University of Southern California Dissertations and Theses

Obesity and healthy food accessibility: case study of Minnesota, USA

(USC Thesis Other)

Obesity and healthy food accessibility: case study of Minnesota, USA

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Obesity and Healthy Food Accessibility: Case Study of Minnesota, USA

by

Elliott Wayne Ingram Jr.

A Thesis Presented to the
Faculty of the USC Graduate School
University of Southern California
In Partial Fulfillment of the
Requirements for the Degree
Master of Science
(Geographic Information Science and Technology)

December 2019

Copyright © 2019 by Elliott Wayne Ingram Jr.

To my queen, Shoreé, two princesses, Erynn and Sekai, my father, my mother, and my siblings.
You all have sacrificed much and guided me through this journey. Thank you and I love you all!

iii
Table of Contents

List of Figures .................................................................................................................................v

List of Tables .................................................................................................................................vi

Acknowledgements .......................................................................................................................vii

List of Abbreviation .....................................................................................................................viii

Abstract ..........................................................................................................................................iv

Chapter 1 Introduction ..................................................................................................................10

1.1 What is Obesity? ........................................................................................................10

1.2 Study Area .................................................................................................................13

1.3 Socioeconomics and Sociodemographics in Minnesota ............................................13

1.4 Healthy Food Accessibility in Minnesota ..................................................................14

1.5 Obesity Disparities Associated with Socioeconomics and Sociodemographics ........15

Chapter 2 Background and Literature Review ..............................................................................16

2.1 GIS-based Analysis of Obesity ...................................................................................17

Chapter 3 Methodology ................................................................................................................23

3.1 Data Acquisition .........................................................................................................24

3.2 Data Preparation ..........................................................................................................26

3.2.1. Data Aggregation ........................................................................................26

3.2.2. Healthy Food Sources .................................................................................26

3.2.3. Correlation of the Dependent Variable and Explanatory Variables............27

3.3 Regression Analysis ArcGIS and ArcMap .................................................................27

3.3.1. Ordinary Least Squares ...............................................................................27

3.3.2. Exploratory Regression Analysis ................................................................28

iv
3.3.3. Spatial Autocorrelation ...............................................................................29

Chapter 4 Results ..........................................................................................................................31

4.1 Excel Correlation of Explanatory Variables and Dependent Variable .......................31

4.2 Regression Modeling ..................................................................................................42

4.2.1. Ordinary Least Squares Results ..................................................................42

4.2.2. Exploratory Regression Analysis Results ...................................................44

4.3 Spatial Autocorrelation Results ..................................................................................47

4.4 Geographically Weighted Regression .........................................................................51

Chapter 5 Discussion and Conclusion ..........................................................................................52

5.1 Summary and Significance of Findings ......................................................................52

5.2 Study Limitations and Future Research ......................................................................53

5.2.1. Study Limitations ........................................................................................53

5.2.2. Future Analyses ..........................................................................................54

References .....................................................................................................................................56

Appendix A Maps of Aggregated Data Used in Analysis ............................................................60

Appendix B Exploratory Regression Model – Raw .....................................................................74

Appendix C Ordinary Least Squares (OLS) Results – Hypothesis ..............................................80

Appendix D Ordinary Least Squares (OLS) Results – Exploratory Regression ..........................83

v
List of Figures
Figure 1 Obesity prevalence in Minnesota: Percentage of Population Per County ...................... 12
Figure 2 Summary of Workflow ................................................................................................... 24
Figure 3 Physical Inactivity vs. Obesity Prevalence Correlation ..................................................32

Figure 4 Diabetes Prevalence vs. Obesity Prevalence Correlation ................................................32

Figure 5 Total Population vs Obesity Prevalence Correlation.......................................................33

Figure 6 Median Family Income vs Obesity Prevalence Correlation ............................................34

Figure 7 Poverty Prevalence vs Obesity Prevalence Correlation ..................................................34

Figure 8 Language Other Than English vs Obesity Prevalence Correlation .................................35

Figure 9 Foreign Born vs Obesity Prevalence Correlation ............................................................36

Figure 10 Unemployment vs Obesity Prevalence Correlation ......................................................36

Figure 11 Associates Degree vs Obesity Prevalence Correlation ..................................................37

Figure 12 Bachelor’s Degree vs Obesity Prevalence Correlation .................................................38

Figure 13 Professional/Master’s Degree vs Obesity Prevalence Correlation ................................38

Figure 14 Source Count vs Obesity Prevalence Correlation .........................................................39

Figure 15 Population Density vs Obesity Prevalence Correlation.................................................40

Figure 16 Source Density vs Obesity Prevalence Correlation .......................................................40

Figure 17 Graphical Summary of Spatial Autocorrelation for Reference .....................................47

Figure 18 Spatial Distribution of OLS Standardized Residuals of Hypothesized Variables .........49

Figure 19 Spatial Distribution of Exploratory Regression Variables Contributing to Obesity .....51

vi
List of Tables
Table 1 Data Type and Sources .................................................................................................... 25
Table 2 Healthy Food Sources in Minnesota ................................................................................ 26
Table 3 Explanatory Variables (Units) ......................................................................................... 29
Table 4 Correlations of Explanatory Variables to Obesity Prevalence Chart .............................. 41
Table 5 OLS Model of Hypothesized Contributing Variables to Obesity .....................................43
Table 6 OLS Diagnostics of Hypothesized Contributing Variables to Obesity ............................44
Table 7 Exploratory Regression Analysis......................................................................................45
Table 8 Diagnostics of Regression Analysis .................................................................................46
Table 9 Exploratory Analysis: Highest Adjusted R-Squared Results ...........................................46
Table 10 Global Moran’s I Summary of Hypothesized Variables ................................................48
Table 11 Global Moran’s I Summary of Exploratory Regression Variables .................................50

vii
Acknowledgements
First and foremost, I would like give praise to the one above all. He has never left or forsaken
me. He gave me the resolve to persevere through this journey. I will forever be grateful for my
family, friends, and peers, as they stabilized my foundation and kept my backbone aligned. It
was an honor to have Dr. Jennifer Bernstein as my faculty advisor. She was genuine,
compassionate, empathetic, sympathetic and understood the haze of the parenting, working, and
continuing education gauntlet. I am also appreciative of the constructive criticism received from
my juries and committee staff of Dr. Su Jin Lee and Dr. Katsuhiko Oda. The feedback steered
me in the right direction to complete this thesis.

viii
List of Abbreviations
ACS American Community Survey
ADJR2 Adjusted R-Squared (R2 )
AICc Alkaike’s Information Criterion
CDC Centers for Disease Control and Prevention
GIS Geographic Information System
GISci Geographic Information Science
JB Jarque-Bera p-value
K(BP) Koenker (BP) Statistic (p-value)
MDA Minnesota Department of Agriculture
MDHFS Minnesota Department of Health and Family Support
MGC Minnesota Geospatial Commons
MDH Minnesota Department of Health
PA Physical Activity
SA Global Moran’s I p-value
SHIP Statewide Health Improvement Partnership
SSI Spatial Sciences Institute
USC University of Southern California
USCB United States Census Bureau
VIF Variance Inflation Factor

ix
Abstract
Since the 1980s, obesity has been categorized as a national and global phenomenon. Although
obesity rates in Minnesota have been consistently lower than the nation’s and neighboring states’
median, the rates have been gradually increasing. The disproportionateness of obesity rates
between Minnesota, Minnesota’s neighboring states, and the United States suggest that aspects
of the Minnesota environment are different. Potential explanatory variables included are linked
to economic opportunity, demographics, healthy food availability, and health policies. Utilizing
the methodology employed by Shresta et. al (2013), this study expanded it by incorporating more
explanatory variables with the intention of building the best model to showcase the impact these
variables have on obesity levels and disparities within the study area. Ordinary Least (OLS) and
Exploratory Regression analyses were used to assess the spatial relationship between explanatory
variables (socio-economics, socio-demographics, and healthy food accessibility) and the
dependent variable (obesity levels) over space in Minnesota. The results suggested that the rate
of obesity correlates weakly with diabetes, median family income, age, education, and healthy
food availability at the county level. The analysis yielded an AICc = 402.068415 and AdjR2 =
0.231832 compared to hypothesis values of AICc = 410.857562 and AdjR2 = 0.162779. The
explanatory variables included in the model did not have a strong relationship with the dependent
variables in space. Given the relatively low correlations between the predicted relationships, the
findings indicate that additional social, cultural, and behavioral factors are required to better
explain the prevalence of obesity within Minnesota.
10

Chapter 1 Introduction
Obesity is an ongoing social and health issue worldwide. The prevalence of the phenomenon is
influenced by many social, cultural, and behavioral variables. This study spatially analyzed and
modeled the relationship between socio-economics, socio-demographics, accessibility to healthy
food options, and their correlations with obesity levels in Minnesota. Socio-economic and
demographic explanatory variables included physical inactivity, population size, education
attainment, income, employment, race and ethnicity, language spoken, poverty, and access to
healthy food. Chapter 1 introduces the problem, presents the study area, and discusses the
motivation behind the research. Chapter 2 highlights previous work wherein authors incorporated
socio-economic and socio-demographic variables and food accessibility to spatially display the
relationships between the factors and how they contribute to obesity. It also articulates gaps in
previous research and discusses how these gaps are addressed by this project. Chapter 3 explains
the methodology behind the research to be conducted in the thesis. Chapter 4 presents the results
of the study. Chapter 5 provides an in-depth summary of the study, discusses its strengths and
weaknesses, and provides direction for future research on this topic.
1.1 What is Obesity?
Obesity is an abnormal or excessive fat accrual that threatens an individual’s health.
It defines individuals with a body mass index (BMI) above 30 kg/m2 (Shrestha et al. 2013).
Consumption of foods saturated with high levels of sodium, added sugars, and sugar-sweetened
beverages contribute to obesity. Individual dietary behaviors such as low intakes of vegetables,
fibers, and milk in children, adolescents and adults also contribute to obesity. In the United
States, 35.7% of adults and 16.9% of children are considered obese (Chi et. al., 2013). In the
United States, approximately 365,000 deaths per year are related to obesity, only second to
11

tobacco (Shrestha et al. 2013). Chronic health conditions like high blood pressure, high
cholesterol, diabetes, coronary heart diseases, strokes, cancer, and poor sexual health are directly
related to obesity. Obesity has negative economic effects, totaling $117 billion dollars in health
care costs in the United States (Shrestha et al. 2013).
1.1.1 Obesity in Minnesota
Minnesotans spend an estimated $2.8 billion each year on obesity related health care costs alone
(MDH, 2017). However, Minnesota’s obesity rates have been consistently lower than the U.S.
median, with exceptions in 2001 and 2002. As of 2017, Minnesota ranks 35th in adult and youth
ages 10-17 obesity rates in the nation. 28.4% of adult Minnesotans are obese, up from 16.4% in
2000 and from 10.3% in 1990 (MDH, 2017). Between 2000-2007, the obesity rate in Minnesota
increased from 17.4% to 26%. From 2007-2017, the obesity trend slowed from 26% to 28.4%
(MDH, 2017). Overall, the state of Minnesota has a lower obesity rate than the U.S. as a whole
(See Figure 1).
Minnesota’s obesity rate followed the U.S. median from 2001-2007, but the rate was
significantly lower compared to neighboring states (Iowa, North Dakota, South Dakota, and
Wisconsin) in 2009 and from 2011-2017 (MDH, 2017). In 2008, the Minnesota obesity rate
diverged from the U.S. median, as did the obesity rates in Minnesota’s neighboring states. These
statistics were affected by sample size and demographic compositions of reported surveys
(MDH, 2017). Economic opportunity, differences in population demographics, and the
availability of healthy food options all are variables contributing to obesity which are different
between Minnesota and neighboring states (MDH, 2017).

