Vous êtes sur la page 1sur 59

Comparative Analysis of Food Desert Mapping Definitions

Ryan Culligan




Project Advisor: Douglas J. Spieles
Department of Environmental Studies
i

Permission to make digital/hard copy of part or all of this work for personal or classroom
use is granted without fee provided that the copies are not made for profit or commercial
advantage, the copyright notice, the title of the work and its date appear, and notice is
given that coping is by permission of the author. To copy otherwise, to republish, to post
on a server, or redistribute to lists, requires prior specific permission of the author and/or
a fee. (Opinions expressed by the author do not necessarily reflect the official policy of
Denison University)








Copyright, Ryan Robert Culligan, 2014


ii

Acknowledgements

I would like to thank my advisor, Dr. Doug Spieles for guiding me through the entire
process of my research from start to finish. It is highly unlikely that I would know how
to use GIS had he not been there to guide me through my research process. I would also
like to thank Dr. Karl Sandin for taking the time out of his busy schedule to be my reader.
Lastly, thank you to the Environmental Studies Department for providing me with the
opportunity to conduct this research.

iii

Table of Contents

Abstract ............................................................................................................................................ 1
Project Overview .............................................................................................................................. 2
Introduction ..................................................................................................................................... 4
Methods ......................................................................................................................................... 22
Results: ........................................................................................................................................... 33
Discussion ...................................................................................................................................... 40
Bibliography ................................................................................................................................... 51







1


Abstract

Food Desert areas are areas with a perceived lack of access to healthy or
affordable food. Issues that have been noted to include mobility disadvantage, income
disadvantage, health and nutritional problems, social or cultural constraints, and lack of
food accessibility. The basic tenets food desert concept have existed for an extended
period of time, however issues of geographic access have only recently been explored.
This study uses GIS to evaluate a breadth of food desert definitions to examine whether
food desert associated issues have a geographic basis. The results of this study suggest
that geographic food deserts exhibit little to no association with demographic
observations such as percentage of households below poverty and percentage of
individuals with access to transportation by car.

2


Project Overview

This study uses census data from 2011 paired with food retailer location data
attained from various databases to identify geographic food desert areas. The goal of
the study was to determine whether poverty within an area had effects on access to
healthy foods at a grocery store level. While a correlation between poverty and lack of
access to foods is often assumed, there is scant evidence to support the notion, as many
studies are inconclusive about the matter (Walker, Keane, & Burke, 2010).
The geographic area of study for this project was the state of Ohio. This area was
chosen, because it was large enough to evaluate food desert definitions intended for
areas as large as the United States. Factors were examined on a census block level
because it was the highest resolution census demographic area. Food desert areas were
created in ArcMap 10.1, and then compared against each other where statistical
comparisons could be made.
Different geographic definitions exhibited different outcomes in their
demographic selections. Food desert definitions from the United States showed
increased poverty in food desert areas, while a definition from the United Kingdom
showed lower levels of poverty in food desert areas. Furthermore statistics could be
easily changed with subtle manipulation of the food desert categorization methodology
employed in ArcMap 10.1.
These results were contrary to the hypothesis that levels of poverty will be higher in
food desert areas. Geographic accessibility is by no means dissociated from the food
3

desert phenomenon. Instead, issues of geographic food accessibility are more likely
symptomatic and often pointed out as a factor of food desert causation within areas
where a plethora of food desert associated factors exist.

4

Introduction

The term food desert was coined by a group of individuals working for the Low
Income Project Team of Nutrition Taskforce in 1995. This task force was operating in
Britain, and their new term was meant to describe areas in which they worked where
foods are too expensive or unavailable to the people living within the areas, for reasons of
systematic built environment structure (Cummins & Macintyre, 1999). The food desert
term attracted both government and media attention and in turn, a new wave of food
desert related studies came about to describe the geographic phenomenon. This media
attention climaxed in the United States with Michelle Obamas pledge to eradicate food
deserts in the United States by 2017 (USDA, 2011). Unfortunately, many of the issues
surrounding food deserts are only loosely related to geographical access, and a majority
of the studies undertaken to address food desert associated problems make omissions
when it comes to the social and economic pressures that underlie food desert problems
in many regions (Donald, 2013).
The geographically delineated food desert definition proposed by Cummins and
Macintyre had broad parameters. Food Deserts within the area of Glasgow they
studied were agreed upon strictly defined as areas with little to no access to grocery
stores. Accessibility problems in Glasgow were speculated to have arisen in the 1980s
due to increased social demands which pressured stores for a greater variety of foods
available for purchase (Cummins & Macintyre, 1999). To meet the demands, grocery
stores with low overhead, due to their location outside the city center, fared well, and
those inside the cities did not, leading to the creation of food desert areas in Glasgow
(Chung & Myers, 1999). Cummins and Macintyres 1999 study took place in urban
5

Glasgow. This area was noted to be the center of the relocation and decentralization of
food retailers away from population centers. The relocation of grocery stores, made
sense for city dwellers with easy access to transportation; however, it left many people
unsatisfied and feeling that they had been left with inadequate access to grocery stores.
In wake of the relocation many social groups including the urban-dwelling elderly, poor,
and youth were left with little to no perceived access to quality fresh and nutritious foods
(Cummins & Macintyre, 1999). Food deserts quickly became a hot topic within the
British media, and were popularized in a manner that pushed for governmental action to
address the growing wealth discrepancies (Cummins & Macintyre, 1999). Despite the
publicity and widespread acceptance that the food desert concept received publically,
relatively few studies had been conducted to pragmatically research characteristics of a
food desert area.
The food access studies that did exist at the time of the definition of a food
desert did not focus on geographical access, but rather the ways that food is acquired by
people in different sociological circumstances. These studies have merits in terms of the
data they collected when it comes to the basic understanding of food access disadvantage
in deprived populations. However, they were not wholly conducive to undertaking swift
actions to fix the problems of geographical access on a large scale. Prior to the food
desert concepts inception, food access studies focused primarily on price research and
low-income or racial economic access to food products in the United States (Goodman,
1968). This research was conducted using store-to-store price comparisons between
predominately areas of different demographic makeups. In these models, nutritional
access was not taken into account (Sexton, 1971). Studies conducted on the previously
6

listed variables had dramatically different outcomes in comparison to one another with
conflicting results in different cities across the United States, and sometimes within the
same city (Sexton, 1971).
In the city of Philadelphia, Dixon and McLaughton (1968) found that food cost in
grocery stores had a positive correlation with neighborhood per capita incomes (Dixon &
McLaughlin, 1968). The same year, Goodman (1968) found that grocery store prices in
Philadelphia depended on store type, and that 92% of the low income population he
studied left their neighborhoods to shop at more price-competitive stores (Dixon &
McLaughlin, 1968) (Goodman, 1968). Discrepancies in the comparative outcomes
between studies were hugely prevalent in the study of food access. While on one hand
the food access imbalance was undeniable when encountered within an area of study, a
metric of access comparison between multiple areas remained elusive (Sexton, 1971).
By the 1990s the study of food access inequality had further diversified as a
field, and was no longer focused primarily on food access inequality by racial
demographics. It had extended to examine problems of food accessibility through a lens
of nutritional availability, systematic reinforcement of food purchasing behaviors, and
psychological conditions associated with living in food desert areas. This expansion in
the breadth of accessibility knowledge paired with technological revolutions gave rise to
new geographic schema to quantify and delineate food deserts on a geographic basis.
To further understanding of the complex issues surrounding geographic access,
many modern studies have found it beneficial to employ GIS-based approaches to
studying food access and food deserts. GIS; short for Geographic Information systems,
is a mapping software which enables placement and manipulation of spatially oriented
7

