USF Libraries
USF Digital Collections

'food deserts' and 'food swamps' in hillsborough county, florida :

MISSING IMAGE

Material Information

Title:
'food deserts' and 'food swamps' in hillsborough county, florida : unequal access to supermarkets and fast-food restaurants
Physical Description:
Book
Language:
English
Creator:
Stein, Dana Beth
Publisher:
University of South Florida
Place of Publication:
Tampa, Fla
Publication Date:

Subjects

Subjects / Keywords:
Built Food Environment
Gis
Obesity
Public Health
Spatial Analysis
Dissertations, Academic -- Geography Public Health Geographic Information Science and Geodesy -- Masters -- USF   ( lcsh )
Genre:
bibliography   ( marcgt )
non-fiction   ( marcgt )

Notes

Summary:
ABSTRACT: Research has shown that the suburbanization of supermarkets has created `food deserts', defined as areas where socially disadvantaged individuals lack access to nutritious food outlets. Additionally, the growing presence of fast-food restaurants has created `food swamps', or areas where socially disadvantaged individuals encounter an overabundance of unhealthy food outlets. While previous studies have analyzed either `food deserts' or `food swamps' using conventional statistical techniques, a more comprehensive approach that includes samples of both healthy and unhealthy entities and considers the variety of available food options is necessary to improve our understanding of the local food environment and related disparities. This thesis addresses several limitations associated with previous geographic research on the built food environment through a case study that examines socio-demographic inequities in access to supermarkets and fast-food restaurants in Hillsborough County, Florida-- an urban area that has been severely affected by the obesity and food crisis plaguing the nation. An important goal is to examine the spatial and statistical association between socioeconomic deprivation and potential access to all supermarkets, healthiest supermarkets, all fast-food restaurants, and unhealthiest fast-food restaurants, respectively. This study utilizes precise locations of food retailers based on government codes, U.S. Census data, GIS-based network analysis, and a combination of conventional statistical measures and exploratory spatial analytical techniques. Specifically, local indicators of spatial association (LISA) are used to visualize how the relationship between socioeconomic deprivation and accessibility to food outlets varies geographically within the county, and identify the locations of food deserts and food swamps based on the statistical significance of spatial correlations. Conventional statistical measures indicate that socioeconomically deprived neighborhoods are significantly less accessible to the healthiest supermarkets and more accessible to all fast-food restaurants. LISA significance maps reveal that food deserts are located in suburban and rural regions, food swamps are located closer to the urban center, and both are found along major highways in Hillsborough County. Logistic regression results show that race and ethnicity play an undeniably pervasive role in explaining the presence and location of both food deserts and food swamps. This research demonstrates the need to explore local variations in statistical relationships relevant to the study of the built food environment, and highlights the need to consider both healthy and unhealthy food outlets in geographic research and public policy initiatives that aim to address the obesity crisis.
Thesis:
Thesis (M.A.)--University of South Florida, 2011.
Bibliography:
Includes bibliographical references.
System Details:
Mode of access: World Wide Web.
System Details:
System requirements: World Wide Web browser and PDF reader.
Statement of Responsibility:
by Dana Beth Stein.
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains 117 pages.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
usfldc doi - E14-SFE0004886
usfldc handle - e14.4886
System ID:
SFS0028143:00001


This item is only available as the following downloads:


Full Text
xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam 22 Ka 4500
controlfield tag 007 cr-bnu---uuuuu
008 s2011 flu ob 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0004886
035
(OCoLC)
040
FHM
c FHM
049
FHMM
090
XX9999 (Online)
1 100
Stein, Dana Beth.
0 245
'food deserts' and 'food swamps' in hillsborough county, florida :
h [electronic resource] /
b unequal access to supermarkets and fast-food restaurants
by Dana Beth Stein.
260
[Tampa, Fla] :
University of South Florida,
2011.
500
Title from PDF of title page.
Document formatted into pages; contains 117 pages.
502
Thesis
(M.A.)--University of South Florida, 2011.
504
Includes bibliographical references.
516
Text (Electronic thesis) in PDF format.
520
ABSTRACT: Research has shown that the suburbanization of supermarkets has created `food deserts', defined as areas where socially disadvantaged individuals lack access to nutritious food outlets. Additionally, the growing presence of fast-food restaurants has created `food swamps', or areas where socially disadvantaged individuals encounter an overabundance of unhealthy food outlets. While previous studies have analyzed either `food deserts' or `food swamps' using conventional statistical techniques, a more comprehensive approach that includes samples of both healthy and unhealthy entities and considers the variety of available food options is necessary to improve our understanding of the local food environment and related disparities. This thesis addresses several limitations associated with previous geographic research on the built food environment through a case study that examines socio-demographic inequities in access to supermarkets and fast-food restaurants in Hillsborough County, Florida-- an urban area that has been severely affected by the obesity and food crisis plaguing the nation. An important goal is to examine the spatial and statistical association between socioeconomic deprivation and potential access to all supermarkets, healthiest supermarkets, all fast-food restaurants, and unhealthiest fast-food restaurants, respectively. This study utilizes precise locations of food retailers based on government codes, U.S. Census data, GIS-based network analysis, and a combination of conventional statistical measures and exploratory spatial analytical techniques. Specifically, local indicators of spatial association (LISA) are used to visualize how the relationship between socioeconomic deprivation and accessibility to food outlets varies geographically within the county, and identify the locations of food deserts and food swamps based on the statistical significance of spatial correlations. Conventional statistical measures indicate that socioeconomically deprived neighborhoods are significantly less accessible to the healthiest supermarkets and more accessible to all fast-food restaurants. LISA significance maps reveal that food deserts are located in suburban and rural regions, food swamps are located closer to the urban center, and both are found along major highways in Hillsborough County. Logistic regression results show that race and ethnicity play an undeniably pervasive role in explaining the presence and location of both food deserts and food swamps. This research demonstrates the need to explore local variations in statistical relationships relevant to the study of the built food environment, and highlights the need to consider both healthy and unhealthy food outlets in geographic research and public policy initiatives that aim to address the obesity crisis.
538
Mode of access: World Wide Web.
System requirements: World Wide Web browser and PDF reader.
590
Advisor:
Chakraborty, Jayajit .
653
Built Food Environment
Gis
Obesity
Public Health
Spatial Analysis
690
Dissertations, Academic
z USF
x Geography Public Health Geographic Information Science and Geodesy
Masters.
773
t USF Electronic Theses and Dissertations.
4 856
u http://digital.lib.usf.edu/?e14.4886



PAGE 1

! Food Deserts' and Food Swamps' in Hillsborough County, Florida: Unequal Access to Supermarkets and Fast f ood Restaurants by Dana Oppenheim Stein A thesis submitted in partial fulfillment of the requirements for the degree of Master of A rts Department of Geography, Environment, and Planning College of Arts and Sciences University of South Florida Major Professor: Jayajit Chakraborty, Ph.D. Russell Kirby, Ph.D. Michael Niedzielski, Ph.D. Date of Approval : March 25, 2011 Keywords: P ublic Health, Obesity, Built Food Environment, GIS, Spatial Analysis Copyright 2011, Dana Oppenheim Stein

PAGE 2

"" ACKNOWLEDGMENTS I would like to thank my major professor, Dr. Chakraborty, whose insightful advice and constant guidance helped me to reach and e xceed numerous academic goals during my tenure as a masters student. Our meetings, phone calls and emails were always educational and inspiring, and truly pushed me to grow and learn both as a student and researcher. He always demanded the absol ute best fr om me and his confidence in my abilities always encouraged me to keep trying. I would also like to thank the other members of my thesis committee, Dr. Kirby and Dr. Niedzielski, for their continual input and feedback during my time as a master's student. Dr. Kirby helped me to understand the theories and methods of spatially related public health research and he was always willing to spend time discussing health related issues and statistical analysis. Dr. Niedielski proved to be an invaluable resource re garding GIS technology and transportation issues. Last but certainly not least, I would like to thank my husband, Josh Stein. Without his unwavering support, continual understanding and ability to always help me see the light at the end of the tunnel, thi s project and degree would not have been possible. He is my biggest cheerleader, my best friend and my rock.

PAGE 3

! i TABLE OF CONTENTS LIST OF TABLES ................................ ................................ ................................ ............. iii LIST OF FIGURES ................................ ................................ ................................ ............. v ABSTRACT ................................ ................................ ................................ .................... vii CHAPTER 1: INTRODUCTION ................................ ................................ ........................ 1 CHAPTER 2: LITERATURE REVIEW ................................ ................................ ............. 8 2.1. Obesogenic Environments and Health Outcomes ................................ ............ 8 2.2. Defining the Built Food Environment ................................ ........................... 11 2.2.1. Creation & Impact of Food Deserts' and Food Swamps' ............. 12 2.2.2. Distribution of Food Outlets & Neighborhood Social Inequities .... 14 2.3. Evaluating the Built Food Environment ................................ ........................ 17 2.3.1. Relevant Food Desert' and Food Swamp' Measurements ............ 18 2.3.2. Spatial Accessibility Techniques ................................ ..................... 19 2.3.3. Exploratory Spatial Data Analysis ................................ ................... 21 2.4. Summary ................................ ................................ ................................ ........ 24 CHAPTER 3: DATA AND METHODOLOGY ................................ ............................... 26 3.1. Study Area ................................ ................................ ................................ ..... 26 3.2. Data S ources and Variables ................................ ................................ ........... 29 3.2.1. Accessibility to Food Outlets ................................ ........................... 36 3.2.2. Socioeconomic Deprivation Index ................................ ................... 39 3.2.3. Racial, Ethnic, and Locational Characteristics ................................ 41 3.3. Statistical Analysis Methodology ................................ ................................ .. 45 CHAPTER 4: STATISTICAL AND SPATIAL ANALYSIS OF ACCESSIBILITY TO ALL SUPERMARKETS AND HEALTHIEST SUPERMARKETS ................................ ................................ ................................ ...... 52 4.1. De scriptive Mapping and Statistics ................................ ............................... 53 4.2. Comparison of Quantile Means ................................ ................................ ..... 57 4.3. Global Statistical Analysis ................................ ................................ ............. 59 4.4. Local Spatial Statistical Analysis ................................ ................................ ... 61 4.5. Characteristics of Food Deserts ................................ ................................ ..... 67 4.6. Summary of Statistical and Spatial Analysis Results ................................ .... 71

PAGE 4

"" ! CHAPTER 5: STATISTICAL AND SPATIAL ANALYSIS OF ACCESSIBILITY TO ALL FAST FO OD RESTAURANTS AND UNHEALTHIEST FAST FOOD RESTAURANTS ................................ ................... 72 5.1. Descriptive Mapping and Statistics ................................ ............................... 73 5.2. Comparison of Quantile Means ................................ ................................ ..... 76 5.3. Global Statistical An alysis ................................ ................................ ............. 79 5.4. Local Spatial Statistical Analysis ................................ ................................ ... 80 5.5. Characteristics of Food Swamps ................................ ................................ .... 86 5.6. Summary of Statistical and Spatial Analysis Results ................................ .... 90 CHAPTER 6: CONCLUSIONS ................................ ................................ ........................ 92 REFERENCES ................................ ................................ ................................ .................. 99

PAGE 5

""" ! LIST OF TABLES Table 3.1. North American Industry Classification System Codes for Food Outlets (U.S. Census Bureau 2009) ................................ ................................ ............... 31 Table 3.2. Supermarkets in Hillsborough County, Florida. ................................ ............... 33 Table 3.3. Fast Food Restaurants in Hillsborough County, Florida. ................................ 34 Table 3.4. Healthiest Supermarkets in Hillsborough County, Florida. .............................. 36 Table 3.5. Unhealthiest Fast Food Restaurants in Hillsborough County, Florida. ............ 36 Table 3.6. Definitions for Accessibility to Fo od Outlets ................................ ................... 39 Table 3.7. Additional Explanatory Variables ................................ ................................ .... 45 Table 3.8. Moran's I Scatter Plot Definitions ................................ ................................ .... 49 Table 3.9. Bivariate Moran's I Significance Categorie s ................................ .................... 50 Table 4.1. Summary Statistics for Socioeconomic Deprivation Variables and Accessibility to Healthy Food Outlets ................................ ............................... 57 Table 4.2. Means for SED Index by Accessibility to Healthy Food Outlets ..................... 59 Table 4.3. Global Bivariate Association Between SED Index and Accessibility to Healthy Food Outlets ................................ ................................ ......................... 61 Table 4.4. Moran's I Scatter Plot Definitions ................................ ................................ .... 62 Table 4.5. Bivariate LISA Significance Categories ................................ ........................... 64 Table 4.6. Demographic Composition of the Bivariate Significance Categories: All Supermarkets ................................ ................................ ............................... 68 Table 4.7. Demographic Composition of the Bivariate S ignificance Categories: Healthiest Supermarkets ................................ ................................ .................... 69

PAGE 6

"# ! Table 4.8. Logistic Regression of Food Deserts ................................ ................................ 70 Table 5.1. Summary Statistics for Accessibility to Unhealthy Food Outlets .................... 76 Table 5.2. Means for SED Index by Accessibility to Unhealthy Food Outlets ................. 78 Table 5.3. Global Bivariate Association Between SED Index and Access to Unhealthy Food Outlets ................................ ................................ ..................... 80 Tabl e 5.4. Moran's I Scatter Plot Definitions ................................ ................................ .... 81 Table 5.5. Bivariate LISA Significance Categories ................................ ........................... 84 Table 5.6. Demographic Composition of the Bivariate Significance Categories: All Fast food Resta urants ................................ ................................ .................. 87 Table 5.7. Demographic Composition of the Bivariate Significance Categories: Unhealthiest Fast food Restaurants ................................ ................................ ... 88

PAGE 7

# ! LIST OF FIGURES Figure 3.1. Hil lsborough County, Florida ................................ ................................ .......... 29 Figure 4.1. Locations of All Supermarkets (n=115) and Accessibility to All Supermarkets by Block Group (n=795), Hillsborough County, Florida, 2010 ................................ ................................ ................................ .................. 54 Figure 4.2. Loc ations of Healthiest Supermarkets (n=47) and Accessibility to Healthiest Supermarkets by Block Group (n=795), Hillsborough County, Florida, 2010 ................................ ................................ ...................... 55 Figure 4.3. Socioeconomic Deprivation (SDE) Index by Block Group, Hillsboroug h County, Florida (n=795), 2000 ................................ .................. 56 Figure 4.4. Means for SED Index by Quintiles of Accessibility to All Supermarkets ................................ ................................ ................................ .... 58 Figure 4.5. Means for SED Index by Quintiles of Accessibility to Healthies t Supermarkets ................................ ................................ ................................ .... 58 Figure 4.6. Bivariate Moran's I between SED Index and Accessibility to All Supermarkets ................................ ................................ ................................ .... 63 Figure 4.7. Bivariate Moran's I between SED Index and Accessibility to Healthiest Su permarkets ................................ ................................ .................. 63 Figure 4.8. LISA Significance Analysis of Spatial Correlation between SED Index and Accessibility to All Supermarkets ................................ ............................. 65 Figure 4.9. LISA Significance Analysis of Spatial Correlati on between SED Index and Accessibility to Healthiest Supermarkets ................................ .................. 67 Figure 5.1. Locations of All Fast food Restaurants (n=513) and Accessibility to All Fast food Restaurants by Block Group (n=795), Hillsborough County, Flor ida, 2010 ................................ ................................ ...................... 74 Figure 5.2. Locations of Unhealthiest Fast food Restaurants (n=34) and Accessibility to Unhealthiest Fast food Restaurants by Block Group (n=795), Hillsborough County, Florida, 2010 ................................ ................. 75

PAGE 8

#" ! Figure 5.4. Means for Socioeconomic Deprivation (SED) Index by Quintiles of Accessibility to All Fast food Restaurants ................................ ...................... 77 Figure 5.5. Means for SED Index by Quintiles of Accessibility to Unhealthiest Fast food Restaurants ................................ ................................ ....................... 77 Figure 5.6. Bivariate Moran's I between SED Index and Accessibility to All Fast food Restaurants ................................ ................................ ............................... 82 Figure 5.7. Bivariate Moran's I between SED Index and Accessibility to Unhealthiest Fast fo od Restaurants ................................ ................................ 83 Figure 5.8. LISA Significance Analysis of Spatial Correlation between SED Index and Accessibility to All Fast food Restaurants by Block Group ..................... 85 Figure 5.9. LISA Significance A nalysis of Spatial Correlation between SED Index and Accessibility to Unhealthiest Fast food Restaurants by Block Group ................................ ................................ ................................ ............ 86

PAGE 9

#"" ! ABSTRACT Research has shown that the suburbanization of supermarkets has created food deserts' defined as areas where socially disadvantaged individuals lack access to nutritious food outlets. Additionally, the growing presence of fast food restaurants has created food swamps', or areas where socially disadvantaged individuals encounter an overab undance of unhealthy food outlets. While previous studies have analyzed either food deserts' or food swamps' using conventional statistical techniques, a more comprehensive approach that includes samples of both healthy and unhealthy entities and conside rs the variety of available food options is necessary to improve our understanding of the local food environment and related disparities. This thesis addresses several limitations associated with previous geographic research on the built food environment through a case study that examines socio demographic inequities in access to supermarkets and fast food restaurants in Hillsborough County, Florida an urban area that has been severely affected by the obesity and food crisis plaguing the nation. An import ant goal is to examine the spatial and statistical association between socioeconomic deprivation and potential access to all supermarkets, healthiest supermarkets, all fast food restaurants, and unhealthiest fast food restaurants, respectively. This study utilizes precise locations of food retailers based on government codes, U.S. Census data, GIS based network analysis, and a combination of conventional statistical measures and exploratory spatial analytical techniques. Specifically, local indicators of sp atial association (LISA) are used to visualize how the

PAGE 10

#""" ! relationship between socioeconomic deprivation and accessibility to food outlets varies geographically within the county, and identify the locations of food deserts and food swamps based on the statist ical significance of spatial correlations. Conventional statistical measures indicate that socioeconomically deprived neighborhoods are significantly less accessible to the healthiest supermarkets and more accessible to all fast food restaurants. LISA s ignificance maps reveal that food deserts are located in suburban and rural regions, food swamps are located closer to the urban center, and both are found along major highways in Hillsborough County. Logistic regression results show that race and ethnicit y play an undeniably pervasive role in explaining the presence and location of both food deserts and food swamps. This research demonstrates the need to explore local variations in statistical relationships relevant to the study of the built food environme nt, and highlights the need to consider both healthy and unhealthy food outlets in geographic research and public policy initiatives that aim to address the obesity crisis.