12

Figure 1 Obesity Prevalence in Minnesota: Percentage of Population Per County
13

1.2 Study Area
The state of Minnesota (MN), USA is the focus of this research. Of the 48 contiguous of the
United States, Minnesota is the northernmost state in the country. Located in the upper Midwest,
it lies north central in the United States. Minnesota borders Canada, Iowa, Wisconsin, North
Dakota, and South Dakota. Geographically, it is over 400 miles in length and 200-350 miles in
wide. Minnesota was ranked as the 12th largest state in the United States (ACS, 2017). Minnesota
has 87 counties, with being a major component of scope of this research. Minnesota experienced
an incremental population growth of 0.92% in 2018 and accounts for 1.72% of the United States
total population (ACS, 2017).
1.3 Socioeconomics and Sociodemographics in Minnesota
Socioeconomics is the social science that studies how economic activity affects and is shaped by
social processes (Hellmich, 2015). It analyzes how societies progress, stagnate, or regress
because of their local or regional economy, or the global economy (Hellmich, 2015).
Sociodemographics are characteristics of a population. Sociodemographic factors including age,
race, ethnicity, as well as language and socio-economic variables of income and education all
influence health outcomes. It is easy to assume that poverty stricken and low-income
communities, for example, are more susceptible to obesity. However, there exist disparities
within each study and what constitute contributing variables to obesity (CDC, 2019). Obesity
also varies geographically (CDC, 2019).
The total population of Minnesota is 5,303,925 (USCB, 2018). Of the total population,
83.75% are White, 5.95% Black or African American, 4.66% Asian, and the remaining are other
races (ACS, 2017). 67% of all Minnesotans are employed, with an unemployment rate of 4%.
The median household income of all Minnesotans is $68,400 (ACS, 2017). Minnesota’s overall
14

poverty rate was 10.8% in 2017, a slight increase from 10.2% in 2015. However, over 500,000
Minnesotans live below the poverty threshold (ACS, 2017). At 48%, Minnesota ranks 2nd
nationally with the percentage of the population age 25-64 earning an associate degree or higher.
However, there are major disparities in degree attainment among racial and ethnic population
groups over age 25, with only Asian (50%) and white (44%) Minnesotans exceeding the state
average. In 2012, 70% of Minnesota adults had at least some college or higher (ACS, 2017).
1.4 Healthy Food Accessibility in Minnesota
1.6 million Minnesotans have low levels of access to healthy food sources (Mattessich, 2016).
235,000 Minnesotans live more than 10 miles from a large grocery store or supermarket
(Mattessich, 2016). 49% of Minnesotans report that not having a store nearby that sells healthy
food directly impacts what they eat (Mattessich, 2016). Price and distance create barriers to
healthy food options. Around 341,000 Minnesotans encounter this barrier (Mattessich, 2016).
Approximately 16% of Minnesota’s census tracts are considered food deserts, defined as areas
with a high proportion of residents who live far from a full-service grocery store and a high
proportion of residents who are low-to-moderate income (Mattessich, 2016). Counties in rural
Minnesota have a disproportionate number of food deserts relative to their population and
geographic area (Mattessich, 2016). It is stated that rural residents, low-income residents, senior
residents, and residents of color have relatively low access to healthy food in their communities
(Mattessich, 2016). This indicates that thousands of Minnesotans don’t have access to healthy
food whether it be because of distance, income, or both. These trends continue to contribute to
rising obesity rates in Minnesota.
Minnesota ranks seventh-worst in the nation for the share of residents, about one-third of
its population, with no grocery options close to their homes (Minnesota Department of Health
15

and Family Support 2012). The saturation of fast food restaurants and lack of farmers’ markets,
supermarkets, co-ops, and other stores deemed as providing healthy foods are highly noticeable
in Minneapolis communities. It is vital for people to have access to places providing healthy
foods to help prevent the effects of obesity, including high cholesterol, heart disease, high blood
pressure, and additional risks associated with the phenomenon. Proximity to healthy food has the
potential to mitigate the obesity epidemic.
1.5 Obesity Disparities Associated with Socio-economics and Socio-
demographics in Minnesota
Minnesota is considered one of the healthiest states in the US in terms of obesity trends.
However, obesity disproportionately affects many population groups and communities including
older adults and seniors, areas of low-income, poverty, low education, US-born Blacks,
Hispanics/Latinos, older residents with disabilities, residents with mental illnesses, and female
LGBT’s (Survey, 2010). Per the 2010 census, 38.5% of US-born adult Blacks and 29.5% adult
Hispanic/Latino were obese. 31.4% of high school adults with a high school education, 26.4% of
adults with less than a high school education, 25.5% of adults with some college education, and
15.9% of adults with a college education or higher were obese (United States Census Bureau
2010).
The research project looked at the relationship between accessibility to healthy foods,
socio-economics, and socio-demographics in the hopes of identifying trends. The successes and
failures of this study can provide guidance for researchers wanting to study similar trends within
their communities. The project hopes to ultimately help address how all populations can access
healthy food to mitigate obesity.

16

Chapter 2 Background and Literature Review
Areas with greater access to healthy foods tend to have lower obesity rates. However, research
conducted to analyze relationships between obesity and healthy food accessibility is complex.
Larson et al. (2009) researched and analyzed the presence, nature, and implications of
neighborhood differences in access to food using a snowball sampling strategy. They found that
national as well as local studies in the United States indicate disparities in socio-economics and
demographics and accessibility to healthy resources. The authors emphasized that there are
neighborhood disparities in access to food. Larson et al. (2009) suggest that additional research is
required to address limitations of current studies promote better healthy food accessibility.
Morland et al. (2002) examined the distribution of food stores and food service locations,
sorting each by neighborhood wealth and segregation. The names and addresses of places to buy
food in Mississippi, North Carolina, Maryland, and Minnesota were obtained from their
respective state Departments of Health and Agriculture. The addresses were then geocoded to
census tracts. Median home values were used to estimate neighborhood wealth, while the
proportion of black residents was used to measure neighborhood racial segregation. Their study
showed that there are four times more supermarkets established in predominantly white
communities compared to predominantly black communities. Without access to supermarkets
and healthier food options, which offer a wide variety of foods at lower prices, impoverished and
minority communities may not have equal access to the array of healthy food choices available
to predominantly white and/or wealthy communities.
Boone-Heinonen et al. (2011) conducted a study where they modeled fast food
consumption, diet quality, and adherence to fruit and vegetable recommendations as a function
of fast food chain, supermarket, or grocery store availability over fixed distances. Their models
17

took into consideration gender, individual sociodemographic characteristics, and community
poverty, and tested for interaction by individual-level income. The authors concluded that fast
food consumption was directly proportional to fast food availability among low income
individuals. However, greater supermarket accessibility was not related to diet quality and fruit
and vegetable consumption. Correlations between grocery store availability and individual diets
showed mixed results.
Bressie (2016) wrote a thesis analyzing spatial patterns of food accessibility in Lane
County, Oregon. The goal was to quantify food retail dispersion in the study area of Lane
County, Oregon, in the context of proximity, affordability, diversity (types of food venues),
perception, food supply (availability), and socio-economics. The authors’ methodology was
composed of four steps: (1) food store classification, (2) measurement calculations, (3)
aggregation of areal units, and (4) statistical analysis (Bressie 2016). Using Esri’s Network
Analyst to measure residential proximity to five different food store types over a road network,
the study showed that deprived and minority-dense communities in Lane County, Oregon had
better access to healthy food sources (Bressie, 2016). The results of this study eliminate
stereotypical assumptions of urban and rural food environments and that evaluations of these
areas’ food environments should be conducted separately.
2.1 GIS-based Analysis of Obesity
One way to analyze obesity trends is to use GIS to better understand the role of social and
economic factors. Specifically, “spatially-varying coefficient models such as OLS and GWR
have become statistical methods for identifying local variations in relationships between
outcome and explanatory variables” (Wen et al. 2010, 263). These complex models benefit
researchers who seek to identify spatial variations in relationships.
18

In Pennsylvania USA, it was found that obesity rates were impacted by many factors,
including physical activity, diabetes, and average distance to the nearest healthy food source.
Shrestha et. al (2013) conducted a study using Ordinary Least Squares (OLS) and
Geographically Weighted Regression (GWR) to spatially analyze the relationship between socio-
economic and physical health in the region. The researchers’ goal was to better understand and
regulate obesity trends, incorporating exploratory variables including diabetes, physical
inactivity, and average distance to healthy food stores. The results of OLS and GWR analyses
were compared to determine which method produced the best model to analyze the relationship
between socio-economics and obesity rates. It was concluded that GWR generated the best
results. Because of only three explanatory variables were used in their analysis, the results had
low levels of variance, indicating that additional factors were needed to better explain the
distribution of obesity in Pennsylvania. This analysis influenced the approach of the study
conducted here, suggesting more explanatory variables should be incorporated to better
understand obesity trends in the study area.
Wen et al. (2010) performed a study analyzing 29,273 working adults aged 21-65 years
of age, in which they used GWR to inspect geographical variations in the relationship between
poverty and obesity. The study revealed geographical inequalities in poverty and that poverty
was a key contributor of obesity in Taiwan. Results from the study concluded that poverty and
obesity were prominent in less developed areas and that poverty and obesity were locally
variable. The impact of poverty on obesity was shown to be locally specific. Variables such as
community low income and deprivation increased the prevalence of obesity.
Obesity and socio-environmental variables have been shown to be correlated. Between
the three primary socio-economic factors of employment, education, and income, Chalkais et al.
19