data. Food desert definitions which employ GIS as a tool simplify complex food
desert associated issues by defining idealized groups of quantifiable information. The
studies then use tools contained within the GIS software to quantify information across
geographic regions with demographic significance.
In their early years, many British food desert studies served as reports
conducted in conjunction with the British government to define areas that needed to have
the geographic food environment addressed. These studies took place slightly after food
deserts were popularized in the mid nineteen nineties and focused on establishing
methods that could geographically define food deserts. Research was essential because
in order for the British Government to enact practical solutions to alleviate the problem of
food deserts, it had to create ways to define food access-deprived urban areas. As
previously mentioned, the definition of a food desert was still in its infancy, and
afflicted regions in Britain had been observed by Government Nutrition Advocates, but
never formally defined (Macintyre & Cummins, 2002). The ambiguity in the definition
of food deserts, and the lack of proper evidence to confirm their existence, prompted
these preliminary studies to stick to geographically based conceptualizations of food
deserts focusing to establish whether the food desert problem was as prevalent on-the-
ground as it had come to be known by popular media sources.
In the wave of food desert studies after the British governments call to action,
different studies took different approaches to better understand food deserts.
Geographic methods of exploring food desert phenomena were limited in the objective
data they could utilize to classify food deserts. In most cases these methods involved
local or regional mapping with analysis of theoretical food availability, or in some cases
8

theoretical food availability compared to actual food availability at studied stores. For
the most part, nutritional quality and cultural factors which contribute to food choices
were thought to play a minor role in food accessibility issues and largely ignored.
Unsurprisingly, different methods for mapping food deserts returned different results
on a basis on the metrics they used to define areas that lack readily available or nutritious
food. These different outcomes did not preclude the existence of food deserts on any
scale, but issues associated with identification became more complex, because more often
than not, food desert areas did not align with any predefined geography.
One of these early British government-sponsored food desert studies was The
Location of food stores in urban areas: a case study in Glasgow. This study took a
mapping-based approach to studying food deserts in the city of Glasgow. This
mapping-based approach was conducted by comparing the locations of food outlets in
Glasgow, Scotland with the deprivation categories of the areas they were located in at the
level of postcode regions (Cummins & Macintyre, 1999). The postcode scores were
chosen because they were originally delineated at a scale that most resembled an
appropriate scale for local walking access to residents within a given area, and therefore
could be approximate an appropriate scale for shopping (Cummins & Macintyre, 1999).
Deprivation was calculated using the Carstairs Morriss Deprivation index (DEPCAT).
This index was created in 1991 is commonly used in studies throughout Scotland. The
DEPCAT deprivation score serve as measurements of social and economic deprivation
that take into account overcrowding, male unemployment, low social class, and car
ownership (Carstairs & Morris, 1991). Based on the aforementioned factors, the
DEPCAT5 scores range 1-7 with one being the lowest levels of deprivation and 7 being
9

the most deprived. The spatial distribution of the food stores identified through the
Public Register of Food Premises were mapped against the postcode deprivation areas
using MAPINFO and SPSS (Cummins & Macintyre, 1999). The mapping of food stores
against the DEPCAT5 scores in turn amalgamated a large number of the assumed
constituent components of food deserts, allowing the mapping of population density
and income against regional grocery store access. Well aligned with their null-
hypothesis, slightly more stores were found in the more deprived areas, and generally
speaking, food stores were evenly distributed throughout the studied area. The notable
omission from this food desert assessment was actual in-store food item availability.
While store type was taken into account, and from that data food availability was
inferred, in this study no data was compiled concerning on the ground food availability.
This lack of real world data could have contributed greatly to whether or not a place was
deemed a food desert on a basis that store designation type is not equivalent to store
quality of selection and item availability. This assumption was addressed in a follow up
study by Cummins and Macintyre in 2002, which compiled a list of food items that were
prospected for on a store-by-store basis to explore food item availability in more depth.
Like Cummins and Macintyre 1998, Cummins and Macintyre 2002 used the
DEPCAT deprivation score to determine socio-economic strata by postal code in
Glasgow. Cummins and Macintyre wished to further explore the concept of the food
desert as it concerned nutrition and nutrition access within an area, especially
considering that their previous attempt to define food deserts in Glasgow had not
returned relevant connections between deprivation and food store inaccessibility. The
2002 study food accessibility study was influenced heavily by several studies, which had
10

established that large food outlets are cheaper than their independently owned
alternatives, and healthy food diets are more expensive than their unhealthy
alternative (Wrigley, Warm, & Margetts, 2002). The 325 studied stores were separated
into a 10-fold classification system that would more thoroughly describe a stores niche
as it applied to the specialty of its products, as the alternative governmental store
classification system that only included 4 potential definitions (small, large, specialized,
and non-specialized). After stores were designated, 57 standard food items were studied
at each of the grocery stores selected from a previous study by Nelson et al., which had
compiled a modest but adequate diet for those living in the UK (Nelson, Mayer, &
Manley, 1993). After data was compiled, it was found that 51 of 57 items were available
in more than 90% of the 325 grocers that were studied. The best indicator for food prices
was not in fact location of the food retailer, but rather retailer designation between large
general grocery stores and small specialty stores.
Many food items did not statistically differ in price, however, and the foodstuffs
that did diverge from consistent pricing were low nutrient density, high energy density
and tended to be less expensive in areas with higher deprivation scores. Despite the
inequalities in food availability that had been colloquially observed in the greater
Glasgow area, Cummins and Macintyres 2002 study showed that when looking at the 57
chosen food items, there were no signs of food availability inequality which could be
directly indicative of a food desert. Despite the cost consistency, Cummins and
Macintyre noted that the small number of foods that were less expensive in lower income
areas including tea cakes, sausages, burgers, chocolate, and frozen French Fries, were all
foods that nutritionists recommend people consume less of. In this case, although the
11

price and availability of healthy foods did not differ, the price of low-nutritional value
foods did, pushing low-income consumers towards unhealthy food choices. For the
reason that there was not a substantial difference between food prices or availability
based on geographic or economic geographical strata, Cummins and Macintyre
concluded that although food deserts may exist to a degree in Glasgow and the greater
United Kingdom, they did not appear in a definable way within the context of their study,
because the costs of foods between neighborhoods did not differ enough to statistically or
practically cause healthy or unhealthy foods to become cost prohibitive (Cummins &
Macintyre, 2002). The Cummins and Macintyre 2002 studys outcome was much like
their first study, maybe because there was indeed a high degree of overlap between on-
the-ground food availability and assumed food item availability that was used in their
first study.
Chung and Meyers 1999 took a study approach which mirrored approaches taken
by both Cummins and Macintyre 1998 and 2002. This study looked at both store
location, food item availability, and store type along geographic income strata in
Minnesotas Hennepin and Ramsay counties. Unlike the studies conducted in Glasgow,
Chung and Myers statistically established that all grocery stores were more common in
more affluent areas, and that if stores were found in poorer areas, they were more likely
to be large or chain grocery stores. There was also a statistical difference between chain
and non-chain store food prices, with chain stores coming in at $16.92 less expensive per
standardized staple foods market basket purchase on average than non-chains. Like
other studies, Chung and Meyers concluded that there is little statistical evidence for
12

food deserts in the Twin Cities metropolitan area, as the prices of foods failed to differ
from region to region within Hennepin and Ramsay county food outlets.
The quantitative price and geographical access approach taken by Chung and
Meyers 1999 can be seen as a counterpoint to the study conducted by Smith et al. in 2010
looking at environmental influences on food access and shopping and dietary behaviors
in Minnesotans in the same area. Smith et al. first define food deserts in much the
same way as Chung and Meyers do in their study, using the Cummins and Macintyre
2002 definition, among others, to define a food deserts as poor urban areas where
residents cannot buy affordable, healthy food, and regions lacking close proximity to
food outlet (Cummins & Macintyre, 1999). After this definition, Smith et al. take a
drastic turn towards studying both quantitative and qualitative measurements of diet
quality and food access, specifically within two family shelters in the Minneapolis area.
These shelters lacked some or all means of food preparation and storage such as
refrigerators, freezers, ovens, and stoves, and by virtue of their limitations, and enforced
policies that disallow perishable food items. In turn, many food items which would be
considered to be of high nutritional value were negatively reinforced by shelter residents
in favor of low nutritional value, high energy density foods such as crackers, cookies, and
chips (Smith, Butterfass, & Richards, 2009).
Unlike many other studies that tend to focus on the straightforward delineation of
food deserts on a geographic basis, Smith et al. attempt to look at the relationships
between the proximity of low income shelters to food sources and between family
income, food prices, and their effects on dietary choices and health. Measurements taken
included height and weight data from participants as well as daily food consumption
13

