Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Defining neighborhood for health research in Arizona
(USC Thesis Other)
Defining neighborhood for health research in Arizona
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
Defining Neighborhood for Health Research in Arizona
by
Tiffany ‘Monicque’ Lee
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)
May 2019
ii
Copyright © 2018 by Tiffany ‘Monicque’ Lee
iii
DEDICATION
For my sons, J.R. and Wyatt; I hope this work will help you both understand that no matter what
life throws at you, not to ever give up. To my husband J.R., thank you for your understanding,
support and encouragement throughout this journey. To my Blue Dog, my unwavering study
companion throughout all the very long nights…
You all are my motivators and my motivation.
iv
Table of Contents
List of Figures ................................................................................................................................ vi
List of Tables ............................................................................................................................... viii
Acknowledgements ........................................................................................................................ ix
List of Abbreviations ...................................................................................................................... x
Abstract ..................................................................................................................................... xii
Chapter 1 : Introduction .................................................................................................................. 1
The Ecological Fallacy and Modifiable Areal Unit Problem .............................................4
The Uncertain Geographic Context and Uncertain Point Observation Problems ............6
Rurality ...............................................................................................................................8
Congruence .........................................................................................................................8
Thesis Objective..................................................................................................................9
Thesis Organization ............................................................................................................9
Chapter 2 : Related Work ............................................................................................................. 10
Administrative Boundaries ...............................................................................................11
2.1.1. State and County Level ............................................................................................13
2.1.2. Census Level ............................................................................................................13
2.1.3. ‘Nonstandard’ Administrative Unit Areas ...............................................................17
Buffer Zones .....................................................................................................................21
2.2.1. Circular ....................................................................................................................23
2.2.2. Network....................................................................................................................24
2.2.3. Activity Space & Perceived Environment ...............................................................25
2.2.4. Dasymetric Mapping ................................................................................................27
Chapter 3 : Methods and Data Sources ......................................................................................... 30
Study Area ........................................................................................................................30
v
Hypothesis.........................................................................................................................31
Data Sources .....................................................................................................................33
Methodology .....................................................................................................................35
Chapter 4 : Results ........................................................................................................................ 39
Median Household Income ...............................................................................................39
4.1.1. County Level ............................................................................................................39
4.1.2. Census Tract Level ..................................................................................................41
4.1.3. Census Block Group Level ......................................................................................46
Children and the Elderly ...................................................................................................61
4.2.1. County Level ............................................................................................................61
4.2.2. Census Tract Level ..................................................................................................62
4.2.3. Census Block Group Level ......................................................................................64
Native American Population .............................................................................................76
4.3.1. County Level ............................................................................................................76
4.3.2. Census Tract Level ..................................................................................................77
4.3.3. Census Block Group Level ......................................................................................79
Chapter 5 : Discussion and Conclusions....................................................................................... 91
References ..................................................................................................................................... 96
vi
List of Figures
Figure 1: Relationships between issues affecting the spatial analysis of neighborhood ................ 4
Figure 2: Administrative boundary types used in this thesis ........................................................ 12
Figure 3: Circular and network buffers ......................................................................................... 22
Figure 4: Choropleth maps ........................................................................................................... 28
Figure 5: Dasymetric map ............................................................................................................. 28
Figure 6: Arizona counties and adminstrative areas ..................................................................... 32
Figure 7: Workflow ...................................................................................................................... 36
Figure 8: Median Household Income for Arizona counties ......................................................... 40
Figure 9: Graph, Median household income by county and census tract ..................................... 44
Figure 10: Maps, Statewide census tracts with Metro areas, median household income ............. 45
Figure 11: Graph, Group I, census block group median household income ................................ 48
Figure 12: Graph, Group I (continued), census block group median household income ............. 49
Figure 13: Graph, Group II, census block group median household income ............................... 51
Figure 14: Graph, Group II (continued), census block group median household income ............ 52
Figure 15: Graph, Group III, census block group median household income .............................. 54
Figure 16: Map, Group I, Median Household Income ................................................................. 57
Figure 17: Map, Group I (Continued), Median Household Income ............................................. 58
Figure 18: Map, Group II, Median Household Income ................................................................ 59
Figure 19: Map, Group III, Median Household Income ............................................................... 60
Figure 20: Vulnerable Population by Arizona counties ................................................................ 61
Figure 21: Graph, Arizona county and census tract estimates for vulnerable population ............ 63
Figure 22: Map, Statewide census tracts and Metro areas for vulnerable population .................. 64
Figure 23: Graph, Group I, census block group Vulnerable population ....................................... 66
Figure 24: Graph, Group I (continued), census block group vulnerable population .................... 67
vii
Figure 25: Graph, Group II: Census block group vulnerable population ..................................... 68
Figure 26: Graph, Group II (continued): Census block group vulnerable population .................. 69
Figure 27: Graph, Group III: Census block group Vulnerable Population ................................... 70
Figure 28: Map, Maricopa County Vulnerable Population .......................................................... 72
Figure 29: Map, Pima County Vulnerable Population ................................................................. 73
Figure 30: Map, Coconino County and Pinal Vulnerable Population .......................................... 74
Figure 31: Map, Apache County and Santa Cruz Vulnerable Population .................................... 75
Figure 32: Map, percent Native American Population by Arizona county .................................. 76
Figure 33. Graph, census tract Native American population ........................................................ 77
Figure 34: Map, Native American Population by census tract .................................................... 78
Figure 35: Graph, Group I: Census block group Native American Population ............................ 80
Figure 36: Graph, Group I (Continued): Census block group Native American Population ....... 81
Figure 37: Graph, Group II; census block group Native American Population ........................... 83
Figure 38: Graph, Group II (continued): Census block group Native American Population ....... 84
Figure 39: Graph, Group III: Census block group Native American population ......................... 85
Figure 40: Map, Maricopa County Native American Population ................................................. 87
Figure 41: Map, Pima County Native American Population ........................................................ 88
Figure 42: Map, Coconino and Pinal Counties Native American Population .............................. 89
Figure 43: Map, Apache and Santa Cruz Counties Native American Population ........................ 90
viii
List of Tables
Table 1: Data sets and sources ...................................................................................................... 34
Table 2: Counts and percentages of census tracts with median household incomes .................... 42
Table 3: Counts and percentages of census block groups with median household incomes ........ 53
Table 4: Counts and percentages of census tract Vulnerability .................................................... 62
Table 5: Counts and percentages of census block group Vulnerability ........................................ 71
Table 6: Counts and percentages of census tracts Native American Population .......................... 79
Table 7: Counts and percentages of census block group Native American Population ............... 86
Table 8: Comparison between reporting styles. ............................................................................ 92
ix
Acknowledgements
A very special thank you to John P. Wilson. It has been a great privilege to learn from you.
Thank you for the opportunity.
I would also like to acknowledge all the faculty and staff that I have had the pleasure to work
with throughout the GIST program at USC. It has been a challenging, at times intimidating, and
all together amazing experience.
Thank you to all my Trojan family.
x
List of Abbreviations
ACS American Community Survey
AIANNHAs American Indian, Alaska Native, and Native Hawaiian Areas
AIRs American Indian Reservations
ANSI American National Standards Institute
BAS Boundary and Annexation Survey
BIA Bureau of Indian Affairs
DOI Department of the Interior
EDGE Education Demographic and Geographic Estimates
FIPS Federal Information Processing Series
GIS Geographic Information Science
GIST Geographic Information Science & Technology
MAUP Modifiable Areal Unit Problem
MCU Measured Contextual Unit
MHI Median Household Income
NCES National Center for Education Statistics
SDTSAs State designated tribal statistical areas
TCU True Contextual Unit
TDSAs Tribal designated statistical areas
TSAP Tribal Statistical Areas Program
UGCoP Uncertain Geographic Context Problem
UPOP Uncertain Point Observation Problem
U.S. United States
xi
USC University of Southern California
USPS United States Postal Service
ZCTAs Zip Code Tabulation Areas
ZIP Zone Improvement Plan
xii
Abstract
Defining place in health studies has been a crux for researchers as the definition of neighborhood
is often regarded as adaptable to study needs and/or the preferences of the researcher. Health
researchers commonly rely on measures of neighborhood that default to any number of
predefined spatial administrative units, providing a relatively quick and cost-effective means to
accessing and categorizing population data within a geographic area of interest. This approach to
inferring population statistics assumes that median values for variables are relatively evenly
disbursed across specific geographic areas of varying sizes.
This thesis explores how research outcomes may be affected by the choice of geographic
reporting zones. The primary research goal of this study was to compare geographic reporting
zones within the State of Arizona and to determine how the choice of neighborhood would
influence the resulting values for three commonly utilized social determinants of health; median
household income, numbers of children and the elderly, and the percent Native American
population. This study used administrative boundaries at the county, census tract, and census
block group levels from the 2000 Decennial Census and examined if and what variation occurred
within the resulting outcomes for differing reporting zones within the State of Arizona.
The results of this thesis demonstrate that outcomes cannot be generalized across
administrative units, that spatial aggregation will affect final outcomes, and that the choice of
spatial reporting zone may produce widely different estimates for the same variable within a
given geographic area. This thesis provides the foundation for future work investigating how
choice of neighborhood can affect outcomes for small area studies and sets the framework for
exploring what effects neighborhood definition might have on estimates of social determinants of
health when proximity buffers are applied.
1
Chapter 1 : Introduction
People are constantly trying to define space. We assign invisible spatial boundaries and lines to
the oceans, the Earth, and even the cosmos. Collectively, we tend to hold these boundary lines as
quasi-physical representations of belonging. These artificial boundaries represent claim status,
power, cultural identification, and give inhabitants a sense of place. Boundary zones are
powerful proclamations representative of people and place. Historically, assigning boundary
delineations to an area was an authoritative act. We are taught from an early age to acknowledge,
generally respect, and not question these unseen lines of boundary delineation that fill our daily
routines (Goodchild 2018). Some of these boundaries are often unknowingly assigned: such as
census areas, ZIP codes, voting precincts and school districts. Some of these boundary lines are
fixed; as in national and state borders. Some we simply accept; such as county boundaries,
property lines and street networks. Other boundary lines are more unclear; for example,
determining where the exact geographical break occurs along a demographic transition, or where
a disease outbreak is likely to occur next. Science and governments are constantly using spatial
delineations as measures of process. Possibly nowhere is the issue of boundary delineation more
important than in the realm of spatial epidemiology and human health. Disease does not respect
these quasi-physical boundaries of place - it is often indiscriminate without regard towards the
places or people it affects (Flowerdew et al. 2008).
The merging of GIScience and epidemiology has given rise to a growing branch of health
research examining where and how human well-being is affected by the local context which is
often delineated by the surrounding spatial patterns and their boundaries. Geography is treated as
a potentially direct correlate of human health and well-being in this view of the world by
geographers, sociologists and other social scientists (e.g. Matthews and Yang 2013; Jankowska
2
et al. 2014). Epidemiologists now regard GIS as a powerful tool in evaluating disease occurrence
and transmission throughout specific areas. This new area of health research merging geography,
sociology, and epidemiology is intent on delivering new methods for analyzing the correlations
between population, health, and place (i.e. geography).
The emergence of GIS in health research has allowed researchers the ability to
reconceptualize boundary delineations and evaluate subjects in localized areas, especially as it
applies to examining the health outcomes connecting people and the environment. Health
researchers designate small area boundaries encompassing a subject of interest as a given study
zone. These spatial zones, known as neighborhoods, are vitally important for understanding
population health patterns as a function of location.
The conceptualization and measurement of neighborhood has yielded a dilemma of grand
proportions for researchers. An individual’s perceived definition of neighborhood often varies
significantly from a researcher’s delineation of the same vicinity. Within health research, this
meaning of locale can vary greatly from one study to another depending on what measures and
considerations are selected in defining that neighborhood.
Selecting data derived from differing measures or interpretations of a neighborhood, or
points within, can sometimes generate substantially different results yielding uncertainty in the
final reporting. The attempt to circumvent reporting uncertainty has led researchers on a quest
for the ultimate method in depicting the representativeness of a population. Multiple problems
arise in data reporting when spatial linkage inferences between and within boundary zones and
point observations are inappropriately applied. Researchers have attempted to use numerous
methods including multilevel analysis as a means of dealing with this uncertainty; however,
although beneficial in addressing some aspects of uncertainty, all forms of spatial analysis are
3
still subject to classical and emerging spatial problems requiring deliberation in study design
and/or data analysis (Robertson and Feick 2018).
As the scale or extent of a study area changes, corresponding details of that area also
change. For example, when the scale of a study area gets larger, some details become
indiscernible (Openshaw and Alvanides 2001). Also, moving a neighborhood boundary line may
change the dynamics of the area under study causing a change in demographics and geography
that may affect the relationships being examined and thus the resulting outcomes (Foster and
Hipp 2011). The reporting issues caused by utilizing different spatial areas of different sizes
and/or scales is commonly recognized as the modifiable areal unit problem (MAUP) (Tatalovitch
et al. 2006; Swift et al. 2014; Robertson and Feick 2018).
Sharing some likeness to but separate from the MAUP are the problems associated by
using static boundaries for analysis. Constraining the measure of an individual within a health
study to a predefined static boundary, often assigned by an administrative unit, does not
represent the true measure of a subject’s dynamic traverse and exposures through time and space
(i.e. potentially spanning multiple administrative boundaries over a given temporal period). This
inferential error results in a measurement problem recognized by researchers as the uncertain
geographic context problem (UGCoP) (Kwan 2012b; Robertson and Feick 2018).
Further complicating the issue of data uncertainty caused by spatial misreporting is an
issue brought to the forefront by technological advances and public accessibility to GIS
applications. This problem, the uncertain point observation problem (UPOP), results from
misreporting point locations and subsequently incorrectly linking them to areal units enabling
users to develop aggregated spatial inferences erroneously (Robertson and Feick 2018).
4
The conventional assignment of subjects to a predefined measured contextual unit
(MCU) (e.g. administrative boundary) is not usually indicative of the true contextual unit (TCU)
affecting those individuals and thus potentially creates ecological or atomistic fallacies within the
final reporting. Ecological fallacies make incorrect inferences about individuals based off
aggregated measures at the ‘group-level’; whereas, an atomistic fallacy makes incorrect
assumptions about an aggregated population based from measures at the ‘individual-level.’ Any
time an analysis utilizes data between higher and lower level aggregations, that data becomes
vulnerable to some form of fallacy (Kwan 2012b; Robertson and Feick 2018). The following
diagram (Figure 1) demonstrates the interrelationships that exist within various methods of
neighborhood delineation and analysis.
Figure 1. Relationships between issues affecting the spatial analysis of neighborhood (Source:
Robertson and Feick 2018)
The Ecological Fallacy and Modifiable Areal Unit Problem
There are misreporting issues to consider when analyzing discrete socioeconomic and
demographic data collected at an administrative level yet applied to individuals in small area
studies. The resulting uncertainty creates an ecological fallacy which is found in data where
5
assumptions for the microscale (i.e. individual) are collected at the macro scale (i.e. ‘group
level’) but inferred per individual (Swift et al. 2014; Robertson and Feick 2018). This leads to
the necessity to account for both micro- and macro-levels with some multilevel analysis to bridge
the gap and circumvent or minimize ecological bias (Schule and Bolte 2015; Strominger et al.
2016; Robertson and Feick 2018). The ecological fallacy, however, is also present in multilevel
analysis which affects both static and dynamic neighborhood evaluations that use aggregated
data for individual inference (Figure 1). This renders both the modifiable areal unit problem and
the uncertain geographic context problem vulnerable to the ecological fallacy (Robertson and
Feick 2018).
The misreporting issues that are introduced by utilizing data from differing spatial
aggregations is commonly recognized as the MAUP (Tatalovich et al. 2006; Swift et al. 2014;
Robertson and Feick 2018). This fact and the accompanying complexity render acknowledgment
of the MAUP within health studies to often be little more than a footnote or caveat (Root 2012;
Swift et al. 2014). The practice and problem of ignoring the MAUP and utilizing aggregated data
is that information is lost in the process of aggregation which results in misreporting (Openshaw
and Alvanides 2001). The lack of consideration of the MAUP can undermine otherwise sound
methodology simply by the uncertainty that might be introduced by not accounting for it.
Some spatial analysis methods are superior in escaping or minimizing the MAUP. The
often-cited examples include political administrative units which utilize zone aggregation and
‘official zoning systems,’ such as the hierarchical census units consisting of blocks, block
groups, and tracts used within the U.S. (Openshaw and Alvanides 2001).
