Estimating at-risk population for lead service lines induced lead exposure and their correlation to socioeconomically disadvantaged neighborhoods in Milwaukee, Wisconsin

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Estimating At-risk Population for Lead Service Lines Induced Lead Exposure and Their Correlation to
Socioeconomically Disadvantaged Neighborhoods in Milwaukee, Wisconsin

by

Ariana Sul-ki Kim

A Thesis Presented to the
FACULTY OF THE USC DORNSIFE COLLEGE OF LETTERS, ARTS AND SCIENCES
University of Southern California
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF SCIENCE
(GEOGRAPHIC INFORMATION SCIENCE AND TECHNOLOGY)

December 2021

Copyright © 2021 Ariana Kim

To my Mom, Sam-chun, and my wife Blandine

iii
iii
Acknowledgements
I would like to thank my advisor, Dr. Leilei Duan, for her unending patience and guidance
throughout this whole journey. Thank you for advocating on my behalf and believing in me
when I struggled to believe in myself. Thank you to my committee members Dr. Robert Vos and
Dr. An-Min Wu for their insightful feedback and being willing to meet at strange times due to
wide geographical separation. A special thank you to Dr. Vanessa Osborne for helping me order
the chaos of my thoughts into coherent text.
I am deeply grateful to all my professors and peers in the Spatial Sciences department for
keeping me curious and open to exploring all the capabilities in the realm of GIS. I would like to
acknowledge Kendrick Watson and Maureen Scott for their check-ins and providing me all the
tools to succeed.
Huge shout out and thanks to Thomas Welcenbach at Lead Free MKE for being willing
to speak with me about the Get the Lead Out cause in Milwaukee and for providing information
and data resources. I hope my work will help in the fight to ensure clean drinking water for all.
Finally, I want to thank my family and friends for their encouragement and understanding
through this hectic time. Thank you for being my cheerleaders and emotional support throughout
this process. Words cannot express the depth of my gratitude. I love you all.

iv
iv
Table of Contents
Dedication........................................................................................................................................ii

Acknowledgements ....................................................................................................................... iii
List of Tables .................................................................................................................................. vi
List of Figures ............................................................................................................................... vii
Abbreviations .............................................................................................................................. viii
Abstract ............................................................................................................................................ x
Chapter 1 Introduction ..................................................................................................................... 1
1.1. Study Area: Milwaukee, Wisconsin ................................................................................... 3
1.2. Environmental Inequity ...................................................................................................... 5
1.3. Lead Toxicity ...................................................................................................................... 8
1.4. Thematic Mapping ............................................................................................................ 11
1.5. Racial and Socioeconomic Correlations with Environmental Justice .............................. 12
Chapter 2 Related Work ................................................................................................................ 14
2.1 Environmental Justice: Milwaukee ................................................................................... 15
2.2 Lead and Water .................................................................................................................. 16
2.3 Dasymetric Mapping ......................................................................................................... 18
2.4 Quantifying Socioeconomic Disadvantage ....................................................................... 21
2.5 Environmental Justice Correlation Studies ....................................................................... 23
Chapter 3 Methods ........................................................................................................................ 26
3.1 Project Design ................................................................................................................... 27
3.2 Data Description and Data Processing .............................................................................. 28
3.3 Analysis Tasks Details ...................................................................................................... 34
Chapter 4 Results ........................................................................................................................... 41

v
v
4.1. Dasymetric Mapping Results ........................................................................................... 41
4.2. Bivariate Correlation Analysis Results ............................................................................ 45
Chapter 5 Conclusion .................................................................................................................... 51
5.1. Dasymetric Map of At-Risk Populations ......................................................................... 51
5.1.1. Considerations and Limitations ............................................................................... 53
5.2. Racial and Socioeconomic Correlations ........................................................................... 54
5.2.1. Considerations and Limitations ............................................................................... 56
5.3 Future Study Recommendations ........................................................................................ 56
References ..................................................................................................................................... 59

vi
vi
List of Tables
Table 1 Datasets and sources ......................................................................................................... 32
Table 2 Coefficient of Variation Reliabilities for Census Data.....................................................34

Table 3 Descriptive statistics for all variables...............................................................................45

Table 4 Correlation matrix between explanatory and dependent variables...................................49

Table 5 Correlation matrix between explanatory variable percentages and the dependent
variable...................................................................................................................................49

vii
vii
List of Figures
Figure 1 City boundary of Milwaukee, Wisconsin ......................................................................... 4
Figure 2 Water service line graphic ............................................................................................... 11
Figure 3 Visual representation of the ARA calculation ................................................................20
Figure 4 Dot density of residential addresses with lead service lines............................................29
Figure 5 First spatial join between parcels and tracts with BA summed.......................................35
Figure 6 Second spatial join between new tracts and parcels with “BA_Sum” ...........................35
Figure 7 Process for layer with only parcels with LSL.................................................................36
Figure 8 Final spatial join to yield total population at-risk for lead exposure in each
census tract.............................................................................................................................36

Figure 9 Final workflow of the CEDS method..............................................................................37

Figure 10 Dasymetric map of the estimated people at-risk for lead exposure by census tract .....42
Figure 11 Dasymetric map of the percentage of at-risk people per census tract ..........................43
Figure 12 Neighborhoods in the city that intersect with the hot and cold spots............................44
Figure 13 Bivariate maps displaying each independent variable with the dependent variable.....47
Figure 14 Legend showing the standard deviation cut-off values for each variable.....................48

viii
viii
Abbreviations
ACS American Community Survey
ARA Adjusted Residential Area
BA Building Area
BLL(s) Blood Lead Level(s)
CEDS Cadastral-based Expert Dasymetric System
CDC Centers for Disease Control
EJ Environmental Justice
EJSEAT Environmental Justice Strategic Enforcement Assessment Tool
EPA Environmental Protection Agency
GIS Geographic information system
GISci Geographic information science
GWR Geographically Weighted Regression
LCR Lead and Copper Rule
LSL(s) Lead Service Line(s)
MAUP Modifiable areal unit problem
MHD Milwaukee Health Department
MKE Milwaukee
MOE Margin of Error
MWW Milwaukee Water Works
OLS Ordinary Least Squares
PCA Principal Component Analysis
PPB Parts per billion

ix
ix
RA Residential Area
RU Residential Unit
SAR Simultaneous Autoregressive (model)
SES Socioeconomic status
SSI Spatial Sciences Institute
USC University of Southern California
WHO World Health Organization

x
x
Abstract
The dangers of lead poisoning have posed a real threat to the population of the United
States since the turn of the century. It has a cumulative effect in the human body and can
therefore build up over time, even with low dose exposure. Children are especially susceptible to
lead exposure due to their increased absorption rate of the metal and the lasting health issues that
can persist throughout their lives. Minority communities with low socioeconomic status are
especially vulnerable to exposure because they are more likely to live in close proximity to lead
pollution sources, older homes, and have lower rates of toxicity screenings. Poisoning occurs
primarily when lead is ingested through lead-based paint, lead contaminated water pipes, dust,
and soil. Older cities across the United States are particularly prone to have populations with
increased blood lead levels because lead was a common building material in the early 1900s.
Milwaukee, Wisconsin is one such historical city where around 40% of the city’s active
residential water service lines are constructed of lead. This study quantifies how many people are
at risk for lead poisoning based on the existence of lead service lines in their buildings by census
tract. Given the deeply segregated history of Milwaukee, an issue that still plagues the city to this
day, this study also examines the relationship between the number of at-risk people per census
tract and a variety of socioeconomic indicators. Dasymetric mapping techniques as well as
regression analysis were used to shed light on this environmental justice issue in Milwaukee.
Results show that the number of at-risk people in a census tract has a positive linear relationship
with the race, education level, and poverty status of neighborhoods. In the context of
Milwaukee’s demographics, the issue of lead exposure due to LSL disproportionately affects
poorer communities of color.

1
1

Chapter 1 Introduction
The fight for environmental justice and equity is an ongoing battle across the nation. The
Environmental Protection Agency of the United States defines environmental justice as the effort
to provide environmental equity for all through implementation, enforcement, laws, regulations,
and policies (US EPA 2014). Too often those in minority social groups – i.e., those of low
socioeconomical status, racial minorities, or the unemployed – experience higher levels of
environmental toxicity (Maantay 2002). The causes of these inequities, also known as
environmental burdens, can include poor air quality, proximity to polluting industry, old lead
water mains, or lack of funding to expand green spaces for greater access (Emer et al. 2020). The
city of Milwaukee, Wisconsin is one of the many major urban centers where these environmental
inequities are starkly apparent. The city has a long history of environmental justice issues such as
increased asthma rates, childhood lead poisoning, and lack of access to green space within poor
black and brown communities (Small 2019; Collins 2011).
In 2020, issues of environmental inequity were exacerbated with the advent of Covid-19
and the realization that mostly poor black and brown communities were disproportionately
affected by the virus. The CDC has listed discrimination, access to healthcare, occupation,
education/wealth/income gaps, and housing as social indicators for the increase in risk these
minority groups face when infected by the virus (CDC 2020). The low-income population of
Milwaukee often relies heavily on state run health programs such as BadgerCare and the Special
Supplemental Nutrition Program for Women, Infants, and Children (WIC) for medical
screenings, treatments, and education. Due to the current pandemic, many of these health centers
have limited office visit availabilities. Not only have these closure contributed to the already

2

2
poor access to health care facilities, but it also prevented many minority communities from
receiving vital information about the dangers of Covid-19 and how to prevent the spread of
infection.
A further consequence of lack of access to medical centers is a 34% decrease in lead
screenings in the city (Dang et al. 2021). Given the worsening state of this health crisis, and
Milwaukee’s recent initiative to replace all lead service lines (LSL) (Jannene 2020), the
examination of where these remaining lead pipes exist in the city, how many people are at risk
for lead poisoning from these pipes, and their spatial correlation with neighborhoods of various
socioeconomic statuses would ideally prove useful information to both city and health officials
as well as environmental justice advocates. Therefore, the aims of this project will be two-fold.
The first goal is to determine which census tracts in Milwaukee have the greatest number of
people directly exposed to LSL, and therefore have higher risk for lead poisoning, by creating a
dasymetric map of the city’s population distribution and relating the population density with the
corresponding number of lead service lines that need to be replaced. The second goal is to
characterize regions of the city with high levels of lead exposure risk with racial and
socioeconomic indicators by creating bivariate correlation matrices.
The targeted audience of this study would be policy or decision makers within the
community, as well as grassroots organizations fighting for change. It would provide a tool of
reference that could help inform them of the neighborhoods that would benefit from intervention
to mitigate harmful environmental burdens. This project would also be accessible to the wider
public as an educational tool that could potentially inspire them to become involved in
community projects whose goals are to combat the impacts of environmental injustice or to
contact their local government officials to incite change in harmful policies.

3

3
1.1. Study Area: Milwaukee, Wisconsin
The study area of this project is the city of Milwaukee in Wisconsin. Founded in 1846,
Milwaukee has a population of roughly 590,000 and is located on the western banks of Lake
Michigan (Figure 1) (“U.S. Census Bureau” 2019). The city is known for a plethora of
breweries, Harley Davidson motorcycles, and being the home of the 2021 NBA Champions, the
Milwaukee Bucks. Unfortunately, the city also has a reputation for being one of the most
segregated cities in America. Due to systemically racist policies, such as red-lining, Milwaukee
is one of the most racially and socioeconomically divided in America (Lynch and Meier 2020).
Studies have been done about the increased exposure to lead toxicity, lower air quality, and
inaccessibility to healthy foods in these neighborhoods. Several studies have been conducted
highlighting the disparity between affluent neighborhoods and poorer neighborhoods in
Milwaukee with respect to access to green space, childhood exposure to lead, and inequities with
transport access (Emer et al. 2020; Heynen et al. 2006; Milwaukee Environmental Justice Lab
n.d.).

