) to distance to the nearest supermarket using either census tract– or block group–based centroids. Larger effect sizes were observed using census tract compared to block group boundaries. Interestingly, the neighborhood SES z-score, a summary index, was only significantly associated with the supermarket opportunity measure; however, the direction of this effect was different depending on whether census tracts or block group boundaries were utilized.

Findings from this study demonstrate that food environment measures and associations can differ based on choice of geographic scale used to define a neighborhood. These differences and changes in associations are most likely due to the impact of scale effects, which comprise part of the MAUP properties. To the best of our knowledge, only one study of the food environment has explicitly compared administratively defined neighborhood measures in the context of scale effects (Fan et al., 2014). However, determining the appropriate boundaries to operationalize geographic data should be an important concern, as many public health policies and initiatives are based on spatial measures of supermarkets and other food sources in local communities and neighborhoods (U.S. Department of Agriculture, 2009, 2011; Centers for Disease Control and Prevention, 2009, 2011a, 2011b).

In our study, higher availability and access to supermarkets were associated with low-income, less-educated, high-poverty neighborhoods. This is an uncommon finding in food environment research, as most studies have suggested that non-poor neighborhoods have a greater availability of supermarkets and healthy food options compared to poorer neighborhoods (Alwitt & Thomas, 1997; Morland et al., 2002; Giang et al., 2008; Glanz et al., 2007; Lewis et al., 2005; Powell, Chaloupka & Bao, 2007). However, the relationship between neighborhood demographics and socioeconomics has not been reported conclusively in the literature (Walker et al., 2010; Larson et al., 2009). Discrepancies between our results and those of some previous studies may be due to most published studies having been conducted in urban communities, with little or no variation in urbanicity. However, a study in a six-county rural region of Texas found that the most deprived neighborhoods with the greatest minority composition had better potential spatial access to food stores (Sharkey and Horel, 2008). In comparison, our study area consisted of neighborhood boundaries spanning both urban and rural geographic areas in South Carolina.

Strengths of our study include the use of multiple measures of the food environment—three different accessibility measures—to assess scale effects and examine associations with neighborhood socioeconomic characteristics. This study also utilized two of the most commonly used neighborhood boundaries: census tract and block group designations. In addition, all supermarket measures were based on a ground-truthed, field-validated database (Liese et al., 2010, 2013). Thus, the present study is unlikely to have been affected by neighborhood socioeconomic-related differential accuracy of the secondary data regarding the food environment, which has been shown to be present by several studies (Powell et al., 2011; Liese et al., 2013).

Our study, however, is not without limitations. First, our findings will not be generalizable to all study areas and populations. Many studies have characterized urban population areas exclusively (Moore & Diez-Roux, 2006; Zenk et al., 2005; Morland et al., 2002; Franco et al., 2008), which may differ from our study region consisting of both urban and rural characteristics. However, it is possible that our findings may be generalizable to study areas of similar makeup in the Southeastern region of the United States. Second, when constructing the accessibility measures, we did not consider food venues that were outside our eight-county region. Thus, there is a potential for some “edge effects” when conducting calculations for those population-weighted centroids close to the eight-county region boundaries. Last, we did not account for spatial autocorrelation, and thus we cannot assume that the values of observations are independent of one another. Correlation coefficients and ordinary least squares regressions (OLSs) assume that observations have been selected randomly to predict consequence. However, if the observations are spatially clustered in some way, the estimates obtained from the correlation coefficient or OLS estimator will be biased and overly precise. However, our modeling procedures did include a random effect. Moreover, our primary aim was to demonstrate that measures of supermarket availability and accessibility and the corresponding associations can differ based on scale effects. We did not attempt to evaluate supermarket accessibility measures and/or establish causality between the supermarket measures and the existing neighborhood characteristics.