PAGE 11

1 ! CHAPTER 1 : INTRODUCTION America is facing a widespread obesity crisis that has become the fastest growing cause of disease and death in the nation ( Office of the Surgeon General [OSG] 2003). According to the National Health and Nutrition Examination Survey conducted by the Centers for Disease Control [CDC] (2009), the number of a dults at least 20 years of age classified as obese' rose from 13.4 percent in 1960 1962 to 34.3 percent in 2005 2006. Additionally, 32.7 percent of adults are overweight' and 5.9 percent are extremely obese', totaling a staggering 72.9 percent of all ad ults at least 20 years of age that are classified between overweight' and 'extremely obese.' According to the World Health Organization [WHO] (2006), obesity can be linked to increased eating of foods "that are high in fat and sugar but low in vitamins" ( WHO 2006). The CDC (2004) also finds evidence linking this epidemic to increased "consumption of food away from home; increased consumption of salty snacks, soft drinks, and pizza; and increased portion sizes." Since individual weight gain has been linked to increased caloric intake and decreased nutrient consumption, local environmental factors such as the distribution of food retailers have been documented to play an especially significant role in this growing epidemic. Research has shown that the suburb anization of healthy food stores has created food deserts', areas where socially disadvantaged individuals face barriers to accessing essential nutrients. Supermarkets tend to offer a large assortment of nutritious food at relatively inexpensive prices, w hich can directly influence healthier diet habits amongst

PAGE 12

2 ! customers. In the 1950s, the growth of automobile ownership, construction of interstate highways, and ensuing development of suburbs forced supermarket chains to leave the city center in order to st ay close to their customer base and maintain the retail space necessary to continue offering a variety of products (Larsen and Gilliland 2008). Consequently, residents of inner city and low income neighborhoods have fewer options for purchasing healthy, af fordable food and often must rely on smaller convenience stores that offer limited nutritious options at more expensive prices. Residents of urban neighborhoods often pay 3 to 37 percent more at these local grocers as compared to suburban customers who are purchasing the same items at larger supermarket chains (Morland et al. 2002). Additionally, a study in San Diego found that supermarket chains offer twice the average volume of heart healthy' foods as compared to neighborhood stores, and four times the a verage volume of these foods when compared to convenience stores (Morland et al. 2002). The location of supermarket chains in suburban areas that are far from urban city centers has played a pivotal role in creating inequities in accessibility to healthy f ood sources. T he fast food industry has also been a significant contributor to increased, often unhealthy, food consumption by people nationwide and has lead to the formation of food swamps', areas where socially disadvantaged individuals are overexposed to unhealthy food options This industry has grown rapidly in recent years as the number of fast food restaurants in the U.S. has increased from 30,000 in 1970 to 222,000 in 2001 (Paeratakul et al. 2003). Americans are also eating more fast food with the percentage of total calories consumed nationally growing from 3 to 20 percent within the past 20 years (Block et al. 2004). Eating fast food has been associated with a high fat diet and a higher

PAGE 13

3 ! body mass index (BMI) (Jeffery et al. 2006). Fast food chains are restaurants that offer affordable, convenient, and unhealthy food that many believe have contributed significantly to this epidemic. Recent research also indicates that neighborhoods with a higher fast food restaurant density and a higher ratio of fas t food to full service restaurants are more likely to have residents with higher BMI assessments and who are at a greater risk of being obese (Mehta and Chang 2008). Ultimately, the spatial location and distribution of both healthy and unhealthy food outle ts play a critical role in creating these unhealthy environments. Although many Americans are classified as obese, racial/ethnic minorities and lower income individuals are more likely to suffer from this condition. Blacks are 1.4 times and Hispanics are 1.1 times more likely to be classified as obese compared to non Hispanic Whites (The Office of Minority Health 2009). In terms of socioeconomic status, obesity is more prevalent amongst lower income women and adolescents than higher income women and adoles cents (Office of the Surgeon General [OSG] 2000; Healthy People 2010). These disparate health outcomes can be linked to unequal access to food entities. A study conducted in Los Angeles found that lower income ZIP codes with a predominantly Black populatio n have fewer healthy food options than higher income ZIP codes with a smaller Black population, both in terms of food preparation options and menu choices (Lewis et al. 2005). Additionally, less affluent neighborhoods have one third to one half fewer super markets as compared to more affluent neighborhoods but twice as many smaller grocery stores, implying that low income areas have fewer affordable and nutritious food options (Brown et al. 2008).

PAGE 14

4 ! The effect of the built food environment on adverse health o utcomes remains an important focus of public health policy, especially with regards to the potentially inequitable distribution of retail food stores in neighborhoods containing higher proportions of racial/ethnic minorities and low income individuals. Whi le previous empirical studies have made important strides in identifying and documenting the presence of food deserts' and food swamps', they have been limited in four critical ways. First, prior studies have examined the locations of either healthy or unhealthy food outlets in a specific geographic area, with respect to socioeconomic status or race/ethnicity (e.g., Block et al. 2004; Apparicio et al. 2007; Larsen and Gilliland 2008). An exclusive focus on one type of food source is unlikely to provide d etailed insights on the entire built food environment in a given study area. A more comprehensive analytical approach that includes both positive and negative entities is necessary to understand the adverse health and social implications of both food dese rts' and food swamps'. Second, there are several facets of the built food environment that must be fully evaluated in order to understand the effect of outlets on residents' dietary intake, including the nature and variety of food options available withi n that geographic area. However, only a few studies have employed more than one technique in order to provide complete insight into this complex, dynamic entity (e.g., Apparicio et al. 2007; Larsen and Gilliland 2008). Instead of treating all food outlets as equal in terms of their offerings, it is necessary to differentiate between them on the basis of food availability, nutritional content, and/or pricing.

PAGE 15

5 ! Third, most food desert' and food swamp' studies do not measure potential accessibility to food s tores from adjacent neighborhoods on the basis of the roadways actually used by residents to travel to these stores. Most studies utilize the count technique, which measures the number of food outlets that are coincidentally located within each neighborhoo d, or the buffer technique, which evaluates potential access by extending each neighborhood boundary by a specified distance and assuming that people travel in straight lines to food stores (e.g., Block et al. 2004; Baker et al. 2006; Powell et al. 2007). The use of network based distance methods that consider the spatial arrangement of streets within a neighborhood is necessary to accurately estimate potential geographic access to these food sources along walkable roadways. Lastly, previous research on th e built food environment has employed conventional statistical techniques such as linear correlation or multivariate regression to examine relationships between accessibility to food outlets and socio demographic characteristics of neighborhoods. These tec hniques may not be suitable for analyzing spatial data, because they fail to account for clustering of similar values over space or ignore local variations in statistical relationships within a study area. The use of global and local indicators of spatial association in food desert' and food swamp' research is necessary to fully account for geographic effects or processes that potentially influence the relationship between accessibility to food and neighborhood composition. This thesis aims to address th ese methodological gaps by examining the relationship between socio demographic characteristics of neighborhoods and access to both positive and negative food outlets in Hillsborough County, Florida. An important objective is to determine if and where supe rmarket chains are less accessible and fast

PAGE 16

6 ! food restaurants are more accessible to neighborhoods containing relatively higher proportions of socioeconomically disadvantaged individuals. The specific research questions to be investigated in this thesis are : 1. Is there a significant statistical association between socioeconomic deprivation and access to: (a) all supermarkets, (b) healthiest supermarkets, (c) all fast food restaurants, and (d) unhealthiest fast food restaurants, in this study area? 2. How does the nature and significance of the statistical associations between socioeconomic deprivation and access to: (a) all supermarkets, (b) healthiest supermarkets, (c) all fast food restaurants, and (d) unhealthiest fast food restaurants vary geographically withi n this study area? 3. How do the racial, ethnic, and locational characteristics of neighborhoods classified as food deserts' and food swamps' differ from those in the rest of the study area? This study utilizes precise locations of supermarket and fast foo d chains, census socio demographic data, and a combination of both conventional statistical measures and exploratory spatial data analysis (ESDA). Specifically, local indicators of spatial association (LISA) are used to identify neighborhood clusters that can be classified as food deserts' and food swamps', based on the significance of statistical relationships between accessibility and socioeconomic deprivation. The racial, ethnic, and locational characteristics of these food deserts' and food swamps' are then compared to those of neighborhoods in the rest of the county, to determine if racial/ethnic minorities are more likely to reside in these areas. Food outlet data is categorized based upon government codes assigned within the North American Indust ry Classification (NAICS) system and is obtained from

PAGE 17

7 ! ReferenceUSA, an online database of nationwide commercial business information. Relevant socio demographic information at the block group level is obtained from the 2000 U.S. Census to evaluate potentia l inequities in the built food environment. The first phase of the analysis utilizes traditional statistical techniques such as comparison of means tests and bivariate parametric correlations to explore the relationship between these two variables at the b lock group level. The second phase employs global and local spatial statistical measures for a more detailed examination of the geographic association between food outlet accessibility and socioeconomic deprivation, and to ultimately identify the location of food deserts' and food swamps' in the study area. The third and final phase employs proportional comparisons and binary logistic regression analysis to examine the demographic and locational characteristics of these food deserts' and food swamps', w ith respect to the rest of Hillsborough County.

PAGE 18

8 ! CHAPTER 2: LITERATURE REVIEW This chapter provides a systematic overview of the research literature on spatial accessibility to the food environment and its effect on adverse health outcomes, such as o besity. It begins by investigating the importance of ecologic factors such as obeseogenic environments, and then gives a critical assessment of: (a) studies that explore the creation and impact of the built food environment, and (b) studies that utilize st atistical or spatial analytic techniques to evaluate the b uilt food environment. This review offers insights on social inequities in the distribution of food locations as related to unfavorable health outcomes, and examines quantitative methods that can be used to describe the nature of the relationship between food outlets and neighborhood composition. 2.1. Obesogenic Environments and Health Outcomes Although the science behind individual weight gain can be partially explained by lifestyle choices such as increased caloric intake and decreased physical activity, these individual decisions are insufficient in explaining the surging, nationwide obesity epidemic (Weinsier et al. 1998; Huang and Glass 2008). An increase in BMI on a group level can more likel y be attributed to mass influences that affect the health outcomes of the population as a whole (Rose 2001; Huang and Glass 2008). Factors such as geographic location, social relationships, culture, and nature impact why people eat certain foods (Lake and Townshend 2006). However, while many of these components contribute to this alarming obesity epidemic, behavioral and environmental factors have

PAGE 19

9 ! been found to play an especially significant role in the widespread growth of this nationwide health crisis. N umerous current studies have placed a renewed emphasis on understanding the effect of the local environment (i.e., neighborhoods) on a person's health status, independent of individual lifestyle choices (Pearce et al. 2006). Recent years have seen a shift in public health research approaches regarding the effect of these ecological factors on adverse health outcomes, such as obesity (Lytle 2009). According to Lytle (2009), early studies focused on how individual perception of the surrounding environment aff ected health related choices, based upon work by philosophers Lewin and Bandura. Recent studies tend to concentrate on how the physical and social environment affects individual health either directly or by providing a framework for health related decision making (Lytle 2009; McKinnon et al. 2009). This research has portrayed neighborhoods as an important geographic context within which ecological factors related to the obesity epidemic can be examined (Block et al. 2004; Apparicio et al. 2007; Pearce et al 2007a, 2007b). Current empirical research in health geography has focused on evaluating the impact of unhealthy areas on the increased occurrence of obesity (Mehta and Chang 2008). One way to understand this relationship is to consider obesogenicity', which is defined as "the sum of influences that the surroundings, opportunities, or conditions of life have on promoting obesity in individuals or populations" (Swinburn et al. 1999 p. 564). While being spatially located within or near an obesogenic enviro nment does not guarantee weight gain, it does increase the probability that the individual or a group will become obese (Hill and Peters 1998). According to the Centers for Disease Control and

PAGE 20

10 ! Prevention [CDC], much of America's urban landscape can be char acterized as obesogenic', implying that these environments promote increased food intake, increased consumption of foods lacking essential nutrients, and decreased physical activity (CDC 2010). Therefore, the proliferation of these unhealthy food environm ents has potentially resulted in a nationwide increase of immediate and long term obesity related issues, which can include emotional and social hardships, severe chronic diseases, expensive health insurance costs, and even premature death. Obesogenic are as are created and promoted by a combination of cultural, political, physical, and economic factors that are present at a variety of scales interpersonal networks such as family and friends; local settings such as homes, schools, workplaces and neighborho ods; and larger scale sectors such as government, industry and society (Swinburn et al. 1999; Huang and Glass 2008; World Health Organization 2010). These identified environmental elements help researchers to evaluate the influence of rule structures, atti tude/belief systems, resource availability, and overall expense on these adverse health outcomes (Swinburn et al. 1999). Furthermore, the distribution and social composition of areas can lead to inequities in obesity rates between individuals, between diff erent racial, ethnic and socioeconomic groups, and between groups of individuals located in different neighborhoods (Pearce et al. 2007b). Although all these elements are significant, recent research has suggested that increased availability and consumptio n of food are major influences on the level of obeseogenicity of an area, and that the unequal distribution of the built food environment could expose certain socio demographic groups to these adverse health outcomes (Block et al. 2004).

PAGE 21

11 ! An evaluation of the built food environment to understand its relationship with socio demographic characteristics and its indirect affect on obesity requires an appropriate selection of both: (a) the concepts to be investigated, and (b) the methodology to be utilized. The following sections thus summarize the various theoretical and methodological concepts associated with the unequal distribution of food deserts' and food swamps' amongst socially disadvantaged neighborhoods. The first portion reviews the broader idea of t he built food environment, focusing on the creation of food deserts' and food swamps' as well as the occurrence of neighborhood level social inequities. The second part presents an overview of commonly used measurements, with a specific focus on the use of spatial analytical techniques for assessing potential accessibility. 2.2. Defining the Built Food Environment Researchers have struggled to develop a universal definition for the built food environment because it is a complex and multidimensional ent ity comprised of physical structures and nutritional influences. The built environment (physical structures) consists of buildings, stores, roads, and natural elements wherein people live, work, study, eat, and exercise whereas the food environment (nutrit ional influences) includes factors that impact what, where, and how much groups of people eat (Sallis and Glanz 2006; Story et al. 2008; Glanz 2009). A recent article by Glanz (2009) identified two important measurements associated with food structures at the neighborhood level: (a) community: the quantity, diversity, spatial location and accessibility of food outlets, and (b) consumer: the quality, affordability, and availability of healthy food and food related information within these outlets. It is impo rtant to focus on the built community nutrition

PAGE 22

12 ! environment to understand the accessibility of residents to healthy and unhealthy food in a particular study area. At the neighborhood level, the built food environment consists of two food access pathways: (a) food products that can be purchased for home consumption, and (b) ready made food that either can be eaten outside the home or brought back/delivered to the home to eat (Cummins and Macintyre 2006). Recent studies have evaluated commercial food locatio ns, such as supermarkets and fast food stores, which provide opportunities to buy and consume food in these two different ways (e.g., Cummins and McIntyre 2002; Block et al. 2004; Cummins et al. 2005; Apparicio et al. 2007). These physical locations influe nce the accessibility and availability of healthy and unhealthy food within a community, which can either make it easier or more difficult for residents to adhere to a nutritious diet (Story et al. 2008; Feng et al. 2010). Therefore, the complex built food environment can affect individual dietary intake and impact the likely occurrence of adverse health outcomes, such as obesity (Lewis et al. 2005; Feng, et al. 2010). 2.2.1. Creation & Impact of Food Deserts' and Food Swamps' Since the built food enviro nment varies amongst different neighborhoods, recent studies have focused on identifying disparities in levels of food accessibility and availability. In the early 1990s, a public housing project resident in west Scotland was supposedly the first person to use the term food deserts', according to Cummins and Macintyre (2002). In 1996, the United Kingdom Nutrition Task Force's Low Income Team utilized the term for the first time in a government publication, defining food deserts' as "areas of relative excl usion where people experience physical and economic barriers to accessing healthy food" (Reisig and Hobbiss 2000 p. 138; Cummins and

PAGE 23

13 ! Macintyre 2002). In recent years, researchers, policy makers, and community advocacy groups have used the term to locate an d analyze environments with these barriers, since their presence can make it more difficult for residents to maintain a healthy diet and weight (Lewis et al. 2005). Additionally, the overwhelming availability of unhealthy food in the surrounding environme nt can stimulate consumption, regardless of physical nutrition needs (Cohen and Farley 2008; Strum 2009). Recent literature has suggested the use of the term, food swamps' to define "areas in which a large relative amounts of energy dense snack foods, inu ndate healthy food options" (Rose et al. 2009; p. 2). These authors argue that the use of a metaphor, which has been used recently in media and policy reports to describe inequities in locations of unhealthy food restaurants, helps to highlight the equally strong, dual forces at play in the built food environment. Recent policy reports have embraced this term to understand high obesity rates that may stem from a combined lack of access to health food options and an overabundance of unhealthy food options (U .S. Department of Agriculture [USDA] 2009a; U.S. Department of Agriculture [USDA] 2010). Political legislation, economic dynamics, and residential migration patterns have contributed to the suburbanization of healthy food stores and the growing presence o f fast food restaurants, which have created an increasing number of urban food deserts' and food swamps' in America. The construction of the interstate highway system during the 1950s played a pivotal role in the unequal distribution of these food outlet s. The Interstate Highway Act of 1956 became the largest public works project in the nation's history and resulted in 46,000 miles of freeways being built from coast to coast (Schlosser 2001). This interstate highway system brought increased automobile use

PAGE 24

14 ! travel, and the movement of homes and businesses to the suburbs, ultimately causing widespread urban sprawl and downtown decline (Mormino 2001; U.S. Department of Agriculture [USDA] 2009a). By the 1970s, many businesses and stores including supermarkets had followed their customers in relocating from the city to the suburbs (Larsen and Gilliland 2008). This change in urban spatial structure allowed these healthy outlets to remain close to their customers, increase in size, and offer a larger variety of fo od, ultimately continuing to maximize profits (Larsen and Gilliland 2008). At the same time, the fast food industry spread across the nation as more entrepreneurs saw the benefit of operating restaurants that offered inexpensive food and convenient service handily located alongside customers' travel routes (Schlosser 2001). Nowadays, this industry "embodies the best and worst of capitalism," catering to an customer base seeking quick, cheap food and thriving from low overhead helped by employment of a low paid, unskilled workforce (Schlosser 2001 p. 8). The targeted placement of supermarkets and fast food restaurants within urban landscapes in the U.S. has led to a proliferation of food deserts' and food swamps' that can play a significant role in the ris ing obesity epidemic. 2.2.2. Distribution of Food Outlets & Neighborhood Social Inequities The distribution and strength of food deserts' and food swamps' can spatially vary between neighborhoods, unfairly enforcing or hindering the eating habits of dif ferent socio demographic groups residing in those local areas (Feng et al. 2010). A September 1998 speech by Donald Acheson, chairman of the publication Independent Inquiry into Inequalities in Health used food deserts' as an example of how neighborhoods where the underprivileged locate could be a significant contributor to poor

PAGE 25

15 ! health (Cummins and Macintyre 2002). Current research has shown that racial/ethnic minorities and socioeconomically disadvantaged residents are more likely to be located in areas that can be classified as food deserts' and food swamps' (Block et al. 2004; Apparicio et al. 2007). The barriers present in these areas are often linked with lower socioeconomic status because impoverished individuals have less mobility, both when consi dering short term factors such as access to transportation or long term effects such as inability to relocate neighborhoods (Apparicio et al. 2007). Most recent studies have focused primarily on positive food entities such as supermarkets, health food sto res, and farmer markets. For example, Larsen and Gilliland (2008) found that inner city neighborhoods with lower income populations are least accessible to healthy food outlets. Another study revealed that supermarkets are more likely to be located in most ly White neighborhoods than mostly Black neighborhoods in Maryland, Minnesota, Mississippi, and North Carolina (Morland et al. 2002). Finally, a comprehensive review of the food environment literature published between 1985 and 2008 indicated that communit ies with better access to supermarkets and less access to convenience stores tend to have healthier diets and lower tendencies toward obesity (Larson et al. 2009). Since fast food outlets have been linked to an increase in obesity, researchers have hypoth esized that their geographic location potentially exposes socially disadvantaged groups to unhealthy nutrition choices. Block et al. (2004) found that predominantly Black and lower income neighborhoods in New Orleans are more likely to contain a higher num ber of fast food restaurants per square mile. Another supporting study revealed food environment disparities in St. Louis communities by income and race

PAGE 26

16 ! when considering two factors: dietary guideline adherence by nearby restaurants and access to healthy f ood options (Baker et al. 2006). Finally, Pearce et al. (2007a) found significant and negative statistical associations between access to the nearest fast food locale and social deprivation in New Zealand, both when considering neighborhoods and schools. A lthough a limited number of studies have been conducted, the empirical evidence suggests that fast food restaurants are more likely to be located in and accessible to neighborhoods that contain higher proportions of racial/ethnic minorities and/or low inco me residents. Although only a few studies have evaluated the variety of food options in a particular geographic area, it is also necessary to consider more than the location pattern of all outlets to understand the built food environment. It is important to account for variety in the built food environment by assessing differences in food availability, food quality, and pricing (McKinnon et al. 2009). A neighborhood located near multiple healthy food outlets could potentially have limited access to nutriti ous food if these stores have less food merchandise options, lower quality of food, and/or higher prices than outlets located in other neighborhoods within the same study area. Based on the literature reviewed in this chapter, this thesis evaluates the pr esence and geographic distribution of food deserts' and food swamps', respectively. It examines access to both types of food retailers, supermarkets and fast food restaurants, with respect to socio demographic characteristics of neighborhoods as well as subset of these retailers, healthiest supermarkets and unhealthiest fast food restaurants. This focus provides a better understanding of spatial health inequities related to dietary habits and weight gain.