The MAUP is additionally complicated by the need to also consider the potential effects
of the UGCoP and the UPOP (Robertson and Feick 2018). Although similar in context, both are
6
unique aspects of neighborhood delineation and each requires specific consideration (Kwan
2012b).
The Uncertain Geographic Context and Uncertain Point Observation
Problems
The UGCoP is to be considered when examining neighborhood because the boundary and scale
issues addressed by accounting for the MAUP do not account for the contextual uncertainty
associated with the dynamic activity space of an individual within a given neighborhood's
population. The definition of the neighborhood by administrative boundaries or buffers does not
account for the social characteristics within neighborhoods that can influence health outcomes
(Kwan 2012b). Accounting solely for the MAUP can still distort results when the potential
effects of the UGCoP are ignored because the individual mobility behaviors within a population
are rarely limited to conventionally used neighborhood boundaries. People’s daily routines often
span multiple types of boundary areas within a given spatiotemporal period. Throughout the
course of a day spatial behaviors are diverse and dynamic frequently spanning or are affected by
multiple locales of different administrative boundary types, often simultaneously (Matthews
2011, 35-54).
Recent research has demonstrated that measures of an individual’s actual daily traverse
usually cover greater geographic areas than the conventional static neighborhood definition that
is used. The result of not accounting for the UGCoP is often misreporting due to mismatched
units of analysis, inadvertently omitting important exposure information, and introducing the
ecological fallacy. The typical geographic boundaries used to delineate neighborhoods can over-
and/or under-estimate the real exposure effects experienced by an individual’s activity space.
Additionally, an individual’s activity space consisting of their daily routine, and less frequent
7
trips, may significantly affect that person’s perception of neighborhood. Neighborhood
perception is an important consideration for researchers as it can provide insight into potential
exposures occuring throughout a person’s daily routine that may not be adequately represented
by analyzing spatiotemporal traverse patterns alone (Chaix et al. 2009; Kwan 2012b; Jankowska
et al. 2017).
Equally important is the spatial uncertainty introduced by the UPOP. The error presented
by the UPOP occurs with a mismatch in point observation location data between the MCU and
TCU reported at the individual level as a basis for group level inferences, making UPOP subject
to the atomistic fallacy (Figure 1). The UPOP can result in incorrect aggregated population
characterizations by introducing error and uncertainty in the association between where and how
a point observation is created and connected to a neighborhood (Robertson and Feick 2018).
Another consideration in identifying how a neighborhood is to be defined for a study, yet
often unacknowledged across health research, is that just as the dynamics of people change over
a diurnal period, so do places. The human dynamics influencing an area such as a city street or
park may be very different over the course of 24 hours causing a range of potential exposures.
The issue of time and space, and the activity of both people and places, can greatly affect
exposure outcomes within a given neighborhood. The often overlooked effects of temporal
exposures over hours, days, seasons, and/or years should be considered when selecting how a
neighborhood is to be determined (Xia et al. 2006; Matthews and Yang 2013). Often the physical
neighborhood dynamics of place are not accurately represented using data from administrative
sources, leaving the spatiotemporal aspect and effects of place unacknowledged (Yiannakoulias
2011; Delmelle 2016).
8
Rurality
Another substantial challenge in designating neighborhood within health studies has to do with
the special characteristics of rural areas. Rurality is a challenge for health researchers examining
environmental contextual influences. Rural communities often have low population densities,
generally isolated within or disbursed across larger geographic regions. Infrastructure measures
vary, and frequently street networks are minimal. Administrative boundaries are commonly used
as aggregated proxies of rural neighborhoods and do not reflect the geographic actuality of a
local area. Small area studies in rural settings are prone to ecological bias caused by aggregation
issues, scale effects and inconsistent measures and considerations of neighborhood (Rousseau
1995).
Congruence
Incompatible and mismatched geographic units present a substantial problem for researchers and
complicate the conundrum caused by spatial problems and fallacies. Administrative boundary
types vary in size, can be nested within one another, might overlap, and sometimes share little to
no commonality with one another – yet all are used to represent place. When data from differing
types of boundary systems are overlaid and used together (e.g. spatial interpolation), data is lost
or erroneously created resulting in misreporting and uncertainty. These geographic zones have
been created for various distinct purposes by differing governmental agencies, sometimes
arbitrarily, and facilitate the most convenient and sometimes only means of data acquisition for
researchers (Eagleson 2002; Mennis 2003).
To complicate matters, the selection of these units for health research is often equally
arbitrary. Furthermore, researchers may be limited in data availability for a given area, requiring
9
the use of a geographic unit that might not be the most representative of a subject (Diez Roux
2001; Eagleson 2002; Chaix et al. 2005; Santos et al. 2010; Robertson and Feick 2018).
Thesis Objective
This thesis set out to examine and compare the descriptive statistics which result from using
aggregated data for various neighborhood contexts, similar to what is commonly done in current
health research. The best available geographic units were used to generate measures of
neighborhood and the potential pitfalls that are likely to follow from using this approach to
measure geographic variability for a large region.
The State of Arizona was selected because of its size, socioeconomic and demographic
diversity, and the various population reporting methodologies used for different sub-populations
in the state. Arizona’s population dynamics with its urban and rural distribution and unique racial
and ethnic diversity offer numerous sampling scenarios. Data were derived from Federal and
State reporting for the general population as well as American Indian reporting areas.
Thesis Organization
The next chapter, Chapter Two, details the most common methods used to define neighborhood
in health research within the U.S. This related work sets the framework for the neighborhood
differentiation methods that were used in this thesis. Chapter Three details the methodology for
how this study was conducted. This chapter describes the data sources and spatial analysis that
was performed. Chapter Four details the results and how the variables of interest varied across
the state as a function of both the geographic variability and/or the ways in which geographic
units for collecting and representing people in various parts of the state varied from one another.
Finally, Chapter Five discusses the significance of these results and the implications for planning
neighborhood selection strategies in future work.
10
Chapter 2 : Related Work
The integration of GIS into epidemiology and health research has become a powerful instrument
for analysis and demonstration of the spatial distribution and movement of populations, disease,
health services and their infrastructure. GIS platforms have allowed health researchers the
accessibility to spatial tools that readily enable the measurement of various dimensions
examining the relationships between health and place (Rushton 2003). The area attributes of a
place, in health research, are often categorized within the boundaries of a neighborhood defined
as some measurable assignment to a geographic locale (Diez Roux 2007). The method of
neighborhood delineation has created a quandary for researchers as the measure and use of a
neighborhood, and its relevance to a given topic, is arbitrarily determined within a study design
and non-specific within any given scientific field (Boscoe and Pickle 2003; Duncan et al. 2014;
Perchoux et al. 2016). Researchers have struggled to determine what the best means of
delineating neighborhood are, as different definitions are known to present varying outcomes
even within the same vicinity and which method to use for a given study is customarily left to the
discretion of the researcher (Diez Roux 2001; Swift et al. 2014). The majority of studies that
have examined the role of boundary selection for neighborhood have focused on pre-determined
administrative units which are not necessarily representative of the population and/or area of
interest (Diez Roux 2007; Santos et al. 2010, Foster and Hipp 2011). Recent research into
neighborhood definition has also examined the role of the perceived environment in defining
neighborhood. Such studies are subjective, examining the activity spaces of an individual and
utilizing self-reporting from subjects and ground truthing within an area of interest (Perchoux et
al. 2016; Kirby et al. 2017). The validity of the classical fixed boundary study is now challenged
against sliding boundary studies, and the quest and argument continue as to how a neighborhood
11
should be defined (Chaix et al. 2009). Few studies, however, have examined the variation and
problems caused by this fluid definition of neighborhood. This chapter sets out to review the
methodologies used for neighborhood delineation and explores the various issues associated with
the selected definitions.
Administrative Boundaries
Administrative boundaries are government-defined and universally recognized jurisdictional
divisions of geography (Chang 2010). How a type of administrative boundary zone is defined
varies from one country to another, but their purposes are the same and serve as a way for
political administrations to organize infrastructure, political elections, community and
emergency services, and often by happenstance, the contextual data residing within geographical
units (Sabel et al. 2013). Administrative boundaries have been predominantly used as a measure
for and proxy of neighborhood for health studies (Diez Roux 2007; Santos et al. 2010). In the
U.S., administrative boundaries are categorized by State, County (or Parish, the geographical
equivalent to a county, within the State of Louisiana), District (school district, voting precinct,
and separate judicial district), ZIP code and Census (Mu et al. 2015). For this thesis,
administrative neighborhoods are examined at the county, census, and American Indian Area
levels (Figure 2).
Health researchers are often limited to using data collected at the administrative level for
small area studies such that neighborhood effects are aggregated using differing geographic
scales and extents (Leite et al. 2015). The use of data collected and analyzed at different scales
within a study may become subject to both the ecological fallacy and the MAUP by introducing
significant potential bias into a given study (Tatalovich et al. 2006; Swift et al. 2014).
12
Figure 2. Administrative boundary types discussed in this thesis
These fixed boundary zones can be additionally problematic as they regularly are not
representative of the true contextual environment and often do not correlate to the actual study
area of interest (Chaix et al. 2005; Tatalovich et al. 2006; Root 2012; Perchoux et al. 2013).
Population distribution and geographic obstacles, such as water and terrain or barriers created by
the built environment, may have significant effects on data reported for a given administrative
boundary (Santos et al. 2010).
Administrative boundary areas have further limitations as these zones are generalizations
of the populations they encompass at each level and they may not account for the within zone
variability. Furthermore, scale plays a critical role in the final reporting, especially in small area
studies. Leite et al. (2015) states that it has been speculated that the scale used in a study often
dictates the population inferences determined by researchers, where health is reflected as being a
function of social and economic factors in larger reporting areas, and a result of the individual in
13
small areas. Small areas are susceptible to small, limited sample size which can misrepresent and
misreport the actual variance in a population by potentially leaving out people within that
population that fall outside the sampling such as minorities, children, and the elderly (Institute of
Education Sciences, National Center for Education Statistics 2018b). Administrative boundary
units may not convey the actual geographic and human variability of a neighborhood (Santos et
al. 2010).
2.1.1. State and County Level
Within the administrative boundary system of the U.S., state and county territories are the main
geographic building blocks for data reporting from the U.S. Census (US Census Bureau 1990).
State borders are temporally static boundaries. These lines of delineation do not move and are
not adjusted over time to account for population and/or urban expansion. County lines, however,
are not always permanent and their boundaries can change over time. Counties can experience
‘substantial changes’ where borderlines are moved, also new counties can be created, or existing
counties ‘deleted’ (U.S. Census Bureau 2016). Data collected and standardized at state and
county levels are often not practical for small area studies because of the introduced ecological
fallacy (Swift et al. 2014; Robertson and Feick 2018).
2.1.2. Census Level
The U.S. Census is a decennial survey that counts every resident of the country by constitutional
mandate. Counts and demographic information are collected by a combination of mailed
questionnaire and ground truthing by census workers to ensure complete (100%) coverage of all
areas. Census information is compiled nationally at block, block group, and tract levels (U.S.
Census Bureau 2010c).
14
Census data is used for statistical purposes by both the government and the public,
private, and not-for-profit sectors. The government uses census data to identify geographical
areas in need of infrastructure and community services and to allocate federal support programs
and funding for those communities deemed as worthy recipients by the reported population
statistics. The Census is also used to determine how many seats within the House of
Representatives each State can hold. Census data is available publicly; however, personal
information reported per household address is not available to the public. Only count data per
administrative unit is available for “public” use (U.S. Census Bureau 2010c).
In 2010, the U.S. Census Bureau revamped their questionnaire procedure changing the
way population data is gathered and reported. Previously, the decennial census utilized both
short and long forms of questionnaires to collect population information. The short form contains
basic demographic questions consisting of ‘name, sex, age, date of birth, race, ethnicity,
relationship and housing tenure.’ The long form questionnaire was sent by random sampling to
an estimated one of every six households and contained additional detailed socioeconomic
questions (U.S. Census Bureau 2010c, 2018a).
In 2010, the U.S. Census switched to using only the short form and eliminated the long
form. In place of the long form the U.S. Census implemented the American Community Survey
(ACS), enacted in 2005, that collects detailed demographic and socioeconomic data monthly,
compiled and reported annually, with the goal of collecting continuously updated population
information for all areas across the U.S. The ACS questionnaire uses random sampling
throughout the U.S. to collect detailed population information from a ‘small percentage’ of
households and does not collect data from the same household more than once every five years
(U.S. Census Bureau 2014, 2018a).
15
The ACS has become an important instrument for the U.S. Census Bureau’s efforts to
maintain up-to-date, continuous, population estimates and is a valued resource for providing
small area statistics to data users. Census statistics are ever-changing however, and data from one
reporting decennial census to another can produce substantially different estimates. The temporal
challenges with the decennial census means many have come to regard the ACS as the solution
because the data presented provides a snapshot of dynamic population information available for
small area evaluations during the time between census reports. The issue of sample size is not
often considered when ACS data is reported or utilized. ACS data is the result of a small
percentage sampling versus the decennial census complete population sampling. The data
collection methodology for the ACS creates an inherent atomistic fallacy within the data reported
by how it uses non-majority data and creates statistical assumptions for the whole based on those
values (Sabel et al. 2013; U.S. Census Bureau 2014; Roberson and Feick 2018).
The ACS provides researchers a convenient option to review annual population estimates
for small areas at the ‘census tract and block group’ levels. These small area statistics were
previously only available from the decennial census report. The adoption of the continuously
collected ACS, in place of the traditionally used long-form collected once every ten years, is
often used by researchers as an interim resource for representing changing population conditions
across the country in between decennial reporting’s, notwithstanding potential sampling bias
(U.S. Census Bureau 2014).
Census tracks are nested within counties and are the larger of the two divisions. Census
tracts generally follow observable physical characteristics and consist of populations ranging
from 2,500 to 8,000 residents (US Census Bureau 1990). Census tracts were initially designed to
represent areas with similar socioeconomic and demographic characteristics. Since census tract
16
inception, however, population dynamics throughout the country have changed resulting in
census tracts that are no longer homologous relative to one another.
Census block groups are nested within census tracts and represent the smallest published
spatial division of the Census, with populations under 2,500 residents. These areas often have
physical geographic boundaries and have been considered valuable for their use in small area
studies (US Census Bureau 1990). Census block groups are often thought to be representative of
neighborhood; however, their boundaries are administratively defined and may have no
relevance to the local culture or functionality of a specific zone.
Matthews (2018) has noted that ‘there is a centralized thought throughout the health
sciences that census data defines neighborhood.’ This generalization has led researchers to use
census data as representations of place in health studies with little regard to the actual functional
neighborhood. Census areas are relatively arbitrary with little regard to the actual practicality,
contextual effects, or social structure within a neighborhood creating potential misreporting and
error when used in health research (Sharkey and Faber 2014).
The ability to access a wide array of population data at these predefined administrative
units makes census and ACS information the most easily accessible and cost-effective data
resource for health researchers (Foster and Hipp 2011; U.S. Census Bureau 2018a). Census and
ACS data are often used as a rudimentary proxy for neighborhood. Although Census data is
considered 100% coverage, it is assigned to administrative census units that are subject to
MAUP effects, miscount, and other non-sampling errors. The ACS is also assigned to
administrative census units; however, in contrast to decennial census reporting, the ACS uses a
relatively small sampling of the entire population introducing potential misreporting error. and
although the ACS is commonly used for population insight between census reports it may be
17
only somewhat representative of current population dynamics. The ACS sampling methodology
utilizing small sample size also means some measures of population characteristics may not be
reliable making the resulting data subject to the atomistic fallacy (Spielman and Logan 2012;
Robertson and Feick 2018). Additionally, administrative census units are subject to boundary
adjustments as their populations change between census periods creating incongruent boundary
zones. Utilizing aggregated census and ACS data can lead to misreporting caused by the bias
introduced from zonal effects, the MAUP, the atomistic fallacy, and by using estimated and
continuously changing population information (Openshaw and Alvanides 2001; Diez Roux 2007;
Kwan 2012b).
2.1.3. ‘Nonstandard’ Administrative Unit Areas
The U.S. Census Bureau also tabulates data for administrative units that fall outside of the
conventional census grouping levels. These special localities have been created for different
reasons than standard census areas and consequently share no geographic connection to
administrative census boundaries. These special zones were created either independently from or
in joint effort with the U.S. Census Bureau; however, the U.S. Census Bureau does compile data
within these areas (U.S. Census Bureau 1990).