records and firsthand accounts and food system personal complaints from the participants
of the study (Smith, Butterfass, & Richards, 2009). It was established that 80% of the
studied individuals were clinically overweight. Many people living in the areas felt they
had no easy access to food at a local level in the center of Minneapolis, and also limited
in their food options by price (Smith, Butterfass, & Richards, 2009). Unlike Chung and
Meyers study, this study defined food deserts using data directly gathered from people
who had reportedly lacked access to food while living in inner-city Minneapolis, and in
turn established the existence of a perceived food desert, in spite of a lack of empirical
data to establish higher food costs within an area.
The differences in the Chung and Meyers and Smith et al. studies as to whether or
not food deserts exist in the Minneapolis metropolitan area point towards weaknesses
in the different approaches that have been utilized to study food deserts. On one hand,
there are mapping-based approaches to studying food deserts which generally follow
the approach taken by Cummins and Macintyre 1999. These studies deal with mapping
access to food on a city or regional scale, by looking at food retail outlets, their prices for
individual food items, income levels within the studied area, and other factors particular
to the specific interests of the study authors. While these studies provide insightful data
for the practical application of solutions to food access problems, they rarely find
statistically relevant differences in food cost or accessibility as demonstrated in studies
conducted in Minneapolis in 1999 (Chung & Myers, 1999). A common alternative
means of understanding food deserts is to take a sociological approach to articulating
the problems faced by individuals with low income in areas without accessible foods, as
seen in Smith et al. 2009. In this mode of study, problems with food accessibility are
14

directly visible, but without a statistical basis for understanding the geographies
associated with the lack of food access encountered by said people, there is no viable
geographic definition. Both types of study are essential to the understanding of food
deserts, because without a sociological basis for geographic research, the geographic
research would not take place; however, the conclusions of the two study types are often
at odds with each other.
An attempt to visualize the variation in food cost and availability which
synthesizes both local deprivation, food availability, and cultural tastes can be seen in
Donkin et al. 2000. This research took place in a two square kilometer area of London
with a high Carstairs deprivation score. The British government spurred the studys
undertaking after they called for the development of quantitative methods for defining
and mapping access to healthy foods as part of a political agenda (Findlay & Sparks,
2002). Unlike other geographically- and socially-tied studies of food deserts, Donkin
et al. looked at factors that contributed to a healthy and balanced diet, such as the ethnic
makeup of the area, population food preferences, ethnic variations in shopping practices
(bulk or individual), and the monetary concerns of the people living in the deprived area.
From these concerns, one hundred twenty three food items were considered to be
candidates to be checked for local availability and seventy-one food items were identified
as candidates for price study. One hundred ninety-nine outlets within the two kilometer
range were surveyed, of which only one hundred sixty-six outlets stocked some of the
healthy foods identified by Donkin et al. From the data collected, GIS plots were
created with different symbology corresponding to differing costs and levels of
availability. Unlike other studies of the same kind, this study did not attempt to derive a
15

universal method for the mapping of food deserts, rather, it sought to create a modular
and scalable toolkit that could be readily applied to different geographies. A modular
tool-kit is beneficial to the study of food deserts because it is vitally important to have
replicability of study methods to help better understand the underlying causes of food
access disparities are attributable to lumped factors unique to each case studied location.
The subjective implications of living in a food desert are of equal importance
because unless they are understood, there is no means through which food accessibility
problems can be practically addressed. Studies such as the Cummins and Macintyre
1998, though important to the study of food deserts, use accessibility and deprivation
as the sole factors to predict diet inequalities. This approach is useful but ignores the
sociological aspects of food deserts, such as attitudes towards healthy eating, fear of
compulsiveness and going over budget, and grocery store familiarity and other
preferences. These seemingly secondary factors contribute to slow adoption of closer,
less expensive, or better-stocked grocery stores. This was the case with a store noted by
Wrigley et al. where after two years of grocery store existence in a food desert area,
only 50% of surveyed shoppers had changed their purchasing habits to reflect the
presence of the store (Wrigley et al. 2004). Habits as examined by Wrigley et al. suggest
that issues of poverty are paramount to the study of food deserts. Money is among the
primary influences food purchases, and unless people are incentivized to make healthy
food purchases from an economic standpoint as for prospects of better health, purchasing
habits are unlikely to change (Wrigley, Warm, & Margetts, 2002).
In the wake of the preliminary studies done in the late 90s and early 2000s,
food deserts have become hot topic issues within American society, and a priority for
16

the United States Department of Agriculture (USDA) (Walker et al 2010). The USDA
was first tasked with monitoring and researching the prevalence and causes in 2008 with
the signing of The Food, Conservation, and Energy Act of 2008 (110th Congress, 2008).
With this new responsibility on the part of the USDA came the first nationwide food
accessibility studies undertaken by a country as large as the United States. The USDA
had to wade through many of the inconsistent means of study to define a food desert at a
nationwide scale using a body of research that was not widely meant to scale beyond a
city level, which involved a literature review process that was taken by Walker et al.
2010. The review of existing literature gave many insights into study types that had been
previously conducted in the United States, and what areas of food desert research
needed to be drastically improved. While studies within the United States had thoroughly
explored the concept of geographic access to foods, the real life implications of living in
a food desert had been under-explored by comparison (Walker, Keane, & Burke,
2010).
The imbalance in successful delineation of food desert areas has implications in
the practical changes that the USDA has taken to incentivize the removal of food
deserts in the United States. Based upon the data collected by the USDA, the Obama
administration created the Healthy Food Financing Initiative, which was meant to
eradicate food deserts nationwide by 2017 (Donald, 2013). Part of the strategy to
accomplish this feat is to address built environment problems by partnering with pre-
existing national and regional grocery chains to bring healthier foods to affected areas
(Donald, 2013). In turn, many globalized companies currently plan on entering or re-
entering perceived food deserts in with small-scale box format stores (Donald, 2013).
17

However, there are many critics of this strategy, as the approach propagates the existence
of grocery store chains and approaches the food desert problem in terms of food item
economic accessibility and food regulations and policy.
Although it can be argued that chain stores can proliferate to new regions faster
than independent grocers because of their pre-existing distribution networks, and thereby
address food desert-associated problems on a faster timeline, there are many reasons
why this may not be the case. Many grocery store chains are said to propagate the
economic divides that contribute to the formation of food deserts through practices
involving low-wages, anti-union sentiments and the intense cost-pressures they place on
local competing businesses (Donald, 2013). If these cost-cutting factors are present, they
propagate cost reduction strategies on multiple levels, which can arguably further bisect
the food market between high-end and low-cost food markets. This split in retail markets
is important because the local demands in low income areas may not be in line with
healthy foods, and if the stores follow a hybridized model of retail globalization and
tailor their stocks to local demands, they could propagate the consumption of low
nutritional value foods (Donald, 2013) (Wrigley, Warm, & Margetts, 2002).
Low nutritional value foods with high energy density are less expensive on
average than nutritious foods for reasons including government food policies (Donald,
2013) (Drewnowski et al. 2005). Although energy density and food cost are negatively
related to one another, whether a healthier diet costs more to purchase is a point of
scientific contention with many conflicting reports (Drewnowski et al. 2005). The
division between research outcomes in food price studies has contributed to the choice to
promote nutritional education over the arguably more effective, but much more complex
18

and costly rehashing of United States food policy to promote purchasing habits involving
foods that are of high nutritional value (Drewnowsky et al. 2004). Food policy is left
largely unaddressed in the Healthy Food Financing Initiative when it comes to
subsidization, but not in food taxation efforts. As a means to improve the diets in the
United States, several efforts have been made to raise taxes on unhealthy foods
(Drewnowski et al. 2005). However, these actions have been received poorly, because
rather than address issues of economic inability to purchase healthier foods, this raises
general food costs, effectively penalizing those who cannot afford more nutritious foods
(Drewnowski et al. 2004).
A typical approach to food desert studies is to use one-time experimental data
to determine whether there are correlative relationships between different groupings of
people living in a small city level geographic area (Cummins & Macintyre, 1999)
(Donkin, Dowler, Stevensonn, & Turner, 2000). This approach is critical to the
preliminary study of food deserts and is often inconclusive, but also often serves as a
base on which to enact policy. Follow up studies are rarely conducted in food desert
literature. Very rarely, do researchers go back into an area after actions have been made
to address food desert associated problems. One of the few studies that has assessed
before and after outside intervention in a food desert area is Cummins et al. 2005.
Like other studies by Cummins et al, the experiment studies the greater Glasgow area
using DEPCAT scoring to determine social deprivation, but unlike his other studies, the
2005 study deals with survey data rather than readily available government census data.
Previous studies by Cummins and Macintyre studied food item availability and
proximity to grocery stores as it related to social deprivation, however, the 2005 study
19