PAGE 27

17 ! 2.3. Evaluating the Built Food Environment The bu ilt food environment, which includes positive and negative health sources, is a complex and dynamic entity that can vary based on geographic location and societal influences. Additionally, this field of study is a relatively new consideration within health geography. McKinnon et al. (2009) found that more than 70 percent of all identified articles containing measures of the food environment were published within the last decade, between January 2002 and August 2007. As a consequence, food desert' and food swamp' research has struggled to develop and utilize universal, adequate measurements to evaluate the spatial distribution of food outlets. In 2006, the National Cancer Institute (NCI) formed a group of food environment experts who noted that this lack o f a systematic measurement is a problem that must be addressed in future research (McKinnon et al. 2009). A few years after this endeavor, an article by Sharkey (2009) identified four key challenges to accurately measuring the built food environment: (a) d efining the components of the food environment, (b) identifying all relevant healthy and unhealthy food sources, (c) evaluating variables that can be used to differentiate between the quality of, and access to, food sources, and (d) accurately locating all food sources. A recent review of food environment literature by Feng et al. (2010) also revealed conceptual and methodological limitations within this field because many researchers do not agree on issues related to data sources, food outlet definitions, and spatial extent of neighborhoods. Thus, it is necessary to develop and utilize valid measurement standards, which will help guide the further investigation and application of these results by academics and community groups (McKinnon et al. 2009).

PAGE 28

18 ! 2.3.1. Relevant Food Desert' and Food Swamp' Measurements This section reviews the spatial analytical techniques that have been used to evaluate food deserts' and food swamps' within the local environment in prior studies. These methods can help researcher s understand where individuals are most likely to purchase food (i.e., supermarkets and/or restaurants) and the various types of food to which they are most likely to have access to (Glanz 2009). McKinnon et al. (2009) also noted four parameters that most built food environment research can be categorized within: accessibility, availability, affordability, and quality. While all these categories are considered to be relevant, measuring accessibility between residents and nearby food entities has become a p ivotal focus of recent food desert' and food swamp' research. Handy (1992) believed that accessibility was the most effective way to identify travel patterns of residents within a geographic area (e.g., neighborhood) because this technique measures the a bility to easily reach certain activities and the magnitude of facilities in specific locations. Most geographic research has focused on this concept of accessibility, which has been defined as "the spatial distribution of activities about a point, adjuste d for the ability and desire of people or firms to overcome spatial separation" (Hansen 1959; p. 73). Almost one third (19) of the 63 built food environment studies (e.g. supermarkets, convenience stores, fast food restaurants) that were published between 2001 and 2008 evaluated spatial access to various healthy and unhealthy food outlets (Feng et al. 2010). Sharkey (2009) notes that there are two ways of conceptualizing accessibility: realized access (actual use) and potential access (closeness to facilit ies). Potential access has spatial and non spatial components, which can be understood by examining the cost

PAGE 29

19 ! of reaching these locations such as time, distance, and financial restrictions, or the attractiveness of those destinations (Handy et al. 1992; Sha rkey 2009; Feng et al. 2010). Ultimately, spatial aspects of the food environment (i.e. number, type, distribution, and location of food stores as well as distance to the residential areas) can serve as barriers or facilitators to real access in those area s (Sharkey 2009). Therefore, as explained further in Chapter 3, this thesis uses geographic analysis to investigate potential accessibility, in order to better understand the built food environment in Hillsborough County, Florida. 2.3.2. Spatial Accessibil ity Techniques According to Larsen and Gilliland (2008), quantitative food desert' studies often use one of two indices related to potential geographic access a container method' (identifying food stores within a geographic area) or a buffer approach (identifying food stores within a certain radius around a geographic area). Powell et al. (2007) used counts of full service restaurants and fast food restaurants for U.S. ZIP codes to examine the relationship between the number of outlets and various so cio demographic variables. A study by Baker et al. (2006) employed spatial clustering techniques to find the density of supermarkets and fast food restaurants in St. Louis, calculating the total amount of stores per 1,000 people within each area. Lastly, B lock et al. (2004) used a measure of areal density by calculating the number of fast food restaurants per square mile within a 0.5 mile and 1 mile buffer around census tracts in New Orleans. However, these techniques have been criticized for not adequatel y accounting for how individuals actually travel to these food retail locations. Larsen and Gilliland (2008) note that the container method' has been referred to as spatial coincidence' because it only measures the food outlets that are unintentionally l ocated within a chosen area or

PAGE 30

20 ! pre defined administrative unit. Meanwhile, the buffer approach' has been criticized because it works on the basis that people travel in straight line paths or as the crow flies', without accounting for the presence of real life barriers (Larsen and Gilliland 2008). Therefore, recent studies have aimed to evaluate proximity between residences and the built food environment by utilizing street networks within the network distance method. Many studies measure network accessib ility by utilizing either the geographic center or population weighted centroid of the spatial area. Pearce et al. (2007b) calculated accessibility as the travel time between the population weighted centroid of each neighborhood in New Zealand and each foo d resource, while factoring variations in speed limits, road surface, and topography. Another study conducted along the Texas Mexico border also used the population weighted centroid as a reference point, computing the shortest network distance to the near est food store as a measurement (Sharkey et al. 2009). Lastly, Zenk et al. (2005) calculated the distance between the geographic center of the census tract and the nearest supermarket to evaluate food accessibility in Detroit. However, there are certain li mitations associated with both these techniques. The geographic center method ignores the spatial distribution of the population inside each unit of analysis and assumes the entire population of the spatial unit to be located at or near the centroid, while the use of the population weighted centroid does not consider the accessibility of less populated areas in those spatial units (Apparicio et al. 2008). A few studies have employed a combination of techniques, including network based distance or time, in order to quantify accessibility within the dynamic food

PAGE 31

21 ! environment more comprehensively. Apparicio et al. (2007) employed three measures to evaluate accessibility of healthy food outlets in Montreal: distance to the closest supermarket, number of supermar kets within 1,000 meters (defined as a walkable distance), and average distance to three closest different supermarket chains. Another study conducted in Canada also used three measurements: percent of neighborhood geographic centers that fall within super market service areas for walking and public transit, distance to the closest supermarket, and number of supermarkets within 1,000 meters (Larsen and Gilliland 2008). Ultimately, this thesis uses a combination of techniques to evaluate the potential access to food sources in Hillsborough County, Florida. 2.3.3. Exploratory Spatial Data Analysis Recent geographic literature has focused on the need to implement specialized techniques that are more suited for the analysis of spatial data, variables, and relatio nships. In the past, many studies utilized Exploratory Data Analysis (EDA), a set of statistical techniques that help to reveal existing patterns, highlight unusual or interesting features, distinguish accidental from important occurrences and guide hypoth esis formation (Anselin and Getis 1992; Haining et al. 1998). Location, a central component to spatial data both in terms of its absolute location (e.g. associated latitudinal/longitudinal coordinates) and its relative location (e.g. relationship to surrou nding administrative units), affects how analysis can be conducted (Anselin 1993). Spatial data sets frequently conform to Tobler's First Law (TFL) of Geography that says "everything is related to everything else, but near things are more related than dist ant things" (Tobler 1970, p. 236). The practical implication of TFL is that observations from

PAGE 32

22 ! nearby locations are often more similar than would be expected on a random basis (Chakraborty 2011). This concept, often called spatial dependence (positive spati al autocorrelation), increases the likelihood for finding similar values between neighboring entities within the same study area (Charreire and Combier 2008). In the presence of significant spatial autocorrelation, units of analysis do not satisfy the key assumptions of independence and homogeneity necessary to implement classical statistical techniques (Anselin 1993). In order to better evaluate and account for these intrinsic spatial traits, recent research has employed Exploratory Spatial Data Analysis ( ESDA) techniques. ESDA is defined as a statistical study of phenomena that manifest themselves in space" focused on aspects such as location, area, topology, spatial arrangement, distance, and interaction (Anselin 1993). ESDA is an extension of EDA that helps expose overarching spatial patterns, formulate hypotheses based on/about the geography of the data and measure spatial models (Haining et al. 1998). According to Anselin (1993), this methodology is beneficial as it helps researchers to find patterns of spatial association (i.e., clustering and dispersion), identify atypical observations (i.e., outliers) and suggest forms of spatial instability (i.e., non stationarity). ESDA methodology falls into two categories: (a) global techniques that focus on the entire study area to help identify spatial dynamics such as clustering, and (b) local techniques that focus on the subsets of the study area in order to uncover neighborhood properties such as the location of clusters (Haining et al. 1998). Although globa l statistics can measure clustering in the study area, they fail to account for different levels of spatial autocorrelation occurring across different neighborhoods, especially when there are a

PAGE 33

23 ! large number of observations (Anselin 1993, Anselin 1995). Loc al indicators of spatial association (LISA) are often seen as a way to identify these disparities by analyzing the relationship between two variables to highlight statistically significant clusters and outliers at a local level (Hare and Barcus 2007). Acco rding to Anselin (1995), a LISA is any statistic that meets the following two requirements: (a) the LISA indicator for each unit of analysis gives an indication of the amount of spatial clustering of similar values around that area, and (b) the sum of all LISA indicators in the study area is proportional to the global spatial indicator. Localized spatial techniques help researchers to explore individual subsets as related to the overall geographic unit and to discover areas of similar or dissimilar values, helping uncover complicated spatial relationships (Unwin and Unwin 1998). Use of these statistics helps to increase confidence in interpreting spatial patterns of data (Hare and Barcus 2007). Few empirical studies on the built food environment have examine d the geography of store location or access using LISA or other localized spatial statistical approaches. Zenk et al. (2005) used Moran's I to test for spatial autocorrelation and moving average spatial regression to adjust for this spatial clustering in o rder to identify localized food deserts'. Another study also employed Moran's I the Geographical Analysis Machine (GAM), and geographically weighted regression (GWR) to reveal spatially variability in food consumption and food poverty in Ecuador (Farrow et al. 2005). Apparicio et al. (2007) identified unusual areas of low accessibility to supermarkets and high social deprivation in Montreal with Moran's I and hierarchical cluster analysis but did not examine local variations in this study area. To add to this small but growing built food environment literature, this thesis employs bivariate

PAGE 34

24 ! measures of both global and local spatial statistics to further explore the nature of the relationship between food outlet accessibility and socioeconomic status in Hil lsborough County, and to classify neighborhoods as food deserts' and food swamps' based on the significance of the spatial association between these variables. 2.4. Summary This literature review has explored the ecological factors that affect the obe sity epidemic, specifically focusing on the conceptualization and evaluation of the built food environment. Although previous studies have made important strides in identifying and documenting the presence of food deserts' and food swamps' in specific ur ban areas, this research has been limited by three methodological limitations. First, most studies have not compared healthy and unhealthy food outlets in a geographic study area, providing a limited or partial understanding of the built food environment. Second, recent studies have relied on the count or buffer techniques, and have not utilized network based proximity measures to evaluate potential access based on walking distances and routes. Third, researchers have failed to use a combination of methods to analyze the variety of nutritious items and prices available in the multifaceted built food environment. Finally, built food environment research has utilized standard statistical methods, instead of local ESDA techniques that are more appropriate for a nalyzing geographic data and relationships. This thesis aims to address these four important gaps in previous food desert' and food swamp' research by including healthy and unhealthy food entities in the sample, and evaluating the variety of the built f ood environment in Hillsborough County.

PAGE 35

25 ! Additionally, the research makes an important empirical contribution to food desert' and food swamp' research by employing multiple network based measurements to evaluate spatial accessibility. Finally, the thesis utilizes bivariate measures of global and local spatial statistical techniques to evaluate the relationship between access and socioeconomic deprivation, and to classify food desert' and food swamp' locations more accurately. The following chapter outlin es the data sources, variables, and methods used in this case study of food deserts' and food swamps', respectively.

PAGE 36

26 ! CHAPTER 3: DATA AND METHODOLOGY This chapter describes the study area, data sources, variables, and methods utilized for the study The first section introduces the study area evaluated in this study. The second section provides the operational definitions used to derive the set of supermarket and fast food chains in this study area, as well as the subset containing the healthiest su permarket and unhealthiest fast food chains in this study area. The third section describes the various dependent, explanatory and descriptive variables, and their data sources. Statistical techniques for exploring the relationship between the dependent an d explanatory variables are discussed in the final section. 3.1. Study Area T here is a growing need to examine the relationship between food deserts, food swamps, and socioeconomic characteristics in metropolitan areas of Florida an area that is relativ ely understudied in terms of its food environment and related health implications. Previous studies have focused on other national and international communities, but few have investigated the obesity crisis and its causes in the populous Sunbelt Region. Fl orida's growth has been extraordinary as its population surged by 76 percent between 1970 and 1990, compared to the nation's population growth of 21 percent during the same time period (Mormino 2002). Florida's growing population has also been affected by the recent food crisis epidemic. According to the Food Research and Action Center [FRAC] (2010), Florida's

PAGE 37

27 ! national rank jumped from 24th in 2008 to 12th in 2009 when considering food hardship' (the inability to pay for food). A recent U.S. Department of Agriculture report [USDA] (2009b) found that between 2004 2006 to 2007 2009, Florida had one of the largest increases in prevalence rates nationwide of low food security', which is defined as a household's consistent access to the food necessary to maint ain a active, healthy lifestyle. Additionally, food stamp usage in Florida has increased by a staggering 70 percent between 2007 and 2009 (Bloch et al. 2009). The state also is plagued by the obesity epidemic currently facing the nation. According to data from the State Behavioral Risk Factor Surveillance System (BRFSS), the prevalence of obesity among adults at least 20 years of age in Florida nearly doubled in recent years, from 9.8 percent in 1986 to 19.4 percent in 2002, which is consistent with the na tional figures (Florida Department of Health, 2004). In the same time period, the percent of Florida adults at least 20 years of age classified as overweight' increased from 35.3 percent in 1986 to 57.4 percent in 2002, which is also consistent with the n ational increase. Additionally, Florida was one of a few states that had large percentage increased in obesity prevalence amongst adolescents, according to results from the 2003 and 2007 National Survey of Children's Health (NSCH) (Singh et al. 2010). The study area for this thesis, Hillsborough County, is a likely focal point of this statewide food crisis and is shown in Figure 3.1. Hillsborough County is the fourth most populous county in Florida and the one with the largest population in the Tampa Bay (T ampa/St. Petersburg/Clearwater) metropolitan statistical area (Office of Economic & Demographic Research [EDR] 2009). While this urban county accounts for almost 2 percent of Florida's land area, it is contains over 6 percent of its population, according t o

PAGE 38

28 ! the 2000 U.S. Census. Hillsborough County also finds itself at the center of the national and state food crisis. According to the Florida Department of Children and Families, the number of households in Hillsborough County receiving food stamps more than double from 2004 to 2010, increasing from 60,721 to 151,802 (Hillsborough Community Atlas 2011). Furthermore, according to the 2007 County BRFSS, 39.4 percent of adults at least 20 years of age were classified as overweight' and 24.8 percent of adults we re classified as obese', totaling 64.8 percent of the population (Florida CHARTS 2010). These figures place Hillsborough County in the top 50 75 percent of all Florida counties with regards to the total percent of adults who fall in range of overweight' to obese' (Florida CHARTS 2010). Therefore, Florida, and more specifically Hillsborough County, is a pivotal location for a case study of healthy and unhealthy built food environments.

PAGE 39

29 ! Figure 3.1. Hillsborough County, Florida 3.2. Data Sources and Var iables An important first step was to develop a consistent operational definition for both healthy and unhealthy food retailers in Hillsborough County, Florida. Both the supermarket and fast food outlets for the study were defined by utilizing their North American Industry Classification System (NAICS) Description code, a standard created and employed by the federal government to catalog business establishments since 1997 (U.S. Census Bureau 2009). The U.S. Census Bureau, which is the primary agency respon sible for collecting and analyzing statistical data on the nation's economy, assigns one NAICS code to each establishment based on its highest revenue generating activity

PAGE 40

30 ! (U.S. Census Bureau 2009). However according to the U.S. Census Bureau, other agenci es have adopted the NAICS system and allow businesses to be classified under different or multiple codes for their own purposes. While it would be ideal to use only the official identification code administered by the U.S. Census Bureau, this information i s not publicly available. To ensure that the built food environment in Hillsborough County was adequately captured, both the Primary NAICS and NAICS 1 Description were used to define food retailers within the study area. Parameters established in recent s tudies of the food environment were employed to identify supermarkets and fast food restaurants by their 2007 NAICS codes (e.g., Morland et al. 2002; Bader et al. 2010; Stein and Chakraborty 2010). As seen in Table 3.1 below, retailers with a Primary NAICS or NAICS 1 Description code of 44510 (Supermarkets) were classified as a supermarket', while businesses needed a Primary NAICS or NAICS 1 Description code of 722211 (Limited Service Restaurants) to be classified as a fast food restaurant' (U.S. Census B ureau 2009). If one location of a food chain that fit this criterion had a different code in either or both fields, it was viewed as a bureaucratic mistake, and the location and the chain were retained. In the same vein, if one location of a food chain did not fit this criterion, it was view as a bureaucratic error and the location was removed.

PAGE 41

31 ! Table 3.1. North American Industry Classification System Codes for Food Outlets (U.S. Census Bureau 2009) Industry Group 2007 NAICS Code 2007 NAICS Code Definitio n Supermarkets 44510: Supermarkets This industry comprises establishments generally known as supermarkets and grocery stores primarily engaged in retailing a general line of food, such as canned and frozen foods; fresh fruits and vegetables; and fresh an d prepared meats, fish, and poultry. Included in this industry are delicatessen type establishments primarily engaged in retailing a general line of food. Fast food Restaurants 722211: Limited Service Restaurants This industry comprises establishments pri marily engaged in providing food services (except snack and nonalcoholic beverage bars) where patrons generally order or select items and pay before eating. Food and drink may be consumed on the premises, taken out, or delivered to the customer's location. Some establishments in this industry may provide these food services in combination with selling alcoholic beverages. Next, each chain was called to verify that they did not fall in other NAICS categories as identified by Bader et al. (2010): ethnic gr ocery stores (small specialized food stores, such as those selling food from a specific part of the world) or full service restaurants (food service locations whose customers can order from a wait staff). If a chain appeared to conduct most of its business in one of these other NAICS categories, it was also removed. Additionally, to be included in the samples as national corporate owned chains, these retailers needed at least one store location outside Hillsborough County, verified using the company website s (e.g., Morland et al. 2002; Block et al. 2004). This criterion resulted in two samples that include the majority of the healthy and unhealthy food market in Hillsborough County.