2.1.3.1. ZIP Code
Zone Improvement Plan (ZIP) code divisions are completely independent of other types of
administrative boundaries. ZIP codes do not correspond with census tracts but instead were
devised by the U.S. Postal Service (USPS) as a means of identifying the primary mail delivery
system area associated with a municipality or postal office. The system was devised by the
Postmaster General to accommodate increasing mail flow to growing populations across the U.S.
and follow a network of mail delivery routes throughout a neighborhood (US Census Bureau
18
2018c). ZIP codes evolve with a neighborhood as it grows and are an independent zoning system
that are frequently spatially incompatible with other types of traditional administrative
boundaries.
ZIP Code Tabulation Areas (ZCTAs) are reported by the U.S. Census Bureau as spatial
delineations of USPS routing zones. These reporting areas are designated by determining the
most recurrent ZIP code inside of a census block and then delegating it to the entire census
block. ZCTAs are often different than the actual ZIP code they represent. Additionally, very
rural areas may not have a ZIP code assignment, leaving a void in ZCTA coverage (US Census
Bureau 2018c).
2.1.3.2. School Districts
School districts are administrative units created and determined by local governments and
maintained in joint effort between the U.S. Census Bureau and the National Center for Education
Statistics (NCES), a part of the Institute of Education Sciences within the U.S. Department of
Education (U.S. Census Bureau 2018b; Institute of Education Sciences, NCES 2018c). School
districts are independent from standard census areas and serve as a demographic and economic
measure of the school-aged population inside the geographic area within district boundaries.
School district boundaries are updated biennially by the U.S. Census Bureau’s Geography
Division for socio-demographic reporting and are used to create spatial layers for census
TIGER/Line files (U.S. Census Bureau 2018b).
School districts are boundary ‘catchment’ areas that dictate which area schools residents
can attend (Institute of Education Sciences, NCES 2018b). The NCES uses school district data
compiled from both the ACS and the decennial census for their Education Demographic and
Geographic Estimates (EDGE) program (Institute of Education Sciences, NCES 2015).
19
The EDGE program uses geographic information to inform policy makers and the public
about the correlation between schools and the people and area they serve. The EDGE program
compiles data to create school district locales, used to classify school districts into four
categories; urban, suburban, town, or rural and three subsequent subtypes. The function of the
EDGE locale is to provide spatial data for research and analysis that gives researchers the ability
to customize detailed investigations of the dynamics occurring between the social and physical
properties of a locale area (Institute of Education Sciences, NCES 2015).
The EDGE locale boundary function also offers a ZCTA locale file where researchers
can assign ‘NCES indicators’ to ZCTAs for use with TIGER/Line files. The ZCTAs are
determined by the boundaries of the NCES locales (Institute of Education Sciences, NCES
2018a).
School districts are not static and are relatively arbitrarily defined. They often conflict
with other types of administrative units such as counties, census tracts, and ZIP codes. The
potential incompatibility of school districts with other administrative unit types causes school
districts to sometimes have multiple, simultaneous, spatial unit assignments, i.e. spanning
multiple counties and/or having varying ZIP codes within the same school district (Institute of
Education Sciences, NCES 2018c). The use of mismatched spatial units may automatically
introduce bias by potentially creating or eliminating important data (Eagleson 2002). Combining
potentially mismatched spatial units with aggregated ACS data adds to uncertainty and creates
significant possibilities for misreporting and error.
2.1.3.3. American Indian, Alaska Native, and Native Hawaiian Areas (AIANNHAs)
The U.S. Census Bureau collects and provides population data for Native American Indian lands
situated throughout the U.S. as well. The Bureau of Indian Affairs (BIA), a branch of the U.S.
20
Department of the Interior (DOI), keeps a record of American Indian tribes and maintains
governance over tribal affairs. The U.S. Census Bureau compiles population data from Indian
area census reporting zones for the collection and quantification of populations within tribal
geographic boundaries and off-reservation trust lands (U.S. Environmental Protection Agency
2016; U.S. Census Bureau 2010a).
Tribal areas known as American Indian Reservations (AIRs) and off-reservation trust
areas are Federally and/or State recognized Indian use territories independent from any other
form of administrative boundary. The geographic boundaries of AIRs reside within the borders
of the U.S. and were determined on an individual tribal basis by final tribal treaties or judicial
orders agreed upon with the appropriate State Government(s) and the Federal Government. Each
tribe has governing authority over their people which consists of their independent tribal
governments, local laws, and boundary zones. The U.S. Federal Government maintains ultimate
federal jurisdiction over AIRs (U.S. Census Bureau 2010a).
Federal AIR borders can span all forms of classical administrative boundaries including
state and county borders (U.S. Environmental Protection Agency 2016). State AIRs have a
government-appointed intermediary that determines and reports State recognized tribal
boundaries to the U.S. Census Bureau. State AIRs cannot cross State boundaries but can cross
county lines (U.S. Census Bureau 2010a)
The U.S. Census Bureau works with tribal government agencies to annually identify and
update reservation boundary lines and features through the Boundary and Annexation Survey
(BAS). The main function of the BAS is to make certain that legal tribal boundaries are
documented correctly so that population data is accurately recorded and reported to local, tribal,
county, and federal agencies (US Census Bureau 2017).
21
The Tribal Statistical Areas Program (TSAP) is a decennial survey that provides tribes
the option of delineating physical boundaries within their specific AIR to create tribal Census
tabulation areas. TSAP data is utilized by the U.S. Census Bureau and the American Community
Survey (ACS) to provide statistical socioeconomic and demographic data for tribal, federal and
state agencies. Tribes can designate boundary lines creating the reporting areas for each State
identified tribal statistical area (SDTSA), Tribal designated statistical area (TDSA), Tribal
census tract, and Tribal block group (U.S. Census Bureau 2010b; U.S. Department of Homeland
Security 2017).
The U.S. Census Bureau designates a unique numeric tribal census code, alphabetically
by tribe, for each federal AIR across the U.S. independent from any non-reservation coding
system. Additionally, Federal Information Processing Series (FIPS) and American National
Standards Institute (ANSI) codes are also assigned, and are unique, to each tribe by State.
Although appointed alphabetically by State, federal AIRs for the same tribe can have completely
different FIPS codes if they cross a State line (U.S. Census Bureau 2010a).
Buffer Zones
GIS serves as a valuable tool for facilitating the ability to measure proximity around subjects of
interest (Perchoux et al. 2013). Proximity zones measure the extent of a neighborhood's
environment and commonly use buffers as a measure of locality (Faber and Sharkey 2015).
Buffers allow researchers flexibility to examine neighborhood exposures at an individual level in
contrast to being constrained by data aggregated at the scale of traditionally used administrative
boundaries (Spielman and Logan 2012; Matthews and Parker 2013).
There are several types of buffers used in health research. Three of the more common
buffer methods are circular, street network, and activity space. Figure 3 compares circular and
22
network buffers (Chaix et al. 2009). Ego-centric buffers are often considered the best approach at
representing a subject’s immediate neighborhood relying on aggregated proxies combining
contextual characteristics of place with an individual’s geocoded location as a representation of
physical address (Spielman and Logan 2012; Matthews and Parker 2013).
Within these different types of buffer zones, boundaries may be sharp or fuzzy depending
on methodology (Chaix et al. 2009; Spielman and Logan 2012). Buffer sizes are subjective and
are often determined by following those used in prior research. Measurement within a buffer
zone may be linear or not and may not always account for obstacles created by physical
surroundings (Leite et al. 2015).
Figure 3. Circular and network buffers around a point of interest (Source: Hall et al. 2007)
23
2.2.1. Circular
Circular buffers create an isotropic neighborhood defined by a specified distance extending out
equally in every direction from the center (Chaix et al. 2009). Frequently, circular study zones
are used in health research as a means of quantifying factors within a specified boundary
distance. These ego-centric zones are often a preferable, all-inclusive, measure of the relevant
neighborhood and are used as an alternative to fixed administrative delineations representative of
a locale (Perchoux et al. 2013).
Circular buffer sizes are not standard or fixed, allowing researchers to adjust the sizing to
whatever threshold deemed most appropriate for a study (Hall et al. 2007). There is no consensus
as to what the buffer size should be; however, recent health studies have used a radius of half a
mile around a specific point of interest (Perchoux et al. 2013). Research has shown however that
the buffer size selected directly affects the reported functionality of a buffer zone for land use
(Strominger et al. 2016).
Circular buffers around a point of interest can introduce bias into a study depending on
what type of contextual variable is being measured. Isotropic neighborhoods can be valuable for
examining environmental and epidemiological exposures around a point of interest; however,
they may have limited efficacy in assessing health issues influenced by an individual’s daily
traverse.
Often circular buffers utilize Euclidian distance as a measure of contextual variables
within a neighborhood. This straight-line distance is measured from point A to point B and
commonly referred to ‘as the crow flies’ which generally does not account for barriers
introduced by physical geography or the built environment. Many of the aspects relating to
health research do not pertain to exposures occurring ‘as the crow flies’ (Chaix et al. 2009; Hall
et al. 2007). Euclidian distance does not account for the interaction between population and
24
physical environment; however, its usefulness and relevance depend on what proxy is being
calculated. Notwithstanding these potential problems, health research frequently uses circular
buffer areas and straight-line distance as a measure of proximity (Foster and Hipp 2011). The
built environment, street networks, and physical geography can directly affect the walkability
and otherwise traversable extent of a neighborhood (Boruff et al. 2012; Perchoux et al. 2016).
2.2.2. Network
Street network buffers are line-based and defined by some measure of the road system from a
feature of interest (Hall et al. 2007). Street network buffer sizes, like circular buffers, are not
standardized and threshold distances vary depending on study purpose. Areas experiencing
higher population densities have demonstrated the need for an increase in buffer size compared
to lesser populated areas (Perchoux et al. 2016).
In contrast to circular buffers, street network buffers are theorized to be a more accurate
representation of neighborhood since they provide a course of navigable passage around and
through the physical and built environment (Hall et al. 2007). Street network buffer zones are
often considered a more human-oriented means of measuring neighborhood as they reflect routes
of regular commute and use, translating to routine environmental exposures at an individual level
(Perchoux et al. 2013).
Past research has attempted to operationalize network buffers by street pattern, viewing
major roadways as buffer zone borders and minor road systems as potential routes of traverse
within a neighborhood. Use of this methodology, although centrally important to the concept of
neighborhood and the use of street networks as buffer zones for analysis, is considered a building
block for delineating neighborhood units (Cutchin et al. 2011). It is theorized that using minor
street network connectivity as a measure of social interactions representative of the
25
neighborhood will reduce measurement bias by minimizing intra-neighborhood variability and
amplifying neighborhood differentiation (Foster and Hipp 2011).
Network distances in urban settings often use a grid-based, Manhattan method, of
distance measurement where the distance between points A and B is determined using right
angles along an axis or on a grid. Manhattan distance is different from Euclidean and network
distance because Manhattan distance is the measure of the two right-angle sides of a triangle
(Apparicio et al. 2017).
This method is commonly used to measure two points on a municipality map and often
represents ‘city block distance’ (Charreire et al. 2010). Grid networks can be a useful measure of
traverse and walkability in certain environments; however, their usefulness is limited to regular
and relatively close urban street networks. When used in environments with sparse or irregular
street networks, grid networks can introduce substantial measurement error (Hall et al. 2007).
Street-based buffer zones are often identified as walkability zones, usually determined
from preceding studies to be within a 15-minute walk from a residence or feature of interest,
translating to roughly 1,000 m from the point of interest (Hall et al. 2007; Perchoux et al. 2016).
Street network buffer distances are commonly determined by a researcher and
subsequently, they do not provide accurate representations of the actual physical area, or
direction, most frequented by a resident. Street network buffers have a multitude of applications;
however, they are not always realistic of an individual’s daily traverse exposures or the
anisotropic nature of human routine (Chaix et al. 2009; Perchoux et al. 2016).
2.2.3. Activity Space & Perceived Environment
One of the primary limitations to the classical measures of neighborhood is their inability to
represent the true dynamic nature of the human element. Traditional measures of neighborhood
26
focus on place of residence and are defined by either fixed administrative boundaries or set
buffers theorized as representative of a resident’s daily exposures. This conventional view of
neighborhood neglects that the human routine is not static but more often variable and changing
(Macintyre et al. 2002; Matthews and Yang 2013).
Twenty-first century society has created an increasingly mobile culture (Chaix et al.
2009; Matthews and Yang 2013). Economy and opportunity have added to the transient nature of
residence. The incorporation of and advancements in GIS and related geospatial technologies
now provide health researchers with new methods of capturing the complex dimensions of life
outside an individual’s dwelling (Perchoux et al. 2013).
The concept of activity space follows a person’s daily traverse, spatially and temporally,
assessing direct and potential exposures relative to that individual (Kwan 2012b; Perchoux et al.
2013). The range of actual exposures an individual experiences throughout their average daily
routine and traverse are often far greater than what has been deemed by classical measures of
neighborhood (Chen and Kwan 2015).
Activity spaces tend to demonstrate a directionally oriented neighborhood composed of
residential, public, and institutional places. Activity spaces vary by age group and residential
type based on socioeconomic status (Perchoux et al. 2013). Current methodologies commonly
utilized in recording an individual’s routine activity space include GPS tracking, reporting
through questionnaires, self-reporting, or other means of volunteered geographical information
(Matthews and Parker 2013; Perchoux et al. 2013). Geoprocessing tools are then employed to
evaluate data and create representative activity spaces (Perchoux et al. 2013).
The perceived neighborhood can be highly subjective often differing from an actual
traversed activity space by encompassing areas thought of as preferable even if outside of a
27
person’s true realm of daily use (Flowerdew et al. 2008; Perchoux et al. 2016). The issue of
differentiated geographic delineations between individual activity spaces is further complicated
by the temporal aspect affecting a perceived or actual neighborhood (Kwan 2012b). The
variability of this form of spatial evaluation creates scale and zonal effects which can introduce
congruence problems and aggregation bias (Swift et al. 2014).
2.2.4. Dasymetric Mapping
Dasymetric mapping is an area-based mapping technique that constructs population information
from multiple aspatial, areal, and linear datasets (Holt et al. 2004; Swift et al. 2014). Dasymetric
maps use ancillary data to create dimensional zones of measure disaggregated from the confines
of administrative boundaries. Dasymetric mapping shares similarities with choropleth mapping
techniques with some substantial differences (Mennis 2003; Mennis and Hultgren 2006; Nelson
2010).
The use of ancillary data in dasymetric mapping creates a more realistic depiction of real-
world attributes compared to choropleth maps. Dasymetric maps follow existing spatial patterns
and are not subject to the contrasting population differences often depicted in the arbitrarily
defined administrative boundaries used in choropleth mapping (Figures 4 and 5) (Mennis 2003;
Nelson 2010).
Figures 4 and 5 demonstrate the contrast between choropleth and dasymetric mapping
techniques. Figure 4 shows how choropleth maps depict population density throughout a county
using traditional administrative boundaries. The choropleth maps show distinct, sharp variations
in area population density between administrative boundary zones and are likely not
representative of the actual population distribution over those areas. Figure 5 depicts the same
28
county using dasymetric mapping. The use of ancillary data creates a more probable
representation of that county’s actual population distribution throughout the area (Nelson 2010).
Figure 4. Guilfrod County, North Carolina census tract and block group population density
(Source: Nelson 2010)
Figure 5. Dasymetric map of Guilford County, North Carolina population density
(Source: Nelson 2010)
29
Dasymetric mapping also shares close similarities with areal interpolation as its function
is to transform geographic data from one boundary system to another (Mennis 2003; Mennis and
Hultgren 2006). The inherent problem of utilizing different boundary systems together is the
inability to analyze and combine data of differing boundary types together without losing and or
biasing the data where population counts are neither fabricated nor eliminated in the final results
(Eagleson 2002; Mennis 2003). A spatial discrepancy, as is found in trying to use incompatible
boundary systems, can also be a result of temporal effects as is often experienced when
comparing census data from different reporting periods (Zandbergen and Ignizio 2010).
An additional complication is that ancillary data often contain geographic and/or attribute
errors, further biasing results and can be difficult to account for because of contextual effects.
Associating population data with area attributes can be problematic and can introduce
uncertainty that is usually not included in reporting (Nagle et al. 2013).