5. Conclusions

The use of food environment measures has increased dramatically over the past decade and has contributed to many public health policies and initiatives (U.S. Department of Agriculture, 2009, 2011; Centers for Disease Control and Prevention, 2009, 2011a, 2011b). In this study, we explored how scale effects can alter the magnitude of associations between predictors and outcomes. Specifically, we found a small increase in the effect size of our associations between supermarket accessibility measures and individual socioeconomic characteristics when using census tract boundaries compared to block groups. In addition, we observed a change in the direction of our association using when using the two difference scales to assess the relationship between neighborhood SES z-score and the supermarket opportunity measure. However, despite our findings, we cannot definitively state whether one scale—census tract or block group—is the most appropriate for all study regions and analyses. Moving forward, researchers should consider carefully their choice of geographic scale. Sensitivity analyses by researchers should serve as an approach to determine what neighborhood boundaries, whether administrative-based (e.g., census tract and block group) or individual level–based, best fit the specific neighborhood experiences (Vallée and Shareck, 2014). Moreover, researchers must be aware that selecting a geographic scale that is either too large or too small in the context of communities and an individual’s actual daily routines and activities can alter measures of association.

Research Highlights.

  • Food environment studies rarely address scale effects

  • Scale affects analytic results due to the size and number of geographic units

  • Our findings demonstrate supermarket measures can differ by choice of scale

  • Scale also affect associations between supermarket and socioeconomic measures

  • Researchers should consider the scale used in food environment studies carefully

Acknowledgments

The authors would like to thank Denise M. Hodo, Kristopher Corwin and Dr. Andrey Bortsov for assisting in the fieldwork related to the establishment of the food outlet database within our eight-county study region in South Carolina. The study was supported by award number R21CA132133 from the National Cancer Institute. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Cancer Institute or the National Institutes of Health. T.L.B. conducted statistical analyses and developed the manuscript; D.E.P. provided geographic expertise; N.C. provided epidemiologic expertise; J.D.H. participated in acquisition of data, geocoded the data and conducted GIS-based data management; A.D.L. wrote the funding application, developed the study aims and assisted in preparing the manuscript. All authors reviewed, edited and approved the final manuscript.