4

4

Figure 1: City boundary of Milwaukee, Wisconsin.
1.1.2 Milwaukee and Lead
The issue of lead poisoning in the city of Milwaukee is a long-standing problem that
medical professionals, city leaders, and activist groups have been fighting for years. Being an
older city, Milwaukee has a good number of homes that were built in the late 1800s to the early
1900s, a time where lead was a common construction material. The pipes that brought water to
homes were made of lead, and lead-based paint was the go-to for decorating homes (City of
Milwaukee 2016). Despite spreading awareness across the U.S. about the dangers of lead
poisoning, especially among small children, Milwaukee continued to have a higher than the
national average of children with elevated blood lead levels (BLL) based on data collected by the
Milwaukee Health Department’s (MPH) Childhood Lead Poisoning Prevention Program
(CLPPP) (Public Health Foundation 2020). As a response to this alarming statistic, the
Community Lead Outreach Project was started in 1995 by the Sixteenth Street Community

5

5
Health Center. Their goal was to decrease childhood lead poisoning by reaching out to the
poorest communities in the city that live in the oldest homes. The project was successful in
achieving this goal and health studies from the time showed a decline in children testing positive
for elevated BLL from 46% to 23% over the years 1996-1999 (Schlenker et al. 2001). As of
2019, 9-10% of children in Milwaukee test positive for elevated BLL.
The Community Lead Outreach Project provided a strong case for the importance of
primary intervention when it came to reducing childhood lead poisoning. They found that
parental education as well as fostering close relationships with the communities they were
working with helped contribute to the overall success of the project (Schlenker et al. 2001).
Despite all the progress Milwaukee has made to reduce childhood lead poisoning, the fact
remains that exposure to lead is a persistent threat to the overall health of the city’s residents. To
date, no such study has been conducted attempting to estimate the actual number of people
exposed to lead via lead service lines, nor has research examining the correlation between lead
exposure and socioeconomic indicator variables within neighborhoods. Considering there are
still over 70,000 active residential lead service lines, this study could be useful for the decision
makers of the city, lead-free activist organizations, and for the general public as a tool to educate
themselves about an issue that impacts their daily lives.
1.2. Environmental Inequity
The causes of environmental exposure disparities are not as simple as residential
segregation of minority communities. It is an issue that is deeply engrained into the fabric of
society through years of discriminatory legislation and social conditioning. From the era of Jim
Crow to the deeply problematic policy of redlining, racial discrimination in laws has lasting
consequences today. In cities like Chicago and Milwaukee, neighborhoods can be distinguished

6

6
based on their racial make-up as a result of redlining in the 1960s. Since black and brown people
were denied mortgage loans to buy property is certain parts of the city, they were forced to
specific areas where they could afford housing. Often the parts of the city that were available
were undesirable for the affluent society because they were near polluting factories. Based on
proximity to hazardous byproducts coming from these factories, the minority population that
were forced to settle in the neighborhood become exposed to environmental toxins at higher rates
(Shrader-Frechette 2002; Hillier 2003; Lester 2018).
From an economic standpoint, environmental inequities – access to urban green spaces
for example – can be linked to the capitalistic commodification of these elements. Urban forests
have become part of the process of production where those who can afford to “consume” more of
the commodity are able to dictate where these spaces are introduced. Therefore, the distribution
of the green spaces become uneven and neglected in the parts of a city that are of a lower
socioeconomic status (Heynen et al. 2006). Commodification of urban spaces is likely at the root
of many cases of environmental injustices minority communities face. The reality is people with
low socioeconomic status in society lack the financial and social resources to combat their
adverse environmental exposures (Kelly-Reif and Wing 2016). The task of mitigating these
hazards is left to local and federal governments who, in the best-case scenario, can commission
studies to be carried out to identify the exact issue at hand and the most effective strategy to
combat the environmental injustice.
1.2.1 Quantifying and Analyzing Environmental Injustice
When studying communities to identify environmental injustice, there are three
components that are regarded. First, there is an exposure assessment for locations within a
geographic region. Second, there must be some way to quantify sociodemographic variables

7

7
across the whole geographic region in question. And third, there is a presence of disease or other
detriments to human health in the region (Waller et al. 1997). The methods behind studying the
cause and effects of environmental injustice have become more sophisticated over the years.
From collecting health data incorporating remote sensing and machine learning, the research on
this subject has expanded significantly since its inception amongst the Civil Rights movement in
the 1960s. The development of remote sensing technology increased data resources for
environmental justice analysis by providing highly detailed views of urban landscapes. These
data can be used to analyze multiple environmental burdens such as heat islands, air pollution,
and access to green spaces (Weigand et al. 2019). Researchers have utilized principal component
analysis (PCA) to create neighborhood deprivation indices based on social data taken from the
US Census Bureau. Once these indices are created, regression analysis is used to examine
correlations between the spatial phenomena being studies and neighborhoods that have low
deprivation index scores. The results of the analysis can then be visualized on a map of various
spatial scales and a hotspot analysis can be used to identify areas of potential intervention
(Padilla et al. 2014).
A major consideration when assessing a geographic area for exposure to an
environmental toxin is the appropriate level of spatial scale. The modifiable areal unit problem
(MAUP) can be an issue when choosing the appropriate scale at which to conduct the analysis
(Mennis 2003). If the spatial resolution is too big, the nuances of the environmental phenomena
being studied can be lost. If the resolution is too fine, particularly in the case of census data, one
runs the risk of having data with high levels of error and thus skewing the results of the analysis.
Specifically in health studies, it is understood that environmental justice health issues are often
spatially autocorrelated. It is therefore critical to understand and consider MAUP and its roll in

8

8
potentially skewing the results of aggregated data by introducing spatial bias (Swift, Liu, and
Uber 2008). Researchers can ensure that their results are as accurate, valid, and transparent as
possible by maintaining the integrity of their methods and disclosing limitations in existing data.
1.3. Lead Toxicity
In the late 1800s, lead became the industry standard for the construction of water
distribution pipes in the United States due to its ability to corrode at a slower rate than iron, as
well as its superior malleability. Despite the slower corrosion rate, lead is still prone to break
down over time. These particulates eventually leach into the water supply, and the residents who
consume the contaminated water see a rise in their blood lead levels (BLLs) (Brown and
Margolis 2012). The dangers of lead service lines (LSL) have been widely known with articles
being published about their toxic effects as early as 1859. However, lead was too convenient of a
material source to prohibit from its use to be mandated until 1986 when Congress passed the
Safe Drinking Water Act Amendments (Rabin 2008). Despite this ban, hundreds of thousands of
Americans are still at risk for lead poisoning due to existing lead service lines. In Milwaukee
specifically, upwards of 70,000 residential structures still have LSL in use (Lewis et al. 2017).
When Milwaukee’s original water services lines were placed in the early 1900s, they
were made out of lead. In 2017, Milwaukee launched an initiative to raise awareness about the
dangers of lead poisoning and to replace the lead laterals, the pipes that bring water to homes
from the water main (Milwaukee Water Works n.d.). Due to the Covid-19 pandemic, the city has
fallen far behind in their goal to replace all the city’s lead laterals and is now projected to
complete the project in 70 years (Jannene 2020). This delay is ultimately the most harmful to
those in the lowest socioeconomic classes, primarily racial minority communities, in Milwaukee
(Emer et al. 2020).

9

9
1.3.1 Blood Lead Levels and Exposure
Lead poisoning has a cumulative effect on the human body. The Centers for Disease
Control (CDC) have stated that the maximum threshold of BLL is five micrograms per deciliter
(μg/dL), however no amount of lead is safe for the human body, especially in young children
(CDC 2021; Miranda et al. 2002). The general symptoms of short-term lead poisoning include
abdominal pain, constipation, exhaustion, headaches, irritability, loss of appetite, memory loss,
tingling in hands and feet, and feeling weak (CDC 2020). The long-term effects of lead exposure
are similar to the short-term effects but also include mood disorders, decreased fertility, and
difficulties concentrating. Arguably the more alarming and damaging impacts of lead poisoning
occur in young children. Exposure to lead has been linked to developmental delay, learning
difficulties, weight loss, hearing loss, and seizures (Mayo Clinic 2019). These symptoms can
severely impact a child’s quality of life, long past the time of their initial exposure. Research has
shown that even low-level exposure can have adverse effects on cognitive development in
children (Hou et al. 2013). In other words, children with lead exposure show decreases in their
IQ scores and academic achievement when compared to children not exposed (Sorensen et al.
2019).
Children and adults can be contaminated with lead poisoning by means of paint, soil,
dust, air, and water (Lynch and Meier 2020). Children who are exposed to lead in the United
States usually live in a structure that was built prior to 1940, an era where it was common to use
lead as a construction material in both the water pipes as well as in the paint (Chisolm 1971).
Despite legislation and efforts to minimize the risk of lead contamination through water
consumption, 10-20% of children and 40-60% of infants’ lead intake can be traced to potable
sources (Rabin 2008). Studies have proven that childhood lead exposure is a spatially correlated

10

10
issue. These exposures also tend to disproportionately impact communities of color, especially
those communities with families living below the poverty line or with low socioeconomic status
(Oyana and Margai 2007). Due to lack of resources, communities with a low socioeconomic
status are also more likely to be unable to afford the cost of replacing their LSL even if they
wanted to (Sampson and Winter 2016).
1.3.2 Lead Service Lines
Lead service lines (LSL) are water pipes that run from the water main to the house
(Figure 2). As mentioned earlier, lead was the most common construction material in the late
1800s to early 1900s. Unless replacements have occurred, areas of the city that were built around
this time will have LSL that supply the houses with water from the main line. According to
Milwaukee’s water department, the water that leaves the water treatment facilities contains no
lead. However, lead can leach into the water simply transporting water using an old LSL
(Milwaukee Water Works 2021). According to Milwaukee Water Works (MWW), the city’s
water supply has been in compliance since 1996 with the EPA’s 1991 Lead and Copper Rule
(LCR) – legislation that sought to regulate lead and copper levels in drinking water. To be
compliant with the LCR, there must be less than 15 parts per billion (ppb) concentration of lead
found in tap water (US EPA 2015). While 15 ppb of lead is an extremely low concentration, it is
a consensus that no level of lead is safe to consume. The EPA has recently updated the LCR to
include earlier intervention to detect lead in communities’ drinking water, push for complete
LSL replacements as opposed to partial replacements, require lead level testing at schools and
childcare facilities, and to make the locations of existing LSL available to the public (US EPA
2020). The obvious and most logical solution to preventing lead exposure by LSL is to replace
the whole service line.

11

11

Figure 2: Water service line graphic (Denver Water 2021).
1.4. Thematic Mapping
When studying a spatial phenomenon, researchers often aggregate their data into polygon
areal units, such as census tracts. From there, a thematic choropleth map can be generated to
determine any spatial patterns in the mapped phenomena. Choropleth maps assume that the data
are spread homogenously across the chosen areal unit. When studying population related data, it
is known that the data are more heterogeneously dispersed. Further, it can prove challenging to
choose the optimal areal unit to aggregate the data to begin with. Government drawn boundaries
can be arbitrary and are for administrative or political reasons. They often do not have any
relation to any underlying spatial occurrence – i.e., crime rates, public health issues, and land use
(Maantay et al. 2007). It is important to note that policies are put in place to protect the privacy
of people when data can contain sensitive information. For this reason, it can be difficult to
obtain accurate data at a detailed level (Kennedy and Kennedy 2004).
Dasymetric mapping, first developed in the early 1900s as an alternative thematic
mapping technique to choropleth maps, involves the division of data into homogenous zones that
represent the underlying statistical surface (Eicher and Brewer 2001). Population density within
census tracts is a prime example of the potential issues with arbitrary unit partitioning.

12

12
Populations within a census tract are assumed to be homogenously distributed across the areal
unit. Depending on the size of the areal unit, this can lead to fallacious analytical claims that do
not accurately represent the population distribution within the unit (Mennis 2003).
Administrative boundaries drawn for bureaucratic purposes can often mask the underlying issue
being studied. Therefore, when studying environmental justice issues that emphasize the
importance of accurate population distribution, it is critical that a reliably detailed spatial scale is
used. Dasymetric mapping techniques can redistribute the population within an aerial unit like a
census tract by disaggregating the population data and repopulating it with ancillary data (such as
residential parcels). This provides a much finer spatial resolution and more precisely represent
the actual population distribution within the study area (Eicher and Brewer 2001; Maantay et al.
2007).
1.5. Racial and Socioeconomic Correlations with Environmental Justice
When traveling through any major city in the United States, it is usually quite apparent
which areas of town are wealthier than others. Just by means of observation, one can
differentiate between these neighborhoods by their cleanliness, property values, access to healthy
and a variety of grocery stores, etc. Drive-by observations aside, researchers have developed
ways to quantify the socioeconomic disparities in order to perform meaningful analysis on either
the reasons for such stark differences within one city or the impacts of these divides, both direct
and indirect, on peoples’ lives. Research has determined that certain social indicators, such as
race, unemployment, poverty, and education level, tend to aggregate at the neighborhood level
(Messer et al. 2006). With this understanding, there can be a multitude of variables that can
indicate economic disadvantage. It is a complex, multi-faceted issue where one or many of these
social indicators are linked to disparities in public health (Eicher and Brewer 2001). Studying the

13

13
relationship between socioeconomic variables and the at-risk population for lead exposure in
Milwaukee can provide greater context and insight into the underlying issues that are
contributing to this environmental inequity.
Correlation analyses are used in environmental justice studies to examine relationships
between a dependent variable and explanatory socioeconomic factors (Jerrett et al. 2001;
Raddatz and Mennis 2013). According to Tobler’s first law of geography, “everything is related
to everything else, but near things are more related than distant things” (Tobler 1970, 236). The
concept of spatial autocorrelation as outlined by Tobler’s famous law can pose problems to
spatial statisticians when attempting to run regression or correlation tests because their base
assumption lies in the independence of the observations and errors (Chakraborty 2011).
Therefore, it is highly important to take spatial autocorrelation and multicollinearity between
variables into account when running correlation tests because of the bias they create in the data.
In the following chapter, studies related to the topics of environmental justice, lead
poisoning, dasymetric mapping techniques, and regression analyses will be explored. Concepts
will be reviewed, methodology outlined, and a case for the justification of this study will be
addressed as well.