Dasymetric mapping attempts to alleviate these situations by utilizing aggregated data
and transforming those values by combining it with ancillary data by reaggregation into smaller
spatial zones that more closely depict the actual population distribution throughout a given area,
as was illustrated on Figures 4 and 5 (Mennis and Hultgren 2006).
Dasymetric maps are helpful in illustrating heterogeneity within a specific population and
are a valuable means of visualizing cluster events throughout a geographic area (Barrozo et al.
2016). Dasymetric mapping is also theorized as providing a truer representation of populations
dispersed throughout small areas, such as block group population distribution in both urban and
rural environments, making this technique a valuable tool for researchers conducting small area
studies (Mennis 2003; Mennis and Hultgren 2006).
30
Chapter 3 : Methods and Data Sources
The state of Arizona was chosen as the study area for this thesis because of the population
distribution and diversity found throughout the state. This type of population dispersion was an
important factor in being able to show how choosing varying forms of neighborhood for a study
can affect the reporting outcomes. Administrative boundaries at the county, census tract, and
census block group levels were compared to evaluate what reporting differences existed between
reporting types for specified variables. Within these variables, specific populations of concern
were identified and focused on to determine if and how these neighborhood variations lead to
poor representation of specific population groups. The temporal aspects of this study used U.S.
Census Bureau data from 2000 and 2015. The 2000 decennial U.S. Census was utilized for the
majority of the analysis as it was the last census using 100 percent reporting. The 2010-2015
ACS estimates published in 2015 were used in the latter part of the study for further comparison
as it was the last ACS to include data for areas within the state that only report once every five
years (as is common for AIRs). Census tracts and census block group boundaries were also
adjusted for the appropriate temporal period, as boundary changes affected the delineation of
reporting units used by the ACS. The population variables examined were median household
income, age, and ethnicity. Data were downloaded from the U.S. Census Bureau. Microsoft
Excel 2016 was utilized for data conversion from .csv and file format preparation. ArcGIS
Desktop (versions 10.5.1 and 10.6) were used for data analysis.
Study Area
The state of Arizona is the sixth largest state in the U.S., covering a total area of 113,990.30
square miles. There are 15 counties and by 2010 the U.S. Census Bureau reported there being
1,526 census tracts and 4,178 census block groups dispersed throughout the state (U.S. Census
31
Bureau 2015). The state is also home to 22 federally recognized American Indian Nations.
Figure 6 shows the largest of the aforementioned geographic units in the study area. The maps in
Figure 6 also show the study area by (a) Arizona counties; (b) census tracts; (c) census block
groups; and (d) AIR’s with their associated area’s minimum, mean, and maximum geographic
extent.
The U.S. Census Bureau estimated the Arizona population in July of 2017 to be at
7,016,270 (U.S. Census Bureau 2018d). Approximately 68% of the state’s population is located
within the Phoenix Metro area (Maricopa County), and 15% in the Tucson Metro area (Pima
County). That leaves the remaining 17% dispersed throughout the predominantly rural,
remaining 13 counties. The variability found throughout the state provides important insight into
how neighborhood delineations might affect outcomes in both traditionally examined
metropolitan zones as well as in less frequently studied rural areas where aggregation issues may
have a greater effect.
Hypothesis
Data acquired from the U.S. Census Bureau is a common resource for health researchers. These
data are reported per administrative unit. The null hypothesis presented and assumed in many
research studies is that median values for variables are relatively unchanging, and equivalent,
regardless of the administrative unit utilized. The null hypothesis used for the current study
offers that aggregation has little statistical relevance throughout a geographic neighborhood,
regardless of population dispersion. The alternative hypothesis presented in this thesis is that
actual values vary across administrative units and that aggregation can affect resulting outcomes.
This variability dictates that that consideration of how a neighborhood is selected is important
and affects how outcomes are reported.
32
(a)
(b) (c) (d)
Figure 6. Area maps by: (a) Arizona counties; (b) census tracts; (c) census block groups; and (d)
AIR’s. The colors show the geographic extent by minimum, mean, and maximum areas for each
administrative unit.
33
The hypothesis is tested using three variables that are commonly used in research as
social health determinants: Median household income, vulnerable population represented by the
percentage of children and the elderly, and ethnicity represented by percentage of Native
American population. These variables were then examined at three different administrative units
to evaluate how and if there are significant variations in terms of their estimation depending on
the size of the geographic reporting unit.
For this thesis the administrative units used were county, census tract, and census block
group. The results presented in Chapter 4 examine decennial census data from the last 100
percent reporting period (i.e. 2000 - 2010) and investigates what variation occurs for selected
variables within differing administrative units. The second phase of this analysis that is briefly
referred to in Chapter 5 examined the variability of selected variables using ACS data for three
of Arizona’s 15 counties. The first ACS reporting that fulfills the needs of this study is from the
period 2010-2015.
Data Sources
Shapefiles and attribute data was downloaded from the U.S. Census Bureau and imported into
ArcGIS (Table 1). Initially, Esri Business Analyst was used; however, it was determined that
Esri utilized numerous sources for their reports and values, potentially confounding results. This
required sourcing data exclusively from the U.S. Census Bureau’s American Factfinder for data
consistency. This was crucial to ensure that the appropriate reporting survey type (decennial vs.
ACS), administrative unit, and temporal period was used for each of the variables examined in
this thesis.
34
Table 1. Census data sets and sources.
Dataset
Description File Type Data Type Temporal
Reporting
Period
Source
Counties
Arizona county boundaries .shp files Vector data
(polygon)
Vintage: 2007
(for 2000
Census)
2015
U.S. Census Bureau
TIGER/Line files
U.S. Decennial Census
ACS 2015
Census Tracts Boundary lines for all census
tracts in Arizona
.shp files Vector data
(polygon)
Vintage: 2007
(for 2000
Census)
2015
U.S. Census Bureau
TIGER/Line files
U.S. Decennial Census
ACS 2015
Census Block
Groups
Boundary lines for all census
block groups within Arizona
.shp files Vector data
(polygon)
Vintage: 2007
(for 2000
Census)
2015
U.S. Census Bureau
TIGER/Line files
U.S. Decennial Census
ACS 2015
Demographic
Profile
Reporting for population by age
group per administrative unit
.csv converted
to .xlsx and
.dbf
Vector data
(discrete,
point)
2000
2015
U.S. Census Bureau
2000 Census;
Demographic Summary
file 1
ACS 2015
Ethnicity
Separate datasets reporting Non-
Hispanic & Hispanic ethnicity
counts per administrative unit
.csv converted
to .xlsx and
.dbf
Vector data
(discrete,
point)
2000
2015
U.S. Census Bureau
2000 Census; Ethnicity
non-hispanic; Hispanic
Ethnicity
Summary file 1
ACS 2015
Median
household
income
Median household income per
administrative unit
.csv converted
to .xlsx and
.dbf
Vector data
(discrete,
point)
2000
2015
U.S. Census Bureau
2000 Census; Summary
file 3
ACS 2015
35
Data from the 2000 decennial census was determined to be the last decennial census to
use 100 percent reporting and that would accurately represent population dynamics for this
analysis. The boundaries for census reporting units were also different in 2000 as compared to
the 2010 decennial Census and later ACS tabulations. The U.S. Census Bureau converted all
2000 boundary files to shapefile compatible format in 2007 requiring usage of a 2007 vintage
file for acquisition of the 2000 boundary shapefiles (Table 1).
The 2010-2015 ACS data was utilized for the second phase of analysis in this thesis. The
appropriate temporally corresponding vintage shapefiles were also downloaded to support this
part of the analysis (see Table 1 for additional details).
Methodology
Data were downloaded from the U.S. Census Bureau website. Variable data were selected per
administrative unit within Arizona, downloaded as a .csv file and converted to Excel files.
Spreadsheets were then designed in Excel and imported into ArcGIS Desktop. In ArcGIS, the
Excel files were converted into database files that were then joined with their corresponding
administrative units (see Figure 7 for workflow).
Administrative boundaries for three commonly used census reporting units were utilized
for comparison. Arizona’s 15 counties were the basis for examining health determinant variables
and if and how results varied between the census reporting unit that were used. Three census
reporting units were used in this study: County level, Census Tract level, and Census Block
Group level. All of the maps displayed throughout the thesis use the UTM Zone 12N map
projection; however, transformations were conducted using ArcGIS, ArcToolbox Data
Management toolset to convert to the most appropriate visualization projection from the U.S.
Census Bureau source data which utilized GCS North American 1983.
36
Figure 7. The workflow
This study examines the geographic variability of three variables that are commonly used
in health studies: (1) Wealth by median household income; (2) race and ethnicity, the percent of
the population that was Native American in this instance; and (3) vulnerable population (the
percent of population) over age 65 and under age 16 in this instance.
The data for these three social determinants of health were imported into ArcGIS and
threshold values were determined for each variable. Then scatterplots were built for each
variable and color coded by threshold bracket. Scatterplots were used to evaluate thresholds set
at ≥ 125%, ≤ 75%, and ± 25% of the corresponding county and census tract values. The
Background
research
•Research previous works to develop research design
Data
Collection
•Locate and download suitable data sources consisting of census tables and the accompanying shapefiles
Data
Preperation
•Convert.csv files to .xlsx and build spreadsheets
•Import data into ArcGIS
Data
Analysis
•Join files and build vizualization datasets in ArcGIS
•Calculate data threshold values
•Build scatter plots for each per variable per census reporting unit
Results
•Report results and discuss the importance and significance of their geographic varability for future health
studies.
37
calculations were conducted using ArcGIS and selecting by attributes where the formulas for
each variable threshold were as follows:
A. County median values as reported by the U.S. Census Bureau were used for counties and
census tracts using the following rates.
(1) ≤ 75%: Census Tract value ≤ (.75*County value)
(2) ≥ 125%: Census Tract value ≥ (1.25*County value)
(3) ± 25%: CT value ≥ (.75*County value); and ≤ (1.25*County value)
where the three classes used to classify the census tract values express the variability
relative to the county estimates for each variable and where < 75% is depicted throughout
in blue, > 125% is depicted throughout in red, and ± 25% is depicted throughout in green.
B. The same approach was also used for census tracts and census block groups using the
following rates:
(4) ≤ 75%: Census Block Group value ≤ (.75*Census Tracts value)
(5) ≥ 125%: Census Block Group value ≥ (1.25* Census Tracts value)
(6) ± 25%: CBG value ≥ (.75*CT value); and ≤ (1.25*CT value)
where the three classes used to classify the census block group values express the
variability relative to the census tract estimates for each variable and where < 75% is
depicted throughout in blue, > 125% is depicted throughout in red, and ± 25% is depicted
throughout in green.
The resulting units classified as outliers were grouped by county using population as a
basis for the groups. Group I contain the two most populated counties in the state, Maricopa and
Pima with populations of 3,072,149 and 843,746 respectively. Group II consists of seven
counties with modest populations ranging from 97,000 to 179,000: Pinal, Yavapai, Yuma,
38
Mohave, Cochise, Coconino, and Navajo. Lastly, Group III contains the six most rural and least
populous counties with totals ranging from 8,000 to 69,000: Apache, Gila, Santa Cruz, Graham,
La Paz, and Greenlee. There were no counties with a population that fell between Groupings.
Graphs were made for each variable by county to describe the variability from threshold
values representing units classified as outliers and introduced by the choice of census reporting
unit. Maps were also prepared but are displayed for just six of Arizona’s 15 counties in the thesis
itself. Two counties from each of the aforementioned groups were selected for this purpose as
follows: Maricopa and Pima in Group I, Coconino and Pinal in Group II, and Apache and Santa
Cruz in Group III.
The following chapter reports the results and reveals information that has far reaching
consequences for health researchers who want to learn how their choice of census reporting unit
may potentially influence the resulting outcomes of studies that consider social determinants of
health.
39
Chapter 4 : Results
This chapter is detailed by section and investigates how the spatial measurement of a locale can
affect the geographic reporting of health determinants. Three independent variables are
compared at three commonly utilized variations of neighborhood: county, census tract, and
census block group. In 2000, Arizona possessed 15 counties, 1,108 census tracts, and 3,570
census block groups. Following are the trends discovered when examining each variable by
neighborhood for that temporal reporting.
Median Household Income
Median Household Income is a significant indicator of human well-being and a key health
determinant. Theoretically, median household income reflects a household’s ability to access
resources, such as sustenance and medical care, and indicates overall quality of life.
4.1.1. County Level
Median Household Income by county is displayed in Figure 8. The values demonstrated are as
reported by the U.S. Census Bureau’s 2000 decennial census. Arizona is depicted throughout this
thesis at a scale of 1:4,275,000.
The counties demonstrating the highest median household incomes, depicted on the map
in red, have multiple factors contributing to their higher incomes (Figure 8). Maricopa County
contains the greater Phoenix Metro area, the largest metropolitan area in Arizona. The greater
Phoenix Metro is a major service hub not just for Arizona but also the entire southwestern U.S.
providing population and infrastructure to support an economy that can provide higher wages
and thus higher household incomes in comparison to the State’s rural areas. Coconino and
Greenlee, the remaining two counties demonstrating high median household income, reflect their
40
affluence from industry. Median household income in these counties is not necessarily associated
with high population or service areas.
Figure 8. Median Household Income for each Arizona county
The counties in green represent the mid-level classification and for the purposes of this
thesis follow the State of Arizona’s Median Household Income. There are seven counties that
comprise this level, making this grouping the largest of the three classes for the variable.
The counties with the lowest median household incomes, colored blue on the map, have
limited service centers and county populations that did not exceed 100,000 (Figure 8). These five
counties are more rural in nature and do not possess any substantial form of industry. Services
are limited and few even across large geographic areas. Additionally, these counties also have
substantial portions allotted as Native American Tribal Areas (Figure 6d).
41
County Median Household Income varies considerably from the low economic class
(counties in blue) to the high economic class (counties in red). The counties in blue are primarily
the most rural; however, rurality does not dictate income level as Coconino and Greenlee
counties are largely rural and yet are still classified as two of the State’s wealthiest counties. The
most populous county, Maricopa County, is also one of the most prosperous.
4.1.2. Census Tract Level
Viewing median household income by census tract provides an entirely different perspective of
the household income distribution throughout the various counties and the State of Arizona
(Table 2). A comparison of Figures 8 and 10 reveals that census tracts with median household
incomes classified as outliers > 125% (depicted in red) and < 75% (depicted in blue) are
relatively confined to specific areas within each county and are not equally distributed
throughout those counties.
Three counties; Maricopa, Pima, and Santa Cruz counties, exceed the States’ average of
52.2% of census tracts classified as outliers (Table 2). Two of these counties, Maricopa and Pima
contain the two largest metropolitan areas within the State, the Phoenix and Tucson Metros,
respectively. Maricopa County has a population of 3,072,149 and Pima County has a population
of 843,746. Santa Cruz County, however, has a relatively low county population of 38,381 and is
the smallest county in terms of area within Arizona (U.S. Census Bureau 2015).
The remainder of the counties fall under the 52.2% state average; however, five counties
have fewer than 20% of their census tracts classified as outliers (Table 2). Four of these counties
fall between 12.5 and 20%. Graham County has a population of 33,489 with 12.5% of its census
tracts classified as outliers. Yavapai has a population of 167,517 with 15.4% of its census tracts
classified as outliers. Mohave contains a population of 155,032 and has 16.7% of its census tracts
42
classified as outliers, and Gila has a population of 51,335 with 20% of its census tracts classified
as outliers.
Table 2. Counts and percentages of census tracts with median household incomes ≤ 75% and ≥
125% of county median values.
County Population No. of census
tracts
No. of census
tracts with
MHI
Nos. and % of outlier census
tracts *
Apache 69,423 14 14 5 + 1 = 42.9
Cochise 117,755 21 21 4 + 5 = 42.9
Coconino 116,320 28 27 8 + 5 = 46.4
Gila 51,335 15 15 2 + 1 = 20
Graham 33,489 8 8 1 + 0 = 12.5
Greenlee 8,547 3 3 0, 0
La Paz 19,715 6 6 0 + 2 = 33.3
Maricopa 3,072,149 663 659 181 + 194 = 56.6
Mohave 155,032 30 30 3 + 2 = 16.7
Navajo 97,470 23 23 6 + 3 = 39.1
Pima 843,746 198 198 52 + 70 = 61.6
Pinal 179,727 33 32 9 + 4 = 39.4
Santa Cruz 38,381 7 7 2 + 2 = 57.1
Yavapai 167,517 26 25 3 + 1 = 15.4
Yuma 160,026 33 32 6 + 6 = 36.4
Totals 5,130,632 1,108 1100 296 + 282 = 52.2
* No. of census tracts with MHI <75%, >125%, and the sum of the two classes of outliers as a
percentage of total.