(2013) revealed that education was the most significant indicator of increased obesity rates in
Athens, Greece, among 18,296 children 8-9 years if age. Using GWR, Chalkais et al. (2013)
concluded that low educational level, high population density, low family income, and green
space availability constituted an “obesogenic” environment Chalkais et al. (2013). Although
findings by Chalkais et al. (2013) displayed a significant relationship between childhood obesity
and socio-economic heterogeneity, further research was needed to understand how socio-
economics and environmental factors interact with one another to better understand the obesity
epidemic. Chalkais et al. (2013) ultimately called for preventative tactics to combat childhood
obesity including changes in diet and physical inactivity.
In a similar study, Drewnowski et al. (2014) linked low socio-economic status to high
obesity rates in both Paris and Seattle, despite differences in urban form, food environments, and
health care systems. The objective of the study was to compare the relationship between the food
environment at the individual level, socio-economic status, and obesity in Paris and Seattle. The
researchers collected sociodemographic data, geocoded home addresses and food source
locations, and calculated the distance between home and supermarkets. A Modified Poisson
regression model was used to test the association between socio-economic status, food
environmental variables, and obesity. Results of the study concluded that distance to
supermarkets did not have a direct link to obesity; however, low income and education, coupled
with low property values and shopping at lower cost stores were directly correlated with high
obesity rates.
Obesity and other chronic conditions linked with low levels of physical activity (PA) are
associated with deprivation of accessibility to recreational physical activities (Ferguson et. al
2013). Ferguson et al. (2013) used GIS car and bus networks in Scotland to determine the
20

number of PA facilities accessible within travel times of 10, 20, and 30 minutes. The
accessibility by car to recreational physical activity facilities greatly exceeded that by bus
(Ferguson et. al 2013). Low income communities were deprived of access to facilities that offer
recreational activities (Ferguson et. al 2013). It was found that access to physical activity
facilities by car was much more significant for the most affluent quintiles of area-based income
deprivation than for most affluent quintiles in small towns and rural areas. Facilities were much
less accessible compared to bus travel for the most affluent quintile than for other quintiles in
urban areas and small towns. The most disadvantaged groups were those without access to a car
in rural areas (Ferguson et. al 2013).
Low accessibility to healthy foods and greater access to unhealthy foods are variables in
dietary habits leading to obesity. Cubbin et al. (2012) found that neighborhoods that have
experienced long-term poverty have the greatest access to both healthy and unhealthy food
sources compared to more economically advanced neighborhoods in Alameda County,
California. This is counter to stereotypical assumptions that minorities and urban areas have less
access to healthy food sources. Blacks and Latino neighborhoods had the greatest access to
healthy food sources. The results of their study suggested that spatial relationships between
sociodemographic characteristics and healthy food accessibility at the community level depends
on place and level of urbanization (Cubbin et al., 2013).
The suburbanization of food retailers in North America and United Kingdom have
contributed to urban food deserts (Larsen and Gilliland, 2008). Larsen and Gilliland (2008) used
GIS and multiple network analyses were implemented to assess supermarket accessibility in
relation to location, socio-economic characteristics, and access to public transit. They found that
21

residents in urban communities with low economic status have the lowest levels of access to
supermarkets and that spatial inequality have increased.
Obesity continues to rise and will grow with the increase in of obesity among younger
people (Daniel et al. 2009). This trend is due to the consumption of high-dense energy food,
reduced energy expenditure, and failure to meet daily fruit and vegetable intake. Daniel et al.
(2009) looked at the density of fast food outlets and stores selling fruits and vegetables. Socio-
demographic predictors including income, household structure, language, education, and urban
form measures (road and highway densities) were assigned. A regression analysis showed that
socio-demographic and urban form measures accounted for 60% and 73% of the variance
densities of fast food outlets and stores selling fruits and vegetables, respectively (Daniel et al.
2009). Fast food outlets were more prevalent in areas with full-time students and households
without fluent speakers of French or English. Stores selling fruits and vegetables were more
prevalent in communities with high proportions of single-status residents and university-
educated residents.
As this literature review shows, obesity is a complex phenomenon that is influenced by
multiple factors. Previous studies provided the blueprint on the best approach to analyzing the
prevalence of obesity, including socio-economics, socio-demographics, and accessibility to
healthy food outlets. Based on past studies, this study anticipated that physical inactivity, median
family income, poverty prevalence, unemployment, and healthy food source density would be
the greatest contributors obesity in Minnesota. 14 explanatory variables, including the five
variables listed above were included in regression analyses to test which factors most contributed
to obesity and constituted the best regression model. This builds on past research by determining
22

which variables, of which there are many, are the strongest predictors of obesity. Ideally, the
results of the study will be used to mitigate obesity in the future.

23

Chapter 3 Methodology
This chapter explains the process of data acquisition, data preparation, and the regression
analyses used in this study. The approach to this analysis followed the study conducted by
Shrestha et al. (2013). First, socio-economic and socio-demographic variables were acquired and
aggregated from non-spatial data in census and county databases. This data was inserted into an
Excel spreadsheet for analyses. The non-spatial data was later formatted and aggregated to the
state level as polygons, with each polygon representing a single county in ArcMap. In ArcMap, a
projected coordinate system was established to best display the data across the study area (NAD
1983 (2011) StatePlane Minnesota Central FIPS 2202 (US feet)). Data on businesses serving
healthy food and grocery stores were plotted as vector points in ArcGIS software for reference.
In the study, healthy food accessibility was generated by dividing the number of healthy food
sources by the area of each county per square mile. This gave the healthy food density per
county. Population density was calculated to identify possible correlations between population
per square mile and the prevalence of obesity. After identifying the explanatory variables, OLS
analyses were conducted to test the correlation of hypothesized explanatory variables against
obesity and a list of 14 explanatory variables and their correlation with obesity using exploratory
regression. The analyses sought to test the statistical significance of each explanatory variable on
obesity in Minnesota. Figure 2 shows a workflow of the methodology.

24

Figure 2 Summary of Workflow

3.1 Data Acquisition
Data was acquired from the US Census Bureau, Minnesota Department of Health databases,
Minnesota GIS databases, and the CDC. Data was used to map how which variables negatively
or positively impacted the prevalence of obesity the most. Socio-economic, socio-demographic
variables, and healthy food accessibility was needed to be thoroughly investigated to help
mitigate the obesity epidemic in Minnesota.

Data
Acquisition
• USDA Farmers Market Directory
• USDA Supermarkets
• CDC health data
• United States Census Bureau data, American Fact Finder, Minnesota Geospatial
Commons, American Communtity Survey (social ecomomics and demographics)
Data
Preparation
• Define parameters of focus area
• Define Projection
• Gather aspatial data in excel
• Search for healthy food outlets
• Aggregate aspatial data to county level
Regression
Analyes
• Correlation and regression in Excel
• Run Exploratory Regression analysis
• Run Ordinary Least Squares (OLS) regressions
• Generate Spatial Autocrorrelation
• Analyze spatial distribution of OLS
25

Table 1 Data Types and Sources

Category Factors Data Source Geographic
Scale
Data Type
Administrative
boundaries
Census Tract,
County,
Municipality, State
Tiger Line Data,
Minnesota
Geospatial
Commons,
Explore
Minnesota
State/County Vector
polygons
Access to healthy
food
Healthy food
businesses/stores
Exploring Food
Environments
(ESRI), ArcGIS
Online, MetroGIS
DataFinder,
Minnesota
Department of
Agriculture
State/County Vector
points and
polygons
Health Obesity, diabetes,
physical inactivity
Minnesota Public
Health Data,
Center for
Disease Control
and Prevention
(CDC)
State/County Vector
polygons
Socio-economics
and Socio-
demographics
Poverty, income,
population, age,
gender, ethnicity
and race,
employment
education
attainment,
language spoken,
United States
Census Bureau,
American Fact
Finder, American
Community
Survey,
Minnesota
Geospatial
Commons,
Minnesota
Geographic Data
Clearinghouse
Data,
MetroGIS
DataFinder
State/County Vector
polygons

26

3.2 Data Preparation
3.2.1 Data Aggregation
Health, socio-economic, and sociodemographic variables were aggregated in an Excel data sheet
in tabular format and appended as aspatial data into county polygons (see Appendix A). Most of
the aspatial data was aggregated to the county scale and by percentage of the total population per
county. Population density and healthy food source density were calculated by taking the
quotient of total population by the area of the county and quotient of the number of healthy food
sources by the area of the county, respectively.
3.2.2 Healthy Food Sources

Healthy food sources are defined as businesses that offer foods that provide nutrients needed to
sustain health and provide energy (Richardson, 2010). These outlets sell health foods, organic
foods, local produce, and nutritional supplements. Generally, supermarkets, grocery stores, food
co-ops are grouped as one entity and farmer’s markets fall into the category of healthy food
sources (Richardson, 2010). They offer an array of nutritious foods on the food pyramid that are
beneficial to healthy living. Table 2 lists the number of healthy food sources in Minnesota.
Supermarket, grocery store, and food co-op data was attained from the Supermarket Access Map
in ArcGIS Online (Richardson, 2010). Data pertaining to the number of farmer’s markets in
Minnesota was gathered from the Minnesota Department of Agriculture Directory (2019).
Table 2 Healthy Food Sources in Minnesota

Store Types Original Counts
Supermarkets, Grocery Stores, Food Co-ops 1332
Farmer’s Markets 196
Totals 1528

27

3.2.3 Correlation of the Dependent Variable and Explanatory Variables

Correlation was used to test the relationship between the potential explanatory variables and the
dependent variable. In Microsoft Excel, the CORREL function was used to find the correlation
between two variables. A correlation coefficient of +1 indicates a perfect positive correlation,
which means that as x increases, variable y increases and while variable x decreases, variable y
decreases. In contrast, a correlation of -1 indicates a perfect negative correlation, as variable x
increases, variable z decreases and as variable x decreases, variable z increases. When
visualized, the x-axis represents the explanatory variables and the y-axis represents the
dependent variable (in this case obesity prevalence).
3.3 Regression Analysis ArcGIS and ArcMap
3.3.1 Ordinary Least Squares
The relationship between the dependent variable and explanatory variables was examined on a
county-wide basis with a cross-sectional analysis by using Ordinary Least Squares (OLS).
Multicollinearity refers to the state of very high inter-correlations or inter-associations among
independent variables (Shrestha et al. 2013). The OLS model assigns an equation to all the
features being analyzed and predicted. OLS’s purpose is to test the significance of explanatory
variables and potential multicollinearity amongst the variables. Using Variation Inflation Factor
(VIF) and Variable Significance (VS) values, multicollinearity addressed by removing variables
with a VIF over 7.5. OLS was run again to mitigate multicollinearity.
Five potential explanatory variables of physical inactivity, median family income,
poverty prevalence, unemployment, and heathy food source density were selected for regression
analysis. The purpose of OLS was provide a global model of the dependent variable, obesity
prevalence, and try to predict the phenomenon by creating a regression equation to represent the
28

process. In ArcMap, the OLS tool prompted a pop-up screen, in which an Input Feature Class
with Unique ID Field was specified. Obesity prevalence was acknowledged as the Dependent
Variables and the five explanatory variables was listed in the Explanatory Variables section. The
OLS analysis was run and generated an output feature class.
3.3.2 Exploratory Regression Analysis