14

14
Chapter 2 Related Work
The damaging effects of ingesting lead are not as recent a discovery as one might think. Texts
from the second century BC suggest that Hellenistic physicians were aware of the toxic effects of
high lead exposure. It was not until the Industrial Revolution when the chronic effects of lead
exposure were made known to the larger medical community (Riva et al. 2012). Within the last
80 years, research about the damages of lead poisoning, as well as about sources of lead
exposure has exploded. The literature on the subject has expanded from a medical perspective to
encompass environmental points of view. Legislation was introduced in the 20
th
century to
restrict the use of lead-based materials in construction to help lower exposure rates among the
population. Overall, the instances of lead poisoning and exposure have significantly decreased
since the Industrial Revolution due to enhanced research and awareness (Brown and Margolis
2012; Chisolm 1971; Rabin 2008).
Issues of lead in the water and the links to environmental justice have been extensively
researched for the past century. The literature has revealed that lead exposure disproportionately
affects minority communities across America (Sampson and Winter 2016). Public attention to
this injustice increased following the Flint, MI water crisis in 2016. The outcry of poor black and
brown communities, the ones most impacted by lead exposure, was finally gaining national
attention (Butler et al. 2016). Other cities, including Milwaukee, began to put renewed efforts
into their clean water programs and removing remaining lead service lines (Jannene 2020; Public
Health Foundation 2020; Wisconsin DNR 2021). In 2017, Milwaukee’s water department
announced an initiative to replace all remaining LSL in the city. Unfortunately, circumstances
like Covid-19 have drastically slowed their progress toward their goal. The delays mean that

15

15
Milwaukee’s residents, particularly historically black and brown neighborhoods, will continue to
be at-risk of poisoning from exposure to lead in their water supply.
The following chapter explores the related literature researching lead exposure from a
historical and environmental justice lens. It also explores the ways in which dasymetric mapping
can be used to accurately estimate the number of people at-risk. Finally, examples of regression
analysis will be explored to show how these techniques can help shed light onto the underlying
patterns of toxicity exposure in relation to socioeconomic factors.
2.1 Environmental Justice: Milwaukee
Milwaukee has been the case study for environmental justice issues for many years. It is a
city where the social divisions between neighborhoods are starkly apparent and have root in
years of divisive public policy. Mary Collins published a study in the American Journal of Public
Health in 2011 called “Risk-Based Targeting: Identifying Disproportionalities in the Sources and
Effects of Industrial Pollution.” The most notable point in this study is the fact that Collins’s
study area was the City of Milwaukee. The aim of her study was to prove that industrial pollution
from a few key polluters in the region disproportionally affects low-income and minority
communities. She used the risk screening environmental indicators model to conduct her
methodology. She first assessed the efficacy of the current mode of monitoring environmental
burdens in the community, the Environmental Justice Strategic Enforcement Assessment Tool
(EJSEAT). Although it contains 18 variables in its assessment process, race is excluded from the
method. The study sought to add race as a variable in the EJSEAT and include more specific data
to effectively measure environmental justice concerns within Milwaukee. Her methodology for
calculating race and socioeconomic class dissimilarity by using percentages will most likely be
used in this project, as both will be examined as a social context indicator (Collins 2011).

16

16
A 2013 health report for Milwaukee conducted by the Center for Urban Population
Health found significant health disparities between the socioeconomic classes within Milwaukee.
One of the categories that disproportionately affects the lower socioeconomic class is childhood
lead poisoning (Greer et al. 2013). Lead poisoning is a serious issue because it is a cumulative
toxicant that can affect multiple systems in the body. Small children are especially susceptible
because they absorb four to five times more lead than adults when ingested (WHO 2019).
2.2 Lead and Water
Studies in the United States examining the toxicity of lead in the drinking water supply
can be traced back as early as 1845 (Brown and Margolis 2012). Despite the fact that many US
cities were voting to move away from using lead for water pipes by the 1920s, the national
plumbing codes continued to approve lead as a viable material source until well into the 1980s,
around the time Congress passed the Safe Drinking Water Act Amendments (Rabin 2008). With
all these efforts to ensure safe, toxin-free water for all, the Environmental Protection Agency still
allows up to 15 parts per billion (ppb) of lead in drinking water (US EPA 2016). The EPA, in
partnership with the Centers for Disease Control (CDC), recognizes that no amount of lead is
safe in the human body, especially in children (US EPA 2016). There is a worldwide consensus
on this view (WHO 2019). An added layer to the environmental issue of lead exposure is
observing which communities are most at risk. It has been proven time and time again that lead
poisoning disproportionately affects minority communities throughout the United States (Brown
and Margolis 2012; Lewis et al. 2017; Butler et al. 2016; Rabin 2008; Sampson and Winter
2016).

17

17
2.2.1 Flint, Michigan Water Crisis
A landmark example that occurred within the past decade is the Flint, Michigan water
crisis. The case in Flint was overwrought with regulation violations and false reporting. The
House Committee of Oversight and Government Reform formed a bipartisan conclusion that the
city officials had been negligent in their duties to ensure the safety and health of its residents
(Chaffetz et al. 2016). It was revealed that after the city switched water sources, high levels of
lead began to be reported in the city’s drinking water.
When conducting lead and copper rule (LCR) sampling, reports found that the Flint’s
water treatment plant improperly collected samples from homes that were not at high risk for
lead contamination. On top this sampling error, the Michigan Department of Environmental
Quality gave faulty instructions to the residents of the town, telling them to pre-flush their taps
before the sample collected (Butler et al. 2016). This is against the protocol for LCR sampling
and skewed the city’s reported lead levels. A major justification for the delayed and poor
response of the local government to intervene is the direct result of the economic and racial
makeup of the city. The population of Flint is 62.6% people of color and 41.6% of individuals
live below the poverty line (Butler et al. 2016).
The BLLs of children under five were compared before and after the change in water
source for the city, and significant increases in elevated BLLs were recorded. Overall, the results
showed that 2.4% of children had elevated BLLs before the change, and 4.9% were recorded to
have elevated levels after the fact (Hanna-Attisha et al. 2015). In areas identified with high water
lead levels, the jump went from 4% to 10.6%. The same study compared these results against
socioeconomic disadvantage scores and found the areas with elevated BLLS to be statistically

18

18
positively correlated with areas with high levels of socioeconomic disadvantage (Hanna-Attisha
et al. 2015).
2.3 Dasymetric Mapping
When analyzing spatial phenomena utilizing census data, it has become common practice
to use dasymetric mapping techniques to gain an accurate visualization of the topic of study. The
boundaries used in census data are meant to serve government purposes and are not always ideal
when creating a map to study population distribution as an example (Mennis 2003). Traditional
choropleth mapping assumes the data are spread evenly across the chosen areal unit. In the case
of census boundaries, which are drawn with no consideration to any sort of spatial subject,
thematic maps can be misrepresented, and the underlying pattern masked (Maantay et al 2007).
There are several different ways to approach dasymetric mapping depending on the
subject matter being mapped and the research question. A common approach for disaggregating
population is areal interpolation. Areal interpolation involves the transfer of data from a source
dataset to a target dataset of overlapping areal units. It is assumed that population is distributed
evenly across the source layer. When estimating the population based on the overlap between the
source and target zones, the ratio of the overlap between the two datasets is applied to the
population of the source zone, thus yielding the estimated population of the target zone (Maantay
et al. 2007). The simplest version is a binary method of areal interpolation that can be used to
estimate population density based on land-use data that filters out regions of uninhabitable land
(Eicher and Brewer 2001).
There are some shortcomings with areal interpolation methods for population
disaggregation. For one, areal interpolation operates under the assumption that all residential
areas have homogenously distributed population density (Maantay et al. 2007). Residentially

19

19
zoned areas in a municipality do not necessarily have even population density throughout,
therefore a similar fallacy in the analysis could occur to if one used the census boundaries as the
areal unit of disaggregation. The spatial resolution of land-use data may not be as fine-tuned as it
needs to be to present a true representation of population distribution over a given area. When
seeking to build as true of a representation as possible for population distribution within a city, a
more refined approach must be taken that is able to disaggregate population data to the finest
possible resolution possible.
2.3.1 The Cadastral-based Expert Dasymetric System
The Cadastral-based Expert Dasymetric System (CEDS) was first introduced by Maantay
et al.’s (2007) study comparing its novel methodology against different, more established,
dasymetric mapping techniques. Notably, the researchers compared their CEDS disaggregation
method against simple and filtered areal weighting disaggregation methods – two of the most
common ways to disaggregate population data. While the areal disaggregation methods use
remotely sensed land-use data as their ancillary data, the CEDS method uses cadastral data to
estimate population distribution. Maantay et al. (2007) developed a method that disaggregates
data from a more general census block group level down to the tax lot level. The researcher used
residential area (RA) and residential units (RU) as stand-ins for the population distribution in
their calculations. The adjusted residential area (ARA) is the total livable area within a building,
and it is derived by multiplying the tax lot’s building area by the ratio of residential units and
total units. The information about the RU and the calculated ARA were aggregated up from the
tax lot level to the census block group and census track levels (Figure 3).

20

20

Figure 3: Visual representation of the ARA calculation
From there, the census population data can be multiplied by the proxy population unit ratios (RU
or ARA) to derive the final dasymetrically calculated population density. Maantay et al. (2007)
could then determine which proxy unit (RU or ARA) was most appropriate for tract population
estimation by a tract-by-tract basis. To perform this check, population data was disaggregated
back down to the tax lot level, which was then aggregated back up to the block group level. By
doing this disaggregation and aggregation for both RU and ARA proxy units, the researchers
were able to compare the calculated block group population estimate with the census block group
population data. Whichever proxy unit with the smallest difference between the two values was
chosen as the preferred proxy unit for the dasymetric calculations for that census tract. The
argument that makes the CEDS method superior to the areal interpolation disaggregation
methods is that CEDS uses tax-lot information which is a much finer spatial resolution than
anything that the areal interpolation methods work with. Thus, the final dasymetric population
density result can display the nuances in human population distribution better than areal
interpolation methods (Maantay et al. 2007). A slightly modified version of the CEDS method of
dasymetric mapping was used in this study when calculating the number of at-risk people for
lead exposure due to the presence of LSL.

21

21
2.4 Quantifying Socioeconomic Disadvantage
Being able to provide context behind mapping spatial phenomena brings meaning to the
analysis of the issue. The ability to answer the question “why?” in reference to the spatial event
being visualized pushes the research in the direction of problem solving over simply pointing out
a potentially problematic situation. Conducting correlation tests to characterize the regions where
an environmental justice issue is occurring can provide valuable evidence to support the plight of
the victims of such an injustice. Being able to identify areas within a city with shared attributes
that experience environmental inequity can help decision makers and advocacy groups know
where to target their outreach programs and solution actions.
The literature on quantifying socioeconomic disadvantage within neighborhoods contains
a wide variety of different methodologies. One route relies on inductive and deductive reasoning
to select the most appropriate social vulnerability indicators. The concept would be to initially
select the vulnerability indicator variables using deductive reasoning by researching with
background literature. From there, the variables can be verified and selected for analysis with
inductive reasoning – looking at current data and statistical studies that utilize the variables in
question (Hinkel 2011). Building on Hinkel’s methodology for indicator identification, Samuel
Rufat et al. (2015) compiled worldwide case studies on social vulnerability to floods. The
researchers were able to identify seven social indicators based on frequency of appearance in the
125 studied cases. These indicators were: demographic characteristics, socioeconomic status
(SES), health, coping capacity, risk perception, neighborhood quality of life, and land tenure
(Rufat et al. 2015). Hinkel’s (2011) method for social indicator selection is highly labor intensive
and scrupulous research on the subject at hand. It can contextualize the indicators that are
selected and seeks to provide a global standard for social vulnerability indicators.

22

22
To select the socioeconomic indicators for this study, a similar approach to Hinkel (2011)
was taken in that extensive background research was conducted to select out appropriate SES
indicator variables from the literature. Given the historical segregation of the city, Milwaukee
has no shortage of socioeconomic studies as they relate to health and environmental justice
issues. In a 2013 health report, the Center for Urban Population Health conducted a study of
health disparities in Milwaukee by socioeconomic status. The researchers derived a SES index
by using household median income and education data. Each zip code within the city was
assigned a SES index score based on the calculation from the two indicator variables. The results
of this report showed major health disparities within the city and upheld previous research
conclusions that socioeconomic status is one of the most telling indicators of peoples’ health
outcomes (Greer et al. 2013). A more recent study conducted by Lynch and Meier (2020)
examined the intersection of poverty, home ownership, and race on childhood blood lead levels
in Milwaukee. All the SES data in this study were taken from a census tract level while the
childhood blood lead levels were measured as a continuous mean across census tracts. The
results of Lynch and Meier’s (2020) research concluded that socioeconomic and racial minority
neighborhoods have higher average childhood BLL. Additionally, high BLL risk greatly
increases if the neighborhood’s population has multiple risk variables (Lynch and Meier 2020).
These studies were just two key examples from literature studying inequities in Milwaukee based
on socioeconomic status. Based on the precedent these studies provided, the socioeconomic
variables used in this study are median household income, poverty status, race, ownership of the
lived-in residence, and education level.