Greenlee County, with the smallest population in the State of 8,547, was the only county
in Arizona that did not have any census tracts classified as outliers (Table 2).
The remaining seven counties fall into the mid-level classification with the percentages of
census tracts exceeding the county thresholds ranging from 33 to 46% (Table 2). These counties
are as follows: La Paz County, the least populous within this category with a population 19,715
and 33.3% of the census tracts classified as outliers; Yuma County, the second most populous
county with a population of 160,026 and 36.4% of the census tracts classified as outliers; Navajo
County with a population of 97,470 and 39.1% of the census tracts classified as outliers; Pinal
County, the most populous county in this grouping with a population of 179,727 and 39.4% of its
43
census tracts classified as outliers; Apache County, the poorest county in the State with a census
tract median household income average of $21,497.29 and a population of 69,423 and 42.9% of
its census tracts classified as outliers; Cochise County with a population of 117,755 and 42.9%
of its census tracts classified as outliers; and lastly, Coconino County, the wealthiest in this
grouping, with a median household income of $39,066, a population of 116,320, and 46.4% of
its census tracts classified as outliers.
Two counties within Arizona contain 70% of the state’s census tracts. These counties are
Maricopa and Pima with 663 and 198 census tracts, respectively. Government apportioning of
spaces by population dictates that the metro areas have smaller reporting units to account for
their higher populations.
Comparison between county and census tract is implemented by evaluating how median
household income is reflected within specified thresholds of the county median income,
classifying values outside of these threshold values as outliers (Table 2 and Figure 9). Red dots
in Figure 9 reflect census tracts that have values ≥ 125% of their county’s median value,
indicating areas that exceed county median incomes (similar to Figure 8). Blue dots indicate
those census tracts that have a value ≤ 75% of their county’s median value, denoting less wealthy
areas that may be susceptible to poverty. Green dots show which census tracts are within ± 25%
of their county’s median value. Figures 9 and 10 reveal that median household income is not
spread equally across a county and is specific to census tracts within those counties.
Figure 9 reports the variability by census tract and county graphically and reveals that
Arizona possesses some very wealthy census tracts in some counties as well as some very poor
census tracts. Furthermore, by examining the graph it is evident that some of these extremes are
found within the same county, e.g. Maricopa County (Figure 10b).
44
Figure 9. Median household income by census tract within each county.
Figure 10a visually demonstrates the geographic structure of the variability found using
the state’s census tracts as the reporting units. The greater Metro areas are also shown at a larger
scale to reveal the census tract variability within these zones (Figures 10b and 10c).
Figure 10b shows part of Maricopa County’s variability in and near the greater Phoenix
metro region. The large zones that are the poorer areas of the county are the furthest from the
greater metropolitan area. Pima County demonstrates the highest variability within the state
(Table 2 shows 61.6% of the census tracts classified as outliers). This county demonstrates a
high amount of variability and a segregation of economic classes at the census tract level where
the greater Tucson metro area shows distinct groupings of low, middle and high economic
classes (Figure 10c). Again however, as with Maricopa County, the census tracts that
demonstrate the highest incomes are in or near to the greater Metro area with distant census
tracts almost all colored blue
45
Figure 10. Median Household Income for each Arizona Census Tract (a) and the Phoenix Metro (b) and Tucson Metro areas (c).
(a) (b)
(c)
46
Santa Cruz County, a relatively sparsely populated county (population of 38,381), shows
people of wealth near to neighboring Pima County’s Tucson Metro area. The census tracts that
show the highest median household incomes also fall around the major transportation corridor
(not shown) from Tucson to Mexico. The higher incomes are also likely representative of the
regions other historic industry, mining. Cochise County, (population 117,755), bordering Santa
Cruz County also shows considerable variability (42.9% of the census tracts classified as
outliers, Table 2). The wealthiest areas, however, are located near the transportation corridor,
which provides economic opportunity. The remaining parts of the county reflect largely average
median household incomes, with remote zones being the least wealthy.
Apache, Navajo, and Coconino Counties (with populations of 69,432, 97,470, and
116,320, respectively) have the largest geographic areas dedicated as Native American
Reservation Areas (AIRs) and percentages of census tracts classified as outliers ranging from
39.1% (Navajo) to 46.4% (Coconino). These counties are diverse, largely rural with few resource
centers, and economically affected by unique challenges not found in other parts of the state.
Yavapai County has a relatively high population (167,517) with few census tracts
classified as outliers (15.4%). The wealthiest census tracts are located near the primary resource
community, Prescott, providing infrastructure and services. The least wealthy fall between
Prescott and AIRs.
4.1.3. Census Block Group Level
Median household income was also examined at the census block group level. Census block
groups classified as outliers are significant when comparing census tract to census block group
because of the increase in spatial units at census block group level going from 1,108 census
47
tracts and partitioning those locales into 3,570 “neighborhoods” as represented by census block
groups.
Group I consists of Maricopa and Pima Counties, respectively (Figures 11-12). Maricopa
County contains most of the Phoenix Metro area and is the highest populated region in the state.
Pima County encompasses the Tucson Metro area, and although considerably smaller than
Phoenix Metro, is the second largest metropolitan area in Arizona.
Scatterplots vary from one another as the number of census tracts and corresponding
census block groups differ based on county population (Figures 11-14). The scatterplots show
census block groups classified as outliers around the median census tract values. Maricopa
County (Figure 11) contains 663 census tracts and 2,113 census block groups, whereas Pima
County (Figure 12) contains 198 census tracts and 617 census block groups.
Census block groups classified as outliers exceed 24% in both Maricopa and Pima
Counties. Figure 11 shows numerous outliers ≥ 125% of the corresponding census tract value,
indicating specific census block groups of wealth, as well as some census block groups with
income ≤ 75 % of the corresponding census tract values, indicating relatively impoverished
areas. The two Metro areas differ as Pima County and the greater Tucson Metro area do not
demonstrate the same extremes of Median Household Income evident in Maricopa County and
the greater Phoenix Metro area, by both census tract and census block group (Figures 11 and 12).
48
Figure 11. Group I, census block group Median Household Income by
census tracts 1-333 (a) and census tracts 334-663 (b) in Maricopa County.
49
Figure 12. Group I (Continued), census block group median household income by census tract in Pima County.
50
Group II includes the seven Arizona counties with medium sized populations and
between 21 and 33 census tracts. These seven counties, in order of highest to lowest populated
county, are Pinal, Yavapai, Yuma, Mohave, Cochise, Coconino, and Navajo Counties
respectively (Figures 13a-d and 14e-g). Each of the counties in this grouping have several census
block groups with incomes exceeding ≥ 125% of the corresponding census tract. Pinal, Yavapai,
Yuma, and Coconino Counties include census block groups with zero values which indicate
census block groups which lacked sufficient resident reporting income data (Figures 13a-c and
14f).
Four of the seven counties reported outlier census block groups over the state average of
24.8%. Those counties were Coconino (34%), Pinal (31%), Navajo (29.7%), and Cochise
Counties (27.8%). The remaining three counties that fell under the state average were Yuma
(24.5%), Yavapai (18.6%), and Mohave (16.8%) Counties (Table 3).
Coconino County, the county that had the largest number of outliers, also had the highest
outlier census block group median household income, with two outlier census block groups
reporting > $80,000 (Figure 14f). Mohave County, the county with the smallest number of
outlier census block groups, had the lowest census block group median household income within
Group II with no outlier census block groups reporting > $50,000 (Figure 13d).
51
Figure 13. Group II, census block group Median Household Income by census tract for: (a) Pinal; (b) Yavapai; (c) Yuma; (d) Mohave
Counties.
52
Figure 14. Group II (Continued), census block group Median Household Income by census tract for: (e) Cochise; and (f) Coconino;
and (g) Navajo Counties.
53
Group III includes the counties with the smallest populations, (< 70,000) which contain
between three and 15 census tracts. All counties in this grouping have at minimum one census
block group that exceeds the ≥ 125% threshold; however, Greenlee County (Figure 14f) does
not have any census blocks groups ≤ 75% of the corresponding census tract median value. None
of the Group III census block groups exceed $60,000 annual median household income.
There is lesser amount of variability when census block groups are used in place of
census tracts (Table 3) than exists between census tracts and county median values (Table 2).
The largest numbers of census block groups classified as outliers occurred in Apache (46.3%),
Coconino (34%), and Santa Cruz (65%) counties. Greenlee (12.5%), La Paz (17.4%), and
Mohave (16.8%) counties had the least number of outliers. The two counties containing
Arizona’s metro areas, Maricopa and Pima, are the closest to the State’s average.
Table 3. Counts and percentages of census block groups with median household incomes ≤ 75%
and ≥ 125% of census tract values.
County Population No. of census
block groups
No. of census
block groups
with MHI
Nos. and % of outlier
census block groups *
Apache 69,423 54 54 11 + 14 = 46.3
Cochise 117,755 72 72 10 + 10 = 27.8
Coconino 116,320 106 104 18 + 18 = 34
Gila 51,335 55 55 7 + 4 = 20
Graham 33,489 27 27 2 + 3 = 18.5
Greenlee 8,547 8 8 1 + 0 = 12.5
La Paz 19,715 23 22 2 + 2 = 17.4
Maricopa 3,072,149 2,113 2,088 301 + 206 = 24
Mohave 155,032 101 101 8 + 9 = 16.8
Navajo 97,470 74 74 13 + 9 = 29.7
Pima 843,746 617 617 93 + 56 = 24.1
Pinal 179,727 116 111 15 + 21 = 31
Santa Cruz 38,381 20 20 7 + 6 = 65
Yavapai 167,517 86 85 10 + 6 = 18.6
Yuma 160,026 98 97 14 + 10 = 24.5
Totals 5,130,632 3,570 3,535 512 + 374 = 24.8
* No. of census block groups with MHI <75%, >125%, and the sum of the two classes of outliers
as a percentage of total.
54
Figure 15. Group III, census block group median household income by census tracts for (a) Apache, (b) Gila, (c) Santa Cruz, (d)
Graham, (e) La Paz, and (f) Greenlee Counties in Arizona.
55
The scatterplots show how median household income was distributed throughout counties
in 2000 using census tracts and census block groups as the geographic reporting unit. When
census block groups are used in place of census tracts, the trends discovered often change, as
seen in Figures 16-19. The color scheme remains the same throughout where red depicts census
block groups with ≥ 125% above the Median Household Income in the corresponding census
tracts, green indicates census block groups with Median Household Income values within ± 25%
of the corresponding value and blue represents census block groups with Median Household
Income values that are ≤ 75% of census tract value.
The two counties in Group I with the highest populations in Arizona are displayed in
Figures 16 and 17. The counties are depicted at a scale of 1: 2,000,000 for the county maps and
the Metro areas are depicted at a scale of 1:700,000 in the Metro area inset maps. As
neighborhoods become smaller using census block groups as the reporting unit, the Median
Household Income estimates display patterns that were indiscernible at the county and census
tract levels.
The Maricopa County maps reproduced in Figure 16 show how the Median Household
Income would be over-and/or under-estimated depending on how a neighborhood is defined.
When median household income is examined at the census block group level and compared to
the corresponding census tract estimate, the census block group estimates tend to display patterns
of wealth and poverty dispersed throughout the greater Metro area and surrounding rural areas.
The Pima County maps reproduced in Figure 16 also show differences resulting from
how Median Household Income is reported. Figure 16a depicts low income census tracts in the
south-central part of the Tucson Metro area and high-income areas to the west, north, and east,
which cover more than 50% of the Metro area. Figure 16b shows how those distributions are
56
representative of specific census block groups and not the standard across large portions of the
greater Tucson Metro area.
Group II is represented in Figure 18 by Coconino and Pinal counties. At the census tract
level, Coconino County shows a small central area of higher Median Household Income, a
pronounced area of low income to the east, and large areas falling near the county mean (Figure
18a). The map which shows Median Household Income by census block group on the other hand
reveals a dissimilar, varied and dispersed pattern throughout the county (Figure 18b).
The Pinal County maps reproduced in Figures 18c and 18d show different locations of
high and low incomes using census tracts and census block groups as reporting units
notwithstanding the presence of mid-range values covering much of this county.
Group III (Figure 19) is represented by Apache and Santa Cruz counties. Apache County
contains census tracts classified as outliers at the north and southern ends of the county (Figure
19a) but the substitution of census block groups for census tracts reveals a diverse pattern of
Median Household Income values dispersed throughout the county (Figure 19b).
Santa Cruz County reveals similar trends (Figures 19c and d). At the census tract level
Median Household Income is split regionally (Figure 19c). High income census tracts are found
in the northern and eastern portions of the county, with a large mid-range area in the southwest
and one small area of low income values within the midrange in the southcentral part of the
county. Figure 19d once again shows that using Census Block Groups as the reporting unit
produces a more varied and dispersed pattern of Median Household Income across the county
and/or individual census tracts.
57
Figure 16. Group I, Median Household Income in Maricopa County by: (a) census tract; and (b) census block group. Both pairs of
maps show the entire county on the top with the greater Phoenix Metro highlighted and shown at a larger scale on the bottom.
58
Figure 17. Group I (Continued), Median Household Income in Pima County by: (a) census tract; and (b) census block group. Both
pairs of maps show the entire county on the top with the greater Tucson Metro highlighted and shown at a larger scale on the bottom.
59
Figure 18. Group II, Median Household Income in: (a) Coconino County by census tract; (b)
Coconino County by census block group; (c) Pinal County by census tract; and (d) Pinal County
by census block group.
60
Figure 19. Group III, Median Household Income in: (a) Apache County by census tract; (b)
Apache County by census block group; (c) Santa Cruz County by census tract; and (d) Santa
Cruz County by census block group.
61
Children and the Elderly
The children (< 16 years old) and elderly (≥ 65 years old) represent the more vulnerable
population given that this group often requires more public services than the non-vulnerable
population, such as school, medical, and transportation. For the purposes of this study,
vulnerability is conveyed by percent.
4.2.1. County Level
The size of the vulnerable populations by county is shown in Figure 20. The counties with the
highest percentages of children and the elderly are predominantly rural and with one exception,
cover the northern two-thirds of Arizona. The three counties that display the lowest vulnerability
are dispersed and six of the eight counties that comprise the southern third of the state have
similar percentages of children and elderly to the State of Arizona as a whole (Table 4).
Figure 20. Vulnerability represented by the percentages of children and the elderly by county in
the State of Arizona.
62
4.2.2. Census Tract Level
The percentages of children and the elderly by census tract are displayed in Figure 21 for the
whole state as well as the Phoenix and Tucson Metro areas, respectively. The maps provide a
stark contrast to the regional values displayed at the county level in Figure 20. As Figure 22
demonstrates, there are many census tracts with estimates < 75% (depicted in blue) or ≥ 125%
(depicted in red) of the values at the county level. The most conspicuous outlier was a Census
Tract in which all of the residents were 65 years or older, notwithstanding the county (Coconino)
was one of the least vulnerable counties in Arizona.
Table 4 shows there were fewer census tracts classified as outliers when evaluating
Vulnerability in place of Median Household Income in most of Arizona’s counties as well as the
state as a whole.
Table 4. Counts and percentages of census tract Vulnerability ≤ 75% and ≥ 125% of county
median values.
County Population No. of census
tracts
No. of census
tracts with
Vulnerability
Nos. and % of outlier census
tracts *
Apache 69,423 14 14 0 + 2 = 14.3
Cochise 117,755 21 21 1 + 1 = 9.5
Coconino 116,320 28 27 6 + 7 = 46.4
Gila 51,335 15 15 0 + 2 = 13.3
Graham 33,489 8 8 1 + 2 = 37.5
Greenlee 8,547 3 3 0,0
La Paz 19,715 6 6 1 + 3 = 66.7
Maricopa 3,072,149 663 658 57 + 138 = 29.4
Mohave 155,032 30 30 0 + 1 = 3.3
Navajo 97,470 23 23 0 +1 = 4.3
Pima 843,746 198 198 18 + 52 = 35.4
Pinal 179,727 33 31 2 + 7 = 27.3
Santa Cruz 38,381 7 7 0,0
Yavapai 167,517 26 24 0 + 3 = 11.5
Yuma 160,026 33 33 4 + 15 = 57.6
Totals 5,130,632 1,108 1,098 90 + 234 = 29.2
* No. of census tracts with Vulnerability <75%, >125%, and the sum of the two classes of
outliers as a percentage of total.