PAGE 42

32 ! Individual food retail locations were obtained from ReferenceUSA, a web bas ed resource that compiles residential and business data nationwide from more than 5,000 public sources. This database, accessible via the University of South Florida's library website, allows the user to search within 26 NAICS codes and by geographic locat ion and includes the latitude/longitude coordinates for all retail locations. Current data (updated within the past 12 month time period 2010) for all locations within Hillsborough County and its five bordering counties (Hardee, Manatee, Pasco, Pinellas, a nd Polk) was downloaded from the website. Extraneous information was eliminated and only locations that fit the NAICS code and location criteria discussed above remained. The analytical capabilities of geographic information software (ArcGIS version 9.3.1) were then be used to geocode the location of each outlet, based on these street level latitude and longitude coordinates. Supermarket and fast food chain locations that were either inside or within 1,000 meters of the county boundary were included in the final sample to account for the fact that outlets located immediately across the border are likely to be visited by residents of Hillsborough County. Previous studies have recommended a 1,000 meter buffer around census units as the preferred walking dista nce for analyzing access to fast food restaurants (e.g., Smoyer Tomic et al. 2006; Apparicio et al. 2007; Larsen and Gilliland 2008). According to Apparicio et al. (2007), 1,000 meters is an approximately 15 minute walk for an adult living in a metropolita n area. As shown in Tables 3.2 and 3.3, this analysis included 11 supermarket chains and a total of 115 supermarket outlets along with 40 fast food restaurant chains and total of 513 fast food outlets located in this area.

PAGE 43

33 ! Table 3.2. Supermarkets in Hill sborough County, Florida. Supermarket Chain Name Number Albertsons 3 ALDI 4 Bravo Supermarkets 2 Publix 43 Save A Lot 10 Sweetbay Supermarket 30 The Fresh Market 2 U Save Supermarket 5 Walmart Neighborhood Market 3 Whole Foods Market 1 Winn Dix ie 12 Total 115

PAGE 44

34 ! Table 3.3. Fast Food Restaurants in Hillsborough County, Florida. Restaurant Chain Name Number A&W Restaurant 1 Arby's 12 Baja Fresh 1 Blimpie 12 Boston Market 8 Burger King 23 Charley's Grilled Subs 2 Checkers 16 Chick fil A 13 Chipotle Mexican Grill 4 Church's Chicken 4 CiCi's Pizza 8 Domino's Pizza 16 Firehouse Subs 7 Five Guys 11 Godfather's Pizza 9 Hardee's 4 Hungry Howie's Pizza 26 Jersey Mike's Subs 3 Jet's Pizza 1 Jimmy John's 4 KFC (Kentucky Fried C hicken) 19 Krystal 2 Lenny's Sub Shop 3 Little Caesars 14 Long John Silver's 4 Maryland Fried Chicken 1 McDonald's 56 Panda Express 4 Panera Bread 9 Papa John's Pizza 15 Pizza Hut 24 Popeyes 9 Quiznos 9 Sbarro 3 Sonic 11 Subway 90 Taco Be ll 21 Wendy's 30 Zaxby's 4 Total 513

PAGE 45

35 ! While all food stores were treated equally in the first phase of the study sample the second sample evaluate d neighborhood accessibility to the healthiest' supermarket and unhealthiest' fast food and chains, def ined in terms of product quality and variety. Current media resources were utilized to select these subsets because these sources have adequately evaluated the quality of these national chains. The three healthiest supermarket chains in Hillsborough County were selected utilizing a Health magazine article, featured on the Today show, which identified the top 10 healthiest supermarkets in America (Paul 2008). A panel of six renowned health experts reviewed the 35 largest chains in the nation and evaluated th e following: taste of prepared food, freshness of produce, healthiness of packaged goods, and availability of nutritional information. The top three healthiest supermarkets from that list, located in Hillsborough County, were: Whole Foods (1 st in the natio n), Albertsons (6 th in the nation), and Publix Super Markets (8 th in the nation). The three unhealthiest fast food chains in Hillsborough County were selected using a classification scheme from a prominent book, Eat This, Not That!: The Best (& Worst) Food s in America! (Zinczenko and Goulding 2009). In an updated version of the study by Zincenko (2009), 66 major chain restaurants were graded from A to F based on the total number of calories per entrŽe along with an analysis of specific menu items: fruits/ve getables, whole grain bread, trans fat food, and high fat desserts. The recent article gave the grade of F to the lowest ranked chains and/or businesses that did not provide nutritional information. Following this approach, the final subset of unhealthiest fast food restaurants in Hillsborough County based on those that provided nutritional information comprised of: Baja Fresh (graded D ), Pizza Hut (graded D ), and

PAGE 46

36 ! Panera (graded D). The 47 healthiest supermarkets and 34 unhealthiest fast food chains in th e study area are listed in Tables 3.4 and 3.5. Table 3.4. Healthiest Supermarkets in Hillsborough County, Florida. Supermarket Chain Name Number Albertsons 3 Publix 43 Whole Foods Market 1 Total 47 Table 3. 5 Unhealthiest Fast Food Restaurants in H illsborough County, Florida. Restaurant Chain Name Number Baja Fresh 1 Panera Bread 9 Pizza Hut 24 Total 34 3.2 .1 Accessibility to Food Outlets The food environment, which includes positive and negative health entities, is a complex and dynamic e ntity that can vary based on geographic location and societal influences. Sharkey (2009) notes that there are several methods used to analyze the built food environment: density that includes ratio of food outlets by geographic unit, proximity that include s distance to these outlets, and variety that includes differences in price, menu, and preparation. A study by Feng et al. (2010) also found that studies that defined neighborhoods with administrative boundaries (i.e. census units) used one of three spatia l access methods identified by Sharkey. Another literature review by McKinnon et al. (2009) indicated that food environment studies employing geographic analysis techniques (which account for 68 out of 137 articles published from January 1990 to August 200 7) utilized one or more of the aforementioned Sharkey's accessibility

PAGE 47

37 ! measurements. Therefore, both proximity (percent of each block group accessible to supermarket and fast food chains) and variety (percent of each block group accessible to the healthiest supermarket chains and unhealthiest fast food) were selected as the most appropriate measurements to represent the dependent variables for this study. A total of four dependent variables were analyzed, as summarized in Table 3.6. Census block groups repr esent the unit of analysis for the study because it is the smallest unit or finest spatial resolution at which the U.S. Census publishes data on the socioeconomic characteristics of the residential population. Previous studies have used this geographic uni t to evaluate accessibility to healthy and unhealthy food entities at a neighborhood level (e.g., Raja et al., 2008; Sharkey et al. 2009; Feng et al. 2010). Geographic accessibility to both healthy and unhealthy food outlets was determined by calculating the service area surrounding each store, which is a spatial representation of walking distance between the residents and their nearest food sources. For this purpose, it was important to use a realistic walking distance to accurately reflect travel pattern s and account for edge effects. In this study, distance along the road network was utilized to account for the way people actually travel on streets as compared to the Euclidean straight line distance, which often provides an inaccurate representation of p otential access (Witten et al. 2003). Additionally, this method accounts for edge effects, the possibility that a facility could be located so close to the edge of a census unit that its immediate and effective neighborhood contains portions of other neigh boring block groups. Furthermore, the specific use of a walking distance instead of a driving distance helps to negate the effect of private car ownership, which can be linked to lower socioeconomic standing, on access to food outlets.

PAGE 48

38 ! The service area fo r each location was calculated using a distance of 1,000 meters from the store along each roadway (excluding Interstate highways), which is approximately a 15 minute walk for an adult in an urban setting (Apparicio et al. 2007). Network based distance has been identified as the most suitable measurement between two points by foot and is frequently used to analyze food access (Apparicio et al. 2007; 2008). Interstate highways were excluded from this analysis since they do not contain sidewalks or provide wal king access to the food retail locations. The network analysis capabilities of GIS software were utilized to construct polygons that follow street segments (1,000 meter distance) around each food outlet, representing the area that is accessible by foot to these facilities. This analysis was conducted using the Network Analyst extension in the ArcGIS software (version 9.3.1) and the most recent street network data, 2008 Census TIGER/Line Files, were obtained from the Florida Geographic Data Library (2010). For each census block group, the proportion of the block group area contained within the service areas surrounding food stores was calculated and used as a measure of access. This analysis was conducted for the supermarket and fast food chains, and their r espective subsets. These four variables provide a detailed and comprehensive assessment of a neighborhood's potential accessibility to the healthy and unhealthy built food environment in Hillsborough County, Florida, as indicated in Table 3.6. At the block group level, food outlet accessibility varies from 0 (no access to that type of food outlet in that block group) to 1 (100 percent areal access to that type of food outlet in that block group).

PAGE 49

39 ! Table 3.6. Definitions for Accessibility to Food Outlets De pendent Variable Definition Accessibility to Supermarkets Area of each block group that is within 1,000 meters of a supermarket divided by the total block group area (km 2 ) Accessibility to Healthiest Supermarkets Area of each block group that is within 1,000 meters of the top three healthiest' supermarkets divided by the total block group area (km 2 ) Accessibility to Fast food Restaurants Area of each block group that is within 1,000 meters of a fast food restaurants divided by the total block group a rea (km 2 ) Accessibility to Unhealthiest Fast food Restaurants Area of each block group that is within 1,000 meters of the top three unhealthiest' fast food restaurants divided by the total block group area (km 2 ) 3.2 .2 Socioeconomic Deprivation Index The primary explanatory variable for the analysis of food outlet accessibility was the Socioeconomic Deprivation (SED) Index, which was used to determine the presence of food deserts and food swamps, respectively. The data used to formulate this index wa s derived from the 2000 U.S. Census at the block group level. Although the socio demographic information comes from a different year (2000) than the food retail coordinates (2010), it is considered to be the most reliable data source for this research by p revious food environment studies (e.g., Block et al. 2004; Bader et al. 2010). Several socioeconomic variables related to wealth, income, and economic status are necessary to understand the relationship between the food environment and neighborhood compos ition. While previous studies have incorporated these factors as separate independent variables, recent studies on food deserts and food swamps have used an index to more accurately account for the complexities inherent in measuring and analyzing a neighbo rhood's socioeconomic status (e.g., Apparicio et al. 2007; Larsen and

PAGE 50

40 ! Gilliland 2008; Sharkey et al. 2009). Using the parameters established by Apparicio et al. (2007) as a guideline, the study combined five census variables to formulate a SED index at the block group level: (a) median household income, (b) proportion single parent household, (c) proportion unemployment (individuals who are at least 16 years old, in the civilian labor force and are jobless), (d) proportion lower education (individuals who a re at least 25 years old and have a 9 th grade education or less), and (e) proportion recent immigrants (foreign born individuals who arrived between 1995 and March 2000). The indexing methodology suggested by Cutter et al. (2000) was used to standardize t hese individual variables on a scale ranging from 0.0 to 1.0. Values of all census variables, except median household income, were standardized based on the following steps: (a) the total number in each block group was computed (X) ; (b) the total number in the county was summed (Y) ; (c) the total number in the block group (X) was divided by the county total (Y) to estimate a proportion for each block group (Z) ; (d) the highest or maximum value of this proportion across all block groups in the country was id entified ( Zmax ); and (e) the proportion in each block (Z) was divided by the maximum ratio in the county (Zmax) Since median household income is not reported as an absolute number at the block group level, this variable was standardized as follows: (a) me dian household income of each block group was subtracted from the median household income of the county (X) ;

PAGE 51

41 ! (b) the difference (X) was added to the absolute value of the maximum difference (X) to account for negative values (Y) ; (c) the highest or maximum value of this sum across all block groups in the country was identified ( Ymax ); and (d) the absolute value of the sum (Y) was divided by the maximum possible value of the sum (Ymax) Following Apparicio et al. (2007), the five standardized variables were then summed as done by Apparicio et al. (2007) to create an aggregated SED index ranging from 0.0 (minimum deprivation) to 5.0 (maximum deprivation). The use of this index helps to account for multicollinearity between individual explanatory variables and provide a more comprehensive depiction of neighborhood socioeconomic distress. 3.2.3. Racial, Ethnic, and Locational Characteristics This study included two other categories of explanatory variables: race/ethnicity and neighborhood locational characteris tics. The definitions for these variables are provided in Table 3.7. To estimate the racial and ethnic composition of neighborhoods, it was necessary to include both Black and Hispanic populations as they constitute the two largest minority groups in Hills borough County (Mormino 2002). These racial and ethnic groups comprise almost 22 percent and 17 percent of the county's population in 2009, respectively (U.S. Census Bureau 2010). Previous studies on the local food environment have also analyzed the presen ce of both minority groups, who are more likely to be subjected to food location disparities (e.g., Mehta and Chang 2008; Larson et al. 2009). Other minority populations (e.g., Asians) were not included because they comprised less than 6 percent of the tot al county population (U.S. Census Bureau 2010). The White

PAGE 52

42 ! population was used as a reference race/ethnicity variable when analyzing characteristics of the food deserts and food swamps (e.g., Metha and Chang 2008). The proportion of individuals in each bloc k group identifying themselves as non Hispanic White (e.g., Bader et al. 2009), non Hispanic Black (e.g., Moore and Diez Roux 2006) and Hispanic were included for the descriptive statistics and regression analysis. Three variables describing locational c haracteristics were included to control for the role played by neighborhood population and local planning strategies in the relationship between food accessibility and socioeconomic factors. These variables are: population density (persons per square mile) the presence of major highways and proportion of commercial zoning. The inclusion of population density is supported by previous studies suggesting that food outlets, which require a certain customer base to be profitable, are more likely to be located i n neighborhoods that are more densely populated (e.g., Apparicio et al. 2007; Larsen and Gilliland 2008). Additionally, lower population density has been linked with neighborhood deprivation, minority population and ultimately, poor eating habits and a hig her risk of obesity (Larson et al. 2009). The population density variable was calculated as the total block group population divided by the area of each block group in square kilometers. Highways are also an important consideration for this research as th ey may influence the location of unhealthy and healthy food entities (Block et al. 2004). The fast food industry began and grew alongside the interstate highway system boom of the 1950s, which also propelled supermarkets to migrate towards the suburbs wher e there is more land and more convenient access to customers traveling by car (Schlosser 2001; Pothukuchi 2005). A qualitative variable was used to account for the presence of

PAGE 53

43 ! Interstate, U.S. and/or State highways within 1,000 meters of each block group ( Apparicio et al. 2007). The digital representation of the highway network for Hillsborough County was the most recent Census TIGER/Line Street Files (2000 and 2002), which classifies each street segment by type. Following the parameters utilized by Block e t al. (2004), this dichotomous variable was coded as 1' if the block group was intersected by at least one type of highway and coded as 0' if no highways were present. It was also necessary to evaluate land use more explicitly to understand how governm ent regulations and local bureaucratic processes affect the distribution of food deserts and food swamps in Hillsborough County. Zoning, one of the government's most important tools for planning and the protection of public health, determines where retaile rs may locate and can play a significant role in the equitable distribution of health resources such as food stores (Maantay 2001; Mair et al. 2005). However, a Congressional report by the U.S. Department of Agriculture [USDA] (2009a) found that local zoni ng regulations may be more burdensome and expensive in densely populated, lower income urban areas, and therefore may pose barriers to food retailers seeking to locate in underserved neighborhoods. Healthy food retailers have cited zoning as one of the top challenges to their industry, claiming that these requirements factor significantly into their site selection processes. Fast food restaurants are also more likely to be located in predominantly commercial areas due to zoning restrictions (Block et al. 20 04; International Council of Shopping Centers and Social Compact 2008; New Orleans Food Policy Advisory Committee 2008). Although zoning data has not been used in food desert research, previous studies emphasize its importance by mentioning it as a study l imitation and/or advocating for these laws to be changed to address food location

PAGE 54

44 ! disparities (e.g., Papas et al. 2007; Pearce et al. 2007a; Powell et al. 2007). The use of zoning data in this thesis was clearly necessary to conduct a comprehensive assessm ent of how the locations of food retailers affect neighborhood accessibility, and to advocate for future policy reform. The most recent zoning shapefiles and background information were obtained from the Hillsborough County's Planning and Growth Managemen t Division, the Hillsborough County City County Planning Commission (City of Plant City data), the City of Plant City's Planning & Zoning Division and GIS Coordinator in the Engineering Division, the City of Tampa's GIS Section and its Land Development Coo rdination Division, and the City of Temple Terrace's GIS Specialist in the Community Services Division. Zoning classification varied amongst the four entities in Hillsborough County, the county and the three cities within its boundaries (Plant City, City o f Tampa, Temple Terrace). The ultimate goal was to isolate the primarily commercial zoning areas in each geographic area. In Hillsborough County, zoning had to be classified as Commercial General', Commercial Intensive' or Commercial Neighborhood' in the Commercial/Office/Industry' category. Plant City's neighborhoods had to be classified as General Commercial District (C 1 or C 2)' or Neighborhood Business District (C 1A, C 1B or C 1C). In the City of Tampa, only the following classifications w ere used: Commercial General (CG)', Commercial Intensive (CI)', Community Neighborhood (CN)', or Community Commercial (CC)'. And lastly for Temple Terrace, areas had to be Commercial General (CG)' or Commercial Office (CO)'. Since zoning areas do not have the same geometry as census block groups, the final variable was the proportion of the area of each block group classified as a commercial' zone. Although block groups

PAGE 55

45 ! are predominantly zoned for commercial activity, they may still contain large res idential populations and/or affect neighboring residential areas (Maantay 2001). Table 3.7. Additional Explanatory Variables Explanatory Variable Definition Percent non Hispanic White Percent of non Hispanic individuals identifying their race as White in each block group. Proportion non Hispanic Black Number of non Hispanic individuals identifying their race as Black divided by the total block group population Proportion Hispanic Number of self identified Hispanic/Latino residents (of any race) divided by the total block group population Population Density The total population of a block group divided by the block group area in square kilometers Highways Presence The presence (coded as 1) or absence (coded as 0) of a Interstate, U.S. and/or State high ways within 1,000 meters of each block group Commercial Land Use Total area of each block group classified as primarily a commercial' zone divided by the block group area in square kilometers. 3.3 Statistical Analysis Methodology To examine the rela tionship between the SED index and access to fast food restaurants and supermarkets, and to identify the location and demographic characteristics of food deserts and food swamps, this thesis uses a combination of traditional parametric and spatial statisti cal analysis. The first phase of the analysis focuses on exploring the relationship between the SED index and accessibility to the four types of food outlets, describing the distribution and nature of each variable. Graphs and independent sample t tests de pict how the mean value of SED index differs by level of accessibility to (quintile of) food outlets. Additionally, bivariate parametric correlations