used mailed surveys to conduct a psychological inquiry of individual health before and
after large scale food retailers moved into deprived areas of Glasgow (Cummins et al.
2005). The pre- and post-intervention surveys were sent to individuals living in the
studied areas and asked questions that pertained to vegetable and fruit consumption, as
well as psychological health. It was hypothesized that improving geographical access to
foods would positively impact consumption of the healthy foods, and negatively impact
consumption of more shelf stable unhealthy foods. Although one study is hardly enough
to give a decisive answer to whether food desert solutions seeking to improve geographic
access are effective, Cummins et al. 2005 would suggest that this type of solution did not
improve diet quality for people in the intervention regions (Cummins et al. 2005). The
only statistical difference in survey results before and after intervention was a small
improvement in psychological health for those immediately involved in the food desert
intervention projects (Cummins et al 2005).
Cummins et al. 2005 did not come up with any relevant differences in food habits
pre- and post-food desert intervention. Regardless of this outcome, the United States
government has adopted much the same strategies in dealing with addressing food
desert problems, when it partnered with nationwide food outlets to increase availability
(Donald, 2013). The adoption of the built-environment approach to alleviating food
desert concerns could be seen as a warning sign that the governments approach may
help to remove food deserts; however, the government is also pushing food education,
and little to no research has been conducted to determine how built environment and
education approaches work in tandem to alleviate food desert pressures. It is of key
20

importance that food deserts are approached from both empirical and subjective angles,
as they can be defined equally well through either means.
An approach that holds promise to the study of food deserts is concept
mapping. This research focuses on making subjective data measures that are of value to
the study of food deserts understandable through programs meant to visually map
relevant yet discreet contributing factors (Walker, Keane, & Burke, 2010). This concept
mapping process is not meant primarily to test previously existing hypotheses, but rather
to serve as an exploratory tool to develop a more complex understanding of the processes
and practices that contribute to the observed food desert phenomena (Jackson & Trochim,
2002). Concept mapping analysis doesnt necessarily show a geographic base to the
food desert problem, but can nonetheless help to assign statistical relevance to
subjective data as well as to model interconnected issues in food supply and demand web.
When combined with geographic mapping, concept-mapping methods of models of food
deserts can provide the information that has been commonly overlooked in the food
desert literature used to inform many policy decisions in recent history (Donald, 2013).
The existence of food deserts has been acknowledged by a multitude of people
dwelling in what they call food deserts, and authorities who have prioritized the
irradiation of the food desert problem. Despite the acknowledgement of food
deserts, practical solutions to food desert associated problems and a tendency to
unilaterally address the problems compound the ill effects of the food deserts.
Although each method of study to understand individual aspects of food desert-
associated problems has its merits, the umbrella structure of the policies to address food
desert issues is typically based on a model of geographic access (Donald, 2013). When
21

the United States government pledged to eradicate food deserts by 2017 in the United
States, it became critically important to define what a food desert was so action could
be taken to address the issue (110th Congress, 2008) (Donald, 2013). Without objective
criteria to judge effectiveness, the effort to eradicate food deserts cannot be successful.
This research projects seeks to implement multiple food desert research
definitions in GIS. As previously mentioned, food desert studies have been
inconsistent in their ability to show relevant differences between food desert and non-
food desert areas. The intent of food desert delineation is to reveal differences
between locations in the food landscape. The world is not a uniform place, and by virtue
of this demographic delineations used for food desert research may not be
representative of the intended selection of classification schemes. To evaluate the
effectiveness of different food desert definitions, definitions will be evaluated on a
basis of their ability to select for demographic inconsistencies that are related to an
idealized food desert concept, including:
Rates of poverty are higher in food desert areas.
Access to transportation is lower in food desert areas.
There are less grocery stores in food desert areas.
There are higher numbers of fast food retail locations in food desert
areas.
Although these metrics are simplistic, they hold the potential to reveal weaknesses in
food desert classification schemes.

22

Methods

Table 1:

The basic method of this study was to identify food desert and non-food
desert areas in Ohio. Food deserts were defined by multiple definitions to allow for
comparisons (Table 1). Areas were then compared to one another using their
socioeconomic attributes as a means of evaluation for the validity of each definition.
To analyze and compare geographically based food desert definitions, a consistent
methodology was required to maintain consistency of selection throughout testing.
Different model studies had varying data collection techniques, which used inconsistent
data sources. This study sought to compare multiple definitions of food deserts, and
generalized databases were used to ensure that data sources remained as consistent as
possible. Multiple definitions required the testing of various hypotheses; the hypotheses
tested in this study include:
Definition: Area of Study:
USDA Low Access:
Rural: 10 miles from grocery store
Urban: 1 mile from grocery store
United Kingdom:
Urban: 500 meter from grocery store
USDA Per Capita Access:
Grocery Stores per 1,000 pop. Census Block
Fast Food Location per 1,000 pop. Census Block

Table 1: A list of food desert delineation schema tested in this study. Euclidean distances were
used for distance delineation. Per Capita definitions used census block delineations.

23

1. Census blocks in food desert areas have a higher number of households below
the poverty line and lower percentage of the population with access to
transportation.
2. Access to car transportation will be lower in areas classified as food deserts, as
transportation is necessary to get to food in the absence of public transportation
3. The number of grocery stores per 1,000 individuals is lower in food desert areas
than it is in non-food desert areas.
4. The number of fast food restaurants per 1,000 individuals is higher in food
desert areas than in non-food desert areas.
The state of Ohio was selected as the site of focus in this study. Smaller regions were
considered in preliminary research, but to encompass both urban and rural areas a larger
scale which encompassed both types of area was selected. The breadth of the
information contained in the state of Ohio databases that were compiled also represented
a better analogue for the USDA definition of a food desert, which seeks to model the
Unites States as a whole.
Ohio census demographic and delineation data was acquired from the U.S. Census
Bureau shapefile and linefile, TIGER database 2011. First, Ohio border files were
downloaded from the U.S. Census State Geodatabase file. U.S. Census block data was
collected to enable the comparison of demographic data across the census block areas in
the state of Ohio. The U.S. Census block data is the smallest census delineation with
statistical data, and contains many metrics that are necessary to the creation of economic
based food desert delineation systems, and useful to the cross-comparison of alternate
food desert definitions.
24

After the collection of Census Block data, it was critical to several food desert
definitions to be able to differentiate between urban and rural areas. The U.S. Census
Bureau urban areas 2010 map was used to differentiate between urban and rural areas.
This classification distinguished census blocks on a basis of whether their centroid was
within the Urban Areas shapefile. This shapefile is defined by the presence of urban
clusters of at least 2,500 individuals and urban areas of more than 50,000 individuals with
a population density exceeding 1,000 individuals per square mile. Centroid selection
was used to select census blocks which fell within the aforementioned urban areas.
Business locations and main lines of business were key components of all geographic
food desert definition schemes. To compile a consistent database for the entirety of the
sampled area of Ohio, a consistent data sampling technique was required. Business
information was attained in its entirety from two databases: DatabaseUSA and
ReferenceUSA (ReferenceUSA, 2014); (DatabaseUSA, 2014). Database entries included
business name, latitude and longitude coordinates, main line of business, square footage,
and annual income, as well as various other comparative metrics. The businesses in Ohio
were sampled from databases as defined by their business type designated by their retail
designation codes. Database entries were compiled in a parallel .csv file to enable use in
JMP, GIS, and Excel. Although individual database format was similar, entries were
modified to fit a consistent order before placement in a common .csv file. Business
listings were filtered by address to remove any duplicate entries. Removal was
conducted by hand to verify that no mistake had been made. The main line of business
was compiled for each store in an Excel spreadsheet using five distinct categories:
restaurant, fast food establishment, convenience store, grocery store, and food bank.
25