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

References

  1. Alwitt LF, Thomas DD. Retail stores in poor urban neighborhoods. The Journal of Consumer Affairs. 1997;31:139–164. [Google Scholar]
  2. Apparicio P, Cloutier MS, Shearmur R. The case of Monreal’s missing food deserts: evaluation of accessibility to food supermarkets. International Journal of Health Geographic. 2007;12:4. doi: 10.1186/1476-072X-6-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Bader MD, Purciel M, Yousefzadeh P, Neckerman KM. Disparities in neighborhood food environments: implications of measurement strategies. Journal of Economic Geography. 2010;86:409–430. doi: 10.1111/j.1944-8287.2010.01084.x. [DOI] [PubMed] [Google Scholar]
  4. Beaulac J, Kristjansson E, Cummins S. A systematic review of food deserts, 1966–2007. Preventing Chronic Disease. 2009;6:A105. [PMC free article] [PubMed] [Google Scholar]
  5. Bilková B, KriŸan F. Mapping of grocery stores in Slovak countryside in context of food deserts. Acta Universitatis Agriculturae et silviculturae Mendelianae Brunensis. 2015;63:1633–1638. [Google Scholar]
  6. Block D, Kouba J. A comparison of the availability and affordability of a market basket in two communities in the Chicago area. Public Health Nutrition. 2006;9:837–845. doi: 10.1017/phn2005924. [DOI] [PubMed] [Google Scholar]
  7. Caspi CE, Sorensen G, Subramanian SV, Kawachi I. The local food environment and diet: a systematic review. Health & Place. 2012;18:1172–1187. doi: 10.1016/j.healthplace.2012.05.006. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Centers for Disease and Control and Prevention (CDC) Children's Food Environment State Indicator Report, 2011. Department of Health and Human Services; 2011a. [Google Scholar]
  9. Centers for Disease Control and Prevention (CDC) [Accessed February 10, 2014];State indicator report on fruits and vegetables, 2013. 2013 Available from: http://www.cdc.gov/nutrition/downloads/State-Indicator-Report-Fruits-Vegetables-2013.pdf.
  10. Centers for Disease Control and Prevention (CDC) [Accessed February 10, 2014];Communities Putting Prevention to Work. 2011b http://www.cdc.gov/CommunitiesPuttingPreventiontoWork.
  11. Charreire H, Casey R, Salze P, Simon C, Chaix B, Banos A, Badariotti D, Weber C, Oppert JM. Measuring the food environment using geographical information systems: a methodological review. Public Health Nutrition. 2010;13:1773–1785. doi: 10.1017/S1368980010000753. [DOI] [PubMed] [Google Scholar]
  12. Diez-Roux AV, Kiefe CI, Jacobs DR, Jr, Hann M, Jackson SA, Nieto FJ, Paton CC, Schulz R. Area characteristics and individual-level socioeconomic position indicators in three population-based epidemiologic studies. Annals of Epidemiology. 2001a;11:395–405. doi: 10.1016/s1047-2797(01)00221-6. [DOI] [PubMed] [Google Scholar]
  13. Diez-Roux AV, Merkin SS, Arnett D, Chambless L, Massing M, Nieto FJ, Sorlie P, Szklo M, Tyroler HA, Watson RL. Neighborhood of residence and incidence of coronary heart disease. New England Journal of Medicine. 2001;345:99–106. doi: 10.1056/NEJM200107123450205. [DOI] [PubMed] [Google Scholar]
  14. Fleischhacker SE, Evenson KR, Sharkey J, Pitts SB, Rodriguez DA. Validity of secondary retail food outlet data: a systematic review. American Journal of Preventive Medicine. 2013;45:462–473. doi: 10.1016/j.amepre.2013.06.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Flowerdew R, Manley DJ, Sabel CE. Neighbourhood effects on health: does it matter where you draw the boundaries? Social Science & Medicine. 2008;66:1241–1255. doi: 10.1016/j.socscimed.2007.11.042. [DOI] [PubMed] [Google Scholar]
  16. Fotheringham AS, Wong DWS. The modifiable areal unit problem in multivariate statistical analysis. Environment and Planning A. 1991;23:1025–1044. [Google Scholar]
  17. Franco M, Diez-Roux AV, Glass TA, Caballero B, Brancati FL. Neighborhood characteristics and availability of healthy foods in Baltimore. American Journal of Preventive Medicine. 2008;35:561–567. doi: 10.1016/j.amepre.2008.07.003. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Geurs KT, van Wee B. Accessibility evaluation of land-use and transport strategies: review and research directions. Journal of Transport Geography. 2004;12:127–140. [Google Scholar]
  19. Giang T, Karpyn A, Laurison HB, Hillier A, Perry RD. Closing the grocery gap in underserved communities: the creation of the Pennsylvania Fresh Food Financing Initiative. Journal of Public Health Management & Practice. 2008;14:272–279. doi: 10.1097/01.PHH.0000316486.57512.bf. [DOI] [PubMed] [Google Scholar]
  20. Glanz K, Sallis JF, Saelens BE, Frank LD. Nutrition Environment Measures Survey in stores (NEMS-S): development and evaluation. American Journal of Preventive Medicine. 2007;32:282–289. doi: 10.1016/j.amepre.2006.12.019. [DOI] [PubMed] [Google Scholar]