23

23
2.5 Environmental Justice Correlation Studies
Correlation research between racial and social factors and environmental justice issues
are found in abundance within the literature. Amongst the most studied relationships is that of
pollution with communities of color and/or those experiencing poverty (Banzhaf et al. 2019). In
1992, researchers Mohai and Bryant found that racial and class biases are directly related to
issues of proximity to environmental hazards. This conclusion was reached after studying sixteen
environmental justice case studies and conducting a public perception study in the Detroit area.
While their research showed class to be a significant factor in exposure to environmental toxins,
race was the more strongly related factor with environmental hazard exposure (Mohai and
Bryant 1992). Mohai and Bryant’s 1992 study was just one example of the increasing number of
environmental justice correlation studies that have been conducted over the past two decades.
2.5.1 Understanding Ecological Fallacy
Significance of racial and socioeconomic inequalities with hazard exposure vary in the
studies, some finding the two to be strongly related, while others finding little correlation
(Anderton et al. 1994). A major source of oversight in studies is the failure to consider ecological
fallacy. Ecological fallacy occurs when one attempts to draw conclusions about spatial
relationships by comparing two sets of data with different aggregation scales. Specifically, error
is introduced when assuming variations at a larger scale are the same at a smaller, more
individual scale (de Munck 2005). A way to help account for ecological fallacy in correlation
analyses is to choose units of analysis as small as possible without compromising the integrity of
the data (Banzhaf et al. 2019). The issue remains that when comparing different variables in a
correlation calculation, the aggregation methods of the data are not guaranteed to be the same.
One could measure estimates on the individual level, while another could represent the data

24

24
collectively across the chosen spatial unit. Therefore, a good practice is to describe the overall
characteristics of the population in the region of study, especially when incorporating multiple
social explanatory variables in the study (de Munck 2005).
2.5.2 Hazard Proximity Methods
Another possible reason for the wide variation in statistical uncertainties among
environmental justice correlation studies is a lack of standardized methodology for calculating
environmental toxin exposure proximities. The classical approach is to define geographical units,
such as counties or zip codes, identify those with and without the environmental burden present,
then compare the demographic data between the two sets of units. This method of determining
hazard exposure is referred to as the spatial coincidence approach and it assumes the population
living in the geographic units that contain the environmental toxin are automatically closer to the
source than those who live in units that do not contain the hazard. The logic behind the thought
of the spatial coincidence methodology is flawed because one, it assumes a spatial unit’s
population is evenly distributed in the distance to the hazard source and two, it does not account
for edge effects among the neighboring units.
Distance-based methods of proximity analysis can account for the issues found in the
classical, coincidence approach. With distance-based approaches, the precise locations of the
hazards are included, and demographic variables within a set distance of the environmental
burden from any geographic unit can be compared against those that occur farther away. In other
words, the data are aggregated based on their closeness to the hazard locations instead of by
more arbitrary geographic boundaries (Mohai and Saha 2006).
While distance-based methods of examining correlations between social and racial
variables and an environmental injustice, it relies heavily on very fine spatial resolutions of data.

25

25
The level of detail that would be required to conduct an accurate study with distance-based
methods can be difficult to obtain from public access databases. For this reason, this study will
attempt to quantify the characteristics of the regions of the city with higher levels of lead
exposure risk based on bivariate correlation analysis on demographic indicators obtained from
the Census Bureau. The following chapter explains the methodology used to derive more
accurate population distributions in the city using a dasymetric mapping technique and the steps
of the correlation analysis using the results of the dasymetric map as the dependent variable and
demographic data as the explanatory variables.

26

26
Chapter 3 Methods
The following chapter contains the details for the methods of this project. It begins with an
overview of the project’s design followed by a section containing a description of the data used
in the analysis. The last two sections go through the creation of the dasymetric map using the
Cadastral-Based Expert Dasymetric System methodology and regression analysis using ordinary
least squares.
A major consideration with data selection in this study was selecting the most appropriate
spatial resolution for the data analysis. Utilizing spatial scale that was at the finest resolution
possible without compromising the accuracy of the data was critical to ensure the integrity of the
analyses being conducted. Datasets that are found in the Census Bureau’s website all contain the
variable estimates as well as a calculated margin of error (MOE) to account for variation within
the data. The MOE from the American Community Survey data were calculated at a 90%
confidence interval (Berkley 2017). The smallest spatial scale that population estimates, as well
as the social indicators from the ACS could be obtained were at the census block group level.
Upon further inspection of these data, MOE for each block group was far too high to perform
reliable statistical analysis. In some cases, the MOE exceeded that of the data estimate itself. The
high error found in the census block group data also disqualified it from being used to validate
the results of the CEDS map. The next spatial scale up was the census tract data. These tables
were much more accurate in terms of a lower MOE for each variable estimate. Therefore, the
spatial scale for all tabular data in this study is at the census tract level.
When it came to the explanatory variable selection for the bivariate correlation analysis,
an inductive approach was taken. Results from previous environmental justice and health equity
studies in Milwaukee were examined and the selection of socioeconomic indicator variables

27

27
were compiled as results of the research. There are statistical methods that can be used to select
socioeconomic variables for a regression analysis, but these are beyond the scope of this study
and are addressed in Chapter 5 of this paper.
3.1 Project Design
Given the historical and modern relevance of the damages lead pipes have caused in
Milwaukee, as well as the abundance of data related to the subject, the environmental burden
variable chosen for this study is the presence of lead water mains within the City of Milwaukee.
As mentioned above, it is difficult to conduct an accurate measurement of exposure to a toxin if
the population estimates that are used in the analysis are assumed to be distributed
homogenously across a census tract. Nuances in the living circumstances of the population can
be lost when using such a large spatial unit. Further, it is difficult to tell a precise number of
people who are directly at-risk for lead exposure based on just the total number of LSLs still
active in the city. This is due to the reality that one LSL does not equal one person exposed. A
single LSL can feed water into a house or apartment complex where multiple people reside. If
one wants to see how many people are at-risk by census tract, a simple proportion of number of
LSL addresses to the tract population will not suffice. By using the census tract as the aerial unit
of measure, it assumes that the populous is spread homogenously across the tract. This rationale
is an obvious fallacy as humans disperse heterogeneously across a given space. A dasymetric
population distribution map is a way to solve the issue of spatial scale by disaggregating the data
into smaller units across the study area. This involved a series of joins and field calculations to
disaggregate and reaggregate census tract data down to city parcel resolution, then back up to
tract data. Once the final at-risk population totals were obtained per census tract, a hotspot

28

28
analysis was run to find the statistically significant hot and cold spots by census/ tract across the
city.
The second step in this project will be to determine the relationship of certain social
indicators and the census tracts with the highest number of people exposed to LSLs. The social
context indicators most widely researched in environmental justice issues are low-income and
racial minority groups (Alexeeff et al. 2012). Based on a compilation of socioeconomic
indicators used in other environmental justice studies in Milwaukee, the final list of explanatory
variables was as follows: median household income, poverty status within the last twelve
months, tenure status of the property (is it being rented or is it owned by its occupants), race, and
level of education. An exploratory ordinary least squares regression analysis was then conducted
on the dependent variable (the number of people at-risk for lead poisoning in each census tract)
with the above listed socioeconomic indicators as the explanatory variables. The variables
exhibiting high multicollinearity were thrown out of the model and the remaining explanatory
variables were run through a local bivariate relationship analysis to visualize each independent
variable’s relationship with the dependent variable.
3.2 Data Description and Data Processing
The city lead pipe data was obtained from the water department’s website, which keeps
up-to-date records of the existing pipes. These addresses were geocoded using ArcGIS Pro’s
geocoding service and mapped as a dot density layer (Figure 4). Of the 74,225 addresses that
were converted to points, two landed outside of the city boundary and were thus excluded from
the study. ArcGIS Pro’s geocoding service placed all points on the street facing edge of the
parcel boundaries. On occasion, the point was not quite inside the parcel boundary. For this

29

29
reason, a five-foot buffer was applied to the parcels when spatially joining the LSL points to the
parcel polygons where if the point fell inside of the five-foot buffer, it was joined to the parcel.

Figure 4: Dot density of residential addresses with lead service lines
The census tract boundaries were obtained from the Milwaukee County’s open data
portal in the form of a polygon shapefile. The tracts that were within Milwaukee’s city
boundaries were selected out from the county shapefile. It was found that two of the 211 census
tracts contained no residents so those two were excluded from the study, leaving a total of 209

30

30
tracts used in this study. City boundary and parcel data were obtained from the city’s open data
portal. The parcel layer contained data about the parcel, including a field with the total building
area of the property. The “Building_Area” field was used for the dasymetric mapping portion of
the methods. The “Building_Area” is an attribute found in the parcel layer file and it is defined
as “the total usable floor area of the structure in square feet” (City of Milwaukee 1999, 10). The
residential parcels were selected out of the parcel layer using the select by attribute tool in
ArcGIS Pro.
The census tract population estimates and social indicator data – median household
income, poverty status, education level, race, and tenure – were obtained from the Census
bureau’s website using the American Community Survey (ACS) formatted as comma-separated
values (csv) files. The race and education level indicators had data tables that contained several
different categories within the files. The race domain was broken up into individual races that
people identify as, and those of Hispanic origin were recorded in a separate dataset all together.
In an effort to reduce the number of redundant variables within the study, the data for all non-
white races and Hispanic origin were combined into a people of color (POC), or non-white,
category. So, the race indicator in this study was comprised of two variables, white and non-
white. The education level dataset also contained a multitude of categories where the ACS
created separate columns of data for each level of education attained – from kindergarten to
doctorate degrees. The indicator variable used in this study was the number of people with a high
school degree or less. To create this high school or less education category, the data were
combined in excel from kindergarten to high school or high school equivalent. All the variable
data used in the analysis were extracted from each table and joined to a master csv file with all

31

31
the social indicators. This master csv file was then combined with the map of the city’s census
tracts via a table join. Table 1 lists all the data and their sources that were used in this project.

32
Table 1: Datasets and Sources

Dataset Description Format Data Type
Spatial
Scale
Reporting
Period
Source
City Boundary
Administrative boundary
of the city of Milwaukee
.shp
Vector data
(polygon)
City limits 2021 Milwaukee Open Data
Parcel Polygons
Shapefile containing the
city’s parcel polygons with
master property file data
.shp
Vector Data
(polygon)
City Parcels 2021 Milwaukee Open Data
Census Tracts
Boundaries of Milwaukee
County’s census tracts
.shp
Vector Data
(polygon)
Census
tracts
2020
Milwaukee County Land
Information Office
Population Estimate
Dataset reporting estimated
population size
.csv
Aggregated
census tract
population
Census
Tract
2019
U.S. Census Bureau ACS
data
Education Level over 25
Dataset reporting estimated
education level attainment
for individuals over the age
of 25
.csv
Aggregated
census tract
population
Census
Tract
2019
U.S. Census Bureau ACS
data
Median Income
Dataset reporting estimated
median income level per
household
.csv
Aggregated
census tract
population
Census
Tract
2019
U.S. Census Bureau ACS
data
Poverty Level
Dataset reporting estimated
household poverty status
within the last 12 months
.csv
Aggregated
census tract
population
Census
Tract
2019
U.S. Census Bureau ACS
data
Race
Dataset reporting estimated
numbers of individuals
identifying within racial
categories
.csv
Aggregated
census tract
population
Census
Tract
2019
U.S. Census Bureau ACS
data
Hispanic/Latino Origin
Dataset reporting estimated
individuals that identify as
Hispanic or Latino origin
.csv
Aggregated
census tract
population
Census
Tract
2019
U.S. Census Bureau ACS
data
Lead Service Lines
Master list of residential
addresses within the city
with LSL
.csv
Point data
with text field
addresses
City Parcels 2021 Milwaukee Water Works

33
33
3.2.1 American Community Survey Margin of Error
To address the margin of error in the ACS data, a coefficient of variation (CV) was
calculated for each variable dataset at a 90% confidence level using methodology outlined by
Tufts GIS Center (Parmenter and Lau 2013).The coefficient of variation is a metric used to
establish the reliability of the ACS data estimate and was calculated for all 209 census tracts in
Milwaukee. First, the standard of error (SE) for each tract estimate was calculated using
Equation 1.
SE = MOE ÷ 1.645 (1)
Once the SE was calculated for each census tract variable’s estimate, the CV was derived using
Equation 2 (Parmenter and Lau 2013).
CV =(01 ÷12345637) × 100 (2)
In the cases of the number of non-white people and those with or less than a high school
education, the data for these two variables were aggregated from smaller categories within each
of their respective datasets. The MOEs for the aggregated data variables were calculated using
the formula in Equation 3, where the letter ‘c’ represents each individual data estimate to be
included in the aggregation (CCRPC 2015).

MOE
!""
= ±<∑ >?1
#
$
#
(3)
The CVs were then categorized into high, medium, and low reliability. The thresholds used for
each were taken from ESRI’s guidelines where high reliability CV scores are anything less than
or equal to 12, medium CV scores are between 12-40, and low reliability scores are anything
above 40 (Herries 2021). The totals for each reliability category were then added together and
recorded as percenta0ges. Table 2 shows all the explanatory variables taken from the ACS and
their respective CV reliability percentages.