Exploratory Regression Analysis was used to evaluate all possible combination of the input
variables, searching for OLS models that best explained the dependent variable within guidelines
of criteria specified. The Exploratory Regression tool mined data for all possible combinations of
explanatory variables to see which models passed all the OLS diagnostics. The minimum and
maximum number of explanatory variables in each model was set at 1 and 5 respectively, with
default threshold criteria for Adjusted R2, coefficient p-values, VIF values, Jarque-Bera values,
and spatial autocorrelation p-values. The Exploratory Regression analysis ran OLS on every
possible combination of explanatory variable listed in Table 3, with at least the minimum
number of explanatory variables and no more than the maximum number of explanatory
variables specified. The dependent variable was obesity prevalence. Each model was assessed
against the default threshold criteria. If the model exceeded the specified Adjusted R2 threshold,
had coefficient p-values for all explanatory variables less than the threshold, had coefficient VIF
values for all explanatory variables less than the threshold, and returned a Jarque-Bera p-value
larger than anticipated, the Spatial Autocorrelation tool was run on the model’s residuals. If the
spatial autocorrelation p-value was larger than the specification in the search criteria, the model
was deemed to have passed. A properly specified OLS model is validated with statistically
significant explanatory variables, with small VIF values indicating non-redundancy. The
coefficients reflect the strength of the relationship between the explanatory variables and the
29

dependent variable. Normally distributed residuals indicated a non-biased model, namely a
Jarque-Bera value that is not statistically significant. A properly specified OLS model also has a
random distribution of over and under predictions.
Table 3 Explanatory Variables (Units)
Explanatory Variables (Per County)
Obesity Prevalence (%)
Physical Inactivity (%)
Total Population (#)
Median Family Income ($)
Poverty Prevalence (%)
Language Other Than English in Household (%)
Foreign Born (%)
Unemployment (%)
Population 25 and Over with Associates Degree (%)
Population 25 and Over with Bachelor’s Degree (%)
Population 25 and Over with Master’s or Professional Degree (%)
Source Count (Accumulation of Supermarkets and Farmer’s Markets) (#)
Population Density (#)
Source Density (#)

3.3.3 Spatial Autocorrelation
Spatial autocorrelation measures the correlation between variable in space. Spatial
autocorrelation, also known as the clustering of residuals, is a symptom of misspecification. This
30

occurs when key explanatory variables are missing. Moran’s I was utilized to test for spatial
autocorrelation and verify that systematic patterns and biases did exist in the model.
After running an exploratory regression analysis of the 14 explanatory variables in
relation to the dependent variable, a spatial autocorrelation analysis was conducted. Spatial
autocorrelation indicates whether there was clustering or dispersion in the correlation between
the explanatory variables and dependent variable. Spatial autocorrelation confirmed if there’s a
significant statistical pattern in the data. A positive Moran’s I indicates that the data was
clustered. In contrast, a negative Moran’s I implies that the data was dispersed. The Spatial
Autocorrelation tool in the Spatial Statistics toolbox of ArcMap used Global Moran’s I function
to compute the z-score value of the correlation between the explanatory variable and the
dependent variable. The z-score value helped determine whether Moran’s I should be classified
as positive, negative, or no spatial autocorrelation. Spatial autocorrelation displayed how the
explanatory variables and the dependent variable spatial relationship geographically.

31

Chapter 4 Results
The results of the analysis are highlighted in this chapter. To test the correlations between the
explanatory variables and the dependent variable, explanatory variables that were thought to
contribute most to obesity prevalence were examined. Five explanatory variables were analyzed
in OLS to correlate with obesity prevalence. 14 explanatory variables were analyzed in an
exploratory regression analysis to find the best model that showcased the spatial relationship
between the explanatory variables and the dependent variable. First, each explanatory variable
was correlated with obesity prevalence in Excel to visually show the statistical significance
between the explanatory variables and the dependent variable before running the regression
analysis.
4.1 Excel Correlations of Explanatory Variables and Dependent Variable
Overall, the relationships between the explanatory variables and the dependent variables yielded
low R-Squared values and poor regression line fits, though some relationships of course were
stronger than others.
In Figure 3, the R-Squared value equaling 0.0825 of Physical Inactivity vs. Obesity
Prevalence indicated a very poor regression line fit. Eight percent of the variation in obesity
prevalence is explained by the independent variable physical inactivity.
32

Figure 3 Physical Inactivity vs. Obesity Prevalence Correlation

In Figure 4, the R-Squared value equaling 0.1123 of Diabetes Prevalence vs. Obesity
Prevalence indicated a very poor regression line fit. 11% of the variation in obesity prevalence is
explained by the independent variable diabetes prevalence.

Figure 4 Diabetes Prevalence vs. Obesity Prevalence correlation
33

In Figure 5, the R-Squared value equaling 0.1541 of Total Population vs. Obesity
prevalence indicated a very poor regression line fit. 15% of the variation in obesity prevalence is
explained by the independent variable total population.

Figure 5 Total Population vs. Obesity Prevalence Correlation

In Figure 6, the R-Squared value equaling 0.0182 of Median Family Income vs. Obesity
Prevalence indicated a very poor regression line fit. 2% of the variation in obesity prevalence is
explained by the independent variable median family income.
34

Figure 6 Median Family Income vs. Obesity Prevalence Correlation

In Figure 7, the R-Squared value equaling 0.0133 of Poverty Prevalence vs. Obesity
Prevalence indicated a very poor regression line fit. 1% of the variation in obesity prevalence is
explained by the independent variable poverty prevalence.

Figure 7 Poverty Prevalence vs. Obesity Prevalence Correlation
35

In Figure 8, the R-Squared value equaling 0.0676 of Language Other Than English vs.
Obesity Prevalence indicated a very poor regression line fit. 7% of the variation in obesity
prevalence is explained by the independent variable language other than English.

Figure 8 Language Other Than English vs. Obesity Prevalence Correlation

In Figure 9, the R-Squared value equaling 0.0987 of foreign-born vs obesity prevalence
indicated a very poor regression line fit. 10% of the variation in obesity prevalence is explained
by the independent variable foreign born.
36

Figure 9 Foreign Born vs. Obesity Prevalence Correlation

In Figure 10, the R-Squared value equaling 0.0543 of Unemployment vs. Obesity
Prevalence indicated a very poor regression line fit. 5% of the variation in obesity prevalence is
explained by the independent variable unemployment.

Figure 10 Unemployment vs. Obesity Prevalence Correlation
37

In Figure 11, the R-Squared value equaling 0.1785 of Bachelor’s Degree vs. Obesity
Prevalence indicated a very poor regression line fit. 17% of the variation in obesity prevalence is
explained by the independent variable bachelor’s degree.

Figure 11 Associates Degree vs. Obesity Prevalence Correlation
In Figure 12, the R-Squared value equaling 0.0185 of Associates Degree vs. Obesity
Prevalence indicated a very poor regression line fit. 2% of the variation in obesity prevalence is
explained by the independent variable associate degree.
38

Figure 12 Bachelor’s Degree vs. Obesity Prevalence Correlation

In Figure 13, the R-Squared value equaling 0.1461 of Professional/Master’s Degree vs.
Obesity Prevalence indicated a very poor regression line fit. 15% of the variation in obesity
prevalence is explained by the independent variable professional/master’s degree.

Figure 13 Professional/Master’s Degree vs. Obesity Prevalence Correlation
39

In Figure 14, the R-Squared value equaling 0.1513 of Source Count vs. Obesity
Prevalence indicated a very poor regression line fit. 15% of the variation in obesity prevalence is
explained by the independent variable source count.

Figure 14 Source Count vs. Obesity Prevalence Correlation

In Figure 15, the R-Squared value equaling 0.1272 of Population Density vs. Obesity
Prevalence indicated a very poor regression line fit. 13% of the variation in obesity prevalence is
explained by the independent variable population density.
40

Figure 15 Population Density vs. Obesity Prevalence Correlation

In Figure 16, the R-Squared value equaling 0.1109 of Source Density vs. Obesity
Prevalence indicated a very poor regression line fit. 11% of the variation in obesity prevalence is
explained by the independent variable source density.

Figure 16 Source Density vs. Obesity Prevalence Correlation
41

As displayed in Figures 3-16, all 14 explanatory variables have poor regression line fits,
though some explanatory variables performed better than others. Explanatory variables 25 and
over with a bachelor’s degree, total population, and source count had the highest R-Squared
values, indicating that those three had the strongest relationships with obesity prevalence.
Explanatory variable 25 and over with a bachelor’s degree was positively correlated with obesity
prevalence. In contrast, explanatory variables total population and source count were negatively
correlated with the dependent variable. Although the explanatory variables are not statistically
significant, an objective of the analysis was to correlate the explanatory variables against obesity
prevalence in Minnesota prior to conducting regression analyses. It was necessary to include all
these explanatory variables for comparison between the hypothesized and actual results. A
thorough analysis of the statistics is shown in Table 4.

Table 4 Correlations of Explanatory Variables to Obesity Prevalence Chart
Explanatory
Variable
R-Squared R-Value Correlation
25 and Over with
Bachelor’s Degree
0.1785 0.422492603 -0.422449991
Total Population 0.1541 0.392555729 -0.392602856
Source Count 0.1513 0.388973007 -0.388988064
25 and Over with
Professional/Master’s
Degree
0.1461 0.382230297 -0.382206529
Population Density 0.1272 0.35665109 -0.356698191
Diabetes Prevalence 0.1123 0.335111922 0.335083855
42

Source Density 0.1109 0.333016516 -0.333033709
Foreign Born 0.0987 0.314165561 -0.314123159
Physical Inactivity 0.0825 0.287228132 0.287206316
Language Other
Than English
0.0676 0.26 -0.259958827
Unemployment 0.0543 0.233023604 0.232968462
25 and Over with
Associate’s Degree
0.0185 0.136014705 0.136115005
Median Family
Income
0.0182 0.134907376 -0.134798968
Poverty Prevalence 0.0133 0.115325626 0.11539687

4.2 Regression Modeling
4.2.1 Ordinary Least Squares Results
The second step in the analysis involved running an OLS Regression analysis. The results of the
hypothesized OLS in Table 5 produced a linear model with the 5 explanatory variables that were
thought to have produced the best model to analyze obesity. Physical inactivity, poverty
prevalence, and unemployment all showed positive relationships with the rate of obesity.
Explanatory variables of median family income and source density had indirect relationships
with obesity rates. The t-statistic and probability values for the explanatory variables physical
inactivity, median family income, and source density suggest that those variables are statistically
significant to the model at the 95%, 95%, and 99% confidence level, respectively. All VIF values
for the explanatory variables in the model are indicative of the removal of multicollinearity
43

insofar as they all have VIF values < 7.5. Although explanatory variables of poverty prevalence
and unemployment were not statistically significant, an objective of the analysis was to compare
influence of space on obesity prevalence in Minnesota. It was necessary to include these
explanatory variables for comparison between the hypothesized and actual results.