PAGE 56

46 ! provide preliminary insights on the statistical association between food outlet accessibility and socioeco nomic deprivation at the county scale. In the second phase, ESDA techniques are employed to measure and visualize how varying magnitudes of the SED index are related spatially to varying proportions of accessibility to food outlets. The significance of sp atial correlations between these variables is used to delineate the boundaries of food deserts and food swamps in Hillsborough County. The spatial statistical techniques employed include the bivariate global Moran's I bivariate Moran I scatter plots, and bivariate significance maps that represent a local measure of the Moran's I Spatial contiguity is measured on the basis of the frequently used first order queen criterion, which defines neighbors as adjacent spatial units (block groups) that share a commo n border or corner with the spatial unit (block group) of interest. The bivariate Moran's I a global spatial statistical measure, was first computed to determine the strength and direction of the geographic association between these variables across the study area (Anselin 1995). This statistic ranges in value from 1.0 to +1.0. A high positive value indicates similar values of the SED index and accessibility for neighboring block groups (cluster), a high negative value indicates dissimilarity among neig hboring block groups (outlier), and a value near zero indicates no spatial relationship across neighboring block groups (Voss et al. 2006). The statistical significance of this observed measure is typically calculated in comparison to a reference distribut ion. First, references values are obtained by a random permutation procedure, which recalculates each observed value by a user determined number of permutations

PAGE 57

47 ! (Hare and Barcus 2007). Then, the level of significance is calculated by using the following fo rmula (Anselin 2011): (M + 1) / (R + 1) where R is the number of permutations and M is the number of times a reference value is equal to or greater than the observed value (positive Moran's I ) or equal to or less than the observed value (negative Moran's I ). The result of this calculation is called a pseudo one sided significance measurement, since the outcomes are sensitive to permutation level (Anselin 2011). This study uses 999 permutations to calculate this statistic While the bivariate Moran's I provi des a measure of overall clustering in the spatial association between two variables, it does not indicate where the clusters or outliers are located or what type of spatial correlations are most dominant within a given study area. Local measures of spati al association provide a measure of the correlation for each individual unit and help identify the type of spatial correlation. These are implemented using Local Indicators of Spatial Association (LISA), such as the local Moran's I The local measu re of th e Moran's I used for this study is expressed as: where x and y are the SED index for block group i and the access for neighboring block group j, respectively; and z x i and z y i are the standardized scores of variables x and y, respectively. The spatial w eight matrix w ij is defined as a binary contiguity matrix, which provides the spatial structure for locations (block groups) that are included in the calculation of the local Moran's I. As stated previously, the first order queen contiguity

PAGE 58

48 ! matrix is used to define neighbor relationships for this study. All observations that share a common border or corner have w ij = 1, otherwise w ij = 0 The formulae indicates where there are spatial clusters of high (or low) values of one variable x surrounded by high (o r low) values of another variable y. High or low values in this context are defined as observations that are greater than or less than the mean of the respective variable, respectively. These values help to provide a measure for the spatial association bet ween neighboring variables as defined by the spatial weight matrix. Moran's I scatter plots were constructed to depict the relationship between socioeconomic deprivation within a block group and the accessibility to the four samples of food outlets in nei ghboring block groups, respectively. The data points are shown as standardized deviations from the mean, and the regression line corresponds to the Moran's I statistic (Anselin 2011). The horizontal axis represents the standardized value of the SED index o f a block group, and the vertical axis represents the standardized value of the average accessibility to a food outlet for that block group's neighbors, as defined by the spatial weights matrix (Voss et al. 2006). The placement of the values in the four qu adrants of the scatter plot is explained in Table 3.8:

PAGE 59

49 ! Table 3.8. Moran's I Scatter Plot Definitions Quadrant Definition Upper Right Block groups with above average socioeconomic deprivation that share boundaries with neighboring block groups that have above average accessibility to food outlets Upper Left Block groups with below average socioeconomic deprivation that share boundaries with neighboring block groups that have above average accessibility to food outlets Lower Right Block groups with abo ve average socioeconomic deprivation that share boundaries with neighboring block groups that have below average accessibility to food outlets Lower Left Block groups with below average socioeconomic deprivation that share boundaries with neighboring bloc k groups that have below average accessibility to food outlets The classification of the Moran's I scatter plot however, does not indicate whether the clusters or outliers, as described in the table, are statistically significant. Moran's I significan ce maps are used to enhance these scatter plots and to incorporate information on the significance of local spatial patterns. Permutation methods are typically utilized to determine the significance of the local Moran's I In practice, the observed values are randomly permuted across the entire study area and a local Moran's I statistic is calculated for each new permutation (Anselin 1995). The significance of the local Moran's I statistic is determined by generating a reference distribution using 999 rando m permutations. Based on this randomization, different theoretical standard deviations for Moran's I are obtained, each yielding a different p value as a measure of pseudo significance. A significance level of 0.05 ( p <.05) is used because this study's purp ose is to be exploratory and illustrative (Talen and Anselin 1998). Based on significance of the local Moran's I each block group can be characterized by one of four types of spatial associations listed in Table 3.9 (Charreire and Comber 2008):

PAGE 60

50 ! Table 3.9. Bivariate Moran's I Significance Categories Relationship Definition High High High values of SED index in a block group significantly associated with high values of access in neighboring block groups (positive relationship ). High Low High values of SE D index in a block group significantly associated with low values of access in neighboring block groups ( negative relationship ). Low Low Low values of SED index in a block group significantly associated with low values of access in neighboring block group s ( positive relationship). Low High Low values of SED index in a block group significantly associated with high values of access in neighboring block groups (negative relationship ). No S ignificance Values of SED index in block groups not significantly as sociated with values of access in neighboring block groups (no relationship) The high low' relationship for supermarket accessibility measures is used to spatially define food deserts and the high high' relationship for fast food restaurant accessibil ity measures is used to define food swamps. The third and final phase examines the racial, ethnic and locational characteristics of these food deserts and food swamps, in comparison to the rest of the neighborhoods in Hillsborough County. Proportional com parisons provide initial insights on the overall population and the racial/ethnic composition of areas where the relationship between the SED index and store access is spatially significant ( p<.05 ). Binary logistic regression models were then utilized to a nalyze the relationship between the presence/absence of food deserts or food swamps and exploratory variables that represent race, ethnicity, and locational characteristics. The regression analysis was conducted separately for each of the four dependent va riables: food deserts when considering all supermarkets, food

PAGE 61

51 ! deserts when considering healthiest supermarkets, food swamps when considering all fast food restaurants, and food swamps when considering unhealthiest fast food restaurants. All statistical ana lyses were conducted using IBM SPSS Statistics (version 19) and all ESDA analyses were conducted using OpenGeoDa software (version 0.9.9.1).

PAGE 62

52 ! CHAPTER 4: STATISTICAL AND SPATIAL ANALYSIS OF ACCESSIBILITY TO ALL SUPERMARKETS AND HEALTHIEST SUPERMARKETS This chapter focuses on assessing the statistical relationship between socioeconomic deprivation and accessibility to all supermarkets and healthiest supermarkets, respectively, and exploring how these relationships vary geographically within Hillsborough County. First, descriptive choropleth mapping is used to investigate the spatial distribution of these variables and understand the relative variability of both accessibility measures and socioeconomic deprivation in this study area. Second, independent s ample t tests and bivariate global statistical analyses are conducted to provide preliminary insights on the association between accessibility and socioeconomic deprivation at the county scale. Third, loca l indicators of spatial association (LISA) are empl oyed to analyze and visualize the spatial relationship between these variables within Hillsborough County at the census block group level, and identify the locations of neighborhoods (census block groups) that can be classified as food deserts based on the statistical significance of spatial correlations. Lastly, proportional comparisons and binary logistic regression are utilized to determine if the racial, ethnic and locational characteristics of neighborhoods classified as food deserts are significantly different from the rest of the neighborhoods in this study area.

PAGE 63

53 ! 4.1. Descriptive Mapping and Statistics The location of healthy food outlets and the proportion of each block group that is accessible to these outlets are displayed in Figures 4.1 and 4.2. For the underlying choropleth pattern, block groups are grouped into five quintiles based on accessibility to all supermarkets and healthiest supermarkets chains, respectively. Figure 4.1 shows that neighborhoods with high levels of accessibility to all su permarkets are concentrated primarily near major roadways (e.g., I 4, I 75, I 275, and U.S. 92) that lead from the urban center of Hillsborough County to the suburban areas, where accessibility to all supermarkets is also high. On the other hand, accessibi lity to all supermarkets is lowest in the rural outskirts of the county and in the City of Tampa. This geographic pattern can be explained, in part, by the construction of the interstate highway system in the 1950s, which caused supermarkets to move out of cities and locate along suburban roadways (Schlosser 2001). This relocation was necessary to stay close to their customer base and maintain the space necessary to offer a variety of products and to be reachable by delivery trucks (Larsen and Gilliland 200 8).

PAGE 64

54 ! Figure 4.1. Locations of All Supermarkets (n=115) and Accessibility to All Supermarkets by Block Group (n=795), Hillsborough County, Florida, 2010 Accessibility to healthiest supermarkets is dispersed in a pattern that resembles accessibility to all supermarket chains, except that neighborhoods with high levels of accessibility in Figure 4.2 are located further away from the city of Tampa along these major roadways (e.g., I 4, I 75, I 275 and U.S. 92), in the suburban areas of Hillsborough County. A larger number of rural block groups comprise neighborhoods with the least accessibility to healthiest supermarkets, when compared with access to all supermarkets. This pattern of accessibility to the healthiest food outlets is to be expected, in part, b ecause grocers in urban areas often charge higher prices but offer fewer nutritious options. Additionally, healthy food sources are not evenly distributed or readily available

PAGE 65

55 ! in rural areas (Morland et al. 2002; Sharkey 2009). Thus, accessibility to the h ealthiest supermarkets, as seen in Hillsborough County, is likely to be highest in the suburban areas and lowest in the urban/rural areas. Figure 4.2. Locations of Healthiest Supermarkets (n=47) and Accessibility to Healthiest Supermarkets by Block Grou p (n=795), Hillsborough County, Florida, 2010 The geographic distribution of the primary explanatory variable, the socioeconomic deprivation (SED) index, is displayed in Figure 4.3. Neighborhoods with higher levels of social deprivation are located in de nsely populated urban and suburban areas of Hillsborough County (e.g., Brandon, Citrus Park, Plant City, Riverview, Tampa, Temple Terrace and Town N' County). As expected, this pattern of social deprivation coincides with the overall pattern of population density in Hillsborough County.

PAGE 66

56 ! Figure 4.3. Socioeconomic Deprivation (SDE) Index by Block Group, Hillsborough County, Florida (n=795), 2000 Summary statistics for the accessibility variables, the SED index, and the variables used to calculate the S ED index are provided in Table 4.1. On average, almost one sixth (16.8 percent) of a block group's area in this county is within walking distance to a supermarket chain, while less than half of that amount (7.8 percent) is within walking distance to a heal thy supermarket chain. As can be expected, due to differences in sample sizes between all supermarkets and healthiest supermarkets (there are almost 2.5 times as many supermarkets as healthiest supermarkets), a significantly higher level of coverage and va riability can be observed for accessibility to all supermarkets. Summary statistics for each individual socioeconomic variable comprising the SED index suggest little variability across block groups in this study area. The index helps to control

PAGE 67

57 ! multicolli nearity between these five individual socioeconomic status variables, and to increase the deprivation and variability captured by this measure across the study area. Block groups in the county, on average, show a value of 1.042 for the SED index, which is low considering that the index ranges from 0.0 (least deprived) to 5.0 (most deprived). Table 4.1. Summary Statistics for Socioeconomic Deprivation Variables and Accessibility to Healthy Food Outlets Variables Min Max Mean SD Accessibility to All S upermarkets 0.000 1.000 0.165 0.258 Accessibility to Healthiest Supermarkets 0.000 1.000 0.078 0.186 Socioeconomic Deprivation (SED) Index 0.060 2.854 1.042 0.364 Variables Used to Calculated SED Index: Median Household Income 0.000 1.000 0.728 0.13 4 Single Parent Households 0.000 1.000 0.138 0.128 Unemployment Rate 0.000 1.000 0.015 0.042 Lower Education Individuals 0.000 1.000 0.101 0.113 Recent Immigrants 0.000 1.000 0.061 0.109 4.2. Comparison of Quantile Means The mean values of the SED i ndex associated with the quintiles for accessibility to all supermarkets and healthiest supermarkets are depicted in Figures 4.4 and 4.5, respectively. For supermarkets, the graph reveals a minor increase in the average socioeconomic deprivation of block g roups with increasing levels of accessibility, which appears to stabilize at the fourth quintile and then slightly decrease at the fifth quintile. Neighborhoods with moderate access to supermarket chains (third and fourth quintiles) appear to have the high est average level of socioeconomic deprivation. There appears to be no significant change in the average socioeconomic deprivation of neighborhoods with increasing accessibility to the healthiest supermarkets, although the most accessible block

PAGE 68

58 ! groups (fou rth and fifth quintiles) appear to be more socioeconomically deprived that those that are less accessible. These charts do not suggest a strong linear association between social deprivation and accessibility to healthy food retailers in Hillsborough County Figure 4.4. Means for SED Index by Quintiles of Accessibility to All Supermarkets Figure 4.5. Means for SED Index by Quintiles of Accessibility to Healthiest Supermarkets 0 0.2 0.4 0.6 0.8 1 1.2 Q1 Q2 Q3 Q4 Q5 SED Index Accessibility to all Supermarkets 0 0.2 0.4 0.6 0.8 1 1.2 Q1 Q2 Q3 Q4 Q5 SED Index Accessibility to Healthiest Supermarkets

PAGE 69

59 ! To examine the influence of accessibility to healthy food retailers on socioeco nomic status in more detail, the differences in average SED index between the first and fifth quintiles are compared in Table 4.2. The mean value of this index increases by 0.052 between the most accessible and least accessible neighborhoods to all superma rkets, but decreases by 0.059 when evaluating the difference between levels of accessibility to healthiest supermarkets. Independent sample t test results confirm that the differences in mean values of the SED index between the highest and lowest accessibi lity quintiles are not statistically significant. Table 4.2. Means for SED Index by Accessibility to Healthy Food Outlets Accessibility Mean SED Index for 1 st quintile Mean SED Index for 5 th quintile Difference t test Accessibility to All Supermarket s 1.003 1.055 0.052 1.477 Accessibility to Healthiest Supermarkets 1.052 0.993 0.059 1.298 These results are somewhat consistent with previous research on food deserts outlined in Chapter 2 for the subset of healthiest supermarkets, but not for the e ntire set of supermarkets. It appears that block groups that have the highest access to supermarkets are more socioeconomically deprived, while social deprivation decreases as accessibility to healthie st supermarkets increases. Neither one of these differe nces is statistically significant, which implies that the findings are not conclusive at this phase in the analysis. 4.3. Global Statistical Analysis The next phase utilizes both traditional and spatial tests in order to better understand how varying mag nitudes of socioeconomic deprivation are related to varying

PAGE 70

60 ! levels of accessibility to healthy food outlets at the county scale. First, bivariate parametric correlations are used to investigate the nature and direction of the statistical relationship betwe en access to supermarkets and the SED index at the block group level. Pearson's correlation coefficients ( r values), presented in Table 4.3, indicate that an increase in socioeconomic deprivation results in a slight increase in accessibility to all superma rkets but a slight decrease in accessibility to the healthiest supermarkets. The correlation coefficient for accessibility to healthiest supermarkets is statistically significant ( p <.10) while the correlation coefficient for accessibility to all supermarke ts is not statistically different from zero ( p >.10). These results indicate a strong negative association between access to healthiest stores and the SED index at the county scale, but a weak linear association between access to all stores and the SED inde x at the county scale that is inconsistent with the theoretical expectations. Next, the global and bivariate Moran's I statistic was employed to assess the strength and significance of the spatial association between the two variables. The relationship be tween accessibility to all supermarkets and socioeconomic deprivation yields a small positive value of the Moran's I test statistic that is significant ( p <.05), indicating a clustering of similar values for adjacent block groups in Hillsborough County. Thi s result implies that accessibility to supermarkets in a given block group is significantly and positively associated with the SED index in neighboring block groups. For the relationship between accessibility to healthiest supermarkets and the SED index, t he Moran's I statistic is again significant ( p <.01) but negative, suggesting clusters of dissimilar values among adjacent block groups in Hillsborough County. Accessibility to healthiest supermarkets in a given block group is significantly and negatively a ssociated

PAGE 71

61 ! with the SED index in neighboring block groups. These findings justify the need to control for spatial dependence amongst neighboring areas and explore local spatial variations in the relationship between access to healthy food outlets and socioe conomic deprivation. Table 4.3. Global Bivariate Association Between SED Index and Accessibility to Healthy Food Outlets Variables Pearsons' r p value Moran's I p value Accessibility to All Supermarkets 0.041 0.250 0.049 0.012** Accessibility to Healthi est Supermarkets 0.067 0.058* 0.065 0.002*** *** p <.01; ** p <.05; p <.10 (two tail) 4.4. Local Spatial Statistical Analysis Although the global Moran's I suggests non randomness in the overall spatial pattern, this measure cannot be used to determine wh at type of spatial correlation is most dominant and where the clusters or outliers are located within the study area. The next step was to utilize a local measure of spatial association (local Moran's I ) to examine geographic variability in the relationshi p between accessibility to healthy food and socioeconomic deprivation at the block group level. Moran's I scatter plots, depicted in Figures 4.6 and 4.7, are used to visualize the distribution of the local Moran's I statistic and explain the different type s of spatial correlations that are present at the block group level. The horizontal axis represents the standardized value of the SED index for each block group, the vertical axis represents the standardized value of the average accessibility to healthy fo od outlets for neighboring block groups, and the regression line corresponds to the global Moran's I statistic for each bivariate relationship. The four quadrants of the scatter plot are defined below in Table 4.4:

PAGE 72

62 ! Table 4.4. Moran's I Scatter Plot Definit ions Quadrant Definition Upper Right Block groups with above average socioeconomic deprivation that share boundaries with neighboring block groups that have above average accessibility to supermarkets Upper Left Block groups with below average socioeco nomic deprivation that share boundaries with neighboring block groups that have above average accessibility to supermarkets Lower Right Block groups with above average socioeconomic deprivation that share boundaries with neighboring block groups that hav e below average accessibility to supermarkets Lower Left Block groups with below average socioeconomic deprivation that share boundaries with neighboring block groups that have below average accessibility to supermarkets Figure 4.6 shows that a large proportion of block groups in Hillsborough County are in quadrants (upper left and lower right) that do not match the overall positive relationship between access to all supermarkets and socioeconomic deprivation suggested by the global Moran's I statisti c (0.049). A majority of block groups in Figure 4.7 are located in the upper right quadrant and correspond to overall negative association between the SED index and access to the healthiest supermarkets that was suggested by the global Moran's I statistic ( 0.065). However, this scatter plot also indicates that a large number of block groups in Hillsborough County are in quadrants (e.g., upper left and lower right) that do not match the negative correlation between access to healthiest supermarkets and soci oeconomic deprivation. These results suggest that a global measure of spatial association may be inadequate in representing local variations in the nature of dependence between the SED index and access to supermarkets within this county. A mix of similar a nd dissimilar associations implies that these bivariate relationships are not consistent over the study area, and thus must be mapped at the local or block group level.