34
34

Table 2: Coefficient of Variation Reliabilities for ACS Data
CV
Categories
Population Rents
Median
Income
Non-
White
Owns White
HS
Education
Poverty
Status
High 93.78% 73.68% 54.55% 50.24% 45.19% 43.27% 23.92% 1.91%
Medium 6.22% 25.84% 44.02% 49.76% 51.92% 42.79% 74.16% 91.39%
Low 0.00% 0.48% 1.44% 0.00% 2.88% 13.94% 1.91% 6.70%
3.3 Analysis Tasks Details
The methodology for the analysis tasks of this project were broken into two main
sections. The first being the CEDS mapping methodology to estimate the number of
Milwaukeeans at-risk more accurately for lead poisoning based on the presence of LSL at their
place of residence. The second part of the study examined the relationship between the number
of people at-risk for lead exposure per census tract with social indicators obtained from the
Census Bureau’s American Community Survey data. The idea is to quantify the characteristics of
census tracts that have a higher number of individuals at-risk for lead exposure. A bivariate
correlation matrix was constructed using Microsoft Excel to determine the nature of the
relationships between the independent, social indicator variables and the dependent, number of
people at-risk variable.
3.3.1 Cadastral-based Expert Dasymetric System
The Cadastral-Based Expert Dasymetric System (CEDS) methodology was used in this
study to create a dasymetric population distribution estimate in the City of Milwaukee. This
method disaggregates the census tract data and redistributes the population estimates by
assuming the number of individuals residing in a parcel is proportionate to the size of the
building area within that parcel. It should be noted that in the original methodology put forth by
Maantay et al. (2007), the spatial unit for disaggregation was tax lot data. In this study, parcel

35
35
information is used as a synonym for tax lot because the data obtained from city websites were
named “parcels” rather than “tax lot.”
In ArcGIS Pro, the ACS population estimates per census tract layer were joined by table
to the county census tract shapefile. Census tracts within the city boundaries were selected out
and created into a separate layer. In the city parcels shapefile, the parcels coded “residential”
were selected out and spatially joined to the city census tracts layer. In the spatial join, the merge
rule for the parcels layer was set to “sum” for the “Building_Area” (BA) field.

Figure 5: First spatial join between parcels and tracts with BA summed
The city census tracts with the BA Summed field were then spatially joined back down to the
parcels layer.

Figure 6: Second spatial join between new tracts and parcels with “BA_Sum”
A new field called “Building_Area_Proportion”(BA_Prop) was created and the calculate field
tool was used to divide each individual parcel’s “Building_Area” by the tract’s “BA_Sum.”
BA_Prop = BA / BA_Sum (1)
A new field, “Parcel_Population”(Parcel_Pop) was also created and calculated by multiplying
the BA_Prop value with the tract’s population (Tract_Pop) field to yield the estimated population
per parcel.

36
36
Parcel_Pop = BA_Prop * Tract_Pop (2)
The lead pipe address point layer was then joined to the parcel layer containing the
“Parcel_Population” field with another spatial join and the parcels containing lead pipes were
selected into a new layer.

Figure 7: Process for layer with only parcels with LSL
A “sum” summarize field calculation with run for the Parcel_Pop field in the
“Only_Parcels_with_LSL” layer to get the total estimated number of people at-risk for lead
exposure. Parcels containing the parcel population and with a lead pipe associated address were
spatially joined back up to the census tracts with ACS data layer with the merge rule for
Parcel_Pop being “sum.”

Figure 8: Final spatial join to yield total population at-risk for lead exposure in each census tract
The last join yielded the final result of the dasymetric map, visualizing the estimated
number of people at-risk for lead poisoning due to the presence of lead pipes per census tract.

37
37
The field containing the estimated people at-risk was summed to yield the overall total of people
at-risk for lead exposure for the whole city.

Figure 9: Final workflow of the CEDS method
A Getis-Ord Gi* statistic, or hotspot analysis, was then run to highlight the statistically
significant hot spots and cold spots of the at-risk census tracts. The tool examines each feature
within a neighborhood of features to determine statistically significant hot and cold spots. A
calculation of the local sum of the feature values is compared with the total sum of all the

38
38
features in the study area. If the difference between the local and total sums is great enough to
not be considered a random occurrence, a statistically significant z-score is calculated. ArcGIS
Pro has an Optimized Hotspot Analysis geoprocessing tool which automatically calculates the
best settings, such as an appropriate distance band value, to produce the most accurate result. The
positive significant z-scores represent hot spot clusters amongst the features, and the negative
significant z-scores represent cold spots. For the purposes of this study, hot spots would
represent areas of high numbers of people at-risk for lead exposure, while cold spots would
represent areas of low numbers of people at-risk.
3.3.2 Bivariate Correlation Analysis
To establish the characteristics of the census tracts with the highest number of people at-
risk for lead exposure, a bivariate regression analysis was run for seven social indicator
explanatory variables obtained from the Census Bureau’s ACS data.
An exploratory ordinary least squares analysis was conducted to assess multicollinearity
amongst the explanatory variables, as well as to see if the data were spatially autocorrelated.
Because OLS assumes normal distribution, the data of the dependent and all seven independent
variables were checked for a normal distribution curve. The at-risk population dependent
variable, median household income, poverty percentage, percentage of white residents, and
percentage of owner residents’ data were found to be not normally distributed. The square-root
transformation was applied to all five of the not normally distributed datasets because it more
accurately corrected for the skew in the data than a logarithmic transformation. While OLS is
used to model explanatory variable relationships with a dependent variable, the OLS tool in
ArcGIS Pro’s spatial statistics toolbox also includes VIF calculations for each individual
explanatory variable. Redundancy amongst the explanatory variables can lead to difficult to

39
39
interpret results as well as the risk of overfitting the model. The OLS model calculates variance
inflation factor (VIF) values for each of the explanatory variables to measure any redundancy
(ESRI 2021). If the VIF value is too high, it would suggest said variable has multicollinearity
with another variable. The suggested cutoff for this value is anywhere from 5-10 (Craney and
Surles 2002). The OLS analysis revealed minimal multicollinearity amongst the explanatory
variables as all had VIF values within the acceptable less than 10 range.
A Global Moran’s I test was run on the residuals of the OLS model to check if the
variables were spatially autocorrelated. K nearest neighbors was selected as the
conceptualization of spatial relationships parameter with 25 being set as the number of
neighbors. The number of neighbors was derived by running the average nearest neighbor tool in
ArcGIS Pro on the studied census tracts. The results of the Moran’s I showed the residuals were
clustered and therefore the variables were spatially autocorrelated.
After the exploratory OLS analysis, a bivariate correlation matrix using Pearson’s r was
constructed for the dependent and seven independent variables in Excel. Two matrices were
constructed, one with the explanatory variables inputted as proportions per census tract, and
another with the actual numbers for each explanatory variable. The only explanatory variable not
calculated as a proportion was median household income. While the at-risk population data were
obtained through dasymetric mapping, there is no way of knowing the same distribution
information with the explanatory variable data. Therefore, the proportions are used to
characterize the nature of each census tract and how these characterizations relate to the number
of people at-risk. Equation 4 is the Pearson’s r formula where r is the correlation coefficient, @
%

represents the values of the x-variable, @̅ is the mean of the x-variable values, B
%
is the y-variable
values, and B C is the mean of the y-variable values.

40
40
D =
∑((
!
) (̅)(,
!
) , -)
.∑((
!
) (̅)
"
∑(,
!
) , -)
"
(4)
The correlation matrix computes the Pearson’s r coefficient for each variable, dependent and
independent, against each other thus calculating the type of linear relationship between each
variable (Lane et al. 2013).

41
41
Chapter 4 Results
The threat of lead poisoning due to exposure from lead service lines is an issue that thousands of
Milwaukeeans face daily. Dasymetric mapping methods help to more accurately display
population distribution over a given aerial unit by disaggregating the population data and
transferring it into a finer unit of analysis. The methodology explored in this study is the
cadastral-based expert dasymetric system which uses breaks down census population estimates
into tax lot data. The results of the dasymetric map highlighting which census tracts have the
greatest number of people at-risk for lead poisoning are shown in this chapter.
Cases of environmental injustices have been seen time and time again in the historically
segregated city of Milwaukee. As shown in Chapter 2, black and brown communities in
Milwaukee exhibit higher rates of poverty, lack of access to healthcare, and are more likely to be
exposed to environmental toxins. To determine if the topic of lead exposure is indeed another
environmental justice issue, bivariate correlation analyses were conducted on seven
socioeconomic variables to examine their relationship with the number of at-risk people for lead
exposure in each census tract. The results of the analyses are explained in this chapter.
4.1. Dasymetric Mapping Results
The results of the dasymetric map of number of people at risk for lead poisoning per
census tract are based on the estimated number of residents living in a building that contains a
lead service line. The rationale behind a dasymetric approach being often the generalized results
of census tract data can mask an underlying environmental injustice. Figure 10 shows the
estimates of people at-risk for lead exposure per census tract, while Figure 11 is a visualization
of the at-risk population percentages across census tracts. A graduated color scheme was used
with natural breaks to symbolize both maps. The results are nearly identical between the two

42
42
maps. The final total for the estimated at-risk population for the entire city was 296,327.13
residents. The estimated total population of the city is 594,772 people. According to these
findings, 49.8% of Milwaukee’s residents are getting their water from LSL and therefore are at
higher risk for lead poisoning.

Figure 10: Dasymetric map of the estimated people at-risk for lead exposure by census tract.

43
43

Figure 11: Dasymetric map of the percentage of at-risk people per census tract.
Just by looking at the two maps, it is clear that there are certain areas in the city with
higher numbers of people at-risk for lead exposure. However, a statistical significance test of
these hot and cold spots is useful to support the initial observation. The result of the optimized
hotspot analysis run on the at-risk population estimate per census tract map can be seen in Figure
12. The statistically significant cold spots symbolized in shades blue, hot spots in shades of red,
and not significant in white.

44
44
Most of the cold spots can be found where there are no lead service lines present. The
cold spots seen in the middle of the city are in the Lower East Side, Haymarket, Yankee Hill, and
Juneau Town neighborhoods (Figure 12). When compared with the dasymetric at-risk population
map, these neighborhoods also have a lower number of people exposed to LSL. The hotspots in
the city can be seen on the northern and southern sides of the city center. The hotspots can also
be seen to have higher numbers of people exposed to LSL when compared with the dasymetric
maps. Figure 12 also labels the neighborhoods that intersect with the hot and cold spot regions of
the city.

Figure 12: Neighborhoods in the city that intersect with the hot and cold spots

45
45
4.2. Bivariate Correlation Analysis Results
In order to characterize the census tracts with high numbers of people at-risk, seven racial
and socioeconomic variables were assessed to determine their relationship with the dependent
variable. As explained in the methods, an OLS analysis was conducted as the initial exploratory
step to quantify the nature of the relationship between the dependent and independent variables
as well as assess any bias between the explanatory variables themselves. The results of the
analysis revealed strong multicollinearity between some of the explanatory variables and an
overall model bias as evidenced by a p < 0.01 Jarque-Bera statistic. When the residuals were run
through a Moran’s I test, the p-value was found to be statistically significant, and the z-value was
positive indicating the residuals are not dispersed at random but rather are found to be spatially
clustered. Therefore, the residuals from the OLS were found to be spatially autocorrelated
rendering the model a bias one.
Before the bivariate correlations were calculated, descriptive statistics of the minimum,
maximum, mean, and standard deviation for each of the variables were determined and can be
viewed in Table 3.

Table 3: Descriptive statistics for all variables
Variable Minimum Maximum Mean Standard Deviation
At-Risk Pop 0.00 5174.17 1417.83 1251.22
Non-White 6.76 100.00 64.83 27.06
HS Education 0.59 47.67 28.61 10.45
Owns 0.00 96.47 40.19 19.47
Rents 3.53 100.00 59.81 19.47
Poverty Status 2.20 59.92 26.40 14.03
White 0.00 93.24 35.17 27.06
Median Income 7917.00 113375.00 42320.37 17714.77

46
46
From there, the explanatory variables were thematically mapped against the dependent variable
to provide an initial visualization of geographic patterns between each individual independent
and the dependent variable. In these bivariate maps, the darker red regions show where the
dependent and independent variable values were both high, and the lighter areas where both
values were low. An additional element of transparency was added in the symbology where
census tracts with higher populations were opaquer and those with lower populations more
transparent (Figure 13). Each of the variables were symbolized in tertiles with cutoff values one
standard deviation more and less from the mean of the dataset.

47
47

Figure 13: Bivariate maps displaying each independent variable with the dependent variable

48
48

Figure 14: Legend showing the standard deviation cut-off values for each variable
The results of the bivariate correlation matrices can be seen in Tables 4 and 5. The first
matrix (Table 4) was calculated using actual population estimates while the second matrix (Table
5) was calculated using the proportions of the population estimates over the total population in
census tract for the explanatory variables. The values with a color background are the Pearson’s r
coefficients, while the values with a white background are the corresponding p-values. Values
where p ≤ 0.05 are denoted with a single asterisk and those where p ≤ 0.001 are denoted with a
double asterisk. The two correlation matrices are nearly identical with which variables have
positive and negative relationships with the at-risk population according to their Pearson’s r
coefficient value. When comparing the at-risk population dependent variable with the seven
explanatory variables, the only variable that differed between the two matrices was the white
population data. When determining the relationship between the number of at-risk people with
the number of white people, there is a weak positive correlation.