Table 5 OLS Model of Hypothesized Contributing Variables to Obesity
Variable Coefficient t-Statistic Probability VIF
Intercept 24.641593 6.662621 0.000000* ----
Physical
Inactivity
0.437778 2.009759 0.047785* 1.239712
Median Family
Income
-0.000048 -2.029751 0.045664* 1.034531
Poverty
Prevalence
0.035709 0.423837 0.672814 1.166013
Unemployment 0.168059 0.706591 0.481845 1.372532
Source Density -7.504010 -2.794633 0.006484* 1.100765

In Table 6, the values for multiple r-squared and adjusted r-squared were 0.211 and
0.162779 respectively, resulting in an AICc value of 410.857562. These values indicate a less
than optimum model fit. The Koenker statistic, p-value, yielded a value of 8.804676, and also
wasn’t statistically significant. The explanatory variables in the analysis do not have a consistent
relationship with the dependent variable of obesity prevalence in both geographic and data space.
This indicates that the model represents stationarity of the variables. However, the model is
unbiased. The standardized residuals of the model follow a normal distribution.

44

Table 6 OLS Diagnostics of Hypothesized Contributing Variables to Obesity
Number of Observations 87
Multiple R-Squared [d] 0.211
Joint F-Statistic [e] 4.344152
Joint Wald Statistic [e] 23.006088
Koenker (BP) Statistic [f] 8.804676
Jarque-Bera Statistic [g] 1.586573
Akaike’s Information Criterion (AICc) [d] 410.857562
Adjusted R-Squared [d] 0.162779
Prob(>F), (5.81) degrees of freedom 0.001492*
Prob(>chi-squared), (5) degrees of freedom 0.000337*
Prob(>chi-squared), (5) degrees of freedom 0.117113
Prob(>chi-squared), (2) degrees of freedom 0.452356

4.2.2 Exploratory Regression Analysis Results
An Exploratory Regression analysis was run after the OLS Regression analysis. The results of
the exploratory regression analysis produced a linear model using the four explanatory variables
that produced the best model to analyze obesity (see Table 7). These explanatory variables
accounted for most of the variance observed for obesity prevalence. Diabetes prevalence was the
lone variable to have positive relationship with the rate of obesity. Explanatory variables of
median family income, adults 25 and over with a bachelor’s degree, and source density all
showed indirect relationships with obesity rates. The t-statistic and probability values for the
explanatory variable source count suggest that it was statistically significant at the 95%
45

confidence level. All VIF values for the explanatory variables in the model were indicative of the
removal of multicollinearity. All have VIF values < 7.5. Although explanatory variables of
diabetes prevalence, median family income, and adults 25 and over with a bachelor’s degree are
not statistically significant, as a reminder, the objective of the analysis was to compare influence
of space on the obesity prevalence in Minnesota. It was necessary to include these explanatory
variables for comparison between the hypothesized and actual results.

Table 7 Exploratory Regression Analysis
Variable Coefficient t-Statistic Probability VIF
Intercept 32.103069 9.911752 0.000000* ----
Diabetes
Prevalence
0.328595 1.346891 0.181730 1.433286
Median Family
Income
-0.000044 -1.950838 0.054492 1.012067
Bachelor’s
Degree
-0.124009 -1.800979 0.075386 1.758920
Source Count -0.017784 -2.245523 0.027421* 1.355902

The diagnostics of the regression analysis is shown in Table 8. The values for multiple r-
squared and adjusted r-squared were 0.267561 and 0.231832 respectively, resulting in an AICc
value of 402.068415. These values are conclusive of a less than ideal model fit. The Koenker
Statistic, p-value, yielded a value of 5.376321, and isn’t statistically significant. The explanatory
variables from the results of the exploratory analysis do not have a consistent relationship with
the dependent variable of obesity prevalence in both geographic and data space. This indicates
that the model represents stationarity of the variables. However, the model, like the hypothesized
46

model, is unbiased. The standardized residuals of the model follow a normal distribution. Table 9
shows a simplified synopsis of the statistics associated with Exploratory Regression analysis.

Table 8 Diagnostics of Regression Analysis
Number of Observations 87
Multiple R-Squared [d] 0.267561
Joint F-Statistic [e] 7.488662
Joint Wald Statistic [e] 176.735692
Koenker (BP) Statistic [f] 5.376321
Jarque-Bera Statistic [g] 2.139391
Akaike’s Information Criterion (AICc) [d] 402.068415
Adjusted R-Squared [d] 0.231832
Prob(>F), (4,82) degrees of freedom 0.000034*
Prob(>chi-squared), (4) degrees of freedom 0.000000*
Prob(>chi-squared), (4) degrees of freedom 0.250817
Prob(>chi-squared), (2) degrees of freedom 0.343113

Table 9 Exploratory Analysis: Highest Adjusted R-Squared Results
AdjR2 AICc JB K(BP) VIF SA Model
0.23 402.07 0.34 0.25 1.76 0.89 Diabetes
Prevalence,
Median
Family
Income*,
Bachelor’s
Degree*,
Source
Count***

47

4.3 Spatial Autocorrelation Results
A graphical summary of the Spatial Autocorrelation Report was generated as an HTML file after
running the tool for both OLS and Exploratory Regression analyses in ArcMap as shown in
Figure 17. Given a set of explanatory variables and a dependent variable for each analysis. The
Spatial Autocorrelation tool evaluated whether the pattern expressed was clustered, dispersed, or
random. The tool also calculated the z-score and p-value to verify the significance of the
contributing explanatory variables to the dependent variable. Figure 17 was used as reference.

Figure 17 Graphical Summary of Spatial Autocorrelation for Reference
The statistical output of Moran’s I for the hypothesis trial is shown in Table 10. Yielding
a z-score of 0.491087, the pattern does not appear to be significantly different than random, with
reference to Figure 17. The illustration in Figure 18. shows the standardized distribution of
residuals across Minnesota’s counties. The figure does not display spatial patterns, hence the
48

results agreeing Moran’s I. The model performed decently in 4 of the 87 counties, with those
counties correlating with a salmon color, being under predicted (having a standard deviation of
residual between 1.5 – 2.5).

Table 10 Global Moran’s I Summary of Hypothesized Variables

Global Moran’s I Summary – Hypothesis
Moran’s Index 0.014294
Expected Index -0.011628
Variance 0.002786
z-score 0.491087
p-value 0.623365
49

Figure 18 Spatial Distribution of OLS Standardized Residuals of Hypothesized Variables
The statistical output of Moran’s I for the Exploratory Regression analysis of explanatory
variables that calculated the best model is shown in Table 11. Yielding a z-score of 0.599203, the
pattern does not appear to be significantly different than random, with reference to Figure 17.
The illustration in Figure 19 shows the standardized distribution of residuals across Minnesota’s
counties. The figure does not display spatial patterns, hence (similar to the results in the
hypothesized analysis) concurs with the findings from Moran’s I. The model performed decently
in 8 of the 87 counties in Minnesota, with those counties being under predicted (having a
standard deviation of residual between 1.5 – 2.5) (See Figure 19).

50

Table 11 Global Moran’s I Summary of Exploratory Regression Variables

Figure 19 Spatial Distribution of Exploratory Regression Variables Contributing to Obesity
Global Moran’s I Summary – Exploratory Regression
Moran’s Index 0.020037
Expected Index -0.011628
Variance 0.002793
z-score 0.599203
p-value 0.549037
51

4.4 Geographically Weighted Regression
The Koenker test for both analyses were statistically insignificant, implying non-stationarity of
the relationship between the explanatory variables and dependent variable. Therefore,
Geographically Weighted Regression (GWR) was not necessary.

52

Chapter 5 Discussion and Conclusion
5.1 Summary and Significance of Findings
Obesity is a significant public health issue and approaching the topic spatially may provide
stakeholders with direction as to how address the phenomenon. It is also a complicated topic that
has defied many researchers attempts to understand it. This study showed that obesity rates in
Minnesota are impacted by various factors according to the OLS model. The AICc and r-squared
values for the hypothesized model were 410.857562 and 0.211 respectively. The AICc and r-
squared values for the model after running an exploratory regression of all explanatory variables
were 402.068415 and 0.23 respectively. These values suggest that under a quarter of the variance
in obesity rates can be explained using this model. Consequently, these models are poor, and
reflect the challenges in modeling the spatial relationship between obesity and other
demographic and economic factors.
One explanation of the low variances in these models could be that additional factors are
required to explain the distribution of obesity rates across Minnesota. This could be because only
14 variables were included in this study, and many other social, cultural, and behavioral
variables were left out. The scale of the analysis, the county level, may have generalized the data.
Looking at the problem using a different scale of analysis, such as the census tract, may yield
more conclusive results. Using disaggregated data could make future analysis more statistically
robust. In addition, studies have shown that obesity is directly influenced seasonal eating habits,
a temporal scale that this study did not take into account. This could be incorporated into future
analyses. Other factors such as healthy food affordability, purchasing decisions, transportation,
walkability, technical and regulatory protocol may also help explain the spatial distribution and
prevalence of obesity, but were not included in this study.
53

5.2 Study Limitations and Future Research
5.2.1 Study Limitations
Although this study examined the spatial distribution of obesity at the county level, studies that
examine obesity at the community scale have been recommended for prevention and intervention
purposes. To do so, a cross-sectional spatial data analysis is ideal. Results would yield
geographical variations in obesity between rural and urban communities. Understanding
community scale obesity trends would better highlight associated behavioral determinants like
diet and physical activity, as well as built environments, socio-economics, and how each
determinant contributes to obesity.
Of the 5.611 million residents of Minnesota, 3.6 million live in Hennepin and Ramsey
Counties, accounting for ~65% of the total population of the state (ACS, 2017). Yet the state of
Minnesota consists of 87 counties. The aggregated data of the explanatory variables could be
subject to the Modifiable Areal Unit Problem (MAUP), where the choice of analytical entities
may influence the spatial patterning and variability of the data and any ensuing interpretations
(Sharkey et al. 2009). Hence, the presence or absence of healthy food outlets would vary greatly
across the study area as more healthy food sources are directly proportional to population. In this
study, proportional comparisons of socioeconomics and sociodemographics revealed disparities
on the county level.
The results of this study suggest the complexity of anticipating phenomenon like obesity,
and that methodological and analytical approaches must be carefully chosen. The study only
analyzed accessibility to healthy food sources and did not include accessibility to unhealthy food
sources, which provides a less comprehensive analysis. The study only measured access to
healthy food sources based on count instead of incorporating roadways and routes. The study
54