PAGE 73

63 ! Figure 4.6. Bivariate Moran's I between SED Index and Accessibility to All Supermark ets Figure 4.7. Bivariate Moran's I between SED Index and Accessibility to Healthiest Supermarkets

PAGE 74

64 ! In order to visualize how the statistical relationship between the two variables varies across Hillsborough County and identify block groups where posi tive or negative spatial correlations are statistically relevant, the significance of the bivariate LISA Moran's I statistic was calculated and mapped. To create these maps, 999 random permutations and a significance level of 0.05 ( p <.05) were used. Figure s 4.8 and 4.9 show these significant local patterns of spatial correlation (bivariate Moran's I ) between the SED index and accessibility to all supermarkets and healthiest supermarkets, respectively. Each block group is classified based on the type and sig nificance of the statistical association between socioeconomic deprivation found in a block group and accessibility to healthy food outlets found in its neighboring block groups. Block groups are classified into one of four categories that are explained in Table 4.5: Table 4.5. Bivariate LISA Significance Categories Relationship Definition High High High values of SED index in a block group significantly associated with high values of access to supermarkets in neighboring block groups (positive relationshi p). High Low High values of SED index in a block group significantly associated with low values of access to supermarkets in neighboring block groups (negative relationship). Low Low Low values of SED index in a block group significantly associated with low values of access to supermarkets in neighboring block groups (positive relationship). Low High Low values of SED index in a block group significantly associated with high values of access to supermarkets in neighboring block groups (negative relations hip). The local spatial distribution of the statistical association between socioeconomic deprivation and accessibility to all healthy food outlets shows interesting and complex patterns across Hillsborough County, as seen in Figure 4.8. Positive relati onships are

PAGE 75

65 ! clustered in centralized urban areas that are primarily closer to the City of Tampa, while negative relationships are spread along the interstates to the suburban areas of Hillsborough County. Block groups where significantly high socioeconomic deprivation coincides spatially with significantly low accessibility to all supermarkets (high low) are classified as food deserts, because these are neighborhoods where socially disadvantaged individuals face barriers to healthy food. As expected, these food deserts are mostly located in and to the north of the City of Tampa and in rural areas of Hillsborough County (e.g., Lithia, North Tampa, Plant City and Riverview). Interestingly, there are more isolated food deserts located in Tampa near MacDill Air Force Base and near Tampa International Airport. Figure 4.8. LISA Significance Analysis of Spatial Correlation between SED Index and Accessibility to All Supermarkets

PAGE 76

66 ! As shown in Figure 4.9, significantly high values of both socioeconomic deprivation a nd accessibility to the healthiest supermarkets tend to be clustered in suburbs (e.g., Brandon, Carrollwood and West Tampa), while low values of both variables are found immediately surrounding downtown Tampa and some of its outlying areas (e.g., Riverview and Town N' Country). Food deserts, defined here as block groups with significantly high socioeconomic deprivation and significantly low accessibility to the healthiest supermarkets (high low), are predominantly located in the City of Tampa, and along ma jor roadways (e.g., I 4, I 75, I 275) that lead to outlying neighborhoods (e.g., Plant City and Temple Terrace) and nearby institutions (e.g., MacDill Air Force Base, Tampa International Airport and the University of South Florida). Interestingly, it appea rs as if State Road 60 acts as a primary barrier between these food deserts (high low) and the areas with low socioeconomic deprivation/high accessibility to healthiest supermarkets (low high). The next section focuses on estimating the racial, ethnic and locational characteristics of these food deserts and comparing them to other neighborhoods in the study area.

PAGE 77

67 ! Figure 4.9. LISA Significance Analysis of Spatial Correlation between SED Index and Accessibility to Healthiest Supermarkets 4.5. Characterist ics of Food Deserts The distribution of individuals in relevant population subgroups across the five categories of significant spatial correlations between the SED index and access to healthy food outlets were first examined to understand the racial and e thnic composition of food deserts. These results for accessibility to all supermarkets and accessibility to healthiest supermarkets, respectively, are summarized in Tables 4.6 and 4.7. When all supermarkets are considered, food deserts comprise only about 8 percent (66 out of 795) of the block groups, as well as approximately 8 percent of the total population of Hillsborough County. However, these block groups contain 18.7 percent of the entire Black population in Hillsborough County, compared to only 6.0 p ercent of the entire White population. Although block groups in three of the other four categories of spatial correlation contain higher proportions of the total White population than the Hispanic population, the reverse

PAGE 78

68 ! is true for the category of block g roups classified as food deserts. These results reveal that racial and ethnic minorities are disproportionately located in food deserts based upon all supermarket chains in Hillsborough County. Table 4.6. Demographic Composition of the Bivariate Significa nce Categories: All Supermarkets Class N Population Whites Blacks Hispanics Difference White Black Difference White Hispanic Not Sig. 567 73.1% 75.5% 63.3% 72.1% 12.2% 3.4% High High 57 9.5% 7.5% 12.6% 13.5% 5.1% 6.0% Low Low 52 3.7% 4.3% 1.8% 3.3% 2.5% 1.0% Low High 53 5.6% 6.8% 3.6% 3.2% 3.2% 3.6% High Low 66 8.1% 6.0% 18.7% 8.0% 12.7% 2.0% Note: Class in red represents food deserts. Both racial and ethnic minorities are disproportionately located in areas classified as food deserts based o n accessibility to healthiest supermarkets. As shown in Table 4.7, these block groups contain a staggering 46 percent and 22 percent of the entire county's Black and Hispanic populations, respectively. Compared to the other four categories of spatial corre lation, food deserts contain both the largest negative White Black difference and White Hispanic difference. A considerably larger percent of these racial and ethnic groups are located in these food deserts as compared to the majority White population. The se results suggest substantial racial and ethnic disparities in terms of those residing at the intersection of low accessibility to healthiest supermarkets and higher socioeconomic deprivation.

PAGE 79

69 ! Table 4.7. Demographic Composition of the Bivariate Significan ce Categories: Healthiest Supermarkets Class N Population Whites Blacks Hispanics Difference White Black Difference White Hispanic Not Sig. 535 69.7% 75.8% 46.4% 66.4% 29.4% 9.4% High High 23 4.3% 3.8% 4.0% 5.5% 0.2% 1.7% Low Low 66 5.5% 6.3% 2.7% 4. 7% 3.6% 1.6% Low High 18 1.6% 1.9% 0.8% 1.1% 1.1% 0.8% High Low 153 19.0% 12.1% 46.1% 22.4% 34.0% 10.3% Note: Class in red represents food deserts. Next, binary logistic regression analysis was employed to investigate the simultaneous effects of ra ce, ethnicity, and locational characteristics on the presence of food deserts and understand how these block groups differ from others in the study area. The dichotomous dependent variable for this regression was coded as 1' if a block group was classifie d as a food desert (high socioeconomic deprivation low accessibility to healthy food outlets category) and 0' if a block group was not classified a food desert (all other categories). Logistic regression models were estimated for the probability of a bloc k group being classified as a food desert, as a function of the explanatory variables describing racial, ethnic, and locational characteristics described in Chapter 3 (Table 3.7). These results are summarized in Table 4.8. For the model evaluating food de serts based on accessibility to all supermarkets, the Nagelkerke R squared (0.220) suggests a relatively high goodness of fit while the chi square tests indicates overall significance ( p <.01). While both racial/ethnic variables indicate a positive and stat istically significant ( p <.05) effect on the probability of a block

PAGE 80

70 ! group being classified as a food desert after controlling for locational factors, the Black proportion yields a much higher odds ratio than the Hispanic proportion. Locational characteristi cs also play a role in the location of food deserts, although population density is the only location variable that is statistically significant ( p <.10). The model concerning the presence of food deserts based on accessibility to healthiest supermarkets s hows an increase in model fit, from 0.222 to 0.364. Race and ethnicity variables remain statistically significant ( p <.01) and substantially increase the likelihood that block groups will be classified as food deserts. Highway density is the only locational variable that is statistically significant ( p <.10) with food deserts that are based on accessibility to the healthiest supermarkets. Table 4.8. Logistic Regression of Food Deserts Variables Supermarkets Healthiest Supermarkets Proportion non Hispanic Bl ack 47.758 98.028 (60.014)*** (108.199)*** Proportion Hispanic 8.166 25.929 (6.291)** (27.416)*** Population Density 0.000 0.000 (4.135)** (1.036) Highway Presence 0.691 0.869 (2.658) (6.708)* Commercial Land Use 0.152 3.156 (0.015) (2. 032) N 795 795 Nagelkerke R squared 0.220 0.364 Chi square Test 80.219 *** 204.924*** Note: Odds ratio with Wald's Chi square statistic in parentheses *** p <.01; ** p <.05; *p<.10 (two tail)

PAGE 81

71 ! 4.6. Summary of Statistical and Spatial Analysis Results In sum mary, traditional or non spatial statistical measures do not suggest a significant association between socioeconomic deprivation and accessibility to all healthy food outlets at the county scale. However, access to the healthiest supermarkets is significan tly lower in neighborhoods with higher socioeconomic deprivation. While the global Moran's I indicates a significant spatial relationship between the SED index and accessibility to all supermarkets and the healthiest supermarkets, respectively, the use of LISA reveals that the nature and significance of these relationships differs substantially across block groups in Hillsborough County and underscores the need to explore local variability in statistical results. Food deserts identified based on accessibili ty to all supermarkets and healthiest supermarkets, respectively, tend to be located in the rural areas of Hillsborough County and in or near the urban center of Tampa. Proportional comparisons and binary logistic regression analyses clearly indicate that higher proportions of racial and ethnic minorities are significantly associated with the prevalence of both types of food deserts in the study area, even after controlling for various locational characteristics. These results are consistent with previous s tudies conducted in other places that found that lower income or minority neighborhoods tend to have less geographic accessibility to healthy food retail locations (e.g. Morland et al. 2002, Apparicio et al. 2007).

PAGE 82

72 ! CHAPTER 5: STATISTICAL AND SPATIAL ANALYSIS OF ACCESSIBILITY TO ALL FAST FOOD RESTAURANTS AND UNHEALTHIEST FAST FOOD RESTAURANTS This chapter focuses on assessing the statistical relationship between socioeconomic deprivation and accessibility to all fast food restaurants and unhealthiest fast food restaurants, respectively, and exploring how these relationships vary geographically within Hillsborough County. First, descriptive choropleth mapping is used to investigate the spatial distribution of access to unhealthy food outlets and unders tand the relative variability of both accessibility measures in this study area. Second, independent sample t tests and bivariate global statistical analyses are conducted to provide preliminary insights on the association between accessibility and socioec onomic status at the county scale. Third, local indicators of spatial association (LISA) are employed to analyze and visualize the spatial relationship between these variables within Hillsborough County at the census block group level, and identify the loc ations of neighborhoods that can be classified as food swamps based on the statistical significance of spatial correlations. Lastly, proportional comparisons and binary logistic regression are utilized to determine if the racial, ethnic and location charac teristics of neighborhoods classified as food swamps are significantly different from the rest of the neighborhoods in this study area.

PAGE 83

73 ! 5.1. Descriptive Mapping and Statistics The location of unhealthy food outlets and the proportion of each block group' s area that is accessible to these outlets are displayed in Figures 5.1 and 5.2. For the underlying choropleth patterns, block groups are grouped into five quintiles based on accessibility to all fast food restaurants and unhealthiest fast food restaurants respectively. As shown in Figure 5.1, neighborhoods with high levels of accessibility to all fast food restaurants are concentrated primarily in downtown Tampa and near major roadways that connect the city to the suburbs (e.g., I 4, I 75, I 275, Route 41 Route 92 and Route 301). On the other hand, accessibility to all fast food restaurants is lowest in the rural outskirts of Hillsborough County, and to the immediate north and south of downtown Tampa. This geographic patterns can be explained, in part, by the construction of the interstate highway system in the 1950s, which caused entrepreneurs to see the benefit of operating quick service restaurants alongside customers' travel routes (Schlosser 2001).

PAGE 84

74 ! Figure 5.1. Locations of All Fast food Restaurant s (n=513) and Accessibility to All Fast food Restaurants by Block Group (n=795), Hillsborough County, Florida, 2010 Neighborhoods with high accessibility to unhealthiest fast food restaurants (Figure 5.2) are dispersed in a similar pattern as the areas with high accessibility to all fast food restaurants, concentrated in suburban areas and along major roadways (e.g., I 4, I 75, I 275 and Route 92). The primary difference is that the City of Tampa and a greater number of rural neighborhoods are found to l east accessible to unhealthiest fast food restaurants as compared to the entire fast food restaurant sample. This pattern of accessibility to the unhealthiest food outlets is unexpected, in part, because these restaurants are more likely to be located in p redominantly commercial areas due to zoning restrictions, which tend to be in more urbanized areas (Block et al. 2004). Additionally, rural areas tend to have disproportionate low access to unhealthy food

PAGE 85

75 ! (Sharkey 2009). Thus, accessibility to the unhealth iest fast food restaurants, as opposed to the trend seen in Hillsborough County, would be highest in the urban and rural neighborhoods. Figure 5.2. Locations of Unhealthiest Fast food Restaurants (n=34) and Accessibility to Unhealthiest Fast food Restau rants by Block Group (n=795), Hillsborough County, Florida, 2010 Summary statistics for the two accessibility variables are provided in Table 5.1. On average, almost one third (33.1 percent) of a block group's area in this county is within walking distan ce to any fast food restaurant, while less than one fifth that amount (5.9 percent) is within walking distance to one of the unhealthiest fast food restaurant. As can be expected, due to differences in sample sizes between all fast food restaurants and unh ealthiest fast food restaurants (there are almost 15 times as many fast food restaurants

PAGE 86

76 ! as unhealthiest fast food restaurants), a significantly higher level of coverage and variability can be observed for accessibility to all fast food restaurants. Tabl e 5.1. Summary Statistics for Accessibility to Unhealthy Food Outlets Variables Min Max Mean SD Dependent: Accessibility to All Fast food Restaurants 0.000 1.000 0.331 0.348 Accessibility to Unhealthiest Fast food Restaurants 0.000 1.000 0.059 0. 164 5.2. Comparison of Quantile Means The mean values of the SED index associated with the quintiles for accessibility to all fast food restaurants and unhealthiest fast food restaurants are depicted in Figures 5.4 and 5.5, respectively. For all fast fo od restaurants, the graph reveals that average socioeconomic deprivation increases with increasing levels of accessibility, stabilizing at the fourth quintile. Neighborhoods with the highest access to fast food restaurants (fourth and fifth quintiles) appe ar to have the highest average level of socioeconomic deprivation. There appears to be no significant change in the average socioeconomic deprivation of neighborhoods with increasing accessibility to the unhealthiest fast food restaurants. There does not a ppear to be a strong linear association between the level of social deprivation and accessibility to the unhealthiest food retailers in this county.

PAGE 87

77 ! Figure 5.4. Means for Socioeconomic Deprivation (SED) Index by Quintiles of Accessibility to All Fast fo od Restaurants Figure 5.5. Means for SED Index by Quintiles of Accessibility to Unhealthiest Fast food Restaurants 0 0.2 0.4 0.6 0.8 1 1.2 Q1 Q2 Q3 Q4 Q5 SED Index Accessibility to All Fast-food Restaurants 0.00 0.20 0.40 0.60 0.80 1.00 1.20 Q1 Q2 Q3 Q4 Q5 SED Index Accessibility to Unhealthiest Fast-food Restaurants

PAGE 88

78 ! To examine the influence of accessibility to unhealthy food retailers on socioeconomic status in more detail, the differences in avera ge SED index between the first and fifth quintiles are compared in Table 5.2. The mean value of this index increases by 0.209 between the most accessible and least accessible neighborhoods to all fast food restaurants, but decreases by 0.026 when evaluatin g the difference between levels of accessibility to unhealthiest fast food restaurants. Independent sample t test results reveal that the difference in mean values of the SED index between the highest and lowest accessibility quintiles is statistically sig nificant ( p <.01) for all fast food restaurants, but is not statistically significant for unhealthiest fast food restaurants. Table 5.2. Means for SED Index by Accessibility to Unhealthy Food Outlets Accessibility Mean SED Index for 1 st quintile Mean SED Index for 5 th quintile Difference t test Accessibility to All Fast food Restaurants 0.904 1.113 0.209 5.449*** Accessibility to Unhealthiest Fast food Restaurants 1.041 1.015 0.026 0.490 *** p <.01 (two tail) These results are consistent with previ ous research on food swamps outlined in Chapter 2 for the entire set of fast food restaurants, but not for the subset of unhealthiest fast food restaurants. It appears that block groups that are most accessible to fast food restaurants are significantly mo re socioeconomically deprived than those that are least accessible. However, there is no evidence to suggest a similar relationship for accessibility to the unhealthiest fast food restaurants.

PAGE 89

79 ! 5.3. Global Statistical Analysis The next phase utilizes both traditional and spatial tests in order to better understand how varying magnitudes of socioeconomic deprivation are related to varying levels of accessibility to unhealthy food outlets at the county scale. First, bivariate parametric correlations are used to investigate the nature and direction of the statistical relationship between access to fast food restaurants and the SED index at the block group level. Pearson's correlation coefficients ( r values), presented in Table 5.3, indicate that an increase in socioeconomic deprivation results in an increase in accessibility to all fast food restaurants but and slight decrease in accessibility to the unhealthiest fast food restaurants. The correlation coefficient for accessibility to all fast food restaurants i s statistically significant ( p <.01) while the correlation coefficient for accessibility to unhealthiest fast food restaurants is not statistically different from zero ( p >.10). These results indicate a strong positive association between access to all store s and the SED index at the county scale, but a weak linear association between access to unhealthiest stores and the SED index at the county scale that is inconsistent with theoretical expectations. Next, the global and bivariate Moran's I statistic was e mployed to assess the strength and significance of the spatial association between the two variables. The relationship between accessibility to all fast food restaurants and socioeconomic deprivation yields a positive value of the Moran's I test statistic that is significant ( p <.01), indicating a clustering of similar values for adjacent block groups in Hillsborough County. This result implies that accessibility to fast food restaurants in a given block group is significantly and positively associated with the SED index in

PAGE 90

80 ! neighboring block groups. For the relationship between accessibility to unhealthiest fast food restaurants and the SED index, the Moran's I statistic is again significant ( p <.01) but negative, suggesting clusters of dissimilar values among adjacent block groups in Hillsborough County. Accessibility to unhealthiest fast food restaurants in a given block group is significantly and negatively associated with the SED index in neighboring block groups. These findings justify the need to control for spatial dependence amongst neighboring areas and explore local spatial variations in the relationship between access to unhealthy food outlets and socioeconomic deprivation. Table 5.3. Global Bivariate Association Between SED Index and Access to Unhea lthy Food Outlets Variables Pearsons' r p value Moran's I p value Accessibility to All Fast food Restaurants 0.173 0.000*** 0.139 0.001*** Accessibility to Unhealthiest Fast food Restaurants 0.038 0.290 0.068 0.001*** *** p <.01 (two tail) 5.4. Local Spatial Statistical Analysis Although the global Moran's I suggests non randomness in the overall spatial pattern, this measure cannot be used to determine what type of spatial correlation is most dominant and where the clusters or outliers are located w ithin the study area. The next step was to utilize a local measure of spatial association (local Moran's I ) to examine geographic variability in the relationship between accessibility to unhealthy food and socioeconomic deprivation at the block group level Moran's I scatter plots, depicted in Figures 5.6 and 5.7, are used to visualize the distribution of the local Moran's I statistic and explain the different types of spatial correlations that are present at the block group level. The horizontal axis repre sents the standardized value of the SED index for each

PAGE 91

81 ! block group, the vertical axis represents the standardized value of the average accessibility to unhealthy food outlets for neighboring block groups, and the regression line corresponds to the global M oran's I statistic for each bivariate relationship. The four quadrants of the scatter plot are defined below in Table 4.4: Table 5.4. Moran's I Scatter Plot Definitions Quadrant Definition Upper Right Block groups with above average socioeconomic depriv ation that share boundaries with neighboring block groups that have above average accessibility to fast food restaurants Upper Left Block groups with below average socioeconomic deprivation that share boundaries with neighboring block groups that have ab ove average accessibility to fast food restaurants Lower Right Block groups with above average socioeconomic deprivation that share boundaries with neighboring block groups that have below average accessibility to fast food restaurants Lower Left Block groups with below average socioeconomic deprivation that share boundaries with neighboring block groups that have below average accessibility to fast food restaurants A majority of block groups in Figure 5. 6 are located in the upper right and lower le ft quadrants, corresponding to overall positive association between the SED index and access to all fast food restaurants that was suggested by the global Moran's I statistic ( 0.139 ). However, this scatter plot also indicates that a sizeable number of bloc k groups in Hillsborough County are in quadrants (e.g., upper left and lower right) that do not match the positive correlation between access to all fast food restaurants and socioeconomic deprivation. Figure 5.7 shows that a large proportion of block grou ps in Hillsborough County are in quadrants (e.g., upper right and lower left) that do not match the overall negative relationship between access to unhealthiest fast food restaurants and socioeconomic deprivation suggested by the global Moran's I statistic ( 0.068).