49
Table 4: Correlation matrix between explanatory and dependent variables

HS MedianIncome Poverty White Non-White Owns Rents At-Risk_Pop
HS 1.0000 0.151 0.000** 0.001** 0.0000** 0.0000** 0.0002** 0.005*
MedianIncome -0.0997 1.0000 0.0000** 0.0000** 0.0000** 0.0000** 0.108 0.138
Poverty 0.5428 -0.5685 1.0000 0.010 0.000** 0.051* 0.000** 0.001**
White 0.2376 0.6298 -0.1774 1.0000 0.150 0.000** 0.002 0.3562
Non-White 0.8181 -0.2995 0.7321 -0.0999 1.0000 0.003* 0.000** 0.000**
Owns 0.4996 0.5954 -0.1352 0.6452 0.2066 1.0000 0.423 0.611
Rents 0.2532 -0.1116 0.4512 0.2121 0.3399 0.0557 1.0000 0.393
At-Risk_Pop 0.1932 -0.1030 0.2370 0.0641 0.3061 -0.0353 0.0594 1.0000

Table 5: Correlation matrix between explanatory variable percentages and the dependent variable

Column1 HS MedianIncome Poverty White Non-White Owner Rent At-Risk_Pop
HS 1.0000 0.000** 0.000** 0.000** 0.000** 0.793 0.793 0.043
MedianIncome -0.4745 1.0000 0.000** 0.000** 0.000** 0.000** 0.000** 0.138
Poverty 0.3942 -0.8331 1.0000 0.000** 0.000** 0.000** 0.000** 0.122
White -0.5829 0.7250 -0.6494 1.0000 NA 0.000** 0.000** 0.342
Non-White 0.5829 -0.7250 0.6494 -1.0000 1.0000 0.000** 0.000** 0.342
Owner -0.0182 0.6839 -0.6603 0.3458 -0.3458 1.0000 NA 0.431
Rent 0.0182 -0.6839 0.6603 -0.3458 0.3458 -1.0000 1.0000 0.431
At-Risk_Pop 0.1403 -0.1030 0.1074 -0.0661 0.0661 -0.0548 0.0548 1.0000

50
Meanwhile, the relationship with the variable when it is represented as a percentage of the
population, it is shown to have a weak negative correlation with the dependent variable.
This could be because there are high numbers of white residents in census tracts across the city,
but proportionally, there are certain census tracts where white people are the overwhelming
minority. Using percentages to represent the explanatory demographic data can standardize the
difference in estimates across the categories. None of the p-values for the coefficients calculated
for the explanatory variable percentages and the dependent variable returned a significant value.
However, in the real number correlation results, high school education, poverty, and the non-
white population variables were all shown to have statistically significant relationships with the
dependent variable.
When comparing the explanatory variables against each other, there were a few more
differences between the two matrices in respect to positive and negative linear relationships.
Further, the strength of the linear relationships between certain variables shifted between the real
numbers and percentages. One notable example is the relationship between those with a high
school education or less and the number of people who own their place of residence. In the real
numbers, the two variables have a positive linear relationship with a Pearson’s r coefficient of
0.4996. However, when the numbers are converted to percentages, the relationship between the
two variables have a slightly negative correlation (Pearson’s r = -0.0182). The shift not only in
the direction of the relationship from positive to negative, but also in the strength of correlation
is a testament to the difference calculating Pearson’s r with real numbers versus percentages.
The results of the dasymetric mapping and analysis revealed patterns similar to other
environmental justice research with toxin exposure and social indicators. In depth explanations
and implication of the results will be discussed in Chapter 5 of this paper.

51
Chapter 5 Conclusion
The two main questions that this project sought to answer were one, how many people run the
risk of being exposed to lead due to the presence of LSL at their place of residence, and two,
what is the relationship between SES indicator variables and the number of people at-risk per
census tract. In response to the first question, the results of the dasymetric map of lead exposure
risk throughout the city based on presence of LSL reveal drastic differences in number of people
at-risk per census tract. There are significant hot and cold spots of exposure by census tract. The
results of the regression analyses helped shed light on the answers to the second question. Non-
white individuals, those with a high school education, and those who have experienced poverty
have a higher chance of being exposed to lead contaminated water.
In the following chapter, the implications of the dasymetric map are discussed as well as
the results from the bivariate correlation analyses. Understanding how one’s race and
socioeconomic status in the community can raise the likelihood of lead exposure is a crucial step
in the fight for environmental equity in Milwaukee. The results of this project represent just the
beginning of understanding lead exposure due to LSL as an EJ issue in Milwaukee. The
limitations, and there were many, are reflected on and recommendations for future research are
presented at the end of the chapter.
5.1. Dasymetric Map of At-Risk Populations
The results of the dasymetric map paint a more precise picture of how many people are at
risk for lead exposure due to the presence of lead pipes on their property. Upon visual inspection,
there are certain census tracts that have a much higher number of people exposed to LSL, and
these high-risk census tracts are geographically close together. Upon conducting a Hot Spot
analysis, the hot spots and cold spots for LSL exposure are confirmed. As discussed in chapter 4,

52
the estimated at-risk population for the entire city is 296,327.13 residents out of a total
population of 594,772. Based on these population estimates, 49.8% of Milwaukee’s residents are
at-risk for lead exposure due to living in a residence linked to a lead service line. Nearly 50% of
the city’s population being at-risk for lead exposure seems like a large proportion considering
MWW’s claims that 40% of active service lines are made of lead. The final total of at-risk
residents of Milwaukee being greater than the percentage of LSL in the city is confirmation that
population density greatly varies in certain census tracts over others.
The method of disaggregating the population data and redistributing the population
estimates (as a function of proportion of building area within each parcel) yielded a more precise
estimate of at-risk people due to the fine spatial resolution of each parcel within a census tract.
The methods used in this study were a slight variation on the CEDS method outlined by Maantay
et al. (2007) due to differences in data availability. The residential unit and adjusted residential
area values put forth in the original methodology were swapped for building area data found in
the Milwaukee parcels schema. The building area per parcel served as the proxy unit for
population dispersion and the ratio of each parcel’s building area to the census tract’s total
building area yielded the estimated number of people living in each individual parcel. Because
the tract data could be disaggregated down to such a fine resolution, the nuances of population
variance within each census tract were accounted for when totaling the number of individuals
living in buildings with LSLs. Despite the slight variance in the methodology, the CEDS
technique of dasymetric mapping is a highly effective way to redistribute population to reflect
real-world situations more accurately in Milwaukee (Maantay et al. 2007).
When examining the hot spot map, it is evident that the hot spots of the most people at-
risk for lead poisoning can be found in the northern and southern parts of the city center.

53
Focusing in on the northern neighborhoods such as Harambee, Park West, Sherman Park, and
Franklin Heights, it can be seen that the these historically black neighborhoods are at the center
of the hot spot area. Franklin Heights for example, saw a major rise in its African American
population in the 1980s after the construction of the freeway led to more affluent, white residents
to move away from the area. Today, it is one of the poorer neighborhoods in the city with a
primarily black population (Gurda 1999; Nelsen 2016). The southern hot spots are found
primarily in the Hispanic neighborhoods of the city. Specifically, Walker’s Point, Clark Square,
Silver City, Burnham Park, Muskego Way, Historic Mitchell Street, Lincoln Village, Polonia,
Southgate, and Layton Park all have majority Latinx populations (Johnson 2020). Many of these
neighborhoods also have lower than average median household income when compared to
Milwaukee as a whole and have less than a high school education (City-Data 2019). There is an
isolated cold spot found in the middle of the city that intersects the Lower East Side, Haymarket,
Yankee Hill, and Juneau Town neighborhoods. This is a far more affluent part of Milwaukee,
with a majority white population, people with college degree educations, and median household
incomes more than double that of the city overall (City-Data 2019). The demographic statistics
of the neighborhoods that intersect with the hot and cold spots in Milwaukee closely reflect those
obtained in this study and will be explored further in section 5.2 of this chapter.
5.1.1. Considerations and Limitations
The dasymetric methods used in this study were a solid starting point for assessing at-risk
populations within the city. However, there were a few considerations and limitations that should
be acknowledged. A major limitation was access to cadastral metadata. While the City of
Milwaukee makes their data easily accessible for anyone to download, deciphering the coded
schema often proved a challenge. Metadata documents were found in separate locations than the

54
data portal and were unclear in referencing field codes between the documentation and the actual
attribute table. An example is the metadata document for zoning codes for all the parcel data in
the city. Beyond the general “residential” designation, according to the zoning ordinances
volume two subchapter five, the residential zones within the city were broken down into 22
subcategories. These subcategories would distinguish between single family, two family, multi-
family, etc. places of residence, but these coded categories would not always be present in the
parcel attribute data (Legislative Reference Bureau 2020). Given the limited timeframe of this
study, it was not possible to relate the zoning and parcel tables to then cross-reference the
specific zoning code to estimate population distribution. Further, just because a residence is
designated as a single-family home does not mean a single family is occupying the property. One
of the indicators of a lower socioeconomic status is overcrowding in one’s place of residence
(Galobardes 2006).
5.2. Racial and Socioeconomic Correlations
One must be vigilant when conducting research on potential correlations between
dependent and explanatory variables. As the saying goes, correlation does not necessarily mean
causation. The purpose of performing the bivariate correlation analyses in this study was to
illuminate some of the social indicators that could explain the number of at-risk people exposed
to lead in each census tract. To perform this analysis, a set of explanatory variables were selected
from existing literature studying socioeconomic disparities within Milwaukee. Based on the
background research, the explanatory variables selected for this study were median household
income, poverty status, educational attainment, tenure status, and race. According to the resulting
correlation matrices, those with a high school education, experiencing poverty, and non-white
residents all have statistically significant positive linear relationship with the number of at-risk

55
people (Tables 4 and 5). Meanwhile, the number of white residents and wealthier census tracts
where people are property owners are shown to have negative linear relationships (Tables 4 and
5). The visualization of these relationships can be seen in the bivariate map series where each
explanatory variable was mapped with the dependent variable. The overlapping variables with
positive linear correlations follow a similar pattern to that of the hotspot analysis map. The
results are further confirmed when examining the demographic characteristics of the
neighborhoods that intersect the hot and cold spots described in section 5.1. Therefore, it can be
concluded that the census tracts with higher numbers and percentages of non-white individuals,
lower education, and low socioeconomic status are more likely to have high numbers of people
at-risk for lead exposure due to an LSL at their place of residence.
It tracks that the number of non-white people, high school education level, and poverty
status would be the variables with the most impact on the number of at-risk individuals per
census tract. The three variables are shown to have strong positive linear relationships amongst
each other. The overlap between race and socioeconomic status is a well recorded phenomenon
in the United States (Raddatz and Mennis 2013). In the case of Black Americans, it has been
found that indicators such as income level, education, and occupation are not as strong at
predicting a Black American’s socioeconomic status as it would a White American’s. This is
most likely due to a general lack of intergenerational wealth and not seeing the same financial
returns for higher education than White Americans (Wolff et al. 2010). In other words, the
strongest predictor of a Black American’s socioeconomic status would be the fact that they are
Black and navigating a world that is systemically not made for their benefit. In Milwaukee’s
case, historically racist policies such as redlining shaped the urban geographic landscape that is
seen today where neighborhoods are for the most part still segregated by race. Discriminatory

56
housing markets drove people of color, those of whom were primarily black, to settle in regions
of the city where the buildings were older and therefore more likely to have LSL (Trotter 1985).
The ripple effect of discriminatory legislation from years ago can be seen in health disparities,
lack of social mobility, and increased exposure to environmental toxins within Milwaukee today.
The results of this study confirm this pattern of inequity amongst poorer, less education
communities of color within the city.
5.2.1. Considerations and Limitations
A limitation in this study was the issue of the margin of error in the ACS census tract
data. It was found that the more people surveyed in the census tract, the lower the margin of
error, while the inverse was true when fewer people were surveyed. While most of the tracts had
MOE within an acceptable range, there were a few outliers with low population counts that
caused the MOE to become very large. The coefficient of variation was calculated for each ACS
demographic category to disclose which had higher levels of error (Table 2). The variable with
the highest percentage of low reliability CV scores was the estimates for the number of white
people per census tract. In future iterations, steps could be taken to further reduce the MOE by
aggregating contiguous census tracts in areas with hot and cold spots.
5.3 Future Study Recommendations
The results of the current study were sufficient as an initial step when examining the
environmental justice issue of populations within Milwaukee exposed to lead pipes. The
questions that this study uncovered could lead to a plethora of future research in this field. For
example, it would be interesting to construct a socioeconomic index using principal component
analysis (PCA) to select the indicator variables from a wider range of domains. The advantage of
this method is it keeps the decision making to a minimum and uses statistics to select out the best

57
fit variables for an SES index. Once an index is created, values can be assigned to each census
tract based on how their SES index scores. The values can then be used to compare SES with the
at-risk for lead exposure population data using a generalized additive model. Such methodology
is outlined in Padilla et al. and Lalloué et al., who both researched environmental justice issues in
relation to air pollutants in France (Lalloué et al. 2013; Padilla et al. 2014). It would also be
interesting to compare lead exposure between lead paint and LSL within Milwaukee. According
to a grassroots coalition called Lead Free Milwaukee, city officials have been pressing the
narrative that lead paint is the primary source for lead exposure as opposed to lead service lines
leaching the metal into residents’ drinking water. The coalition asserts that the city is
downplaying the severity of the issue of lead in the water and is trying to redirect attentions and
efforts to mitigating lead paint (Washington and Welcenbach 2019). Finally, given the recent
revisions to the Lead and Copper Rule by the EPA, a study on the efficacy of partial versus full
lead service line replacements on reducing lead concentrations in the water supply could be
useful to residents of the city. When the city replaces an LSL, usually a partial replacement
approach is carried out where the city will replace only the public portion of the LSL, and leave
the private portion as is. Research on the effects of partial service line replacements has shown
that in only replacing part of the LSL, more lead is found in the water supply due to the
disturbance of the service line (Lewis et al. 2017).
The fight for environmental equity is a long, uphill battle. In characterizing the
neighborhoods that experience the highest levels of exposure, decision makers in Milwaukee will
know where to focus their mitigation efforts and activist groups can know where to go to educate
people about the dangers of lead exposure. GIS is a powerful tool that can contribute to the
growing research on EJ through techniques such as dasymetric mapping and correlation analysis.