  21. Handy SL, Niemeier DA. Measuring accessibility: an exploration of issues and alternatives. Environment and Planning A. 1997;29:1175–1194. [Google Scholar]
  22. Guy CM. The assessment of access to local shopping opportunities: a comparison of accessibility measures. Environment and Planning B: Planning and Design. 1983;10:219–239. [Google Scholar]
  23. Haynes R, Daras K, Reading R, Jones A. Modifiable neighbourhood units, zone design and residents' perceptions. Health & Place. 2007;13:812–825. doi: 10.1016/j.healthplace.2007.01.002. [DOI] [PubMed] [Google Scholar]
  24. Jackson AL, Davies CA, Leyland AH. Do differences in the administrative structure of populations confound comparisons of geographic health inequalities? BMC Medical Research Methodology. 2010;10:74. doi: 10.1186/1471-2288-10-74. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. KriŸan F, Bilková K, Zubriczky G, Riška M, Barlík P. Identification and Mapping of Food Deserts in Rural Areas: A Case Study From Slovakia. Geographia Technica. 2014;9:54–59. [Google Scholar]
  26. Larsen K, Gilliland J. Mapping the evolution of ‘food deserts’ in Canadian city: supermarket accessibility in London, Ontario, 1961–2005. International Journal of Health Geographics. 2008;18:16. doi: 10.1186/1476-072X-7-16. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Larson N, Story M. A review of environmental influences on food choices. Annals of Behavioral Medicine. 2009;38:S56–S73. doi: 10.1007/s12160-009-9120-9. [DOI] [PubMed] [Google Scholar]
  28. Larson NI, Story MT, Nelson MC. Neighborhood environments: disparities in access to healthy foods in the U.S. American Journal of Preventive Medicine. 2009;36:74–81. doi: 10.1016/j.amepre.2008.09.025. [DOI] [PubMed] [Google Scholar]
  29. Lewis LB, Sloane DC, Nascimento LM, Diamant AL, Guinyard JJ, Yancey AK, Flynn G. African Americans' access to healthy food options in South Los Angeles restaurants. American Journal of Public Health. 2005;95:668–673. doi: 10.2105/AJPH.2004.050260. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Liese AD, Barnes TL, Lamichhane AP, Hibbert JD, Colabianchi N, Lawson AB. Characterizing the food retail environment: Impact of count, type and geospatial error in two secondary data sources. Journal of Nutrition Education and Behavior. 2013;45:435–442. doi: 10.1016/j.jneb.2013.01.021. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Liese AD, Colabianchi N, Lamichhane AP, Barnes TL, Hibbert JD, Porter DE, Nichols MD, Lawson AB. Validation of 3 food outlet databases: completeness and geospatial accuracy in rural and urban food environments. American Journal of Epidemiology. 2010;172:1324–1333. doi: 10.1093/aje/kwq292. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Liu JL, Han B, Cohen DA. Beyond neighborhood food environments: distance traveled to food establishments in 5 U.S. cities, 2009–2011. Preventing Chronic Disease. 2015;12:E126. doi: 10.5888/pcd12.150065. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Kelly B, Flood VM, Yeatman H. Measuring local food environments: An overview of available methods and measures. Health & Place. 2011;17:1284–1293. doi: 10.1016/j.healthplace.2011.08.014. [DOI] [PubMed] [Google Scholar]
  34. Martikainen P, Kauppinen TM, Valkonen T. Effects of the characteristics of neighbourhoods and the characteristics of people on cause specific mortality: a register based follow up study of 252, 000 men. Journal of Epidemiology & Community Health. 2003;57:210–217. doi: 10.1136/jech.57.3.210. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Moore LV, Diez-Roux AV. Associations of neighborhood characteristics with the location and type of food stores. American Journal of Public Health. 2006;96:325–331. doi: 10.2105/AJPH.2004.058040. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Morland K, Filomena S. Disparities in the availability of fruits and vegetables between racially segregated urban neighbourhoods. Public Health Nutrition. 2007;10:1481–1489. doi: 10.1017/S1368980007000079. [DOI] [PubMed] [Google Scholar]
  37. Morland K, Wing S, Diez-Roux AV, Poole C. Neighborhood characteristics associated with the location of food stores and food service places. American Journal of Preventive Medicine. 2002;22:23–29. doi: 10.1016/s0749-3797(01)00403-2. [DOI] [PubMed] [Google Scholar]
  38. Oliver LN, Hayes MV. Does choice of spatial unit matter for estimating small-area disparities in health and place effects in the Vancouver Census Metropolitan Area? Canadian Journal of Public Health. 2007;98:S27–S34. doi: 10.1007/BF03403724. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Openshaw S. The Modifiable Areal Unit Problem. Vol. 38. Norwich, England: Geobooks; 1983. [Google Scholar]