PAGE 92

82 ! These results suggest that a global measure of spatial association may be inadequate in representing local variations in the nature of dependence between the SED index and access to fast food restaurants in this county. A mix of similar and dis similar associations implies that these relationships are not consistent over the entire study area, and thus must be mapped at a local or block group level. Figure 5.6. Bivariate Moran's I between SED Index and Accessibility to All Fast food Restauran ts

PAGE 93

83 ! Figure 5.7. Bivariate Moran's I between SED Index and Accessibility to Unhealthiest Fast food Restaurants In order to visualize how the statistical relationship between the two variables varies across the county and identify block groups where posi tive or negative spatial correlations are statistically relevant, the significance of the bivariate LISA Moran's I statistic was calculated and mapped. For these maps, 999 random permutations and a significance level of 0.05 ( p <.05) were used. Figures 5.8 and 5.9 show these significant local patterns of spatial correlation (bivariate Moran's I ) between the SED index and accessibility to all fast food restaurants and unhealthiest fast food restaurants, respectively. Each block group is classified based on th e type and significance of the statistical association between socioeconomic deprivation in a block group and accessibility to unhealthy food outlets found in its neighboring block groups. Block groups are classified into one of four categories that are ex plained in Table 5.5:

PAGE 94

84 ! Table 5.5. Bivariate LISA Significance Categories Relationship Definition High High High values of SED index in a block group significantly associated with high values of access to fast food restaurants in neighboring block groups (positive relationship). High Low High values of SED index in a block group significantly associated with low values of access to fast food restaurants in neighboring block groups (negative relationship). Low Low Low values of SED index in a block group significantly associated with low values of access to fast food restaurants in neighboring block groups (positive relationship). Low High Low values of SED index in a block group significantly associated with high values of access fast food restaurants in neighboring block groups (negative relationship). The local spatial distribution of the statistical association between socioeconomic deprivation and accessibility to all unhealthy food outlets shows interesting and complex patterns across Hillsborough County (Figure 5.8) Positive relationships are primarily clustered in urban and suburban areas of Hillsborough County, while negative relationships are predominantly located in suburban and rural area of the county. Block groups where significantly high socioeconomic deprivation coincides spatially with significantly high accessibility to all fast food restaurants (high high) are classified as food swamps, because these are neighborhoods where socially disadvantaged individuals are overexposed to unhealth y food. These food swamps are mostly located immediately to the west and north of the urban center of the county along I 275 and Route 92 (e.g., Seminole Heights, Temple Terrace, Town n' Country and West Tampa). Interestingly, it appears as if West Kennedy Boulevard near I 275 acts as a primary barrier between these food swamps (high high) and the areas with low socioeconomic deprivation/high accessibility to fast food restaurants (low high).

PAGE 95

85 ! Figure 5.8. LISA Significance Analysis of Spatial Correlation between SED Index and Accessibility to All Fast food Restaurants by Block Group As shown in Figure 5.9, significantly high values of both socioeconomic deprivation and accessibility to the unhealthiest fast food restaurants tend to be clustered further to the west and north of the City of Tampa (i.e., Temple Terrace and West Tampa), while low values of both variables are found in more affluent suburban areas (i.e., Citrus Park and Lithia). Food swamps, defined here as block groups with significantly high socioeconomic deprivation and significantly high accessibility to the unhealthiest fast food restaurants (high high), are scattered to the west and north of the City of Tampa and a few of the suburbs of Brandon. Interestingly, it appears as if State Road 60 acts as a primary barrier between these food swamps (high high) and the areas with low socioeconomic deprivation/high accessibility to unhealthiest fast food restaurants (low high). The next section focuses on estimating the racial, ethnic and

PAGE 96

86 ! locationa l characteristics of these food swamps and comparing them to other neighborhoods in the study area. Figure 5.9. LISA Significance Analysis of Spatial Correlation between SED Index and Accessibility to Unhealthiest Fast food Restaurants by Block Group 5 .5. Characteristics of Food Swamps The distribution of individuals in relevant population subgroups across the five categories of significant spatial correlations between the SED index and access to unhealthy food outlets were first examined to understand the demographic composition of food deserts. These results for accessibility to all fast food restaurants and unhealthiest fast food restaurants, respectively, are summarized in Tables 5.6 and 5.7. When all fast food restaurants are considered, food swamp s comprise only about 9 percent (72 out of 795) of the block groups and approximately 10 percent of the total population in Hillsborough County. However, these block groups (high high) contain 21 percent and

PAGE 97

87 ! 17 percent of the county's Black and Hispanic po pulations, respectively, compared to only about 7 percent of the county's White population. Table 5.6 also reveals that food swamps contain the highest percentage of both Black and Hispanic residents compared to the other categories of significant spatial correlation. This category of block groups also indicates the largest negative White Black and White Hispanic differences. These results suggest that racial and ethnic minorities are disproportionately located in food swamps associated with all fast food r estaurants in Hillsborough County. Table 5.6. Demographic Composition of the Bivariate Significance Categories: All Fast food Restaurants Class N Population Whites Blacks Hispanics Difference White Black Difference White Hispanic Not Sig. 509 65.1% 65.4 % 66.1% 62.3% 0.7% 3.1% High High 72 10.8% 6.6% 20.9% 17.0% 14.3% 10.4% Low Low 91 9.6% 12.6% 3.0% 4.9% 9.6% 7.7% Low High 64 4.3% 5.0% 2.0% 3.7% 3.0% 1.3% High Low 59 10.2% 10.4% 8.0% 12.1% 2.4% 1.7% Note: Classes in red represent food swamps. Both Black and Hispanic populations are not as disproportionately located in food swamps based on accessibility to the unhealthiest fast food restaurants. As shown in Table 5.7, the share of the total, White, and Hispanic populations residing in the high high category are very similar and reasonably small (approximately 3 percent each) but the share of Black populations residing there is relatively high at 6.6 percent. However, this category of block groups indicates the second largest negative difference between White and Black proportions. Interestingly, both Black and Hispanic residents are

PAGE 98

88 ! overrepresented in block groups with high socioeconomic deprivation/low accessibility to the unhealthiest fast food restaurants. Overall, these results suggest substa ntial racial and ethnic disparities in terms of those residing at the intersection of high accessibility to fast food restaurants and higher socioeconomic deprivation. Similar disparities are not clearly evident for food swamps defined by the unhealthiest fast food restaurants. Table 5.7. Demographic Composition of the Bivariate Significance Categories: Unhealthiest Fast food Restaurants Class N Population Whites Blacks Hispanics Difference White Black Difference White Hispanic Not Sig. 567 69.5% 53.4% 66.6% 73.8% 20.4% 7.2% High High 21 3.6% 2.8% 6.6% 3.5% 3.8% 0.7% Low Low 1 0.2% 0.0% 0.0% 0.3% 0.3% 0.3% Low High 55 5.1% 1.7% 3.4% 6.4% 4.7% 3.0% High Low 151 21.6% 38.3% 26.5% 16.6% 21.7% 9.9% Note: Classes in red represent food swamps. Next binary logistic regression analysis was employed to investigate the simultaneous effects of race, ethnicity, and locational characteristics on the presence of food swamps and understand how these block groups differ from others in the study area. The dic hotomous dependent variable for this regression was coded as 1' if a block group was classified as a food swamp (high socioeconomic deprivation high accessibility to unhealthy food outlets category) and 0' if a block group was not classified a food swamp (all other categories). Logistic regression models were estimated for the probability of a block group being classified as a food swamp, as a function of the explanatory variables

PAGE 99

89 ! describing racial, ethnic, and locational characteristics described in Chap ter 3 (Table 3.7). These results are summarized in Table 5.8. For the model evaluating food swamps based on accessibility to all fast food restaurants, the Nagelkerke R squared (0.286) suggests a relatively high goodness of fit while the chi square tests indicates overall significance ( p <.01) for this model. While both racial and ethnic variables indicate a positive and statistically significant ( p <.01) effect on the probability of a block group being classified as a food swamp after controlling for locati onal factors, the Hispanic proportion yielding a much higher odds ratio than the Black proportion. Locational characteristics also seem to play a role in the location of food swamps because both population density and commercial land use are statistically significantly ( p <.01) related. The model concerning the presence of food swamps based on accessibility to unhealthiest fast food restaurants shows a decrease in model fit, from 0.286 percent to 0.131. Race and ethnicity variables are no longer statistical ly significant, and while the Black proportion increases the likelihood that block groups will be classified as food deserts by about 178 percent, the Hispanic proportion decreases the likelihood that block groups will be classified as such by about 31 per cent. Population density and commercial land use remain statistically significant ( p <.05) with food swamps that are based on accessibility to the unhealthiest fast food restaurants.

PAGE 100

90 ! Table 5.8. Logistic Regression of Food Swamps Variables Fast food Restau rants Unhealthiest Fast food Restaurants Proportion non Hispanic Black 9.305 1.798 (19.563)*** (1.756) Proportion Hispanic 66.164 0.307 (29.601)*** (0.053) Population Density 0.000 0.000 (14.011)*** (8.028)*** Highway Presence 0.051 2.142 (0.024) (3.583) Commercial Land Use 237.169 51.255 (23.829)*** (5.577)** N 795 795 Nagelkerke R squared 0.286 0.131 Chi square Test 111.066*** 22.922*** Note: Odds ratio with Wald's Chi square statistic in parentheses *** p <.01; ** p <.05 (two tail) 5.6. Summary of Statistical and Spatial Analysis Results In summary, traditional or non spatial statistical measures suggest a significant association between socioeconomic deprivation and accessibility to all fast food restaurants at the county scale, b ut do not suggest a significant association between socioeconomic deprivation and accessibility to unhealthiest fast food restaurants. While the global Moran's I indicates a significant spatial relationship between the SED index and accessibility to all fa st food restaurants and the unhealthiest fast food restaurants, respectively, the use of LISA reveals that the nature and significance of these relationships differs substantially across block groups in Hillsborough County and underscores the need to explo re local variability in statistical results. Food swamps associated with all fast food restaurants and unhealthiest fast food restaurants tend to be in and near the urban center, the City of Tampa, and along major roadways. Proportional

PAGE 101

91 ! comparisons and bin ary logistic regression analyses clearly indicate that racial and ethnic minorities are significantly overrepresented in food swamps for all fast food restaurants in the study area, even after controlling for various locational characteristics. These resul ts are consistent with previous studies conducted in other places that found that lower income or minority neighborhoods tend to have greater geographic accessibility to unhealthy food retail locations (e.g. Block et al. 2004, Pearce et al. 2007a).

PAGE 102

92 ! C HAPTER 6: CONCLUSIONS Despite a reputation for being one of the wealthiest and most technologically advanced countries in the world, the U.S. faces a growing and alarming health epidemic, obesity. Although almost three quarters of all adults at least 20 y ears of age are either overweight or obese, this crisis is especially prevalent amongst racial/ethnic minorities and low income individuals (CDC 2009). Recent studies have linked these health inequities with disproportionate access to retail food outlets. The suburbanization of supermarket chains in U.S. cities due to urban residential growth has led to the creation of food deserts, or areas where socially disadvantaged individuals lack access to affordable and nutritious food. At the same time, the rapid g rowth of the fast food industry in recent years has led to the creation of food swamps, or areas where these socially disadvantaged individuals are exposed to an overabundance of unhealthy food options. Although recent empirical studies on the built food environment have attempted to identify and analyze both food deserts and food swamps, this research has been hindered in four critical ways. First, a majority of these studies only include either healthy or unhealthy food stores, and are thus unable to pro vide a comprehensive analysis of the entire built food environment in a given urban area and related social inequities. Second, most studies have treated all food outlets as equal entities, and thus failed to account for differences in nutritional offering s and pricing. Third, most studies have measured potential access to food outlets based on the count or buffer techniques, instead

PAGE 103

93 ! of employing the actual roadways or routes used by residents to travel to these stores. Lastly, most of these studies have us ed conventional statistical methods such as correlation or regression to examine the relationship between these food outlets and neighborhood composition. The use of classical statistical techniques may not be suitable for spatially referenced data because they either assume observations are independent or that statistical relationships remain unchanged across all units in a given study area. My thesis addresses these methodological gaps in previous quantitative research on the built food environment throu gh a case study that examined social inequities in access to healthy and unhealthy food retailers in Hillsborough County, Florida. Spatial data for supermarket and fast food restaurants chains were categorized using NAICS codes to ensure that both componen ts were evaluated as part of the built food environment in the study area. Additional information from media sources was employed to identify the healthiest supermarket and unhealthiest fast food restaurant chains and thus, consider the nature and variety of food options within this geographic area. Then, network based distance methods were implemented to accurately estimate potential access to relevant food sources (all supermarkets, healthiest supermarkets, all fast food restaurants, and unhealthiest fast food restaurants). Lastly, in order to address the limitations of traditional bivariate analysis, global and local indicators of spatial association were used to examine the statistical relationship between accessibility to food outlets and neighborhood s ocioeconomic status, as well as to visualize the geographic variability of this relationship within Hillsborough County. The first phase of this study relied on conventional statistical techniques to explore the relationship between socioeconomic deprivat ion and accessibility to food

PAGE 104

94 ! outlets at the county scale. Although the findings from three of the four independent sample t tests were not conclusive, socioeconomic deprivation was found to be significantly higher in neighborhoods that are least accessibl e to healthiest supermarkets and in neighborhoods that are most accessible to all fast food restaurants. Parametric correlation analysis revealed that an increase in socioeconomic deprivation leads to a significant decrease in access to the healthiest supe rmarkets and a significant increase in access to all fast food restaurants. However, results from this analysis did not indicate a significant statistical relationship between socioeconomic deprivation and access to either all supermarkets or unhealthiest fast food restaurants. Bivariate global and local measures of spatial association were then used to understand how the nature and significance of the relationship between socioeconomic deprivation and accessibility to food outlets varies within Hillsborou gh County. The global Moran's I statistic and its permutation based significance test indicated that socioeconomic deprivation is positively associated with accessibility to all supermarkets and all fast food restaurants, and negatively associated with acc essibility to the healthiest supermarkets and unhealthiest fast food restaurants. However, the scatter plots of the local Moran's I statistic suggested that this global measure of spatial association may not adequately reveal the relationship between the S ED index and accessibility to each of the food outlet variables. Bivariate local indicators of spatial association (LISA) were then used to identify the neighborhoods where social deprivation coincides spatially and statistically with lower access to both supermarket samples (food deserts) and higher access to both fast food restaurant samples (food swamps). The LISA significance maps reveal that food deserts tend to be located in suburban and rural areas (e.g., Brandon,

PAGE 105

95 ! Carrollwood, and West Tampa), while food swamps tend to be located immediately surrounding the urban center of Tampa, and that both of these areas are found along major highways (e.g., I 4, I 75 and I 275). Finally, the socio demographic characteristics of neighborhoods classified as food s wamps and food deserts, respectively, were compared to other neighborhoods in the county. Proportional comparisons of demographic information revealed substantial racial and ethnic disparities in terms of those residing in areas with high socioeconomic dep rivation/low accessibility to supermarkets (food deserts) and those residing in areas with high socioeconomic deprivation/high accessibility to fast food restaurants (food swamps). A substantially larger proportion of both Black and Hispanic residents are located in these areas compared to the rest of Hillsborough County. Additionally, logistic regression analyses clearly indicated that race and ethnicity play an undeniably pervasive role in explaining the presence and location of food deserts and food swam ps, respectively. Specifically, higher proportions of Hispanic and Black significantly increase the likelihood that a neighborhood will be classified as a food desert or food swamp, even after controlling for locational characteristics. These results are c onsistent with findings from previous studies conducted in other urban areas that found minority and lower income neighborhoods have increased exposure to food deserts and food swamps (e.g., Block et al. 2004, Pearce et al. 2007). Although this thesis rep resents the first systematic attempt to examine social inequities in potential access to both healthy and unhealthy food outlets in a metropolitan area of Florida, there are several limitations that exist within these improved methodological parameters. Fi rst, only supermarket and fast food restaurant chains are

PAGE 106

96 ! used to represent healthy and unhealthy food outlets within Hillsborough County, following previous research on the built food environment (e.g., Block et al. 2004, Apparicio et al. 2007). Other sm aller food retailers such as specialty stores, ethnic grocers, farmer markets, and snack/beverage shops are not included, although they are part of the food environment in this study area. Future research should incorporate these smaller or independently o wned outlets in the definition and analysis of food deserts and food swamps. Second, data for the socioeconomic deprivation index and several other explanatory variables were derived from the 2000 U.S. Census because it is considered to be the most reliab le source by previous food environment studies (e.g., Block et al. 2004; Bader et al. 2010). This aggregated information could be subject to the modifiable areal unit problem (MAUP), where the choice of analytical entity may influence the spatial patternin g and variability of the data and any ensuing interpretations. Although this analysis may be sensitive to boundary and scale effects, socio demographic data at the smallest available areal unit (finest geographic resolution) or block group level was employ ed to provide appropriate representations of neighborhoods in Hillsborough County (e.g., Raja et al., 2008; Sharkey et al. 2009; Feng et al. 2010). Third, although potential access to food outlets is an appropriate measure that has been used in previous e mpirical research, it is not necessarily a direct representation of actual visitation to these retail stores or actual consumption of food by residents. Individuals may face other barriers that prevent them from visiting these establishments, such as unsaf e walkways, extreme weather conditions and/or physical imparities. Even if they are able to reach these entities, residents may not purchase or eat the food available

PAGE 107

97 ! there because of high prices, a lack of nutritious offerings at these locations, or a lac k of education regarding which items provide the necessary health benefits. Fourth, network based accessibility was computed using a walkable distance threshold with respect to the healthy and unhealthy food outlets because this measurement has been deem ed sufficient by previous built food environment research (e.g., Apparicio et al. 2007). This calculation treats all homes within 1,000 meters of the facility as having equal access, when in reality those who are closer to the store will have greater acce ss that those who are further from the store. Additionally, this measurement fails to account for other travel modes or routes that individuals may use to visit these retail locations. It also important to consider that trips to supermarkets or fast food r estaurants do not always begin or end at home locations. Future research would benefit from incorporating measurements of distance decay, alternative modes of transportation (e.g., public transit, car), and trips that are associated with the journey to wor k, school, shopping, or recreational activities Lastly, this study uses the built food environment to examine the geographic distribution of potential health disparities at the neighborhood level. Unfortunately, health outcome data was unable to be inco rporated into this study as it is not available at the block group resolution, and so potential access acts as a proxy for actual physical well being. Future studies would benefit from including information about Body Mass Index (BMI) as well as other rela ted diseases in an effort to more fully understand how these healthy and unhealthy food outlets contribute to disparities in the obesity epidemic in America.