also only uses statistical methods such as correlation and regression to examine the relationship
between healthy food sources, socio-economics, and socio-demographics, and their correlations
with obesity. This suggests that statistical data assumes observations are independent or the
statistical relationships remain unchanged across the study area.
5.2.2 Future Analyses
Given the relatively weak findings in this study, it is suggested that future analyses implement
average nearest distances to fast food outlets and/or healthy food sources to examine their
prevalence on obesity rates. Shresta et. al (2013) results showed that obesity rates in
Pennsylvania correlated with diabetes, physical inactivity, and average nearest distance to the
nearest healthy food store after running an OLS Regression analysis. The AICc and R-Squared
values were 299.87 and 0.34 respectively, inferring that only 34% of the variance in obesity rates
were explained using OLS. Shresta et. al (2013) also ran a GWR analysis, which yielded AICc
and R-Squared values of 261.59 and 0.45, respectively. Their model suggested that additional
explanatory variables were required to account for the variance in obesity rates in Pennsylvania.
Shresta et. al (2013) approach was different in comparison to this analysis. For example, Shresta
et. al (2013) used the Network Analyst tool in ArcMap to successfully determine food
accessibility and the average nearest health food facilities to the centroid of each census tract in
Pennsylvania. In a network analysis, accessibility can be measured in terms of travel time,
distance, or other criteria. Evaluating accessibility can help answer basic questions such as how
many people live within a 10-minute drive from healthy food outlet? How many people live
within a half-kilometer walking distance from a grocery store? A network analysis was not used
in this study due to the large extent of the study area’s scale, data restrictions, and the overall
time to upload and download massive amounts of data from the ArcMap file. Implementing a
55

network system may have contributed to better results in this analysis. Examining accessibility
helps determine whether a community has access to healthy food outlets. Results would yield
geographical variations in obesity between rural and urban communities. It better highlights
behavioral determinants like diet and physical activity, as well as built environments, socio-
economics, and how each determinant contributes to obesity.
In conclusion, this study attempted to understand the relationship between a variety of
predictive factors and obesity rates within Minnesota. The existing literature was carefully
reviewed and, despite using Shrestha et al. (2013) as a template for the methodology, the results
were suggestive rather than conclusive. That said, diabetes rates, education, and age were all
found to have some relationship with obesity, despite not being statistically significant. The
relationship between obesity, the decisions individuals make about their dietary health, and
geographic proximity is a complex phenomenon. This study, by pointing out which variables
were not explanatory at the county level, can prompt future researchers to choose a wider variety
of variables, a different statistical technique, and/or a different scale of analysis.

56

References
Bressie, Shanna. “Spatial Patterns of Food Accessibility Across Lane County, Oregon in
2015-2016.” Master’s Thesis, University of Southern California, 2016.

Boone-Heinonen, Janne, Kiefe, Catarina I., Gordon-Larsen, Penny. “Fast Food Restaurants and
Food Stores Longitudinal Associations with Diet in Young to Middle-Aged Adults: The
CARDIA Study.” Arch Intern Med 171, (2011): 1162-1170.
doi:10.1001/archinternmed.2011.283.

Centers for Disease Control and Prevention. “Adult Obesity Prevalence Maps.” Center for
Disease Control and Prevention, last modified 25 March 2019, accessed 18 February
2019, 2019, https://www.cdc.gov/obesity/data/prevalence-maps.html

Centers for Disease Control and Prevention. “Physical Activity.” Center for Disease Control and
Prevention, last modified 13 May 2019, accessed 18 February 2019,
https://www.cdc.gov/nccdphp/dnpao/data-trends-maps/index.html

Centers for Disease Control and Prevention. “Diabetes.” Center for Disease Control and
Prevention last modified 30 May 2019, 2019, accessed 18 February 2019,
https://gis.cdc.gov/grasp/diabetes/DiabetesAtlas.html.

Chalkias, Christos, Apostolos G. Papadopoulos, Kleomenis Kalogeropoulos, Kostas Tambalis,
Glykeria Psarra, and Labros Sidossis. “Geographic heterogeneity of the relationship
between childhood obesity and socio-environmental status: Empirical evidence from
Athens, Greece.” Applied Geography 37, (2013): 34-43.
doi: http://dx.doi.org/10.1016/j.apgeog.2012.10.007

Chen, Xiang. “Take the edge off: A hybrid geographic food access measure.” Applied
Geography 87, (2017): 149-159. doi: http://dx.doi.org/10.1016/j.apgeog.2017.07.013

Chi, Sang-Hyun, Diana S. Grigsby-Toussaint, Natalie Bradford, and Jinmu Choi. “Can
Geographically Weighted Regression improve out contextual understanding of obesity in
the US? Findings from the USDA Food Atlas.” Applied Geography 44, (2013): 134-142.
doi: http://dx.doi.org/10.1016/j.apgeog.2013.07.017

Cubbin, Catherine, Jina Jun, Claire Margerison-Zilko, Nicolas Welch, James Sherman, Talia
McCray, and Barbara Parmenter. “Social Inequalities in neighborhood conditions: spatial
relationships between sociodemographic and food environments in Alameda, California.”
Journal of Maps 8, no.4 (2012): 344-348. doi: 10.1080/17445647.2012.747992

Cruz, Hildemar. “A Geospatial Analysis of Income Level, Food Deserts and Urban Agriculture
Hot Spots.” Master’s Thesis, University of Southern California, 2016

57

Daniel PhD, Mark, Yan Kestens, PhD and Catherine Paquet, PhD. “Demographic and Urban
Form Correlates of Healthful and Unhealthful Food Availability in Montréal, Canada.”
Canadian Journal of Public Health 100, no.3 (2009): 189-193. url: https://www-jstor-
org.libproxy1.usc.edu/stable/41995243

Deitz, Shiloh L. “A Spatial Analysis of the Relationship Between Obesity and the Built
Environment in Southern Illinois.” Master’s Thesis, Southern Illinois University
Carbondale, 2014

Drewnowski, A., AV Moudon, J. Jiao, A. Aggarwal, H. Charreire, and B. Chaix. “Food
environment and socioeconomic status influence obesity rates in Seattle and in Paris.”
International Journal of Obesity, no.38 (2014): 308-314. doi: 10.1038/ijo.2013.97

ESRI 2011. ArcGIS Desktop: Release 10. Redlands, CA: Environmental Systems Research
Institute

Ferguson, Neil S., Karen E. Lamb, Yang Wang, David Ogilvie, and Anne Ellaway. “Access to
Recreational Physical Inactivity by Car and Bus: An Assessment of Socio-Spatial
Inequalities in Mainland Scotland.” PLoS ONE 8(2) (2013): e55638.
doi: 10.1371/journal.pone.0055638.

Fleischhacker, S.E., K.R. Evenson, D.A. Rodriguez, and A.S. Ammerman. “A systematic review
of fast food access studies.” Obesity Reviews 12, (2011): e460-e471. doi: 10.1111/j.1467-
789X.2010.00715.x

Gard, Julienne. “A Spatial Accessibility Analysis of the Los Angeles Foodscape.” Doctorate’s
Dissertation, University of Southern California, 2015

Helbich, Marco, Bjorn Schadenberg, Julian Hagenauer, and Maartje Poelman. “Food deserts?
Healthy food access in Amsterdam.” Applied Geography 83, (2017): 1-12.
doi: http://dx.doi.org/10.1016/j.apgeog.2017.02.015

Hellmich, Simon N. “What is Socioeconomics? An Overview of Theories, Methods, and Themes
in the Field.” Forum for Social Economics 46, (2015): 3-25.
doi:10.1080/07360932.2014.999696

Hilmers, Angela, MD, MS, David C. Hilmers, MD, MPH, and Jayna Dave, PhD. “Neighborhood
Disparities in Access to Healthy Foods and Their Effects on Environmental Justice.”
American Journal of Public Health 102, no. 9 (2012): 1644-1654.
doi: 10.2105/AJPH.2012.300865

Horner, Mark W., and Brittany S. Wood. “Capturing individuals ‘food environments using
flexible space-time accessibility measures.” Applied Geography 51, (2014): 99-107. doi:
http://dx.doi.org/10.1016/j,apgeog.2014.03.007

58

Howard, Philip H., Margaret Fitzpatrick, and Brian Fulfrost. “Proximity of food retailers to
schools and rates of overweight ninth grade students: an ecological study in California.”
BMC Public Health 11, no. 68 (2011): 1-8. doi: 10.1186/1471-2458-11-68

Jeffery, Robert W., Judy Baxter, Maureen McGuire, and Jennifer Linde. “Are fast food
restaurants an environmental risk factor for obesity?” International Journal of Behavioral
Nutrition and Physical Activity 3, no. 2 (2006). doi: 10.1186/1479-5868-3-2

Larsen, Kristian and Jason Gilliland. “Mapping the evolution of ‘food deserts’ in a Canadian
city: Supermarket accessibility in London, Ontario, 1961-2005.” International Journal of
Health Geographics 7, no. 16 (2008): 1-16. doi: 10.10.1186/1476-072X-7-16

Larson, Nicole I., PhD, MPH, RD, Mary T. Story, PhD, RD, and Melissa C. Nelson, PhD, RD.
“Neighboring Environments: Disparities in Access to Healthy Foods in the U.S.”
American Journal of Preventive Medicine 36, no. 1 (2009): 74-81.

Malina, Robert M. “Ethnic variation in the prevalence of obesity in North American children and
youth.” Critical Reviews in Food Science and Nutrition 33, (1993): 389-396.
doi: http://dx.doi.org/10.1080/10408399309527637

Mattessich, Paul W., Rausch, Ela J. “Healthy Food Access: A View of the Landscape in
Minnesota and Lessons Learned from Healthy Food Financing Initiatives.” Wilder
Research: (2016). 91-54. url:
https://www.wilder.org/WilderResearch/Publications/Studies/Healthy.pdf.

Minnesota Compass. Wilder Research (2017). “Near-North Neighborhood,”
www.mncompass.org/profiles/neighborhoods/minneapolis/near-north#notes.

Minnesota Department of Agriculture. “Minnesota Grown Directory.”
https://minnesotagrown.com/search-directory/farmers-markets/?gclid=EAIaIQobChMI-
aPZ2pHI5AIVF6SzCh25UQuQEAAYASAAEgIut_D_BwE

Minnesota Department of Health. 2017. “Adult Obesity in Minnesota 2017: Data Brief.”
https://www.health.state.mn.us/people/obesity/docs/obesitybrief.pdf.

Minneapolis Department of Health and Family Support. 2012. “Minneapolis Healthy Corner
Store Program: Making produce more visible, affordable and attractive,”
www.health.state.mn.us/divs/oshii/docs/Mpls_Healthy_Corner.pdf.

“Minneapolis, Minnesota (MN) Poverty Rate Data,” 2017. Avameg Inc. www.city-
data.com/poverty/poverty-Minneapolis-Minnesota.html.

Morland, Kimberly, PhD, Steve Wing, PhD, Ana Diez Roux, MD, PhD, and Charles Poole, ScD.
“Neighborhood Characteristics Associated with the Location of Food Stores and Food
Service Places.” American Journal of Preventive Medicine 22, no.1 (2001): 23-29.

59

Olson, Jeremy. 2016. “Minnesota among 10 worst states for access to fresh food: Leaders look
for Legislature for state funding solutions.” Star Tribune.
http://www.startribune.com/minnesota-among-10-worst-states-for-food-
deserts/375573111/.