Separation methodology was largely conducted using predefined identifier codes, but
some hand modifications were made to the dataset to account for mis-designation by
code (ie. Kroger gas station from Grocery Store to Convenience Store). Business
addresses were then geocoded using the Google Maps API and exported be to a .kml
format. This conversion was completed using Google Fusion Tables, which could then
re-exported into a formatted .kml with unique business identifiers that could be used to
match data with ArcMap 10.1. Business locations were then imported into ArcGIS 10.1
using the .kml to shapefile script.
Projections from all sources were redefined using the batch project script in ArcMap
10.1 in accordance with the UTM 17N projection. Layers from U.S. census data
geodatabases were then processed to lie only within the Ohio border to remove
superfluous information outside the chosen study area. The number of businesses per
given census block corresponding to each type of business category were created from
the geocoded business location layer and census block data using ArcMap 10.1s union
function. This in turn, was joined to the census layer and business layer attribute tables to
enable statistical analysis of both layers concurrently.
Maps which visualized poverty and transportation data used Jenk distributions to
visualize the differences in percentage with or without access to a variable for each map.
Jenk distributions were chosen because they isolate discreet groups in data sets by
maximizing variance between separate groups, and minimizing variance within distinct
groups. This method is best for data visualization of this type, because food desert or
non-food desert groups only make sense within the context of the opposite category.
26

Even sampling fails to show differences between deprived and affluent areas (Map 2,
Map 3).
The USDA and five hundred meter United Kingdom low access geographic food desert
definitions were created using the buffer function in ArcMap 10.1. A ten mile buffer was
used for rural areas and a one mile buffer for urban areas to define low access areas
within the USDA food desert definition. These buffers were created and then clipped
to the area of Ohio. Urban grocery stores were buffered twice, as they needed to account
for both the urban 1 mile and rural 10 mile food desert definitions. Buffers that were
created for the 10 mile area were erased from urban areas to remove any compounded
non-food desert areas. Food desert census blocks were then selected by their
centroid, as defined as non-food desert if it was within the area of a defined non-food
desert areas. Lists of the selected census blocks were compiled from the selected
features and turned into binary classifications in Excel by checking lists of shape
identifiers for their presence in exported arrays using Boolean statements. This
methodology was repeated to select for the 500 meter urban food desert definition.
The above mentioned food desert categorizations were then selected from
compiled census block and business data, and attribute tables were exported as .csv files.
The .csv format allowed data to be used in Microsoft Excel and for the Continuous and
Nominal datasets to undergo bivariate fit-line, T-Test, and ANOVA comparisons
between the data sets in JMP. JMP enabled the statistical testing of otherwise subjective
map data, allowed for the comparison and classification of food desert schema
according to their statistically relevant differences.
27

Fast food restaurants per 1000 individuals and grocery stores per 1000 individuals
were studied as a function of the percentage of households living below the poverty line
using a bivariate fit analysis. Fit-lines were created between these relationships, to
determine whether there was a correlation between households living below the poverty
line and decreased access to grocery stores or increased access to fast food establishments
in a given census block. The dataset was created using grocery store point data that had
been buffered in ArcMap 10.1 to account for distances. This dataset was then used to
create a geometric union in ArcMap 10.1, to derive a count of buffers per given census
block. The count was defined by the presence of a buffer shape overlapping the centroid
of a census block shape to define a non-food desert area.
The USDA food desert classification scheme was the first definition to undergo
statistical analysis. The USDA food desert definition uses a 1 mile buffer for urban
areas and 10 mile buffer for rural areas to delineate low access. To evaluate the low
access component of the USDA food desert definition, ANOVA analysis was
performed to distinguish differences between Urban, Rural, and Combined urban and
rural food desert areas by virtue of their food desert centroid selection. This
comparison used percentage of households below poverty line, percentage of population
with access to a car, and number of grocery or fast food locations per 1000 individuals as
comparative metrics. If food desert definitions had different low access and low-
income components, relevant data was selected and combined to enable ANOVA tests
across paired variables. This statistical review process was repeated for all applicable
classification schemes in which comparison was valid.
28

Map 1: The location of grocery
stores in Ohio
29

Map 2: Percentage of Households
below the poverty line by Census
Block. Percentage delineations are
created using Jenk optimization
algorithms.
30

Map 2: Percentage of Households
below the poverty line by Census
Block. Percentage delineations were
created using Jenk optimization for
best viewing results.
31

Map 3: USDA low access food
desert areas are shown in green. A
one mile buffer was used for urban
areas and ten mile buffer for rural
areas.
32




Map 4: United Kingdom low access
food desert areas are shown in
green. A 500 meter buffer was used
for urban areas. White areas are rural
and therefore not taken into account
within this definition.
33

Results:

Statistics were run to
develop an understanding of
the demographic groups in
Ohio before any food desert
classification schemes were
derived from mapping data.
These statistics compared
urban and rural areas without
any food desert or non-food desert distinction. T-tests and ANOVA tests were
performed to analyze whether there were statistical differences between urban and rural
populations. There was no statistical difference between the percentage of the total
population living below the poverty line between urban or rural areas (p = .6132). There
was a statistically relevant difference in percentage of the population over the age of 16
with access to transportation between urban and rural areas. In urban areas, 90.42% of
residents had access to a car transportation, but in rural areas 91.63% of the population
had access to transportation (p = .001).
The number of grocery stores and fast food establishment counts per 1,000 individuals
showed no statistical difference
in number of either of the establishments per 1000 individuals between urban and rural
census blocks (grocery p = .3335, fast food = .5406).
Rural Urban
Percentage of Households
below the poverty line
15.96 15.66
Percentage of Individuals with
access to transportation by car
91.63* 90.42*
Number of Grocery Stores per
1,000 Individuals
.2032 .2293
Number of Fast Food Retailers
per 1,000 Individuals
.3274 .3488
*demarcates statistical significance within a row (<0.05)

Table 2: No statistically relevant difference existed
between the number of holds below the poverty line
between urban and rural areas. More people in rural areas
of Ohio have access to transportation by car than in urban
areas.

34

To evaluate the effectiveness of food desert categorization, the number of
grocery stores per census block area was examined on a basis of whether a census block
was considered a food desert or non-food desert area for both the USDA and United
Kingdom food desert definitions. There was no statistical difference in the number of
food sources in food-desert areas or non-food desert areas in any selection
methodology. Tested selection methods included block intersection with food desert
area, block centroid in food desert area, and block completely contained within food
desert area. The most statistically relevant selection method was centroid selection
which had extremely an extremely weak difference between the number of grocery stores
per given census block and census block classification (p-value = .1358). The centroid
selection method was used for all following experimentation as it was based on the
USDA definition, and provided better classification of food desert areas than any
feasible alternative.
Table 3:

Definition: Rural Urban Combined
USDA Low Access:
Food Desert 15.83 17.45* 17.10*
Non-Food Desert 20.47 14.98* 15.00*
United Kingdom:
Food Desert: --- 14.96* ---
Non-Food Desert --- 16.64* ---
*demarcates statistical significance within a definition column (<0.05)

Table 3: Differences in percentage of households living below the poverty line between
urban and rural food desert and non-food desert areas. USDA low access combined
urban and rural areas as well as urban areas by themselves showed statistical differences
between percentage of individuals below the poverty line between food desert and non-
food desert. The United Kingdom recommended definition exhibited higher poverty in
non-food desert areas.

35

Urban and rural census blocks designated as USDA food deserts showed a
statistically higher number of households below the poverty line (Table 3). The food
desert census blocks as had a mean of 17.10% of households below the poverty line,
whereas census blocks designated as having high-access to grocery stores had a mean of
15.00% of households below the poverty line (p < .0001).
Urban food deserts and rural food deserts (as defined by the USDA) were
separated from one another for independent analysis. Within the USDA definition, urban
food-desert areas had a statistically higher percentage of households living below the
poverty line. A mean of 17.45% of households in urban food deserts live below the
poverty line, compared to a mean of 14.98% of households in urban non-food desert
areas (Table 2). In rural areas, there was no statistically relevant difference in percentage
of households living below the poverty line regardless of whether the area was
considered a food desert (Table 3).
There was found to be a statistical difference in the percentage of households
living below the poverty line. In food desert areas under the 500 meter United
Kingdom urban food desert definition, a mean of 14.96% of the households were living
below the poverty line (< .0001). In contrast, non-food desert areas had a mean of
16.64% of households living below the poverty line (Table 3).