PAGE 108

98 ! Despite these limitations, this study provides strong evidence that socioeconomically disadvantag ed neighborhoods in Hillsborough County are significantly more likely to lack access to the healthiest supermarkets and to be overexposed to all fast food restaurants. The analyses suggest that both food deserts and food swamps are found along major roadwa ys within Hillsborough County, and that racial/ethnic minorities are significantly more likely to be located in these areas. Unlike previous studies on the built food environment that employed arbitrary classification schemes to identify the location of th ese areas (e.g., Apparicio et al. 2007), locally significant correlations are used to find neighborhoods that lie at the intersection of socioeconomic deprivation and accessibility to food outlets. This thesis demonstrates how local measures of spatial ana lysis can be used to provide a scientifically valid method for the geographic identification of food deserts and food swamps that can be applied to enhance future analyses of the build food environment and other health disparities. Additionally, the findin gs from this case study clearly emphasize the need to consider healthy and unhealthy aspects of the built food environment in formulating public policy solutions that address the obesity epidemic in America.

PAGE 109

99 ! REFERENCES Anselin, L. "Exploratory spatia l data analysis and geographic information systems." DOSES/EUROSTAT Workshop on New Tools for Spatial An alysis, ISEGI, Lisbon, Portugal November 18 20, 1993. Print. 1995. Local indicators of spatial association LISA. Geographical Analysis 27 (2): 93 115. 2011. Glossary of key terms. The GeoDa Center for Geospatial Analysis and Computation. Arizona State University. http://geodacenter.asu.edu/node/390 (last accessed February 27, 2011). Anselin, L., and A. Getis. 1992. Spatial statistical analysi s and geographic information systems. The Annals of Regional Science 26:19 33. Apparicio, P., M. S. Cloutier, and R. Shearmur. 2007. The case of MontrŽal's missing food desert: Evaluation of accessibility to food supermarkets. International Journal of Hea lth Geographics 6 (4). Apparicio, P., M. Abdelmajid, M. Riva, and R. Shearmur. 2008. Comparing alternative approaches to measuring the geographical accessibility of urban health services: Distance types and aggregation error issues. International Journal of Health Geographics 7 (7). Bader, M. D. M., J. A. Ailshire, J. D. Morenoff, and J. S. House. 2010. Measurement of the local food environment: A comparison of existing data sources. American Journal of Epidemiology 171 (5): 609 17. Baker, E. A., M. Scho otman, E. Barnidge, and C. Kelly. 2006. The role of race and poverty in access to foods that enable individuals to adhere to dietary guidelines. Preventing Chronic Disease 3 (3). Bloch, M., J. DeParle, M. Ericson, & R. Gebeloff. 2009. Food stamp usage acr oss the country (map). The New York Times, November 28, 2009. http://www.nytimes.com/interactive/2009/11/28/us/20091128 foodstamps.html (last accessed November 18, 2010). Block, J. P., R. A. Scribner, and K. B. DeSalvo. 2004. Fast food, race/ethnicity, an d income: A geographic analysis. American Journal of Preventive Medicine 27 (3): 211 17.

PAGE 110

100 ! Brown, A. F., R. B. Vargas, A. Ang, A. R. Pebley. 2008. The neighborhood food resource environment and the health of residents with chronic conditions. Journal of Gen eral Internal Medicine 23 (8): 1137 44. Centers for Disease Control and Prevention (CDC). 2004. Trends in intake of energy and macronutrients United States, 1971 2000. http://www.cdc.gov/mmwr/preview/mmwrhtml/mm5304a3.htm (last accessed November 17, 201 0). 2009. NCHS Health E Stat: Prevalence of overweight, obesity and extreme obesity among adults: United States, trends 1960 62 through 2005 06. National Center for Health Statistics, Centers for Disease Control and Prevention. http://www.cdc.gov/nchs /data/hestat/overweight/overweight_adult.htm (last accessed November 17, 2010). 2010. Overweight and obesity. Division of Nutrition, Physical Activity and Obesity, National Center for Chronic Disease Prevention and Health Promotion. Centers for Disea se Control and Prevention (CDC). http://www.cdc.gov/obesity/index.html (last accessed November 18, 2010). Chakraborty, J (in press). Revisiting Tobler's First Law of Geography: Spatial Regression Models for Assessing Environmental Justice and Health Risk Disparities. In J. Maantay and S. McLafferty (eds.) Geospatial Analysis of Environmental Health NY: Springer Verlag. Charreire, H., and E. Combier. 2008. Poor prenatal care in an urban area: A geographic analysis. Health & Place 15:412 419. Cohen, D. A ., and T. A. Farley. 2008. Eating as an automatic behavior. Preventing Chronic Disease 5 (1). www.cdc.gov/pcd/issues/2008/jan/07_0046.htm (last accessed February 23, 2011). Cummins, S., and S. Macintyre. 2002. "Food deserts" Evidence and assumption in he alth policy making. BMJ 325 (7361):436 38. Cummins, S. C. J., L. McKay, and S. MacIntyre. 2005. McDonald's restaurants and neighborhood deprivation in Scotland and England. American Journal of Preventive Medicine 29 (4): 308 10. Cummins, S., and S. Macin tyre. 2006. Food environments and obesity Neighbourhood or nation? International Journal of Epidemiology 35 (1): 100 104. Cutter, S. L., J. T. Mitchell, and M. S. Scott. 1998. Revealing the vulnerability of people and places: A case study of Georgetown Co unty, South Carolina. Annals of the Association of American Geographers 90(4): 713 737.

PAGE 111

101 ! Farrow, A., C. Larrea, G. Hyman, and G. Lema. 2005. Exploring the spatial variation of food poverty in Ecuador. Food Policy 30: 510 531. Feng, J., T. A. Glass, F. C. Curriero, W. F. Stewart, and B. S. Schwartz. 2010. The built environment and obesity: A systematic review of the epidemiologic evidence. Health & Place 16 (2): 175 90. Florida CHARTS. 2010. Florida CHARTS. http://www.floridacharts.com (last accessed Febr uary 23, 2011). Florida Department of Health. 2004. Obesity in Florida: Report of the governor's task force on the obesity epidemic Florida Department of Health. 2009b. Household food security in the United States, 2009 Food Assistance & Nutrition Research Program, Economic Research Service, U.S. Department of Agriculture. Florida Geographic Data Library. 2010. FGDL Metadata Explorer: Search & Download Data. http://www.fgdl.org/metadataexplorer/explorer.jsp (last accessed December 30, 2010). Food Research and Action Center (FRAC). 2010. Food hardship: A closer look at hunger. Food Research and Action Center (FRAC). Glanz, K. 2009. Measuring food environments: A historical perspective. American Journal of Preventive Medicine 36 (4S): S93 98. Hain ing, R., S. Wise, and J. Ma. (1998). Exploratory spatial data analysis in a geographic information system environment. The Statistician 47 (3): 457 469. Handy, S. L. 1992. Regional versus local accessibility: Neo traditional development and its implicatio ns for non work travel. Built Environment 18 (4): 253 67. Hansen, W. G. 1959. How accessibility shapes land use. Journal of the American Planning Association 25 (2): 73 6. Hare, T. S., and H. R. Barcus. 2007. Geographical accessibility and Kentucky's hea rt related hospital services. Applied Geography 27:181 205. Healthy People. 2010. Leading health indicators. http://www.healthypeople.gov/Document/HTML/uih/uih_4.htm ( last accessed November 17, 2010). Hill, J. O., and J. C. Peters. 1998. Environmental co ntributions to the obesity epidemic. Science 280 (5368): 1371 74.

PAGE 112

102 ! Hillsborough Community Atlas. 2011. Hillsborough County demographics. http://www.hillsborough.communityatlas.usf.edu/demographics/default.asp?ID=1 2057&level=cnty (last accessed February 23, 2011). Hillsborough County Government. 2010. GIS home: Search by attribute features. http://gisweb.hillsboroughcounty.org/zoning1/home.cfm (last accessed November 18, 2010). Hu., Z., and K. R. Rao. 2009. Particulate air pollution and chronic ischemic he art disease in the eastern United States: A county level ecological study using satellite aerosol data. Environmental Health 8:26 36. Huang, T. T K., and T. A. Glass. 2008. Transforming research strategies for understanding and preventing obesity. Journal of the American Medical Association 300 (15): 1811 13. Infogroup, Inc. (2010). U.S. Businesses Database Retrieved December 15, 2010, from ReferenceUSA database. International Council of Shopping Centers, and Social Compact. 2008. Inside site selection: Retailers' search for strategic business locations. International Council of Shopping Centers, and Social Compact. Jeffery, R. W., J. Baxter, M. McGuire, and J. Linde. 2006. Are fast food restaurants an environmental risk factor for obesity? Internationa l Journal of Behavioral Nutrition and Physical Activity 3 (2): 2 7. Lake, A., and T. Townshend. 2006. Obesogenic environments: Exploring the built and food environments. Perspectives in Public Health 126 (6): 262 67. Larsen, K., and J. Gilliland. 2008. M apping the evolution of food deserts' in a Canadian city: Supermarket accessibility in London, Ontario, 1961 2005. International Journal of Health Geographics 7 (16). Larson, N. I., M. T. Story, and M. C. Nelson. 2009. Neighborhood environments: Disparit ies in access to healthy foods in U.S. American Journal of Preventive Medicine 36 (1): 74 81.e10. Lewis, L. B., D. C. Sloane, L. M. Nascimento, A. L. Diamant, J. J. Guinyard, A. K. Yancey, and G. Flynn. 2005. African Americans' access to healthy food opti ons in south Los Angeles restaurants. American Journal of Public Health 95 (4): 668 73. Lytle, L. A. 2009. Measuring the food environment: State of the science. American Journal of Preventive Medicine 36 (4S): S134 44.

PAGE 113

103 ! Maantay, J. 2001. Zoning, equity, a nd public health. American Journal of Public Health 91 (7): 1033 41. Mair, J. S., M. W. Pierce, and S. P. Teret. 2005. The city planner's guide to the obesity epidemic: Zoning and fast food. Centers for Disease Control and Prevention (CDC), National Cente r for Environmental Health. McKinnon, R. A., J. Reedy, M. A. Morrissette, L. A. Lytle, and A. L. Yaroch. 2009. Measures of the food environment: A compilation of the literature, 1990 2007. American Journal of Preventive Medicine 36 (4S): S124 33. Mehta, N. K., and V. W. Chang. 2008. Weight status and restaurant availability: A multilevel analysis. American Journal of Preventive Medicine 34 (2): 127 33. Moore, L. V., and A. V. Diez Roux. 2006. Association of neighborhood characteristics with the location and type of food stores. American Journal of Public Health, 96 (2): 325 31. Morland, K., S. Wing, A. Diex Roux, and C. Poole. 2002. Neighborhood characteristics associated with the location of food stores and food service places. American Journal of Prev entive Medicine 22 (1): 23 29. Mormino, G. R. 2002. Sunbelt dreams and altered states: A social and cultural history of Florida, 1950 2000. The Florida Historical Quarterly 81 (1): 3 21. New Orleans Food Policy Advisory Committee. 2006. Building healthy communities: Expanding access to fresh food retail. Centers for Disease Control and Prevention (CDC). Office of Economic & Demographic Research (EDR). (2009). Population and demographic data. Office of Economic & Demographic Research (EDR), The Florida Le gislature. http://edr.state.fl.us/Content/population demographics/data/index.cfm (last accessed November 18, 2010). Office of the Surgeon General (OSG). 2000. Overweight and obesity: At a glance. Office of the Surgeon General (OSG). http://www.surgeonge neral.gov/topics/obesity/calltoaction/fact_glance.html (last accessed November 17, 2010). 2003. The obesity crisis in America: Testimony of Richard H. Carmona, Surgeon General, U.S. Public Health Service, Acting Assistant Secretary for Health, Departm ent of Health and Human Services before the Subcommittee on Education Reform, Committee on Education and the Workforce, United States House of Representatives.

PAGE 114

104 ! http://www.surgeongeneral.gov/news/testimony/obesity07162003.htm (last accessed November 17, 201 0). Paeratakul, S., D. P. Ferdinand, C. M. Champagne, D. H. Ryan, and G. A. Bray. 2003. Fast food consumption among U.S. adults and children: dietary and nutrient intake profile. Journal of the American Dietetic Association 103 (10): 1332 38. Papas, M. A ., A. J. Alberg, R. Ewing, K. J. Helzlsouer, T. L. Gary, and A. C. Klassen. 2007. The built environment and obesity. Epidemiologic Reviews 29 (1): 129 43. Paul, Pamela. 2008. America's 10 healthiest supermarkets. Health Magazine 6 November. http://today.m snbc.msn.com/id/27573342 ( last accessed November 15, 2010). Pearce, J., K. Witten, and P. Bartie. 2006. Neighbourhoods and health: A GIS approach to measuring community resource accessibility. Journal of Epidemiology & Community Health 60 (5): 389 95. P earce, J., T. Blakely., K. Witten, and P. Bartie. 2007a. Neighborhood deprivation and access to food retailing: A national study. American Journal of Preventive Medicine 32 (5): 375 82. Pearce, J., K. Witten, R. Hiscock, and T. Blakely. 2007b. Are sociall y disadvantaged neighbourhoods deprived of health related community resources? International Journal of Epidemiology 36 (2): 348 55. Pothukuchi, K. 2005. Attracting supermarkets to inner city neighborhoods: Economic development outside the box. Economic D evelopment Quarterly 19 (3): 232 44. Powell, L. M., F. J. Chaloupka, and Y. Bao. 2007. The availability of fast food and full service restaurants in the United States: Associations with neighborhood characteristics. American Journal of Preventive Medicine 33 (4S): S240 45. Raja, S., C. Ma, and P. Yadav. 2008. Beyond food deserts: Measuring and mapping racial disparities in neighborhood food environments. Journal of Planning Education and Research 27 (4): 469 82. Reisig, V. M. T., and A. Hobbiss. 2000. Fo od deserts and how to tackle them: A study of one city's approach. Health Education Journal 59 (2): 137 49. Rose, D., J. N. Bodor, C. M. Swalm, J. C. Rice, T. A. Farley, and P. L. Hutchinson. "Deserts in New Orleans?: Illustrations of urban food access an d implications for policy." University of Michigan National Poverty Center/USDA Economic Research Service Research, Understanding the Economic Concepts and Characteristics of Food Access February 2009. Print.

PAGE 115

105 ! Rose, G. 1985. Sick individuals and sick popu lations. International Journal of Epidemiology 30 (3): 427 32. Sallis, J. F., and K. Glanz. 2006. The role of built environments in physical activity, eating, and obesity in childhood. The Future of Children 16 (1): 89 108. Schlosser, R. 2001. Fast Food Nation: The Dark Side of the All American Meal New York: Houghton Mifflin Company. Sharkey, J. R. 2009. Measuring potential access to food stores and food service places in rural areas in the U.S. American Journal of Preventive Medicine 36 (4S): S151 55 Sharkey, J. R., S. Horel, D. Han, and J. C. Huber Jr. 2009. Association between neighborhood need and spatial access to food store and fast food restaurants in neighborhoods of Colonias. International Journal of Health Geographics 8 (9). Singh, G. K., M. D. Kogan, and P. C. van Dyck. 2010. Changes in state specific childhood obesity and overweight prevalence in the United States from 2003 to 2007. Archives of Pediatrics & Adolescent Medicine 164 (7): E1 E10. Smoyer Tomic, K. E., J. C. Spence, and C. A mrhein. 2006. Food deserts in the prairies? Supermarket accessibility and neighborhood need in Edmonton, Canada. The Professional Geographer 58 (3): 307 326. Stein, D.O., and J. Chakraborty. 2010. Racial, ethnic, and socioeconomic disparities in exposure to fast food in Hillsborough County, Florida. Florida Public Health Review 7:83 92. Story, M., K. M. Kaphingst, R. Robinson O'Brien, and K. Glanz. 2008. Creating healthy food and eating environments: Policy and environmental approaches. Annual Review of P ublic Health 29:253 72. Strum, R. 2009. Affordability and obesity: Issues in the multifunctionality of agricultural/food systems. Journal of Hunger & Environmental Nutrition 43 (3 4): 454 465. Swinburn, B., G. Egger, and F. Raza. 1999. Dissecting obesoge nic environments: The development and application of a framework for identifying and prioritizing environmental interventions for obesity. Preventive Medicine 29 (6): 563 70. Talen, E., and L. Anselin. 1998. Accessing spatial equity: An evaluation of meas ure of accessibility to public playgrounds. Environment and Planning A 30:595 613.

PAGE 116

106 ! The Office of Minority Health. 2009. Obesity data/statistics. http://minorityhealth.hhs.gov/templates/browse.aspx?lvl=3&lvlid=537 ( last accessed November 17, 2010). Toble r, W. R. 1970. A computer movie simulating urban growth in the Detroit region. Economic Geography 46:234 240. U.S. Census Bureau. 2009. North American industry classification system (NAICS). http://www.census.gov/eos/www/naics/faqs/faqs.html (last accesse d November 15, 2010). 2010. State and county quickfacts: Hillsborough County, Florida (Data files). http://quickfacts.census.gov/qfd/states/12/12057.html (last accessed November 15, 2010). U.S. Department of Agriculture (USDA). 2009a. Access to aff ordable and nutritious food: Measuring and understanding food deserts and their consequences. U.S. Department of Agriculture, Economic Research Service, Food and Nutrition Service and the Cooperative State Research, Education, and Extension Service. 2009b. Household food security in the United States, 2009 Food Assistance & Nutrition Research Program, Economic Research Service, U.S. Department of Agriculture. 2010. Access to affordable, nutritious food is limited in "food deserts". Amber Waves 8 (1). www.ers.usda.gov/AmberWaves/March10/PDF/FoodDeserts.pdf (last accessed February 23, 2011). ! Unwin, A., and D. Unwin. 1998. Exploratory spatial data analysis with local statistics. Journal of the Royal Statistical Society, Series D (The Statistician) 47 (3): 415 421. ! Voss, P. R., K. J. Curtis White, and R. G. Hammer. (2006). Explorations in spatial demography. In W. A. Kandel & D. L. Brown (Eds.). Population Change and Rural Society (407 429). Netherlands: Springer. Weinsier, R. L., G. R. Hunter, an d A. F. Heini. 1998. The etiology of obesity: Relative contribution of metabolic factors, diet, and physical activity. American Journal of Medicine 105 (2): 145 50. Witten, K., D. Exeter, and A. Field. 2003. The quality of urban environments: Mapping vari ation in access to community resources. Urban Studies 40 (1): 161 77. World Health Organization (WHO). 2006. Obesity and overweight. http://www.who.int/mediacentre/factsheets/fs311/en/index.html (last accessed November 17, 2010).

PAGE 117

107 ! 2010. Controlling th e global obesity epidemic. http://www.who.int/nutrition/topics/obesity/en/index.html (last accessed November 18, 2010). Zenk, S. N., A. J. Schulz, B. A. Israel, S. A. James, S. Bao, and M. L. Wilson. 2005. Neighborhood racial composition, neighborhood pov erty, and the spatial accessibility of supermarkets in metropolitan Detroit. American Journal of Public Health 95 (4): 660 67. Zinczenko, D. 2009. America's unhealthiest restaurants. Yahoo! Health, March 9, 2009. http://health.yahoo.net/experts/eatthis/am erica%E2%80%99s unhealthiest restaurants (last accessed November 18, 2010). Zinczenko, D., and M. Goulding. 2009. Eat This, Not That!: The Best (& Worst!) Foods in America! New York: Rodale, Inc.