58
By conducting these studies, geospatial scholars can aid in giving a voice to the voiceless. The
fruits of their labor can empower those who want to see justice served and insure healthy
communities worldwide.

59
References

Alexeeff, George V., John B. Faust, Laura Meehan August, Carmen Milanes, Karen Randles,
Lauren Zeise, and Joan Denton. 2012. “A Screening Method for Assessing Cumulative
Impacts.” International Journal of Environmental Research and Public Health 9 (2): 648–
59. https://doi.org/10.3390/ijerph9020648.
Anderton, Douglas L., Andy B. Anderson, John Michael Oakes, and Michael R. Fraser. 1994.
“Environmental Equity: The Demographics of Dumping.” Demography 31 (2): 229–48.
https://doi.org/10.2307/2061884.

Banzhaf, Spencer, Lala Ma, and Christopher Timmins. 2019. “Environmental Justice: The
Economics of Race, Place, and Pollution.” Journal of Economic Perspectives 33 (1): 185–
208. https://doi.org/10.1257/jep.33.1.185.

Berkley, Jennifer. 2017. “Using American Community Survey Estimates and Margins of Error
Webinar - Transcript.” Decennial Statistical Studies Division, April 19.
https://census.webex.com/census/ldr.php?RCID=714c57425cd24a5b15fb4eea3da438ec.
Brown, Mary Jean, and Stephen Margolis. 2012. “Lead in Drinking Water and Human Blood
Lead Levels in the United States.” Supplement 61. Morbidity and Mortality Weekly
Report. Centers for Disease Control and Prevention.
Butler, Lindsey J., Madeleine K. Scammell, and Eugene B. Benson. 2016. “The Flint, Michigan,
Water Crisis: A Case Study in Regulatory Failure and Environmental Injustice.”
Environmental Justice 9 (4): 93–97. https://doi.org/10.1089/env.2016.0014.
Cameron, Ryan. 2020. “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.” University of Southern California.
CCRPC. 2015. “Understanding and Using the U.S. Census Bureau’s American Community
Survey.” Champaign County Regional Planning Commission.

CDC. 2020a. “Community, Work, and School.” Centers for Disease Control and Prevention.
February 11, 2020. https://www.cdc.gov/coronavirus/2019-ncov/community/health-
equity/race-ethnicity.html.
CDC. 2020b. “Health Problems Caused by Lead | NIOSH | CDC.” September 10, 2020.
https://www.cdc.gov/niosh/topics/lead/health.html.
CDC. 2021. “Lead (Pb) Toxicity: What Are U.S. Standards for Lead Levels? | Environmental
Medicine | ATSDR.” February 9, 2021.
https://www.atsdr.cdc.gov/csem/leadtoxicity/safety_standards.html.
Census Bureau. 2020. “U.S. Census Bureau QuickFacts: Milwaukee City, Wisconsin.” 2020.
https://www.census.gov/quickfacts/milwaukeecitywisconsin.

60
Chaffetz, Jason, John L Mica, Michael R Turner, John J Duncan, Jim Jordan, Tim Walberg,
Justin Amash, et al. 2016. Examining Federal Administration of the Safe Drinking Water
Act in Flint, Michigan.

Chakraborty, Jayajit. 2011. “Revisiting Tobler’s First Law of Geography: Spatial Regression
Models for Assessing Environmental Justice and Health Risk Disparities.” In Geospatial
Analysis of Environmental Health, edited by Juliana A. Maantay and Sara McLafferty,
337–56. Geotechnologies and the Environment. Dordrecht: Springer Netherlands.
https://doi.org/10.1007/978-94-007-0329-2_17.
Chakraborty, Jayajit, Timothy W. Collins, Marilyn C. Montgomery, and Sara E. Grineski. 2014.
“Social and Spatial Inequities in Exposure to Flood Risk in Miami, Florida.” Natural
Hazards Review 15 (3): 04014006. https://doi.org/10.1061/(ASCE)NH.1527-
6996.0000140.
Chisolm, J. Julian. 1971. “Lead Poisoning.” Scientific American 224 (2): 15–23.
City-Data. 2019. “Lower East Side Neighborhood in Milwaukee, Wisconsin (WI), 53202
Subdivision Profile - Real Estate, Apartments, Condos, Homes, Community, Population,
Jobs, Income, Streets.” 2019. https://www.city-data.com/neighborhood/Lower-East-Side-
Milwaukee-WI.html.

City-Data. 2019. “Muskego Way Neighborhood in Milwaukee, Wisconsin (WI), 53204
Subdivision Profile - Real Estate, Apartments, Condos, Homes, Community, Population,
Jobs, Income, Streets.” 2019. https://www.city-data.com/neighborhood/Muskego-Way-
Milwaukee-WI.html.

City of Milwaukee. 1999. “City of Milwaukee Master Property File Documentation.”

City of Milwaukee. 2016. “Lead-Safe Milwaukee.” 2016.
https://city.milwaukee.gov/LeadSafeMKE.
Collins, Mary B. 2011. “Risk-Based Targeting: Identifying Disproportionalities in the Sources
and Effects of Industrial Pollution.” American Journal of Public Health 101 (Suppl 1):
S231–37. https://doi.org/10.2105/AJPH.2011.300120.
Dang, Do, Morgan Lively, and Alonzo Jalan. 2021. “Lead Poisoning and Racism in the Time of
COVID-19.” Wisconsin Medical Journal, March, 3.
Denver Water. 2021. “Lead Service Lines.” Denver Water. 2021.
https://www.denverwater.org/your-water/water-quality/lead/lead-service-lines.
Eicher, Cory L., and Cynthia A. Brewer. 2001. “Dasymetric Mapping and Areal Interpolation:
Implementation and Evaluation.” Cartography and Geographic Information Science 28
(2): 125–38. https://doi.org/10.1559/152304001782173727.
Emer, Lindsay R., Amy E. Kalkbrenner, Mallory O’Brien, Alice Yan, Ron A. Cisler, and Lance
Weinhardt. 2020. “Association of Childhood Blood Lead Levels with Firearm Violence

61
Perpetration and Victimization in Milwaukee.” Environmental Research 180 (January):
108822. https://doi.org/10.1016/j.envres.2019.108822.
ESRI. 2021. “How Hot Spot Analysis (Getis-Ord Gi*) Works.” 2021.
https://pro.arcgis.com/en/pro-app/latest/tool-reference/spatial-statistics/h-how-hot-spot-
analysis-getis-ord-gi-spatial-stati.htm.

ESRI. 2021. “Ordinary Least Squares (OLS) (Spatial Statistics)—ArcGIS Pro | Documentation.”
April 2021. https://pro.arcgis.com/en/pro-app/latest/tool-reference/spatial-
statistics/ordinary-least-squares.htm.
Galobardes, B. 2006. “Indicators of Socioeconomic Position (Part 1).” Journal of Epidemiology
& Community Health 60 (1): 7–12. https://doi.org/10.1136/jech.2004.023531.
Greer, Danielle M., Dennis J. Baumgardner, Farrin D. Bridgewater, David A. Frazer, Courenay
L. Kessler, Erica S. LeCounte, Geoffrey R. Swain, and Ron A. Cisler. 2013. “Milwaukee
Health Report 2013: Health Disparities in Milwaukee by Socioeconomic Status.” Center
for Urban Population Health: Milwaukee, WI.
Gulledge, Kelly W. 2020. “An Analysis of Racial Disparity in the Distribution of Alcohol
Licenses and Retailers in Orange County, California.” University of Southern California.
Gurda, John. 1999. The Making of Milwaukee. Milwaukee, Wis: Milwaukee County Historical
Society.

Hanna-Attisha, Mona, Jenny LaChance, Richard Casey Sadler, and Allison Champney Schnepp.
2015. “Elevated Blood Lead Levels in Children Associated With the Flint Drinking
Water Crisis: A Spatial Analysis of Risk and Public Health Response.” American Journal
of Public Health 106 (2): 283–90. https://doi.org/10.2105/AJPH.2015.303003.
Herries, Jim. 2021. “A Straightforward Approach to Mapping Margins of Error.” ArcGIS Blog
(blog). June 7, 2021. https://www.esri.com/arcgis-blog/products/arcgis-living-
atlas/mapping/a-straightforward-approach-to-mapping-margins-of-error/.

Heynen, Nik, Harold A. Perkins, and Parama Roy. 2006. “The Political Ecology of Uneven
Urban Green Space: The Impact of Political Economy on Race and Ethnicity in
Producing Environmental Inequality in Milwaukee.” Urban Affairs Review 42 (1): 3–25.
https://doi.org/10.1177/1078087406290729.
Hillier, Amy E. 2003. “Spatial Analysis of Historical Redlining: A Methodological Exploration -
ProQuest.” Journal of Housing Research 14 (1): 31.
Hinkel, Jochen. 2011. “‘Indicators of Vulnerability and Adaptive Capacity’: Towards a
Clarification of the Science–Policy Interface.” Global Environmental Change 21 (1):
198–208. https://doi.org/10.1016/j.gloenvcha.2010.08.002.
Hou, Shuangxing, Lianfang Yuan, Pengpeng Jin, Bojun Ding, Na Qin, Li Li, Xuedong Liu,
Zhongliang Wu, Gang Zhao, and Yanchun Deng. 2013. “A Clinical Study of the Effects

62
of Lead Poisoning on the Intelligence and Neurobehavioral Abilities of Children.”
Theoretical Biology and Medical Modelling 10 (1): 13. https://doi.org/10.1186/1742-
4682-10-13.
Jannene, Jeramey. 2020. “City Hall: Milwaukee on 70-Year Pace To Replace Lead Laterals.”
Urban Milwaukee. March 13, 2020. https://urbanmilwaukee.com/2020/03/13/city-hall-
milwaukee-on-70-year-pace-to-replace-lead-laterals/.
Jerrett, Michael, Richard T Burnett, Pavlos Kanaroglou, John Eyles, Norm Finkelstein, Chris
Giovis, and Jeffrey R Brook. 2001. “A GIS–Environmental Justice Analysis of
Particulate Air Pollution in Hamilton, Canada.” Environment and Planning A: Economy
and Space 33 (6): 955–73. https://doi.org/10.1068/a33137.
Johnson, John D. 2020. “John D. Johnson | Neighborhoods Where Milwaukee Isn’t Segregated.”
June 17, 2020. https://johndjohnson.info/post/milwaukee-neighborhoods-race-2018/.

Kelly-Reif, Kaitlin, and Steve Wing. 2016. “Urban-Rural Exploitation: An Underappreciated
Dimension of Environmental Injustice.” Journal of Rural Studies 47 (October): 350–58.
https://doi.org/10.1016/j.jrurstud.2016.03.010.
Kennedy, Erika, and Leslie W. Kennedy. 2004. “Using Dasymetric Mapping for Spatially
Aggregated Crime Data.” Journal of Quantitative Criminology 20 (3): 243–62.
https://doi.org/10.1023/B:JOQC.0000037733.74321.14.
Lalloué, Benoît, Jean-Marie Monnez, Cindy Padilla, Wahida Kihal, Nolwenn Le Meur, Denis
Zmirou-Navier, and Séverine Deguen. 2013. “A Statistical Procedure to Create a
Neighborhood Socioeconomic Index for Health Inequalities Analysis.” International
Journal for Equity in Health 12 (1): 21. https://doi.org/10.1186/1475-9276-12-21.
Lane, David M., David Scott, Mikki Hebl, Rudy Guerra, Dan Osherson, and Heidi Zimmer.
2013. Introduction to Statistics. Online. Rice University and University of Houston.

Legislative Reference Bureau. 2020. “Zoning 295-501 Subchapter 5 Residential Districts.” In
Milwaukee Code of Ordinances: Volume 2 Building and Zoning Code, 2:598. Common
Council.
Lester, James. 2018. Environmental Injustice In The U.S.: Myths And Realities. Routledge.
Lewis, Carrie M., Lon A. Couillard, Patricia J. Klappa, and Terrence D. Vandenbush. 2017.
“Lead Water Service Lines: Extensive Sampling and Field Protocol Protect Public
Health.” Journal - American Water Works Association 109 (January): 34–41.
https://doi.org/10.5942/jawwa.2017.109.0016.
Lynch, Emily E., and Helen C. S. Meier. 2020. “The Intersectional Effect of Poverty, Home
Ownership, and Racial/Ethnic Composition on Mean Childhood Blood Lead Levels in
Milwaukee County Neighborhoods.” PLoS ONE 15 (6).
https://doi.org/10.1371/journal.pone.0234995.