36


Table 4:
Definition: Rural Urban Combined
USDA Low Access:
Food Desert 91.74 87.47* 88.40*
Non-Food Desert 87.49 91.55* 91.54*
500 meter Urban Definition
Food Desert ---- 92.1 ---
Non-Food Desert --- 88.11 ---
*demarcates statistical significance within a definition column (<0.05)

Table 3: Differences in percentage of households living below the poverty line between urban
and rural food desert and non-food desert areas. USDA low access combined urban and
rural areas as well as urban areas by themselves showed statistical differences between
percentage of with access to transportation by car between food desert and non-food
desert.


Percentage access to vehicular transportation for individuals over the age of 16
was considered for urban and rural census blocks as a function of whether they fell within
the boundaries of a food desert. Census blocks with centroids outside the USDA
delineation had statistically lower access to car transportation than areas within the
USDA delineation (table 4). These low access census blocks had a mean percentage
access to transportation of 88.40%, while the high access areas had a mean access to
transportation of 91.54% (p < .0001) (Table 4).
Urban and rural census blocks were independently analyzed for differences in
percentage of population over 16 with access to transportation by car. There was a
statistical difference in access to transportation between urban food desert and non-
food desert areas (p < .0001). In urban food desert census blocks 87.47% of the
population had access to vehicular transportation compared to 91.55% in non-food
37

desert census blocks (Table 4). There was no difference in access to transportation by
car between rural food deserts and non-food desert areas.
A comparison between access to transportation by car for persons 16 and older
yielded a statistically relevant difference between the food desert and non-food
desert urban census blocks using a 500 meter food desert definition (p < .0001).
Using this definition, food desert census blocks 92.10% of people had access to
transportation, and in non-food desert areas 88.11% of individuals had access to
transportation by car (Table 4).

38


Chart 1: There was a weak positive correlation between number of grocery store
locations per 1,000 individuals and the percentage of households below the poverty
line in a given census block (p < 0.0001)(r
2
= 0.0159) .


Chart 2: There was a weak positive correlation between number of fast food
locations per 1,000 individuals and the percentage of households below the poverty
line in a given census block (p < 0.0001)(r
2
= .0021) .


0
2
4
6
8
10
12
0 20 40 60 80 100 120
N
u
m
b
e
r

o
f

F
a
s
t

F
o
o
d

L
o
c
a
t
i
o
n
s

P
e
r

1
,
0
0
0

i
n
d
i
v
i
d
u
a
l
s

(
C
e
n
s
u
s

B
l
o
c
k
)
Percentage of Households Below the Poverty Line
Number Of Fast Food Locations per 1,000
Individuals vs. Percentage of Households Below
the Poverty Line
0
1
2
3
4
5
6
7
8
9
10
0 20 40 60 80 100 120
N
u
m
b
e
r

o
f

G
r
o
c
e
r
y

S
t
o
r
e
s

P
e
r

1
,
0
0
0

i
n
d
i
v
i
d
u
a
l
s

(
C
e
n
s
u
s

B
l
o
c
k
)
Percentage of Households Below the Poverty Line
Number of Grocery Stores per 1,000 Individuals vs.
Percentage of Households Below the Poverty LIne
39

The relationship between the percentage of households below poverty in a given
census block and the number of grocery stores per 1,000 individuals and in a given
census bock were shown to have a positive directional relationship (p < .0001) (Chart 1).
The positive directional relationship found across the state of Ohio was significant in
both rural and urban areas. This result was in line with Cummins and Macintyre 1999,
which found a higher number of grocery stores in more deprived areas (Cummings &
Macintyre, 1999).
Another positive directional relationship was found when comparing the number
of fast food restaurants in a given census block with the levels of poverty in the same
census block (p < .0001)(Chart 2). This relationship also remained consistently
significant and positive across urban and rural census blocks. Although it had not been
previously reported in Ohio, this finding mirrors what has been reported in many health
related food desert studies such as (Cummings & Macintyre, 1999).

40

Discussion

The geographically defined food desert definitions recreated in this study
attempt to define food deserts on the basis of lack of ready accessibility to healthy
foods. This type of definition is consistent with the original definition of a food desert
as put forward by the Low Income Task Force in the UK, which defined a food desert
as an area with little to no access to food retail provision (Cummings & Macintyre,
1999). However, geographic studies only examine symptomatic and surface level
evidence which may be indicative of problems within an area. In this aspect, geographic
definitions fail to take into account the extraordinary complexity of the food desert
associated problem-set, including mobility disadvantage, income disadvantage, social
constraints, and nutritional problems to name a few (Guy & David, 2005). However
simplistic a geographic definition is, mapping solutions can serve as valuable assets to
food desert identification prior to on-the-ground investigation. However, this is only
the case so long as the definition that is used for mapping is widely applicable to a huge
range of diverse geographies, beyond the area that it was originally identified.
To encompass a wider range prospective food desert areas, many food desert
definitions contain selection caveats to make themselves more widely applicable to a
range of on-the-sground circumstances. To evaluate comparative criteria among
definitions, many subdivisions within individual datasets were necessary to ensure that
variation in testing outcomes was not due to changes in definitions at different scales (ie.
Food Desert Urban 1 mile compared to Food Desert Rural 10 mile within the USDA
definition).
41

It was found that within the USDA food desert definition using centroid
selection, urban food desert areas had lower percentages of the population in urban
areas with access to transportation and higher percentages of households below poverty
compared to urban non-food desert areas. This data confirmed a food desert
hypotheses within urban areas of Ohio and disproved another. While levels of poverty
were greater in food desert areas, access to transportation was lower, suggesting that
even while there may be greater need for transportation, the need may not be satisfied.
However, no statistically relevant differences between car access or percent of
households below the poverty line were found in rural areas. The United Kingdom 500
meter definition showed a statistically higher percentage of the population living below
poverty line in urban non-food desert areas using centroid selection compared to urban
food desert areas. This means that the selection threshold set at 500 meters within the
European food desert definition had opposite effects of the USDA 1 mile definition,
suggesting that it is too strict of a delineation threshold. Statistically relevant
relationships were also observed between the number of census blocks per 1,000
individuals and number of households below poverty. There were more fast food
restaurants and grocery stores in Urban and Rural areas with higher number of
households below poverty than in census block areas with lower number of households
below poverty. This result was in line with what other food desert studies had found in
their experimentation, and advocated for a weakly directional statement that there were
more grocery and fast food locations in more impoverished areas.
The same dataset was used to calculate food desert delineation buffering and
count of grocery stores per 1,000 individuals. Comparative results of the two tests show
42

separate conflicting results. For food desert buffers, census blocks were chosen on the
basis of whether or not they had a centroid in a food desert area. For the number of
grocery stores per 1000 individuals definition, a count of grocery store data points (in the
form of distance buffers) was created for each census block, and the number of grocery
stores per 1,000 individuals created from the grocery store count and census block
population data. These statistically relevant results are odds with one another. Urban
census block areas with high poverty are more likely to have healthy food locations
such as grocery stores nearby, and also have a higher probability of being in a food
desert area, as defined by the centroid of a given census block existing outside the
bounds of a one mile buffer from any food source. This result can be explained on a
basis of the selection biases created by the buffer distance of selection in each definition
scheme. While count per 1,000 individuals takes into account the number of buffered
locations within distance of individuals, accessibility by distance does not.
Geographic accessibility is a fickle measurement, because small changes in
criteria of data selection or alternatively changes in the distance scale for data selection
can have large implications for which data are selected. Different food desert
delineations can be achieved by using spatial intersection as a selection method. This
was demonstrated when separate selection methodologies were tested to determine their
utility in choosing a viable way to designate a census block as food desert or non-food
desert. Of the three options that were tested, none did a satisfying job of matching and
capturing relevant overlaps between the different geographies, should any have existed.
Even within simple geographic food desert definitions, selection methodology
can change statistical outcomes with seemingly small tweaks. Multiple selections are
43