63
Figure 21. Maps showing the percentages of children and the elderly for
each Arizona census tract (a) and the Phoenix Metro (b) and Tucson Metro (c) areas.
(a) (b)
(c)
64
Table 2 shows how 52.2% of the census tracts were classified as outliers for Median Household
Income, whereas just 29.2% of the census tracts were classified as outliers for percentages of
children and the elderly (Table 4).
Figure 22. Arizona census tract estimates within each county for the percentages of children
(≤ 16 years) and the elderly (≥ 65 years).
4.2.3. Census Block Group Level
When Census Block Group was used as the geographic reporting unit, in place of Census Tract,
the pattern continued to change. Figures 23-27 show the census block groups classified as
outliers around the census tract values. The distributions of high and low estimates around
census tract values display which counties contained varying numbers of children and the elderly
over short distances. Coconino County included several census block groups in a single census
tract with all residents > 65 years as expected (Figure 26f).
65
The metrics summarized in Table 5 show that the outliers increased when using census
block groups in place of census tracts, as compared to the pattern with county and census tract
values. Table 5 reports that 45.1% of the census block groups were classified as outliers,
compared to the 29.2% of the census tracts reported in Table 4.
Seven of the counties exceeded the State average of 45.1%: Maricopa (45.9%) and Pima
(52.7%) in Group I, and La Paz (56.6%), Pinal (68.1%), Graham (70.4%), Yuma (78.6%), and
Santa Cruz (85%) in Groups II and III. Table 5 also denotes the sparse population of Greenlee
County in Group III and how it had no census block groups classified as outliers.
Figures 28-31 demonstrate how census tract and census block group outliers for
Vulnerability are dispersed across the 15 counties and the State of Arizona as a whole.
Depending on the geographic reporting unit examined, differing patterns of vulnerability emerge.
When the counties in Group I are visualized at the census tract level, large census tracts
classified as outliers (≥125%) are found at some distance away from the Metro areas (Figures
28a and 29a). When the reporting unit becomes smaller, at the census block group level, patterns
showing areas with high percentages of children and elderly are seen within the Metro areas
(Figures 28b and 29b).
Group II and Group III shows counties that are largely rural with distinct patterns of
vulnerability that reflect proximity to economic centers in both census tracts and census block
groups (Figures 30a-d and 31a-d).
66
Figure 23. Group I: Census block group Vulnerable Population by census tracts 1-333 (a) and census tracts 334-663 (b) in Maricopa
County.
67
Figure 24. Group I (Continued): Census block group vulnerable population by census tract in Pima County.
68
Figure 25. Group II: Census block group vulnerable population by census tract for: (a) Pinal; (b) Yavapai; (c) Yuma; (d) Mohave.
69
Figure 26. Group II (Continued): Census block group vulnerable population by census tract for: (e) Cochise; (f) Coconino; and (g)
Navajo counties.
70
Figure 27. Group III: Census block group Vulnerable Population by census tracts for: (a) Apache; (b) Gila; (c) Santa Cruz; (d)
Graham; (e) La Paz; and (f) Greenlee Counties.
71
Table 5. Counts and percentages of census block group Vulnerability that are ≤ 75% and ≥
125% of the corresponding census tract values.
County Population No. of census
block groups
No. of census
block groups with
Vulnerability
Nos. and % of outlier
census block groups *
Apache 69,423 54 54 2 + 0 = 3.7
Cochise 117,755 72 72 23 + 2 = 34.7
Coconino 116,320 106 106 21 + 5 = 24.5
Gila 51,335 55 55 20 + 0 = 36.4
Graham 33,489 27 27 16 + 3 = 70.4
Greenlee 8,547 8 8 0, 0
La Paz 19,715 23 22 12 + 1 = 56.5
Maricopa 3,072,149 2,113 1,102 901 + 69 = 45.9
Mohave 155,032 101 101 18 + 0 = 17.8
Navajo 97,470 74 74 8 + 2 = 13.5
Pima 843,746 617 617 311 + 14 = 52.7
Pinal 179,727 116 113 71 + 8 = 68.1
Santa Cruz 38,381 20 20 17 + 0 = 85.0
Yavapai 167,517 86 85 7 + 2 = 3.7
Yuma 160,026 98 98 76 + 1 = 34.7
Totals 5,130,632 3,570 2,554 1,503 + 107 = 45.1
* No. of census block groups with Vulnerability <75%, >125%, and the sum of the two classes
of outliers as a percentage of total.
72
Figure 28. Group I, Percent Vulnerable Population in Maricopa County by: (a) census tract; and (b) census block group. Both pairs of
maps show the entire county on the top with the greater Phoenix Metro area highlighted and shown at a larger scale on the bottom.
73
Figure 29. Group I (continued), Percent Vulnerable Population in Pima County by: (a) census tract; and (b) census block group. Both
pairs of maps show the entire county on the top with the greater Tucson Metro area highlighted and shown at a larger scale on the
bottom.
74
Figure 30. Group II, Percent Vulnerable Population in (a) Coconino County by census tract; (b)
Coconino county by census block group; (c) Pinal County by census tract; (d) Pinal County by
census block group.
75
Figure 31. Group III, Percent Vulnerable Population in: (a) Apache County by census tract; (b)
Apache County by census block group; (c) Santa Cruz County by census tract; and (b) Santa
Cruz County by census block group.
76
Native American Population
Ethnicity is also a commonly studied health determinant as it is indicative of specific needs
within certain populations and/or spatial zones. The overall Native American population in
Arizona is relatively small yet significant, evident by the presence of 22 Federally recognized
AIRs that cover a large proportion of the State, as well as the strong cultural identity that their
presence represents in terms of Arizona’s history and heritage.
4.3.1. County Level
Figure 32 displays the county distribution of the State’s Native American population by
percentage. The two counties in red, denoting the counties with the highest Native American
populations, are counties where approximately half of their total area is AIRs. The counties in
blue and green also contain AIRs, however, apart from Pima County, the reservations tend to be
smaller in physical area and are more dispersed throughout these counties.
Figure 32. Native American Population by percent for each Arizona county.
77
4.3.2. Census Tract Level
The scatterplot in Figure 33 shows Native American census tract populations compared to
county population values. Census tracts are colored to highlight census tracts classified as
outliers (where < 75% of census tracts classified as outliers are depicted in blue, > 125% of
census tracts classified as outliers are depicted in red, and ± 25% of census tracts classified as
outliers are depicted in green) and show that numerous census tracts in multiple counties across
the state have Native American population of 50% or more. Eight of the 15 counties have one or
more census tracts with Native American census tract populations at or near 100%.
Figure 34 shows the census tract Native American population distribution relative to
county values across Arizona. When the spatial reporting unit is decreased from county to census
tract, large numbers of census tracts and therefore areas are classified as outliers. The numbers in
Table 6 confirm this result, showing that 50% or more of the census tracts in every county were
classified as outliers and 75% or more of the census tracts in 11 of 15 counties were classified as
outliers (i.e. with substantially smaller or larger Native American populations, than the
appropriate county as a whole).
Figure 33. Arizona Census Tract estimates within each county for Native American population
by percentage.
78
Figure 34. Native American Population by Percentage for each Arizona census tract (a) with separate maps showing the Phoenix
Metro (b), and Tucson Metro (c) areas.
(a)
(b)
(c)
79
Table 6. Counts and percentages of census tract Native American Population by percentage
examining outliers ≤ 75% and ≥ 125% of county median values.
County Population No. of census
tracts
No. of census tracts with
Native American
population
Nos. and % of
outlier
census tracts *
Apache 69,423 14 14 3 + 4 = 50.0
Cochise 117,755 21 21 3 + 11 = 66.7
Coconino 116,320 28 27 5 + 22 = 96.4
Gila 51,335 15 15 2 + 13 = 100.0
Graham 33,489 8 8 1 + 7 = 100.0
Greenlee 8,547 3 3 0 + 2 = 66.7
La Paz 19,715 6 6 2 + 4 = 100.0
Maricopa 3,072,149 663 656 71 + 476 = 82.5
Mohave 155,032 30 30 2 + 25 = 90.0
Navajo 97,470 23 23 11 + 12 = 100.0
Pima 843,746 198 198 10 + 170 = 90.9
Pinal 179,727 33 31 4 + 28 = 97.0
Santa Cruz 38,381 7 7 0 + 5 = 71.4
Yavapai 167,517 26 25 4 + 18 = 84.6
Yuma 160,026 33 32 2 + 23 = 75.8
Totals 5,130,632 1,108 1,096 120 + 820 = 84.8
* No. of census tracts with Native American Population by % < 75%, > 125%, and the sum
of the two classes of outliers as a percentage of total.
4.3.3. Census Block Group Level
The scatterplots in Figures 35-39 show the Native American census block group population
percentages relative to the corresponding census tract values, with colors showing the same
outlier groupings as earlier. The scatterplots show how the distribution of the Native American
population numbers vary across the state when different spatial reporting units are chosen.
The scatterplots for Group I (Figures 35 and 36) show that although these counties are the
two highest populated counties in the State, only three census tracts in Maricopa County had
Native American populations of > 25%. Table 6, however, indicates that that the number of
census block groups classified as outliers were relatively high in both Maricopa (69.1%) and
Pima (63%) counties.
80
Figure 35. Group I: Census block group percentage Native American population by census tract for Maricopa County: (a) census
tracts 1-333; and (b) census tracts 334-663.
81
Figure 36. Group I (Continued): Census block group percentage Native American population by census tract in Pima County.
82
The scatterplots in Figures 37, 38 and 39 show several of the counties with medium-sized
and small populations have one or more census block groups with large Native American
populations (see Figures 37a, 38f and g, and Figures 39a, b, and c for examples) and that the
switch from census tracts to census block groups highlights numerous outliers, or census block
groups with relatively high or low Native American populations relative to the corresponding
census tract estimates.
The three counties with the most census block groups classified as outliers were La Paz
(73.9%), Gila (74.5%), and Graham (81.5%) and those with the lowest number of census block
groups classified as outliers were Apache (14.8%), Navajo (36.5%), and Greenlee (50.0%). All
but Navajo County are in Group III, the group with the smallest populations across the State.
Navajo County is the least populated county in Group II.
Table 7 indicates that there were fewer outliers going from census tract to census block
group (65.9%) than there was going from county to census tract (84.8%) (Table 6) for this
variable. The total number of census block groups classified as outliers is still substantial
compared to the first two variables examined when switching the geographic unit from census
tract to census block group: Median Household Income, Table 3 (24.8%) and Vulnerability,
Table 5 (45.1%).
The maps in Figures 40-43 visually demonstrate the outliers when switching the
geographic reporting units from census tract to census block group and how the patterns vary by
county, population, and AIRs. The Group I counties, Figures 40 and 41, exhibit an increase in
variability at the census block group level and the inset maps show that the outliers are especially
evident in the metropolitan areas in both counties. The Group II counties, Figure 42, demonstrate
similar patterns switching from census tract to census block group.
83
Figure 37. Group II: Census block group percentage Native American Population compared to census tract percentage Native
American Population for: (a) Pinal; (b) Yavapai; (c) Yuma; (d) Mohave.
84
Figure 38. Group II (Continued): Census block group percentage Native American Population compared to census tract percentage
Native American Population for: (e) Cochise; (f) Coconino; and (g) Navajo counties.
85
Figure 39. Group III: Census block group Native American population by percentage compared to census tract Native American
population by percentage for: (a) Apache; (b) Gila; (c) Santa Cruz; (d) Graham; (e) La Paz; and (f) Greenlee Counties.
86
The Group III counties in Figures 43, visually demonstrate the modest numbers of census block
groups classified as outliers compared to the more populated counties.
At the census tract level, there is little diversity within reporting zones, as seen in Figures
40-43, and there are sizable percentages of census tracts classified as outliers (Table 6). As the
reporting unit becomes smaller, variability becomes evident, especially in and around AIRS,
throughout Metro areas and economic centers.
Table 7. Counts and percentages of census block group Native American population by % ≤
75% and ≥ 125% of census tract values.
County Population No. of census
block groups
No. of census block
groups with Native
American population
Nos. and % of
outlier census block
groups *
Apache 69,423 54 54 3 + 5 = 14.8
Cochise 117,755 72 71 17 + 33 = 69.4
Coconino 116,320 106 102 25 + 41 = 62.3
Gila 51,335 55 51 19 + 22 = 74.5
Graham 33,489 27 26 9 + 13 = 81.5
Greenlee 8,547 8 8 2 + 2 = 50.0
La Paz 19,715 23 22 8 + 9 = 73.9
Maricopa 3,072,149 2,113 1,915 594 + 867 = 69.1
Mohave 155,032 101 100 16 + 43 = 58.4
Navajo 97,470 74 74 15 + 12 = 36.5
Pima 843,746 617 578 140 + 249 = 63.0
Pinal 179,727 116 104 27 + 49 = 65.5
Santa Cruz 38,381 20 15 1 + 10 = 55.0
Yavapai 167,517 86 83 19 + 34 = 61.6
Yuma 160,026 98 92 23 + 45 = 69.4
Totals 5,130,632 3,570 3,295 918 + 1,434 = 65.9
* No. of census block groups with Native American population <75%, >125% of the
corresponding census tract value, and the sum of the two classes of outliers as a percentage of
total.
87
Figure 40. Group I, Percent Native American in Maricopa County by: (a) census tract; and (b) census block group. Both pairs show
the entire county on the top with the greater Phoenix Metro highlighted and shown at a larger scale on the bottom.
88
Figure 41. Group I (Continued), percent Native American in Pima County by: (a) census tract; and (b) census block group. Both pairs
of maps show the entire county on the top with the greater Tucson Metro highlighted and shown at a larger scale on the bottom.
89
Figure 42. Group II, Percent Native American population in: (a) Coconino County by census
tract; (b) Coconino County by census block group; (c) Pinal County by census tract; and (d)
Pinal County by census block group.
90
Figure 43. Group III: Percent Native American population in: (a) Apache County by census
tract; (b) Apache County by census block group; (c) Santa Cruz County by census tract; and (d)
Santa Cruz County by census block group.
91
Chapter 5 : Discussion and Conclusions
County, Census Tract, and Census Block Group estimates were compared for three commonly
used social determinants of health. County estimates are often used by researchers and officials
to broadly describe population health. This study tested the efficacy of this supposition.
The findings show that county areas, whether small or large in geographic extent and
with rural and/or metropolitan populations, provide very generalized population descriptions and
as such subject the resulting assumptions and/or the statistical interpretations to both MAUP
affects and ecological fallacy. Rurality, often identified by geographic population density, can
also be denoted by perception-based attributes such as regional lifestyle choices, and also
cultural factors. Geographic areas considered rural within Maricopa and Pima Counties may not
experience the geographic isolation found in some of the low population density counties such as
those in Groups II and III; however, the residents may still experience limited infrastructure and
are equally subject to ecological bias introduced by aggregation issues. Rurality is therefore
found to affect resulting research outcomes in the gamut of rural settings as inconsistencies in
neighborhood definition subject data to potential misreporting error.
When social determinants of health are examined at the census tract and census block
group levels, nuances within larger geographic areas of the county begin to appear. Depending
on the social determinant of health under examination and the geographic reporting unit used,
this variability can sometimes be considerable. This variability occurs regardless of what data
compilation method is utilized and proved present in the 2000 Decennial Census, as
demonstrated in Chapter 4, and the 2010-2015 ACS data, as demonstrated in Table 8 below.
The total number of geographic reporting units changed between the 2000 Decennial
Census and the 2010-2015 ACS creating not only potential MAUP effects but also congruency
92
issues. Table 8 compares data between the two reporting’s for three counties, one from each of
the aforementioned groupings: Maricopa (Group I); Coconino (Group II); and Apache (Group
III) Counties. Census block groups classified as outliers increased in two-thirds of the
comparisons with the most significant change being in Maricopa County where the Median
Household Income estimates increased from 24% in the 2000 Decennial Census to 63.6% in the
2010-2015 ACS.
The 2000 Decennial Census was used for this project; however, temporal issues have
consequences as this static portrait of population means that the estimates that were derived are
somewhat dated. Nonetheless, the ability to use this Census reporting was extremely beneficial
in being able to conduct the spatial unit investigations and comparisons to evaluate if population
reporting techniques might affect research outcomes.