  40. Parenteau MP, Sawada MC. The modifiable areal unit problem (MAUP) in the relationship between exposure to NO2 and respiratory health. International Journal of Health Geographics. 2011;10:58. doi: 10.1186/1476-072X-10-58. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Powell LM, Chaloupka FJ, Bao Y. The availability of fast-food and full-service restaurants in the United States: associations with neighborhood characteristics. American Journal of Preventive Medicine. 2007a;33:S240–S245. doi: 10.1016/j.amepre.2007.07.005. [DOI] [PubMed] [Google Scholar]
  42. Powell LM, Slater S, Mirtcheva D, Bao Y, Chaloupka FJ. Food store availability and neighborhood characteristics in the United States. Preventive Medicine. 2007b;44:189–195. doi: 10.1016/j.ypmed.2006.08.008. [DOI] [PubMed] [Google Scholar]
  43. Powell LM, Han E, Zenk SN, Khan T, Quinn CM, Gibbs KP, Pugach O, Barker DC, Resnick EA, Myllyluoma J, Chaloupka FJ. Field validation of secondary commerical data sources on the retail food outlet environment in the US Health & Place. 2011;17:1122–1131. doi: 10.1016/j.healthplace.2011.05.010. [DOI] [PubMed] [Google Scholar]
  44. Reijneveld SA, Verheij RA, de Bakker DH. The impact of area deprivation on differences in health: does the choice of the geographical classification matter? Journal of Epidemiology & Community Health. 2000;54:306–313. doi: 10.1136/jech.54.4.306. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Ross NA, Tremblay SS, Graham K. Neighbourhood influences on health in Montreal, Canada. Social Science & Medicine. 2004;59:1485–1494. doi: 10.1016/j.socscimed.2004.01.016. [DOI] [PubMed] [Google Scholar]
  46. Sharkey JR, Horel S. Neighborhood socioeconomic deprivation and minority composition are associated with better potential spatial access to the ground-truthed food environment in a large rural area. Journal of Nutrition. 2008;138:620–627. doi: 10.1093/jn/138.3.620. [DOI] [PubMed] [Google Scholar]
  47. Schuurman N, Bell N, Dunn JR, Oliver L. Deprivation indices, population health and geography: an evaluation of the spatial effectiveness of indices at multiple scales. Journal of Urban Health. 2007;84:591–603. doi: 10.1007/s11524-007-9193-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Thornton LE, Pearce JR, Kavanagh AM. Using Geographic Information Systems (GIS) to assess the role of the built environment in influencing obesity: a glossary. International Journal of Behavioral Nutrition and Physical Activity. 2011;8:71. doi: 10.1186/1479-5868-8-71. [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. U.S. Census Bureau. US Census Summary File 1. U.S. Census Bureau; 2000. [Google Scholar]
  50. U.S. Department of Agriculture (USDA) Accessing to Affordable and Nutritious Food: Measuring and Understanding Food Deserts and Their Consequences: Report to Congress. United States Department of Agriculture Economic Research Service; 2009. AP–036. [Google Scholar]
  51. U.S. Department of Agriculture (USDA), Economic Research Service. USDA Food Desert Locator. U.S. Department of Agriculture, Economic Research Service; 2011. [Google Scholar]
  52. Unger E, Diez-Roux AV, Lloyd-Jones DM, Mujahid MS, Nettleton JA, Bertoni A, Badon SE, Ning H, Allen NB. Association of neighborhood characteristics with cardiovascular health in the Multi-Ethnic Study of Atherosclerosis. Circulation Cardiovascular Quality and Outcomes. 2014;7:524–531. doi: 10.1161/CIRCOUTCOMES.113.000698. [DOI] [PMC free article] [PubMed] [Google Scholar]
  53. Vallée J, Shareck M. RE: “Examination of how neighborhood definition influences measurements of youths’ access to tobacco retailers: A methodological note on spatial misclassification”. American Journal of Epidemiology. 2014;179:660–662. doi: 10.1093/aje/kwt436. [DOI] [PubMed] [Google Scholar]
  54. Van Meter EM, Lawson AB, Colabianchi N, Nichols M, Hibbert J, Porter DE, Liese AD. An evaluation of edge effects in nutritional accessibility and availability measures: a simulation study. International Journal of Health Geographics. 2010;9:40. doi: 10.1186/1476-072X-9-40. [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Van Meter EM, Lawson AB, Colabianchi N, Nichols M, Hibbert J, Porter D, Liese AD. Spatial accessibility and availability measures and statistical properties in the food environment. Spatial and Spatiotemporal Epidemiology. 2011;2:35–47. doi: 10.1016/j.sste.2010.09.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. Walker RE, Keane CR, Burke JG. Disparities and access to healthy food in the United States: A review of food deserts literature. Health & Place. 2010;16:876–884. doi: 10.1016/j.healthplace.2010.04.013. [DOI] [PubMed] [Google Scholar]
  57. Zenk SN, Schulz AJ, Israel BA, James SA, Bao S, Wilson ML. Neighborhood racial composition, neighborhood poverty, and the spatial accessibility of supermarkets in metropolitan Detroit. American Journal of Public Health. 2005;95:660–667. doi: 10.2105/AJPH.2004.042150. [DOI] [PMC free article] [PubMed] [Google Scholar]

RESOURCES