needed in each dataset to isolate definition variables, but each selection introduces more
previously unseen bias, necessary to each definition, but not relevant in meaningful ways.
Not surprisingly, If the selection methodology is changed in this study, so too are its
results, in some way significantly.
As previously mentioned, food desert selection criteria can have huge impacts
on which census blocks are selected within a given definition. By changing the census
block selection method from centroid with a food desert area to intersection with a
food desert area, or full-shape containment within a food desert area, the statistical
outcomes of a study can be manipulated. The difference between percentages of
population living below the poverty line in urban food desert and urban non-food
desert areas can be increased, or alternately the difference in percentage of households
below the poverty line between rural food desert and rural non-food desert census
block can be made relevant, by changing selection methodology. These selection
choices are all equally viable ways to select for probable food areas but also completely
negate the effects of a buffer distance on large portions of the population. These effects
are so much so that no selection methodology has a relevant relationship with the number
of grocery stores in a given census block area, defeating the purpose of any geographic
low access definition.
Centroid selection is weakest in rural areas. Rural census blocks in Ohio have a
higher land area than urban census blocks in Ohio. Centroids in ArcMap 10.1 are created
by a heuristic algorithm used to find the geometric center of a two dimensional figure, in
this case a shape file. To define an area as a food desert in centroid selection, the
centroid must be encompassed in a buffered area. As areas get larger for census blocks,
44

centroid selection becomes less accurate, by virtue of the fact that the centroid is less
applicable to the relative distances in the combined area. Intersection selection is far less
picky, and any intersection with shapefile boundaries will designate a census block as a
non-food desert, but intersection has no bearing whatsoever on percent cover.
Containment is by comparison to the other two selection methodologies the most strict,
but does not work well in small buffered non-food desert areas, which despite large
percent coverage of census blocks, are not enough to contain shapes.
Food desert definitions often use buffered models to select which populations to
study as food desert areas. Studies including Donkin et. al. 1999 use buffers (500m) to
identify portions with of the map within select distances of identified grocery stores
(Donkin, Dowler, Stevenson, & Turner, 1999). This data analysis method has been
widely adopted in United Kingdom at the recommendation of the government (Furey,
Strugnell, & Mclveen, 2001). However, at scale and with non-consistent delineation
shape of census areas, selection methodologies are highly inaccurate. An alternative
selection method which removes selection from buffers is a cellular automata model.
Such a model was used in Ploeg et al. to model United States food deserts.
In this approach a cellular food desert model uses a grid of a set size and shape
to model relationships between variables which represent real world information. Ploeg
et al. uses a cellular automata model with a set resolution of a kilometer to calculate
population weighted distances from grocery stores in the USDA 1 mile urban 10 mile
rural food desert definition. Cellular automata models are meant to increase distance
resolution and create better accessibility maps.
45

Ploeg et al. uses population information from descending census geographic
levels (tract to block) to infer population density on a resolution as defined by a 2D grid
of set mile size model size to model actual geographic areas. Unlike in modeling using
buffers, cellular modeling calculates distances to and from food sources at a scale defined
by each grid piece centroid rather than a distinct census block. Distance to grocery stores
was calculated using a Euclidian distance measure from the centroid of each non-grocery
store grid-piece to the centroid each grocery store grid-piece. This data is averaged for
census blocks and combined with an assumed population distribution. Census areas are
designated as food deserts in accordance with the results of the combined cells
contained in their abstracted cellular area.
Furthermore, distances can be within a cellular automata model such as this one
can be deceptive, as the kilometer grid projection of the cells in Ploeg et al. does not
fully align well with any large geography. Distance weighting also introduces selection
bias in an automata model. In the Ploeg et al study methods, distance to a food source is
weighted by population in the Plough et al. USDA definition (Ploeg, et al., 2010). This is
meant to help infer block and tract information from grid points. However, block
information is the smallest measurement made accessible by the U.S. Census Bureau (U.
S. Census Bureau, 2014). To determine the proportion of a given census blocks
population living below the poverty line, the population was weighted between the
kilometer square cells in each block. This weighting scheme combined with different
distances for Urban and Rural areas showed that a limitation of the methodology was a
selection of low density urban areas towards classification as food deserts (Ploeg, et al.,
46

2010). Cells didnt scale between distances consistently, and depending on distance
thresholds, maps were be extremely different.
There is also reason to suspect the data reported in the Ploeg et al. model. The
median distance to a supermarket reported to be .86 miles per household (Ploeg, et al.,
2010). Simply stated, there are not enough supermarkets in their sample size of 39,502
United States supermarkets to allow the values to be that low. Using a buffering method
and this studys 1,523 Ohio data points, the same median distance would mean that only
2% of Ohios area would be inhabited, assuming a one mile buffer with no overlap. This
means that more than fifty percent of Ohios households would be living within two
percent of its land (and thereby eligible for consideration within a Euclidean distance
calculation that could result in this median). This raises the question of whether the
downcasting methodology in the Plough et al. study actually worked, or if it
concentrated populations into relatively few half kilometer cells that happened to be
within one mile of food sources.
A possible solution to the inconsistencies in both buffering methods and cellular
automata methods are network based accessibility studies. Studies such as McEntee and
Ageyman (2009) use road networks within a city or state to derive distance from food
sources (McEntee & Agyeman, 2009). This removes almost all distance selection bias,
as points are calculated at a household level to the nearest food source over a road
network (McEntee & Agyeman, 2009). Network calculated distances are longer than
buffered or cellular, as they are no longer Euclidian, but rather reflect actual road
pathways taken by a majority of people. In a 2009 study in Vermont the median distance
to a food source was 4.14 miles (urban and rural), as compared to a median of .83 miles
47

(urban and rural) for the United States in 2010 (McEntee & Agyeman, 2009) (Ploeg, et
al., 2010). This inconsistency points to a underestimate of distance to food sources that
use Euclidean distance as a measure. Network maps are also much harder to generate
than other data types from store and census information, and were noted to be too
computationally taxing to be used in the Ploeg et al. study (Ploeg, et al., 2010).
Regardless of computational time, all small scale network studies generate substantially
higher distance estimates, and to address geographic food desert problems, should they
exist, underestimate of distance should be cited in buffering or automata studies.
Selection bias is important to consider when looking into food desert
delineations, but geographic selection bias in census areas hardly captures the realities of
symptomatic food desert problems. Many geographic food desert classification
schemes are by their very nature tied to the locations in which they are experienced.
Geographic classification schemes, take into account limited quantifiable information to
infer problems that may be experienced on by individuals living in the studied area.
No population is homogenous by virtue of its geography, culture, or
demographics, it is this clear that any of the relevant differences in populations
demonstrated in different classification schema are a result of their carefully tailored
selection bias. In United Kingdom studies such as Donkin et al. mapped food
accessibility and pricing indices in GIS, and had success in finding relevant price
differences between low income and low access areas and high income, high access areas
where Cummins and Macintyre hadnt the same year (Cummings & Macintyre, 1999)
(Donkin, Dowler, Stevenson, & Turner, 1999). This success was attributed to taste
preferences and local knowledge that were introduced to pricing indices that went
48

beyond the pricing of staple foods in a given area (Donkin, Dowler, Stevenson, & Turner,
1999). When definitions are transplanted to new locations, there is a high probability that
they will not work as a tool to select for the causes behind symptomatic problems as they
once did.
The United States Department of Agriculture (USDA) food desert definition is
maintained for general use across the United States. This definition is the broader than a
majority of the food desert definitions. It designed to encompass both rural and urban
areas across the geographic area of the United States. As the United States is a large and
diverse geographic area, in an attempt to diversify classification schemes, the USDA
definition takes into account many variables. In their combined form, the definition is
incredibly unruly and complex, but can point to general issues (such as low income or
low access) in certain areas. In the USDA definition geographic access is calculated
using distance thresholds. Geographic food accessibility is measured on a basis of family
income and vehicle availability. Low access can also mean low access to transportation,
or living more than a set distance from any food provider. After qualifications are made,
urban and rural food desert sub-definitions are reorganized into manageable tracks, to
select for low-income and low-access populations in urban and rural areas.
The broad definition of a food desert in the USDA definition provides
flexibility to the USDA definition, as it can be theoretically applied to many areas, and
still remains relevant with respect to varying factors that are presumably symptomatic of
a food desert area. However, grouping quantifiable traits of food deserts together
allows for many compounded variables at a large scale. While mixed methods
definitions are inclusive of many variables, due to variation in the United States
49