Table 8. Specific county variability comparisons at Census Block Group Level between
decennial census data and ACS data.
(a) 2000 Decennial Census
County No. of census
block groups
Median Household
Income - Estimates
<75% or ≥125% of
census tract values
Vulnerability
comparison - Estimates
<75% or ≥125% of
census tract values
Native American
Population - Estimates
<75% or ≥125% of
census tract values
Apache 54 46.3 3.7 14.8
Coconino 106 34 24.5 62.3
Maricopa 2,113 24 45.9 69.1
(b) 2010-2015 ACS
County No. of census
block groups
Median Household
Income - Estimates
<75% or ≥125% of
census tract values
Vulnerability
comparison - Estimates
<75% or ≥125% of
census tract values
Native American
Population - Estimates
<75% or ≥125% of
census tract values
Apache 55 27.3 7.3 18.2
Coconino 98 46.9 32.7 62.2
Maricopa 2,505 63.6 19.5 85.3
93
The thesis project used three separate means of visualizing variability across the spatial
units studied. Each method built on the other; however, all contributed to the variability
assessment in the following ways.
Threshold values were used to identify the places where the social determinant of health
estimates varied by ≥ 25% from the previous estimate at the next highest level of aggregation
(i.e. county for census tract and census tract for census block group). This means of visualizing
areas with little or no change (green ± 25% of median values) and ≥ 25% change (blue
represented as median values classified as outliers < 75% and red represented as median values
classified as outliers > 125%) across different geographic reporting units were used on all maps
and scatterplots to unlock spatial patterns that might otherwise not be apparent. The choice of
this approach was used to highlight the sensitivity of the social determinants of health estimates
to the choice of geographic reporting units across the State of Arizona.
The scatterplots graphs were essential for analyzing the actual values behind the map
visualizations. The scatterplots provided intricate insights as to how many spatial units fell inside
or outside the previously noted thresholds. The color coding of spatial units per threshold bracket
helped as a visual aid to assess the range of variability documented within this thesis. The
scatterplots along with the maps revealed the true variability present within Arizona and
provided story lines that might otherwise not have been evident when looking at a table of values
or a map by itself.
The maps, on the other hand, provide intricate insights into the locations across the state
where the choice of geographic reporting unit fell inside or outside the previously noted
thresholds. This information is helpful in showing these parts of the state where social
94
determinants of health vary over short distances as well as places where similar characteristics
cover large areas.
This study has demonstrated that in general, the variability in social determinants of
health increased as the number of spatial units increased. As spatial units were examined at
differing sizes, and units over a given area were increased by spatial partitioning, the percentages
of census units classified as outliers outside of a median value range also changed as indicated in
the summary tables (Tables 2 - 8) throughout Chapters 4 and 5. This finding suggests that the
null hypothesis should not be accepted since the results showed that social determinants of health
estimates changed and affected research results as smaller (census tracts) and smaller (census
block groups) geographic reporting units were used in place of counties.
This thesis therefore sets a basis for future work to continue in the exploration of how
choice of neighborhood effects research outcomes in health studies. Further investigation into
neighborhood effects might include study into what social determinants of health are the most
likely to show the greatest variability across geographic reporting units such as those across
Native American Indian Areas. Additionally, of interest is if and how the independent social
determinant of health variables as selected for this thesis, might be connected and if there is any
overlap of spatial patterning within areas classified as outliers.
Future work might also use this project as a basis to explore how proximity buffers are
affected by the choice of geographic reporting units for acquisition of census data. Specifically,
how those buffers might report what population has potential exposure to a pollution source, e.g.
a target population within a certain distance to a major roadway or gas and oil well.
Environmental exposure assessments are frequently dependent on proximity factors which are
directly influenced by choice of neighborhood making further investigation on how
95
neighborhood is defined for health research of vital importance for health researchers and
professionals.
96
References
Adams, Paul C., Steven Hoelscher, and Karen E. Till. 2001. Textures of Place : Exploring
Humanist Geographies. NED - New edition ed. Minneapolis: University of Minnesota
Press. doi:10.5749/j.cttttg77. http://lib.myilibrary.com?ID=522749.
Apparicio, Philippe, Jeremy Gelb, Anne-Sophie Dube, Simon Kingham, Lise Gauvin, and Eric
Robitaille. 2017. "The Approaches to Measuring the Potential Spatial Access to Urban
Health Services Revisited: Distance Types and Aggregation-Error Issues." International
Journal of Health Geographics 16. doi:10.1186/s12942-017-0105-9.
Aylin, Paul, Ravi Maheswaran, Jon Wakefield, Samantha Cockings, Lars Jarup, Richard Arnold,
Gideon Wheeler, and Paul Elliott. 1999. “A National Facility for Small Area Disease
Mapping and Rapid Initial Assessment of Apparent Disease Clusters Around a Point
Source: The UK Small Area Health Statistics Unit.” Journal of Public Health Medicine
21 (3): 289-298. doi:10.1093/pubmed/21.3.289.
http://www.ncbi.nlm.nih.gov/pubmed/10528956.
Barrozo, Ligia Vizeu, Reinaldo Paul Pérez-Machado, Christopher Small, and William Cabral-
Miranda. 2016. "Changing Spatial Perception: Dasymetric Mapping to Improve Analysis
of Health Outcomes in a Megacity." Journal of Maps 12 (5): 1242-1247.
doi:10.1080/17445647.2015.1101403.
Bell, B. Sue., Richard Hoskins, Linda Pickle, and Daniel Wartenberg. 2006. "Current Practices
in Spatial Analysis of Cancer Data: Mapping Health Statistics to Inform Policymakers
and the Public." International Journal of Health Geographics 5 (1): 49.
doi:10.1186/1476-072X-5-49.
Beyer, Kirsten M. M., Chetan Tiwari, and Gerard Rushton. 2012. "Five Essential Properties of
Disease Maps." Annals of the Association of American Geographers 102 (5): 1067-1075.
doi:10.1080/00045608.2012.659940.
http://www.tandfonline.com/doi/abs/10.1080/00045608.2012.659940.
Bithell, J. F. 2000. "A classification of disease mapping methods." Statistics in Medicine 19, no.
17‐18 (2000): 2203-2215.
Bolstad, Paul. 2012. GIS Fundamentals: A First Text on Geographic Information Systems. White
Bear Lake, Minn.: White Bear Lake, Minn.: Eider Press.
Boruff, Bryan, Andrea Nathan, and Sandra Nijënstein. 2012. "Using GPS Technology to (Re)-
Examine Operational Definitions of 'Neighbourhood' in Place-Based Health Research."
International Journal of Health Geographics 11: 22. doi:10.1186/1476-072X-11-22.
Boscoe, Francis P. and Linda W. Pickle. 2003. "Choosing Geographic Units for Choropleth Rate
Maps, with an Emphasis on Public Health Applications." Cartography and Geographic
Information Science 30 (3): 237-248. doi:10.1559/152304003100011171.
97
Bravo, Mercedes A., Rebecca Anthopolos, Michelle L. Bell, and Marie Lynn Miranda. 2016.
"Racial Isolation and Exposure to Airborne Particulate Matter and Ozone in Understudied
US Populations: Environmental Justice Applications of Downscaled Numerical Model
Output." Environment International 92-93: 247-255. doi:10.1016/j.envint.2016.04.008.
Chaix, Basile, Juan Merlo, and Pierre Chauvin. 2005. "Comparison of a Spatial Approach with
the Multilevel Approach for Investigating Place Effects on Health: The Example of
Healthcare Utilisation in France." Journal of Epidemiology and Community Health 59
(6): 517. doi:10.1136/jech.2004.025478.
Chaix, Basile, Juan Merlo, David Evans, Cinira Leal, and Sabrina Havard. 2009.
"Neighbourhoods in Eco-Epidemiologic Research: Delimiting Personal Exposure Areas.
A Response to Riva, Gauvin, Apparicio and Brodeur." Social Science & Medicine 69 (9):
1306-1310. doi:10.1016/j.socscimed.2009.07.018.
Chang, Bowon. 2010. The power of geographical boundaries: Cultural, political, and economic
border effects in a unitary nation. Master’s thesis. Iowa State University.
http://lib.dr.iastate.edu/etd/11344
Charreire, Hélène, Romain Casey, Paul Salze, Chantal Simon, Basile Chaix, Arnaud Banos,
Dominique Badariotti, Christiane Weber, and Jean-Michel Oppert. 2010. "Measuring the
Food Environment using Geographical Information Systems: A Methodological Review."
Public Health Nutrition 13 (11): 1773-1785. doi:10.1017/S1368980010000753.
http://journals.cambridge.org/abstract_S1368980010000753.
Chen, Xiang and Mei-Po Kwan. 2015. "Contextual Uncertainties, Human Mobility, and
Perceived Food Environment: The Uncertain Geographic Context Problem in Food
Access Research." American Journal of Public Health 105 (9): 1734.
doi:10.2105/AJPH.2015.302792.
Cruz, Hildemar. 2016. A Geospatial Analysis of Income Level, Food Deserts and Urban
Agriculture Hot Spots. Master’s thesis. University of Southern California
Cutchin, Malcolm P., Karl Eschbach, Christine A. Mair, Hyunsu Ju, and James S. Goodwin.
2011. "The Socio-Spatial Neighborhood Estimation Method: An Approach to
Operationalizing the Neighborhood Concept." Health and Place 17 (5): 1113-1121.
doi:10.1016/j.healthplace.2011.05.011.
Delmelle, Elizabeth C. 2016. “Mapping the DNA of Urban Neighborhoods: Clustering
Longitudinal Sequences of Neighborhood Socioeconomic Change.” Annals of the
American Association of Geographers 106, no. 1 (n.d.): 36–56.
Diez Roux, Ana V. 2001. "Investigating Neighborhood and Area Effects on Health." American
Journal of Public Health 91 (11): 1783-1789. doi:10.2105/AJPH.91.11.1783.
http://ajph.aphapublications.org/cgi/content/abstract/91/11/1783.
Diez Roux, Ana. V. 2007. “Neighborhoods and health: where are we and were do we go from
here?” Reviews in Epidemiology Sante Publique, 55(1), 13–21.
98
Duncan, Dustin T., Ichiro Kawachi, S. V. Subramanian, Jared Aldstadt, Steven J. Melly, and
David R. Williams. 2014. "Examination of how Neighborhood Definition Influences
Measurements of Youths' Access to Tobacco Retailers: A Methodological Note on
Spatial Misclassification." American Journal of Epidemiology 179 (3): 373-381.
doi:10.1093/aje/kwt251.
Eagleson, Serryn, 2002. "Hierarchical Spatial Reasoning Theory and GIS Technology Applied to
the Automated Delineation of Administrative Boundaries." Computers, Environment and
Urban Systems 26 (2): 185-200. doi:10.1016/S0198-9715(01)00040-0.
Eicher, Cory L. and Cynthia A. Brewer. 2001. "Dasymetric Mapping and Areal Interpolation:
Implementation and Evaluation." Cartography and Geographic Information Science 28
(2): 125-138. doi:10.1559/152304001782173727.
Elliott, Paul and Daniel Wartenberg. 2004. "Spatial Epidemiology: Current Approaches and
Future Challenges.(Information Systems: Mini-Monograph)." Environmental Health
Perspectives 112 (9): 998. doi:10.1289/ehp.6735.
Faber, Jacob W., and Sharkey, Patrick. 2015. “Neighborhood Effects.” International
Encyclopedia of the Social & Behavioral Sciences, 443–449.
https://doi.org/10.1016/B978-0-08-097086-8.32189-4
Flint, Ellen, Steven Cummins, and Stephen Matthews. 2013. "Do Perceptions of the
Neighbourhood Food Environment Predict Fruit and Vegetable Intake in Low-Income
Neighbourhoods?" Health & Place 24: 11-15. doi:10.1016/j.healthplace.2013.07.005.
http://www.ncbi.nlm.nih.gov/pubmed/23999578.
Flowerdew, Robin, David J. Manley, and Clive E. Sabel. 2008. "Neighbourhood Effects on
Health: Does it Matter Where You Draw the Boundaries?" Social Science & Medicine 66
(6): 1241-1255. doi:10.1016/j.socscimed.2007.11.042 [doi].
Foster, Kirk A. and J. Aaron Hipp. 2011. "Defining Neighborhood Boundaries for Social
Measurement: Advancing Social Work Research." Social Work Research 35 (1): 25-35.
Goodchild, Michael F. 2018. “Reimagining the history of GIS”, Annals of GIS, 24:1, 1-8, DOI:
10.1080/19475683.2018.1424737
Hall Alexander, W., Schuurman Nadine, and N. Oliver Lisa. 2007. "Comparing Circular and
Network Buffers to Examine the Influence of Land use on Walking for Leisure and
Errands." International Journal of Health Geographics 6 (1): 41. doi:10.1186/1476-
072X-6-41.
Hipp, J. Aaron and N. Chalise. 2015. "Spatial Analysis and Correlates of County-Level Diabetes
Prevalence, 2009-2010." Preventing Chronic Disease 12: E08.
Holt, James B., C. P. Lo, and Thomas W. Hodler. 2004. "Dasymetric Estimation of Population
Density and Areal Interpolation of Census Data." Cartography and Geographic
Information Science 31 (2): 103.
99
Institute of Education Sciences, National Center for Education Statistics. 2015. “Education
Demographic and Geographic Estimates (EDGE) Program; Locale Boundaries”
Retrieved from
https://nces.ed.gov/programs/edge/docs/EDGE_NCES_LOCALE_2015.pdf
Institute of Education Sciences, National Center for Education Statistics. 2018a. Geographic
Data. Retrieved from https://nces.ed.gov/programs/edge/geographic.aspx
Institute of Education Sciences, National Center for Education Statistics. 2018b. School
Attendance Boundary Survey; Frequently Asked Questions. Retrieved from
https://nces.ed.gov/programs/sabs/sabsFaq.aspx
Institute of Education Sciences, National Center for Education Statistics. 2018c. School District
Geographic Relationship Files. Retrieved from
https://nces.ed.gov/programs/edge/geographicRelationshipFiles.aspx
Jacquez, Geoffrey. 2014. "Spatial Clustering and Autocorrelation in Health Events." In
Handbook of Regional Science, 1311-334. Springer Berlin Heidelberg. doi: 10.1007/978-
3-642-23430-9_80
Jankowska, Marta M., Jasper Schipperijn, and Jacqueline Kerr. 2014. "A Framework for using
GPS Data in Physical Activity and Sedentary Behavior Studies." Exercise and Sport
Sciences Reviews 43 (1): 48.
Jankowska, Marta M, Natarajan, Loki, Godbole, Suneeta, Meseck, Kristin, Sears, Dorothy D,
Patterson, Ruth E, and Kerr, Jacqueline. 2017. “Kernel Density Estimation as a Measure
of Environmental Exposure Related to Insulin Resistance in Breast Cancer Survivors.”
Cancer epidemiology, biomarkers & prevention : a publication of the American
Association for Cancer Research, cosponsored by the American Society of Preventive
Oncology 26, no. 7 (2017): 1078–1084.
Kim, Dohyeong, M. A. Galeano, Andrew Hull, and Marie Lynn Miranda. 2008. "A Framework
for Widespread Replication of a Highly Spatially Resolved Childhood Lead Exposure
Risk Model.(Children's Health)." Environmental Health Perspectives 116 (12): 1735.
doi:10.1289/ehp.11540.
Kirby, Russell S., Eric Delmelle, and Jan M. Eberth. 2017. "Advances in Spatial Epidemiology
and Geographic Information Systems." Annals of Epidemiology 27 (1): 1-9.
doi:10.1016/j.annepidem.2016.12.001.
Kwan, Mei-Po. 2012a. "Geographies of Health." Annals of the Association of American
Geographers 102 (5): 891-892. doi:10.1080/00045608.2012.687348.
Kwan, Mei-Po. 2012b. "The Uncertain Geographic Context Problem." Annals of the Association
of American Geographers 102 (5): 958-968. doi:10.1080/00045608.2012.687349.
100
Lam, Nina Siu-Ngan. 2012. "Geospatial Methods for Reducing Uncertainties in Environmental
Health Risk Assessment: Challenges and Opportunities." Annals of the Association of
American Geographers 102 (5): 942-950. doi:10.1080/00045608.2012.674900.
http://www.tandfonline.com/doi/abs/10.1080/00045608.2012.674900.
Leite, Maria Jalila Vieira de Figueirêdo, Aldo Dantas, and Angelo Giuseppe Roncalli. 2015.