Penney, T. L., D. G. C. Rainham, T. J. B. Dummer, and S. F. L. Kirk. “A spatial analysis of
community level overweight and obesity.” Journal of Human Nutrition and Dietetics 27,
no. 2 (2013): 65-74. doi: 10.1111/jhn.12055

Richardson, Karen. “Exploring Food Environments: Assessing access to nutritious food.”
ArcUser Online (2010): 50-52. accessed 1 March 2019
https://www.esri.com/news/arcuser/1010/files/foodataset.pdf.

Sharkey, Joseph R., Cassandra M. Johnson, Wesley R. Dean, and Scott A. Horel. “Focusing on
fast food restaurants alone underestimates the relationship between neighborhood
deprivation and exposure to fast food in a large rural area.” Nutrition Journal 10, no. 10
(2011): 1-14. doi: http://www.nutritionj.com.content/10/1/10

Shrestha, Ranjay, Ron Mahabir, and Liping Di. Healthy Food Accessibility and Obesity: Case
Study of Pennsylvania, USA. 2013.

Stein, Dana Beth. “‘Food Deserts’ and ‘Food Swamps’ in Hillsborough County, Florida:
Unequal Access to Supermarkets and Fast-food Restaurants.” Master’s Thesis, University
of South Florida, 2011.

Todd, Michael, Mark A. Adams, Jonathan Kurka, Terry L. Conway, Kelli L. Cain, Matthew P.
Bunman, Lawrence D. Frank, James F. Sallis, and Abby C. King. “GIS-measured
walkability transit, and recreation environments in relation to older Adults’ physical
inactivity: A latent profile analysis.” Preventive Medicine 93, (2016): 57-63.
doi: http://dx.doi.org/10.1016/j.ypmed.2016.09.019

Xu, Yanqing, Wang, Lei. “GIS-based analysis of obesity and the built environment in the US.”
Cartography and Geographic Information Science 42, no.1 (2015): 9-21.
doi: 0.1080/15230406

Wen, Tzai-Hung, Duan-Rung Chen, and Meng-ju Tsai. “Identifying geographical variations in
poverty-obesity relationships: empirical evidence from Taiwan.” Geospatial Health 4,
no. 2 (2010): 257-265. doi: https://doi.org.10.4081/gh.2010.205

60

Appendix A Maps of Aggregated Data Used in Analysis

61

62

63

64

65

66

67

68

69

70

71

72

73

74

Appendix B Exploratory Regression Model – Raw Results
******************************************************************************
Choose 5 of 15 Summary
Highest Adjusted R-Squared Results
AdjR2 AICc JB K(BP) VIF SA Model
0.23 403.34 0.31 0.50 1.76 0.76 +DIABETES_PREVALENCE -
MEDIAN_FAMILY_INCOME**
+POPULATION_25_AND_OVER_ASSOCIATES_DEGREE -
POPULATION_25_AND_OVER_BACHELORS_DEGREE* -SOURCE_COUNT*
0.23 403.59 0.48 0.19 3.10 0.76 +DIABETES_PREVALENCE
+POPULATION_25_AND_OVER_HIGH_SCHOOL_GRADUATE -
MEDIAN_FAMILY_INCOME* -POPULATION_25_AND_OVER_BACHELORS_DEGREE*
-SOURCE_COUNT*
0.23 403.59 0.46 0.19 2.57 0.68 +PHYSICAL_INACTIVITY_PREVALENCE
+POPULATION_25_AND_OVER_HIGH_SCHOOL_GRADUATE -
MEDIAN_FAMILY_INCOME** -
POPULATION_25_AND_OVER_BACHELORS_DEGREE** -SOURCE_COUNT*
Passing Models
AdjR2 AICc JB K(BP) VIF SA Model

******************************************************************************
Choose 6 of 15 Summary
Highest Adjusted R-Squared Results
AdjR2 AICc JB K(BP) VIF SA Model
0.23 405.18 0.35 0.47 24.91 0.75 +DIABETES_PREVALENCE +TOTAL_POPULATION -
MEDIAN_FAMILY_INCOME* +POPULATION_25_AND_OVER_ASSOCIATES_DEGREE
-POPULATION_25_AND_OVER_BACHELORS_DEGREE* -SOURCE_COUNT
0.23 405.42 0.40 0.26 3.23 0.70 +DIABETES_PREVALENCE
+POPULATION_25_AND_OVER_HIGH_SCHOOL_GRADUATE -
MEDIAN_FAMILY_INCOME* +POPULATION_25_AND_OVER_ASSOCIATES_DEGREE
-POPULATION_25_AND_OVER_BACHELORS_DEGREE* -SOURCE_COUNT*
0.22 405.53 0.53 0.18 24.87 0.75 +DIABETES_PREVALENCE +TOTAL_POPULATION
+POPULATION_25_AND_OVER_HIGH_SCHOOL_GRADUATE -
MEDIAN_FAMILY_INCOME* -
POPULATION_25_AND_OVER_BACHELORS_DEGREE** -SOURCE_COUNT
Passing Models
AdjR2 AICc JB K(BP) VIF SA Model

******************************************************************************
Choose 7 of 15 Summary
Highest Adjusted R-Squared Results
AdjR2 AICc JB K(BP) VIF SA Model
0.22 407.27 0.45 0.23 20.39 0.69 +DIABETES_PREVALENCE
+POPULATION_25_AND_OVER_HIGH_SCHOOL_GRADUATE -
MEDIAN_FAMILY_INCOME** +LANGUAGE_OTHER_THAN_ENGLISH -
75

FOREIGN_BORN -POPULATION_25_AND_OVER_BACHELORS_DEGREE* -
SOURCE_COUNT*
0.22 407.35 0.45 0.25 24.94 0.69 +DIABETES_PREVALENCE +TOTAL_POPULATION
+POPULATION_25_AND_OVER_HIGH_SCHOOL_GRADUATE -
MEDIAN_FAMILY_INCOME* +POPULATION_25_AND_OVER_ASSOCIATES_DEGREE
-POPULATION_25_AND_OVER_BACHELORS_DEGREE* -SOURCE_COUNT
0.22 407.40 0.37 0.23 19.97 0.62 +PHYSICAL_INACTIVITY_PREVALENCE
+POPULATION_25_AND_OVER_HIGH_SCHOOL_GRADUATE -
MEDIAN_FAMILY_INCOME** +LANGUAGE_OTHER_THAN_ENGLISH -
FOREIGN_BORN -POPULATION_25_AND_OVER_BACHELORS_DEGREE** -
SOURCE_COUNT*
Passing Models
AdjR2 AICc JB K(BP) VIF SA Model

******************************************************************************
Choose 8 of 15 Summary
Highest Adjusted R-Squared Results
AdjR2 AICc JB K(BP) VIF SA Model
0.22 409.01 0.33 0.77 177.22 0.90 +DIABETES_PREVALENCE +TOTAL_POPULATION -
MEDIAN_FAMILY_INCOME* +POPULATION_25_AND_OVER_ASSOCIATES_DEGREE
-POPULATION_25_AND_OVER_BACHELORS_DEGREE* -SOURCE_COUNT -
POPULATION_DENSITY +SOURCE_DENSITY
0.22 409.03 0.50 0.50 177.86 0.84 +DIABETES_PREVALENCE +TOTAL_POPULATION
+POPULATION_25_AND_OVER_HIGH_SCHOOL_GRADUATE -
MEDIAN_FAMILY_INCOME -POPULATION_25_AND_OVER_BACHELORS_DEGREE* -
SOURCE_COUNT -POPULATION_DENSITY +SOURCE_DENSITY
0.22 409.10 0.47 0.46 178.13 0.79 +PHYSICAL_INACTIVITY_PREVALENCE
+TOTAL_POPULATION +POPULATION_25_AND_OVER_HIGH_SCHOOL_GRADUATE
-MEDIAN_FAMILY_INCOME* -
POPULATION_25_AND_OVER_BACHELORS_DEGREE** -SOURCE_COUNT -
POPULATION_DENSITY +SOURCE_DENSITY
Passing Models
AdjR2 AICc JB K(BP) VIF SA Model

******************************************************************************
Choose 9 of 15 Summary
Highest Adjusted R-Squared Results
AdjR2 AICc JB K(BP) VIF SA Model
0.22 410.93 0.44 0.59 179.29 0.84 +DIABETES_PREVALENCE +TOTAL_POPULATION
+POPULATION_25_AND_OVER_HIGH_SCHOOL_GRADUATE -
MEDIAN_FAMILY_INCOME +LANGUAGE_OTHER_THAN_ENGLISH -
POPULATION_25_AND_OVER_BACHELORS_DEGREE** -SOURCE_COUNT* -
POPULATION_DENSITY +SOURCE_DENSITY
0.21 411.04 0.38 0.23 25.15 0.70 +PHYSICAL_INACTIVITY_PREVALENCE
+DIABETES_PREVALENCE +TOTAL_POPULATION
+POPULATION_25_AND_OVER_HIGH_SCHOOL_GRADUATE -
76

MEDIAN_FAMILY_INCOME** +LANGUAGE_OTHER_THAN_ENGLISH -
FOREIGN_BORN -POPULATION_25_AND_OVER_BACHELORS_DEGREE* -
SOURCE_COUNT
0.21 411.08 0.40 0.51 179.63 0.79 +PHYSICAL_INACTIVITY_PREVALENCE
+TOTAL_POPULATION +POPULATION_25_AND_OVER_HIGH_SCHOOL_GRADUATE
-MEDIAN_FAMILY_INCOME* +LANGUAGE_OTHER_THAN_ENGLISH -
POPULATION_25_AND_OVER_BACHELORS_DEGREE** -SOURCE_COUNT -
POPULATION_DENSITY +SOURCE_DENSITY
Passing Models
AdjR2 AICc JB K(BP) VIF SA Model

******************************************************************************
Choose 10 of 15 Summary
Highest Adjusted R-Squared Results
AdjR2 AICc JB K(BP) VIF SA Model
0.22 412.49 0.48 0.55 179.86 0.79 +DIABETES_PREVALENCE +TOTAL_POPULATION
+POPULATION_25_AND_OVER_HIGH_SCHOOL_GRADUATE -
MEDIAN_FAMILY_INCOME* +LANGUAGE_OTHER_THAN_ENGLISH -
FOREIGN_BORN -POPULATION_25_AND_OVER_BACHELORS_DEGREE* -
SOURCE_COUNT* -POPULATION_DENSITY +SOURCE_DENSITY
0.21 412.74 0.37 0.50 180.25 0.78 +PHYSICAL_INACTIVITY_PREVALENCE
+TOTAL_POPULATION +POPULATION_25_AND_OVER_HIGH_SCHOOL_GRADUATE
-MEDIAN_FAMILY_INCOME* +LANGUAGE_OTHER_THAN_ENGLISH -
FOREIGN_BORN -POPULATION_25_AND_OVER_BACHELORS_DEGREE** -
SOURCE_COUNT -POPULATION_DENSITY +SOURCE_DENSITY
0.21 413.13 0.38 0.58 180.84 0.87 +PHYSICAL_INACTIVITY_PREVALENCE
+DIABETES_PREVALENCE +TOTAL_POPULATION
+POPULATION_25_AND_OVER_HIGH_SCHOOL_GRADUATE -
MEDIAN_FAMILY_INCOME* +LANGUAGE_OTHER_THAN_ENGLISH -
POPULATION_25_AND_OVER_BACHELORS_DEGREE** -SOURCE_COUNT -
POPULATION_DENSITY +SOURCE_DENSITY
Passing Models
AdjR2 AICc JB K(BP) VIF SA Model