63
Maantay, Juliana. 2002. “Mapping Environmental Injustices: Pitfalls and Potential of
Geographic Information Systems in Assessing Environmental Health and Equity.”
Environmental Health Perspectives 110 (suppl 2): 161–71.
https://doi.org/10.1289/ehp.02110s2161.
Maantay, Juliana Astrud, Andrew R. Maroko, and Christopher Herrmann. 2007. “Mapping
Population Distribution in the Urban Environment: The Cadastral-Based Expert
Dasymetric System (CEDS).” Cartography and Geographic Information Science 34 (2):
77–102. https://doi.org/10.1559/152304007781002190.
Maantay, Juliana, and Andrew Maroko. 2009. “Mapping Urban Risk: Flood Hazards, Race, &
Environmental Justice in New York.” Applied Geography 29 (1): 111–24.
https://doi.org/10.1016/j.apgeog.2008.08.002.
Mayo Clinic Staff. 2019. “Lead Poisoning - Symptoms and Causes.” Mayo Clinic. December
2019. https://www.mayoclinic.org/diseases-conditions/lead-poisoning/symptoms-
causes/syc-20354717.
Mennis, Jeremy. 2003. “Generating Surface Models of Population Using Dasymetric
Mapping*.” The Professional Geographer 55 (1): 31–42. https://doi.org/10.1111/0033-
0124.10042.
Messer, Lynne C., Barbara A. Laraia, Jay S. Kaufman, Janet Eyster, Claudia Holzman, Jennifer
Culhane, Irma Elo, Jessica G. Burke, and Patricia O’Campo. 2006. “The Development of
a Standardized Neighborhood Deprivation Index.” Journal of Urban Health : Bulletin of
the New York Academy of Medicine 83 (6): 1041–62. https://doi.org/10.1007/s11524-
006-9094-x.
Milwaukee Environmental Justice Lab. n.d. “A Community / University Partnership Around
Emancipatory Storytelling and Transformative Actions.” MKEEJLAB. Accessed
February 15, 2021a. https://mkeejlab.weebly.com/.
Milwaukee Environmental Justice Lab. n.d. “Transit History.” Climates of Inequality. Accessed
February 13, 2021b. https://mkeejlab.weebly.com/transit-history.html.
Milwaukee Water Works. 2021. “Lead and Water.” 2021.
https://city.milwaukee.gov/water/WaterQuality/LeadandWater.
Miranda, Marie Lynn, Dana C Dolinoy, and M Alicia Overstreet. 2002. “Mapping for
Prevention: GIS Models for Directing Childhood Lead Poisoning Prevention Programs.”
Environmental Health Perspectives 110 (9): 947–53.
https://doi.org/10.1289/ehp.02110947.
Mohai, Paul, and Bunyan Bryant. 1992. “Environmental Injustice: Weighing Race and Class as
Factors in the Distribution of Environmental Hazards.” University of Colorado Law Review
63 (4): 13.

64
Mohai, Paul, and Robin Saha. 2006. “Reassessing Racial and Socioeconomic Disparities in
Environmental Justice Research.” Demography 43 (2): 383–99.
https://doi.org/10.1353/dem.2006.0017.

Munck, Victor C. de. 2005. “Units of Analysis.” In Encyclopedia of Social Measurement, edited
by Kimberly Kempf-Leonard, 899–907. New York: Elsevier. https://doi.org/10.1016/B0-
12-369398-5/00043-8.

Nelsen, James. 2016. “Franklin Heights.” Encyclopedia of Milwaukee. Retrieved September 2,
2021 (https://emke.uwm.edu/entry/franklin-heights/).

Oyana, Tonny J., and Florence M. Margai. 2007. “Geographic Analysis of Health Risks of
Pediatric Lead Exposure: A Golden Opportunity to Promote Healthy Neighborhoods.”
Archives of Environmental & Occupational Health 62 (2): 93–104.
https://doi.org/10.3200/AEOH.62.2.93-104.
Padilla, Cindy M., Wahida Kihal-Talantikite, Verónica M. Vieira, Philippe Rossello, Geraldine
Le Nir, Denis Zmirou-Navier, and Severine Deguen. 2014. “Air Quality and Social
Deprivation in Four French Metropolitan Areas—A Localized Spatio-Temporal
Environmental Inequality Analysis.” Environmental Research 134 (October): 315–24.
https://doi.org/10.1016/j.envres.2014.07.017.
Parmenter, Barbara M., and Janet Lau. 2013. “Estimating and Mapping Reliability for American
Community Survey Data.” Tufts GIS Center.

Potash, Eric, Rayid Ghani, Joe Walsh, Emile Jorgensen, Cortland Lohff, Nik Prachand, and Raed
Mansour. 2020. “Validation of a Machine Learning Model to Predict Childhood Lead
Poisoning.” JAMA Network Open 3 (9): e2012734.
https://doi.org/10.1001/jamanetworkopen.2020.12734.
Pulido, Laura. 2000. “Rethinking Environmental Racism: White Privilege and Urban
Development in Southern California.” Annals of the Association of American
Geographers 90 (1): 12–40.
Rabin, Richard. 2008. “The Lead Industry and Lead Water Pipes ‘A MODEST CAMPAIGN.’”
American Journal of Public Health 98 (9): 1584–92.
https://doi.org/10.2105/AJPH.2007.113555.
Raddatz, Liv, and Jeremy Mennis. 2013. “Environmental Justice in Hamburg, Germany.” The
Professional Geographer 65 (3): 495–511.
https://doi.org/10.1080/00330124.2012.700500.
Riva, Michele Augusto, Alessandra Lafranconi, Marco Italo D’Orso, and Giancarlo Cesana.
2012. “Lead Poisoning: Historical Aspects of a Paradigmatic ‘Occupational and
Environmental Disease.’” Safety and Health at Work 3 (1): 11–16.
https://doi.org/10.5491/SHAW.2012.3.1.11.

65
Rufat, Samuel, Eric Tate, Christopher G. Burton, and Abu Sayeed Maroof. 2015. “Social
Vulnerability to Floods: Review of Case Studies and Implications for Measurement.”
International Journal of Disaster Risk Reduction 14 (December): 470–86.
https://doi.org/10.1016/j.ijdrr.2015.09.013.
Salm Ward, Trina, Naoyo Mori, Timothy Patrick, Mary Madsen, and Ron Cisler. 2010.
“Influence of Socioeconomic Factors and Race on Birth Outcomes in Urban Milwaukee.”
WMJ : Official Publication of the State Medical Society of Wisconsin 109 (October):
254–60.
Sampson, Robert J., and Alix S. Winter. 2016. “The Racial Ecology of Lead Poisoning: Toxic
Inequality in Chicago Neighborhoods, 1995-2013.” Du Bois Review: Social Science
Research on Race 13 (2): 261–83. https://doi.org/10.1017/S1742058X16000151.
Schlenker, Thomas L, Renee Baxmann, Peter McAvoy, John Bartkowski, and Amy Murphy.
2001. “Primary Prevention of Childhood Lead Poisoning Through Community
Outreach.” Wisconsin Medical Journal 100 (8): 8.
Schwarz, Kirsten, Adam Berland, and Dustin L. Herrmann. 2018. “Green, but Not Just?
Rethinking Environmental Justice Indicators in Shrinking Cities.” Sustainable Cities and
Society 41 (August): 816–21. https://doi.org/10.1016/j.scs.2018.06.026.
Shen, Shiran Victoria. 2018. “Using Machine Learning to Find Environmentally At-Risk
Communities.” SSRN Scholarly Paper ID 3329557. Rochester, NY: Social Science
Research Network. https://doi.org/10.2139/ssrn.3329557.
Shrader-Frechette, Kristin. 2002. Environmental Justice: Creating Equality, Reclaiming
Democracy. Oxford University Press.
Small, Virginia. 2019. “Environmental Justice Advocacy Growing in Milwaukee.” Shepherd
Express. March 5, 2019. https://shepherdexpress.com/api/content/43e8a1ec-3f84-11e9-
b6cb-120e7ad5cf50/.
Sorensen, Lucy C., Ashley M. Fox, Heyjie Jung, and Erika G. Martin. 2019. “Lead Exposure and
Academic Achievement: Evidence from Childhood Lead Poisoning Prevention Efforts.”
Journal of Population Economics 32 (1): 179–218. https://doi.org/10.1007/s00148-018-
0707-y.
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): 134–48. https://doi.org/10.1016/j.compenvurbsys.2008.01.002.
Trotter, Joe William. 1985. Black Milwaukee: The Making of an Industrial Proletariat, 1915-45.
University of Illinois Press.
UN-HABITAT, and World Health Organization. 2020. Integrating Health in Urban and
Territorial Planning: A Sourcebook. Geneva: UN-HABITAT and World Health
Organization.

66
US EPA, OP. 2014. “Environmental Justice.” Collections and Lists. US EPA. November 3,
2014. https://www.epa.gov/environmentaljustice.
US EPA. 2015. “Environmental Justice Timeline.” Collections and Lists. US EPA. April 15,
2015. https://www.epa.gov/environmentaljustice/environmental-justice-timeline.
US EPA, OW. 2015. “Lead and Copper Rule.” Other Policies and Guidance. October 13, 2015.
https://www.epa.gov/dwreginfo/lead-and-copper-rule.
US EPA. 2016. “Basic Information about Lead in Drinking Water.” Overviews and Factsheets.
US EPA. February 2, 2016. https://www.epa.gov/ground-water-and-drinking-water/basic-
information-about-lead-drinking-water.
US EPA. 2020. “Revised Lead and Copper Rule.” Overviews and Factsheets. December 21,
2020. https://www.epa.gov/ground-water-and-drinking-water/revised-lead-and-copper-
rule.
Vila, Peter M, Geoffrey R Swain, Dennis J Baumgardner, Sarah E Halsmer, Patrick L
Remington, and Ron A Cisler. 2007. “Health Disparities in Milwaukee by
Socioeconomic Status.” WISCONSIN MEDICAL JOURNAL, 9.
Waller, Lance A, Bradley P Carlin, Hong Xia, and Alan E Gelfand. 1997. “Hierarchical Spatio-
Temporal Mapping of Disease Rates.” Journal of the American Statistical Association,
12.
Washington, Rodney, and Thomas Welcenbach. 2019. “Meeting Minutes: Transportation, Public
Works and Transit Committee.” Milwaukee County.
http://milwaukeecounty.granicus.com/player/clip/1855?view_id=2&meta_id=406313&re
direct=true.
Weigand, Matthias, Michael Wurm, Stefan Dech, and Hannes Taubenböck. 2019. “Remote
Sensing in Environmental Justice Research—A Review.” ISPRS International Journal of
Geo-Information 8 (1): 20. https://doi.org/10.3390/ijgi8010020.
WHO. 2019. “Lead Poisoning and Health.” August 23, 2019. https://www.who.int/news-
room/fact-sheets/detail/lead-poisoning-and-health.
Wisconsin DNR. n.d. “Lead Service Line (LSL) Replacements | Wisconsin DNR.” Wisconsin
Department of Natural Resources. Accessed August 7, 2021.
https://dnr.wisconsin.gov/aid/documents/EIF/leadServiceLineFunding.html.
Wolff, Lisa S., Dolores Acevedo-Garcia, S.V. Subramanian, Deanne Weber, and Ichiro
Kawachi. 2010. “Subjective Social Status, a New Measure in Health Disparities
Research: Do Race/Ethnicity and Choice of Referent Group Matter?” Journal of Health
Psychology 15 (4): 560–74. https://doi.org/10.1177/1359105309354345.
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. https://doi.org/10.1559/152304010792194985.

Asset Metadata

Creator Kim, Ariana Sul-ki (author)

Core Title Estimating at-risk population for lead service lines induced lead exposure and their correlation to socioeconomically disadvantaged neighborhoods in Milwaukee, Wisconsin

Contributor Electronically uploaded by the author (provenance)

School College of Letters, Arts and Sciences

Degree Master of Science

Degree Program Geographic Information Science and Technology

Degree Conferral Date 2021-12

Publication Date 10/09/2021

Defense Date 08/24/2021

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

Tag cadastral-based expert dasymetric system,dasymetric mapping,environmental inequity,environmental justice,lead pipes,Milwaukee,OAI-PMH Harvest,thematic mapping

Format application/pdf (imt)

Language English

Advisor Duan, Leilei (committee chair), Vos, Robert (committee member), Wu, An-Min (committee member)

Creator Email ariana.sulki.kim@gmail.com,arianak@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC16207867

Unique identifier UC16207867

Legacy Identifier etd-KimArianaS-10156

Document Type Thesis

Format application/pdf (imt)

Rights Kim, Ariana Sul-ki

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 author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

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

Repository Email uscdl@usc.edu

Abstract (if available)

Abstract The dangers of lead poisoning have posed a real threat to the population of the United States since the turn of the century. It has a cumulative effect in the human body and can therefore build up over time, even with low dose exposure. Children are especially susceptible to lead exposure due to their increased absorption rate of the metal and the lasting health issues that can persist throughout their lives. Minority communities with low socioeconomic status are especially vulnerable to exposure because they are more likely to live in close proximity to lead pollution sources, older homes, and have lower rates of toxicity screenings. Poisoning occurs primarily when lead is ingested through lead-based paint, lead contaminated water pipes, dust, and soil. Older cities across the United States are particularly prone to have populations with increased blood lead levels because lead was a common building material in the early 1900s. Milwaukee, Wisconsin is one such historical city where around 40% of the city’s active residential water service lines are constructed of lead. This study quantifies how many people are at risk for lead poisoning based on the existence of lead service lines in their buildings by census tract. Given the deeply segregated history of Milwaukee, an issue that still plagues the city to this day, this study also examines the relationship between the number of at-risk people per census tract and a variety of socioeconomic indicators. Dasymetric mapping techniques as well as regression analysis were used to shed light on this environmental justice issue in Milwaukee. Results show that the number of at-risk people in a census tract has a positive linear relationship with the race, education level, and poverty status of neighborhoods. In the context of Milwaukee’s demographics, the issue of lead exposure due to LSL disproportionately affects poorer communities of color.