"Epidemiology and Place: A Spatial Area for Health Analysis." Holos 8: 357-367.
doi:10.15628/holos.2015.2854.
https://doaj.org/article/576df02d761c4fe390c136ffc31e1b21
Macintyre, Sally, Ellaway, Anne, and Cummins, Steven. 2002. “Place Effects on Health: How
Can We Conceptualise, Operationalise and Measure Them?” Social Science & Medicine
55, no. 1: 125–139. doi://doi-org.libproxy1.usc.edu/10.1016/S0277-9536(01)00214-3.
http://www.sciencedirect.com.libproxy1.usc.edu/science/article/pii/S0277953601002143.
Martin, David, Samantha Cockings, and Samuel Leung. 2015. "Developing a Flexible
Framework for Spatiotemporal Population Modeling." Annals of the Association of
American Geographers, 2015, 1-1.
Matthews Stephen A. 2011. “Spatial Polygamy and the Heterogeneity of Place: Studying People
and Place via Egocentric Methods.” In Burton, L., Matthews, S., Leung, M., Kemp, S., &
Takeuchi, D. Communities, Neighborhoods, and Health Expanding the Boundaries of
Place. New York, NY: Springer New York. https://doi.org/10.1007/978-1-4419-7482-2
Matthews, Stephen A., 2018. “Place and the Goldilocks Principle: Theory, Measurement and the
Relevance of Places in Population and Health.” Lecture. Spatial Sciences Institute of
University of Southern California, February 22
Matthews, Stephen A. and Daniel M. Parker. 2013. "Progress in Spatial Demography."
Demographic Research 28: 10.
Matthews, Stephen A, and Tse-Chuan, Yang. 2013. "Spatial Polygamy and Contextual
Exposures (SPACEs)." American Behavioral Scientist 57, no. 8: 1057-081.
Mennis, Jeremy. 2003. "Generating Surface Models of Population using Dasymetric Mapping."
The Professional Geographer 55 (1): 31-42. doi:10.1111/0033-0124.10042.
Mennis, Jeremy and Torrin Hultgren. 2006. "Intelligent Dasymetric Mapping and its Application
to Areal Interpolation." Cartography and Geographic Information Science 33 (3): 179-
194. doi:10.1559/152304006779077309.
Miranda, Marie, Rebecca Anthopolos, and Douglas Hastings. 2011. "A Geospatial Analysis of
the Effects of Aviation Gasoline on Childhood Blood Lead Levels." Environmental
Health Perspectives 119 (10): 1513-6. doi:10.1289/ehp.1003231.
Mu, Lan, Fahui Wang, Vivien W. Chen, and Xiao-Cheng Wu. 2015. "A Place-Oriented, Mixed-
Level Regionalization Method for Constructing Geographic Areas in Health Data
Dissemination and Analysis." Annals of the Association of American Geographers 105
101
(1): 48-66. doi:10.1080/00045608.2014.968910.
http://www.tandfonline.com/doi/abs/10.1080/00045608.2014.968910.
Mujahid, Mahasin S., Ana V. Diez Roux, Jeffrey D. Morenoff, and Trivellore Raghunathan.
2007. "Assessing the Measurement Properties of Neighborhood Scales: From
Psychometrics to Ecometrics." American Journal of Epidemiology 165 (8): 858-867.
doi:10.1093/aje/kwm040. http://www.ncbi.nlm.nih.gov/pubmed/17329713.
Nagle, Nicholas N., Barbara P. Buttenfield, Stefan Leyk, and Seth Spielman. 2013. "Dasymetric
Modeling and Uncertainty." Annals of the Association of American Geographers.
doi:10.1080/00045608.2013.843439.
Nelson, Elisabeth S. 2010. "Dasymetric Maps." Encyclopedia of Geography, edited by Barney
Warf, 670-671. Thousand Oaks, CA: SAGE Publications, Inc., 2010. doi:
10.4135/9781412939591.n252.
Onubogu, Carl. 2013. Size and Shape Matter: The Modifiable Areal Unit Problem in
Neighborhood Effects Research. ProQuest Dissertations and Theses.
Openshaw, Stan and Seraphim Alvanides. 2001. "Designing Zoning Systems for Representation
of Socio-Economic Data." Time and Motion of Socio-Economic Units. Taylor and
Francis, London.
Perchoux, Camille, Basile Chaix, Ruben Brondeel, and Yan Kestens. 2016. "Residential Buffer,
Perceived Neighborhood, and Individual Activity Space: New Refinements in the
Definition of Exposure Areas - the RECORD Cohort Study." Health and Place 40: 116-
122. doi:10.1016/j.healthplace.2016.05.004.
https://www.sciencedirect.com/science/article/pii/S1353829216300533.
Perchoux, Camille, Basile Chaix, Steven Cummins, and Yan Kestens. 2013. "Conceptualization
and Measurement of Environmental Exposure in Epidemiology: Accounting for Activity
Space Related to Daily Mobility." Health & Place 21: 86.
doi:10.1016/j.healthplace.2013.01.005.
Robertson, Colin, and Feick, Rob. 2018.“Inference and Analysis Across Spatial Supports in the
Big Data Era: Uncertain Point Observations and Geographic Contexts.” Transactions in
GIS 22, no. 2 (n.d.): 455–476. https://doi.org/10.1111/tgis.12321
Root, Elisabeth Dowling. 2012. "Moving Neighborhoods and Health Research Forward: Using
Geographic Methods to Examine the Role of Spatial Scale in Neighborhood Effects on
Health." Annals of the Association of American Geographers 102 (5): 986-995.
doi:10.1080/00045608.2012.659621.
Rousseau, Nicolette. 1995. “What is rurality?” Occasional Paper (Royal College of General
Practitioners), 71, 1–4.
Rushton, Gerard. 2003. "Public Health, GIS, and Spatial Analytic Tools." Annual Review of
Public Health 24: 43.
102
Sabel, Clive E., W Kihal, D Bard, and Christiane Weber. 2013. "Creation of Synthetic
Homogeneous Neighbourhoods using Zone Design Algorithms to Explore Relationships
between Asthma and Deprivation in Strasbourg, France." Social Science & Medicine 91:
110-121. doi:10.1016/j.socscimed.2012.11.018.
http://www.ncbi.nlm.nih.gov/pubmed/23332654.
Santos, Simone M., Dora Chor, and Guilherme Loureiro Werneck. 2010. "Demarcation of Local
Neighborhoods to Study Relations between Contextual Factors and Health."
International Journal of Health Geographics 9 (1): 34. doi:10.1186/1476-072X-9-34.
Schüle, Steffen Andreas and Gabriele Bolte. 2015. "Interactive and Independent Associations
between the Socioeconomic and Objective Built Environment on the Neighbourhood
Level and Individual Health: A Systematic Review of Multilevel Studies." PloS One 10
(4): e0123456. doi:10.1371/journal.pone.0123456.
Sharkey, Patrick and Jacob W. Faber. 2014. "Where, when, Why, and for Whom do Residential
Contexts Matter? Moving Away from the Dichotomous Understanding of Neighborhood
Effects." Annual Review of Sociology 40: 559-579. doi:10.1146/annurev-soc-071913-
043350.
Smith, George Davey. 2002. “Commentary: Behind the broad street pump: Aetiology,
epidemiology and prevention of cholera in mid-19th century Britain.” International
Journal of Epidemiology 31 (5): 920-32.
Spielman, Seth E., and Logan, John R. 2012. “Using High-Resolution Population Data to
Identify Neighborhoods and Establish Their Boundaries.” Annals of the Association of
American Geographers 103, no. 1 (n.d.): 67–84. doi:10.1080/00045608.2012.685049.
Stafford, Mai, Oliver Duke-Williams, and Nicola Shelton. 2008. "Small Area Inequalities in
Health: Are we Underestimating them?" Social Science & Medicine 67 (6): 891-899.
doi:10.1016/j.socscimed.2008.05.028.
Strominger, Julie, Rebecca Anthopolos, and Marie Lynn Miranda. 2016. "Implications of
Construction Method and Spatial Scale on Measures of the Built Environment."
International Journal of Health Geographics 15: 15. doi:10.1186/s12942-016-0044-x.
Swift, Andrew, Lin Liu, and James Uber. 2008. "Reducing MAUP Bias of Correlation Statistics
between Water Quality and GI Illness." Computers, Environment and Urban Systems 32
(2): 134-148. doi:10.1016/j.compenvurbsys.2008.01.002.
https://www.sciencedirect.com/science/article/pii/D0198971508000033.
Swift, Andrew, Lin Liu, and James Uber. 2014. "MAUP Sensitivity Analysis of Ecological Bias
in Health Studies." GeoJournal 79 (2): 137-153. doi:10.1007/s10708-013-9504-z.
Tatalovich, Zaria, John P. Wilson, Joel E. Milam, Michael L. B. Jerrett, and Rob McConnell.
2006. "Competing Definitions of Contextual Environments." International Journal of
Health Geographics 5 (1): 55. doi:10.1186/1476-072X-5-55.
https://www.ncbi.nlm.nih.gov/pubmed/17156433.
103
Tobler, Waldo. R. 1970. “A computer movie simulating urban growth in the Detroit region.”
Economic Geography 46 : 234-40.
U. S. Census Bureau. 1990. Geographic Areas Reference Manual. Retrieved from
https://www.census.gov/geo/reference/garm.html
U.S. Census Bureau. 2010a. Geographic Terms and Concepts - American Indian, Alaska Native,
and Native Hawaiian Areas. Retrieved from
https://www.census.gov/geo/reference/gtc/gtc_aiannha.html
U.S. Census Bureau. 2010b. Overview of the 2010 Census Tribal Statistical Areas Program
(TSAP). Retrieved from
https://www.census.gov/geo/partnerships/tsap2010_overview.html
U.S. Census Bureau. 2010c. What is the Census. Retrieved from
https://www.census.gov/2010census/about/
U.S. Census Bureau. 2014. American Community Survey Design and Methodology (January
2014). Retrieved from https://www.census.gov/history/pdf/acsdesign-
methodology2014.pdf
U.S. Census Bureau. 2015. 2010 Census Tallies of Census Tracts, Block Groups & Blocks.
Retrieved from https://www.census.gov/geo/maps-data/data/tallies/tractblock.html
U.S. Census Bureau. 2016. Substantial Changes to Counties and County Equivalent Entities:
1970-Present. Retrieved at https://www.census.gov/geo/reference/county-changes.html
U.S. Census Bureau. 2017. Boundary and Annexation Survey (BAS). Retrieved at
https://www.census.gov/programs -surveys/bas.html
U.S. Census Bureau. 2018a. American Community Survey. Retrieved from
https://www.census.gov/history/www/programs/demographic/american_community_surv
ey.html
U. S. Census Bureau. 2018b. School District Boundaries. Retrieved from
https://www.census.gov/did/www/schooldistricts/data/boundaries.html
U. S. Census Bureau. 2018c. Zip Code Tabulation Areas. Retrieved from
https://www.census.gov/geo/reference/zctas.html
U.S. Census Bureau. 2018d. Population Estimate Table. Retrieved from
https://www.census.gov/quickfacts/AZ
U.S. Department of Homeland Security. 2017. Indian Reservations. Retrieved at
https://catalog.data.gov/dataset/indian-reservations
U.S. Environmental Protection Agency. 2016. 2016 American Indian/Alaska Native/Native
Hawaiian Areas (AIANNH) Michigan, Minnesota, and Wisconsin. Retrieved from
104
https://catalog.data.gov/dataset/2016-american-indian-alaska-native-native-hawaiian-
areas-aiannh-michigan-minnesota-and-wisconsin
Waller, Lance A., and Gotway, Carol A. 2004. Applied Spatial Statistics for Public Health Data.
Hoboken: John Wiley & Sons, Incorporated. Accessed January 21, 2018. ProQuest
Ebook Central.
Xia, Gangqiang, Marie Lynn Miranda, and Alan E. Gelfand. 2006. "Approximately Optimal
Spatial Design Approaches for Environmental Health Data." Environmetrics 17 (4): 363-
385. doi:10.1002/env.775.
Yiannakoulias, Nikolaos. 2011. "Spatial Aberration vs. Geographical Substance: Representing
Place in Public Health Surveillance." Health and Place 17 (6): 1242-1248.
doi:10.1016/j.healthplace.2011.07.009.
Zandbergen, Paul A. and Drew A. Ignizio. 2010. "Comparison of Dasymetric Mapping
Techniques for Small-Area Population Estimates." Cartography and Geographic
Information Science 37 (3): 199-214. doi:10.1559/152304010792194985.
Abstract (if available)
Abstract
Defining place in health studies has been a crux for researchers as the definition of neighborhood is often regarded as adaptable to study needs and/or the preferences of the researcher. Health researchers commonly rely on measures of neighborhood that default to any number of predefined spatial administrative units, providing a relatively quick and cost-effective means to accessing and categorizing population data within a geographic area of interest. This approach to inferring population statistics assumes that median values for variables are relatively evenly disbursed across specific geographic areas of varying sizes. ❧ This thesis explores how research outcomes may be affected by the choice of geographic reporting zones. The primary research goal of this study was to compare geographic reporting zones within the State of Arizona and to determine how the choice of neighborhood would influence the resulting values for three commonly utilized social determinants of health
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Evaluating the MAUP scale effects on property crime in San Francisco, California
PDF
Integrating spatial visualization to improve public health understanding and communication
PDF
Investigating the association of historical preservation and neighborhood status in Detroit, 1970-2015
PDF
Modeling patient access to point-of-care diagnostic resources in a healthcare small-world network in rural Isaan, Thailand
PDF
Estimating populations at risk in data-poor environments: a geographically disaggregated analysis of Boko Haram terrorism 2009-2014
PDF
Incomplete data & insufficient methods: transgender population health research in the US
PDF
An exploration of the spatiotemporal distribution of snow crab (Chionoecetes opilio) in the eastern Bering Sea: 1982 – 2018
PDF
Selection of bridge location over the Merrimack River in southern New Hampshire: a comparison of site suitability assessments
PDF
Spatial analysis of urban built environments and vehicle transit behavior
PDF
Spatio-temporal analysis of Western snowy plover nesting at Vandenberg Air Force Base
PDF
Creating Hot Streets: developing an automated approach using ModelBuilder
PDF
Mapping future population impacts caused by sea level rise in Huntington Beach and Newport Beach: comparing the cadastral-based dasymetric system to past dasymetric mapping methods
PDF
Spatial analysis of vision services of Kaiser Permanente members
PDF
An examination of close-range photogrammetry and traditional cave survey methods for terrestrial and underwater caves for 3-dimensional mapping
PDF
Finding the green in greenspace: an examination of geospatial measures of greenspace for use in exposure studies
PDF
Using aerial imagery to assess tropical forest cover surrounding restoration sites in Costa Rica
PDF
The role of amenities in measuring park accessibility: a case study of Downey, California
PDF
Filling in the gaps: 3D mapping Arizona's Basin and Range aquifer in the Prescott Active Management Area
PDF
Obesity and healthy food accessibility: case study of Minnesota, USA
PDF
Testing social disorganization theory on violent crime: a case study on Pueblo, Colorado
Asset Metadata
Creator
Lee, Tiffany Monicque
(author)
Core Title
Defining neighborhood for health research in Arizona
School
College of Letters, Arts and Sciences
Degree
Master of Science
Degree Program
Geographic Information Science and Technology
Publication Date
02/14/2019
Defense Date
11/27/2018
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
administrative units,American Community Survey,boundary study,census block group,census tracts,geographic variability,GIS,MAUP,neighborhood,OAI-PMH Harvest,place,population health patterns,small area studies,social determinants of health,spatial analysis,spatial measurement,UGCoP,UPOP
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Wilson, John P. (
committee chair
), Loyola, Laura (
committee member
), Oda, Kirk (
committee member
)
Creator Email
arizona_lees@msn.com,tiffanml@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c89-120240
Unique identifier
UC11675409
Identifier
etd-LeeTiffany-7067.pdf (filename),usctheses-c89-120240 (legacy record id)
Legacy Identifier
etd-LeeTiffany-7067.pdf
Dmrecord
120240
Document Type
Thesis
Format
application/pdf (imt)
Rights
Lee, Tiffany Monicque
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
Tags
administrative units
American Community Survey
boundary study
census block group
geographic variability
GIS
MAUP
population health patterns
small area studies
social determinants of health
spatial analysis
spatial measurement
UGCoP
UPOP