population, a broad definition including many caveats is subject to become useless,
unless basic measurements such as real world network distance are taken into account.
When government agencies seek to define food desert areas, the studies they
conduct studies are purposeful and meant to address relevant inconsistences in the food
landscape (United States Department of Agriculture, 2014). The USDA Food Access
Research Atlas was created as a tool to identify areas to target as a part of what is now
called the Lets Move! initiative (United States Department of Agriculture, 2014).
The Lets Move! initiative has five pillars, one of which is increasing food accessibility
(White House, 2014). After identifying areas of low food accessibility, the Lets Move!
Initiative seeks to develop and equip food retailers of all sorts, which included grocery
stores, convenience stores and farmers markets. However, as shown in this study, in
many areas, there are no differences in accessibility as a function of distance to food
sources. Rather, overall improvement in access to food and food education in all
geographic areas, including food deserts could be a better approach to the food desert
problem.
The problems associated with food deserts are multidisciplinary. Issues that
combine and to create spaces commonly called food deserts include, health and
nutritional problems, income disadvantage, social and cultural constraints, and shortage
of good-quality, low priced food locally (Guy & David, 2005). Food deserts can occur
as a result of one or many of these factors working together to create an area that is in
some observable sense deprived of food availability. Food desert definitions seek to
identify generalized symptoms of low food accessibility. Despite searching for statistical
differences in culture, demographic differences, and distance to food sources, ease of
50

transportation, too many complex variables can be associated with the food desert
problem to calculably find and address specific food desert problems on a large scale.
Food deserts are a broad problem set. Although they were first identified as
geographic areas, it is clear that their definition transcends a simple geographic lack of
food sources. The average percentage of households living below the poverty line in
Ohio is 11.15%, with a median of 11.16%. These values are high enough to warrant
attention regardless of geographic access. Especially considering the best food desert
definitions observed in this study were able to isolate only three percent more households
than the comparative mean. While evidence exists to support a relationship between
distance to food sources and diet quality by means of food choice (Walker, et al., 2011).
A vast body of research supports the relationship between economic food access and diet
quality. Healthy food items cost on average more than energy-dense alternatives
(Drewnowski, Andrienu, & Darmon, 2005). Furthermore, in low income individuals cite
prohibitive costs as a reason for not eating more fruits and vegetables (Jetter & Cassady,
2006). These issues are ones of income and food price rather than geographic location.
However income does come into play with food availability, especially in areas where
residents are dependent on cars for transportation.




51

Bibliography

110th Congress. (2008). Food, Conservation, and Energy Act of 2008. 662.
Carstairs, V., & Morris, R. (1991). Deprivation & Health. Aberdeen: Aberdeen University Press.
Chung, C., & Myers, S. (1999). Do the poor pay more for food? An analysis of Grocery Store
Availability and Food Price Disparities. Journal of Consumer Affairs, 276-296.
Cummings, S., & Macintyre, S. (1999). The location of food stores in urban areas: a case study in
Glasgow. British Food Journal, 545-553.
Cummins, S., & Macintyre, S. (1999). The Location of Food Stores in Urban Areas: A Case Study
in Glasgow. British Food Journal, 545-553.
Cummins, S., & Macintyre, S. (2002). A Systematic Study of an Urban Foodscape: The Price and
Availability of Food in Greater Glasgow. Urban Studies, 2115-2130.
DatabaseUSA. (2014, January 07). DatabaseUSA. Retrieved from www.databaseusa.com
Dixon, D., & McLaughlin, D. (1968). Do the inner city poor pay more for food? The Economic and
Buisness Bulletin, 6-12.
Donald, B. (2013). Food retail and access after the crash: rethinking the food desert problem.
Journal of Economic Geography, 231-237.
Donkin, A., Dowler, E., Stevenson, S., & Turner, S. (1999). Mapping access to food in a deprived
area: the development of price and availability indices. Public Health Nutrition , 31-38.
Donkin, A., Dowler, E., Stevensonn, S., & Turner, S. (2000). Mapping access to food in a deprived
area: the development of price and availability indices. Public Health Nutrition, 31-38.
Drewnowski, A., Andrienu, & Darmon. (2005). Low-cost diets: more energy, fewer nutrients.
European Journal of Clinical Nutrition, 434.
Findlay, A., & Sparks, L. (2002). The Retail Plainning Knowledge Base Briefing Paper 2: Food
Deserts. Stirling: University of Stirling.
Furey, S., Strugnell, C., & Mclveen, H. (2001). An investigation of the potential existence of
``food deserts'' in rural and urban areas of Northern Ireland. Agriculture and Human
Values, 447-457.
Goodman, C. (1968). Do the poor pay more? Journal of Marketing, 18-24.
Guy, C., & David, G. (2005). Measuring physical access to "healthy foods". International Journal
of Consumer Studies, 222-234.
Jackson, K., & Trochim, W. (2002). Concept mapping as an alternative approach for the analysis
of open-ended survey responses. Organizational Research Methods, 307-336.
52

Jetter, K., & Cassady, D. (2006). The availability and cost of healthier food alternatives. Journal of
American Preventative Medicine, 38-44.
Macintyre, S., & Cummins, S. (2002). "Food Deserts" --Evidence and assumption in health policy
making. BMJ, 436-38.
McEntee, J., & Agyeman, J. (2009). Towards the development of a GIS method for identifying
rural food deserts: Geographic access in Vermont, USA. Applied Geography, 1-12.
Nelson, M., Mayer, A., & Manley, P. (1993). The food budget. Budget Standards for the UK, 35 -
64.
Ploeg, M., Breneman, V., Dutko, P., Williams, R., Snyder, S., Dicken, C., & Kaufman, P. (2010).
Access to Affordable and Nutritious Food: Updated Estimates of Distance to
Supermarkets Using 2010 Data. Washington D.C.: USDA.
ReferenceUSA. (2014, January 07). Reference USA. Retrieved from
http://www.referenceusa.com/
Sexton, D. (1971). Comparing the Cost of Food to Blacks and Whites. Journal of Marketing, 40-
46.
Smith, C., Butterfass, J., & Richards, R. (2009). Environment inuences food access and resulting
shoppingand dietary behaviors among homeless Minnesotans living in food deserts.
Agriculture and Human Values, 141-161.
U. S. Census Bureau. (2014, January 07). Unites States Census Bureau Geography. Retrieved
from Tiger/Line Shapefiles and Tiger/Line Files: http://www.census.gov/geo/maps-
data/data/tiger-line.html
United States Department of Agriculture. (2014, March 2014). Food Access Research Atlas.
Retrieved from Economic Research Service: http://www.ers.usda.gov/data-
products/food-access-research-atlas.aspx#.U0m1YflT4R4
USDA. (2011, January 18). CED Data Healthy Food Financing Initiative. Retrieved from Office of
Community Services: http://www.acf.hhs.gov/programs/ocs/resource/healthy-food-
financing-initiative-0
Walker, R., Fryer, C., Butler, J., Keane, C., Krisha, A., & G, J. (2011). Factors influencing food
buying practices in residents of a low-income food desert. Journal of Mixed Methods
Research, 247-267.
Walker, R., Keane, C., & Burke, J. (2010). Disparities and access to healthy food in the United
States: a review of food deserts literature. Health and Place, 876-884.
White House. (2014, April 12). Let's Move! Retrieved from About Let's Move!:
http://www.whitehouse.gov/accessibility
Wrigley, N., Warm, D., & Margetts, B. (2002). Deprivation, diet and food retail access: findings
from the Leeds 'food deserts' study. Environment and Planning, 1-38.
53

Vous aimerez peut-être aussi