******************************************************************************
********* Exploratory Regression Global Summary (OBESITY_PREVALENCE) *********

Percentage of Search Criteria Passed
Search Criterion Cutoff Trials # Passed % Passed
Min Adjusted R-Squared > 0.50 28886 0 0.00
Max Coefficient p-value < 0.05 28886 0 0.00
Max VIF Value < 7.50 28886 12064 41.76
Min Jarque-Bera p-value > 0.10 28886 28886 100.00
Min Spatial Autocorrelation p-value > 0.10 21 21 100.00

------------------------------------------------------------------------------
77

Summary of Variable Significance
Variable % Significant % Negative % Positive
MEDIAN_FAMILY_INCOME 30.45 100.00 0.00
POPULATION_25_AND_OVER_BACHELORS_DEGREE 21.98 100.00 0.00
DIABETES_PREVALENCE 3.02 0.00 100.00
SOURCE_COUNT 2.80 100.00 0.00
TOTAL_POPULATION 1.60 54.19 45.81
POPULATION_25_AND_OVER_GRADUATE_OR_PROFESSIONAL 1.14 56.88 43.12
POPULATION_DENSITY 1.09 99.47 0.53
SOURCE_DENSITY 0.80 44.59 55.41
PHYSICAL_INACTIVITY_PREVALENCE 0.49 0.07 99.93
FOREIGN_BORN 0.42 93.01 6.99
LANGUAGE_OTHER_THAN_ENGLISH 0.19 38.72 61.28
POPULATION_25_AND_OVER_HIGH_SCHOOL_GRADUATE 0.10 28.82 71.18
POPULATION_25_AND_OVER_ASSOCIATES_DEGREE 0.02 0.00 100.00
UNEMPLOYMENT_RATE 0.01 61.84 38.16
POVERTY_PREVALENCE 0.00 0.17 99.83

------------------------------------------------------------------------------

Summary of Multicollinearity
Variable VIF Violations Covariates
PHYSICAL_INACTIVITY_PREVALENCE 2.11 0 --------
DIABETES_PREVALENCE 2.38 0 --------
TOTAL_POPULATION 43.85 7007 SOURCE_COUNT (100.00), SOURCE_DENSITY
(21.19), LANGUAGE_OTHER_THAN_ENGLISH (21.19), POPULATION_DENSITY (21.19),
FOREIGN_BORN (21.19)
POPULATION_25_AND_OVER_HIGH_SCHOOL_GRADUATE 7.42 0 --------
MEDIAN_FAMILY_INCOME 1.13 0 --------
POVERTY_PREVALENCE 1.57 0 --------
LANGUAGE_OTHER_THAN_ENGLISH 22.40 7007 FOREIGN_BORN (100.00),
TOTAL_POPULATION (21.19), SOURCE_DENSITY (21.19), POPULATION_DENSITY
(21.19), SOURCE_COUNT (21.19)
FOREIGN_BORN 18.01 7007 LANGUAGE_OTHER_THAN_ENGLISH (100.00),
TOTAL_POPULATION (21.19), SOURCE_DENSITY (21.19), POPULATION_DENSITY
(21.19), SOURCE_COUNT (21.19)
UNEMPLOYMENT_RATE 2.47 0 --------
POPULATION_25_AND_OVER_ASSOCIATES_DEGREE 1.43 0 --------
POPULATION_25_AND_OVER_BACHELORS_DEGREE 7.47 0 --------
POPULATION_25_AND_OVER_GRADUATE_OR_PROFESSIONAL 6.62 0 --------
SOURCE_COUNT 36.83 7007 TOTAL_POPULATION (100.00), SOURCE_DENSITY
(21.19), POPULATION_DENSITY (21.19), LANGUAGE_OTHER_THAN_ENGLISH (21.19),
FOREIGN_BORN (21.19)
78

POPULATION_DENSITY 184.07 7007 SOURCE_DENSITY (100.00),
TOTAL_POPULATION (21.19), SOURCE_COUNT (21.19),
LANGUAGE_OTHER_THAN_ENGLISH (21.19), FOREIGN_BORN (21.19)
SOURCE_DENSITY 169.61 7007 POPULATION_DENSITY (100.00),
TOTAL_POPULATION (21.19), LANGUAGE_OTHER_THAN_ENGLISH (21.19),
SOURCE_COUNT (21.19), FOREIGN_BORN (21.19)

------------------------------------------------------------------------------

Summary of Residual Normality (JB)
JB AdjR2 AICc K(BP) VIF SA Model
0.999679 0.113295 418.598264 0.559382 21.346142 0.615356 -TOTAL_POPULATION
+POVERTY_PREVALENCE -FOREIGN_BORN +UNEMPLOYMENT_RATE
+POPULATION_25_AND_OVER_ASSOCIATES_DEGREE -SOURCE_COUNT -
POPULATION_DENSITY
0.999625 0.118684 416.670770 0.688669 3.164811 0.615985 +POVERTY_PREVALENCE -
LANGUAGE_OTHER_THAN_ENGLISH +UNEMPLOYMENT_RATE
+POPULATION_25_AND_OVER_ASSOCIATES_DEGREE -SOURCE_COUNT* -
SOURCE_DENSITY
0.999616 0.114588 418.471318 0.639007 14.395781 0.594059 -TOTAL_POPULATION
+POVERTY_PREVALENCE +LANGUAGE_OTHER_THAN_ENGLISH -FOREIGN_BORN
+UNEMPLOYMENT_RATE +POPULATION_25_AND_OVER_ASSOCIATES_DEGREE -
POPULATION_DENSITY

------------------------------------------------------------------------------

Summary of Residual Spatial Autocorrelation (SA)
SA AdjR2 AICc JB K(BP) VIF Model
0.895426 0.218912 409.013337 0.330985 0.770122 177.218120 +DIABETES_PREVALENCE
+TOTAL_POPULATION -MEDIAN_FAMILY_INCOME*
+POPULATION_25_AND_OVER_ASSOCIATES_DEGREE -
POPULATION_25_AND_OVER_BACHELORS_DEGREE* -SOURCE_COUNT -
POPULATION_DENSITY +SOURCE_DENSITY
0.866338 0.209368 413.131510 0.376597 0.578981 180.842188
+PHYSICAL_INACTIVITY_PREVALENCE +DIABETES_PREVALENCE
+TOTAL_POPULATION +POPULATION_25_AND_OVER_HIGH_SCHOOL_GRADUATE
-MEDIAN_FAMILY_INCOME* +LANGUAGE_OTHER_THAN_ENGLISH -
POPULATION_25_AND_OVER_BACHELORS_DEGREE** -SOURCE_COUNT -
POPULATION_DENSITY +SOURCE_DENSITY
0.838562 0.218796 409.026201 0.497854 0.499425 177.860337 +DIABETES_PREVALENCE
+TOTAL_POPULATION +POPULATION_25_AND_OVER_HIGH_SCHOOL_GRADUATE
-MEDIAN_FAMILY_INCOME -POPULATION_25_AND_OVER_BACHELORS_DEGREE*
-SOURCE_COUNT -POPULATION_DENSITY +SOURCE_DENSITY

------------------------------------------------------------------------------

79

Table Abbreviations
AdjR2 Adjusted R-Squared
AICc Akaike's Information Criterion
JB Jarque-Bera p-value
K(BP) Koenker (BP) Statistic p-value
VIF Max Variance Inflation Factor
SA Global Moran's I p-value
Model Variable sign (+/-)
Model Variable significance (* = 0.10; ** = 0.05; *** = 0.01)

------------------------------------------------------------------------------

80

Appendix C Ordinary Least Squares (OLS) Results – Hypothesis

81

82

83

Appendix D Ordinary Least Squares (OLS) Results – Exploratory Regression

84

85

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Finding food deserts: a study of food access measures in the Phoenix-Mesa urban area

PDF

A spatiotemporal analysis of environmental risk factors of Lyme disease in the Northeastern United States

PDF

Investigating the association of historical preservation and neighborhood status in Detroit, 1970-2015

PDF

The spatial effect of AB 109 (Public Safety Realignment) on crime rates in San Diego County

PDF

Estimating populations at risk in data-poor environments: a geographically disaggregated analysis of Boko Haram terrorism 2009-2014

PDF

A spatial narrative of alternative fueled vehicles in California: a GIS story map

PDF

Smart growth and walkability affect on vehicle use and ownership

PDF

Measuring seasonal variation in food access: a case study of Everett, Washington

PDF

Preparing for immigration reform: a spatial analysis of unauthorized immigrants

PDF

The movement of Mexican migration and its impact based on a GIS geospatial database

PDF

Integration of topographic and bathymetric digital elevation model using ArcGIS interpolation methods: a case study of the Klamath River Estuary

PDF

Applying least cost path analysis to search and rescue data: a case study in Yosemite National Park

PDF

Network accessibility and population change: historical analysis of transportation in Tennessee, 1830-2010

PDF

Testing social disorganization theory on violent crime: a case study on Pueblo, Colorado

PDF

Finding the green in greenspace: an examination of geospatial measures of greenspace for use in exposure studies

PDF

A comparison of urban land cover change: a study of Pasadena and Inglewood, California, 1992‐2011

PDF

Use of remotely sensed imagery to map sudden oak death (Phytophthora ramorum) in the Santa Cruz Mountains

PDF

Comparing Landsat7 ETM+ and NAIP imagery for precision agriculture application in small scale farming: a case study in the south eastern part of Pittsylvania County, VA

PDF

Heart, brain, and breath: studies on the neuromodulation of interoceptive systems

Asset Metadata

Creator Ingram, Elliott Wayne, Jr. (author)

Core Title Obesity and healthy food accessibility: case study of Minnesota, USA

School College of Letters, Arts and Sciences

Degree Master of Science

Degree Program Geographic Information Science and Technology

Publication Date 12/05/2019

Defense Date 08/27/2019

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag correlation,exploratory regression,healthy food accessibility,Minnesota,OAI-PMH Harvest,obesity,ordinary least squares

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Bernstein, Jennifer (committee chair), Lee, Su Jin (committee member), Oda, Katsuhiko (committee member)

Creator Email elliottingramjr@gmail.com,ewingram@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-245523

Unique identifier UC11673817

Identifier etd-IngramElli-8000.pdf (filename),usctheses-c89-245523 (legacy record id)

Legacy Identifier etd-IngramElli-8000.pdf

Dmrecord 245523

Document Type Thesis

Rights Ingram, Elliott Wayne, Jr.

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA